Study of a parallel algorithm on pipelined computation of the finite difference schemes on FPGA

Abstract

With the rapid increase of data, stream computation contributes to the improvement of data real time processing. One of characters of stream computing is in the pipeline to exploit parallelism in the MIMD way. Moreover, pipelined parallel implementation is a common problem and often yields much better parallel efficiency with respect to other commonly used methods. Plentiful physical modeling in the field of image processing and machine vision deals with the solution of partial differential equations, typical representative by Poisson's equation. The finite difference schemes as the main method for the solution of partial difference equation for physical modeling can be time-consuming and computationally expensive. Constructing highly parallel computation models can greatly solve the problem and it is important that quickly and efficiently carry out it. The paper mainly studies the parallel implementation of the mix of Jacobi iterative and Gauss-Seidel method for the finite difference schemes to solve Poisson equations in a pipelined fashion on FPGA.