"Implementation of Image Processing Kernels on Reconfigurable Computers: A Comparative Study between SRC and SGI Platforms"

Esam El-Araby1, Mohamed Taher1, Tarek El-Ghazawi1, and Kris Gaj2

1The George Washington University
2George Mason University

Abstract

Digital image processing is a rapidly evolving field with growing application in science and engineering. Until recently the computational burden of digital image processing for the most part had to be handled by either dedicated hardware or pure software solutions. The last decade has seen a move towards reconfigurable computing solutions, due to the advent first of FPGAs and later of general-purpose reconfigurable computers (RCs). The SRC Reconfigurable Computer (SRC-6) and the SGI Altix Reconfigurable Application Specific Computing (RASC) platforms are two examples of these RCs.

In this paper we analyze the similarities and differences between the SRC-6 and SGI-RASC systems by implementing selected image processing kernels, such as Discrete Wavelet Transform (DWT), on both platforms. This study will be performed with respect to system hardware architecture, programming model, and hardware/software libraries. Both systems will be compared in terms of ease of programming, development time, and performance. The speed-up vs. a cluster of microprocessor nodes will be determined, as well as the primary bottlenecks limiting the performance of both systems. The capability to share the codes between the SRC and SGI systems, as well as the paths for porting applications between both platforms will be explored.

 

2005 MAPLD International Conference Home Page