Exploiting VSIPL and OpenMP for Parallel Image Processing

Abstract

VSIPL and OpenMP are two open standards for portable high performance computing. VSIPL delivers optimized single processor performance while OpenMP provides a low overhead mechanism for executing thread based parallelism on shared memory systems. Image processing is one of the main areas where VSIPL and OpenMP can make a large impact. Currently, a large fraction of image processing applications are written in the Interpreted Data Language (IDL) environment. The aim of this work is to demonstrate that the performance benefits of these new standards can be brought to image processing community in a high level manner that is transparent to users. To this end, this talk presents a fast, FFT based algorithm for performing image convolutions. This algorithm has been implemented within the IDL environment using VSIPL (for optimized single processor performance) with added OpenMP directives (for parallelism). This work demonstrates that good parallel speedups are attainable using standards and can be integrated seamlessly into existing user environments.