high performance computing on graphics processing units: hgpu.org

The evolution of high-performance and programmable graphics processing units (GPUs) has generated considerable advancements in graphics and parallel computing. In this paper we present a Roberts filter based on edge detection algorithm using CUDA and OpenGL architectures. The basic idea is to use the Pixel Buffer Object (PBO) to create images with CUDA on a pixel-by-pixel basis and display them using OpenGL. The images can then be processed applying a Roberts filter for edge detection. Finally, it describes the results of an extensive measurement campaign as well as several comparisons among the code performance on CPUs and GPUs. The results are very promising since the GPU parallel version offers much higher performances than the CPU sequential version. The execution time of the GPU parallel version is much lower than the sequential equivalent execution time.