high performance computing on graphics processing units: hgpu.org

This paper describes an approach exploiting the full capabilities of GPU’s to enhance the usability of edge bundling in real applications. Edge bundling, as well as other edge clustering approaches relying on the use of high quality edge rerouting. Typical approach for drawing edge-bundled graph is to render edges as curves. But curves generation can have a relatively high computational costs and do not easily comply with real-time interaction. Furthermore, while edge bundling provides a much better overall readability of a graph, the bundles make it more difficult to recover local information. Our goal was thus to provide fluid interaction allowing the recovery of local information through specific interaction techniques. The system we built offers folklore or classical interaction such as zoom & pan, fish-eye and magnifying lens. We also implemented the Bring & Go technique by Tominski et al. We proposed an approach exploiting the full computing power of GPU’s when rendering graph edges as parametric splines. The gain in efficiency when running all curves computations on the GPU turns bundling techniques into techniques that can be embedded in interactive systems concerned with graphs of several thousands of nodes and edges.