Output details

Parallel graph component labelling with GPUs and CUDA

<12> This article reports in-depth algorithm development for graph component-labelling. This involves identifying which components in a model are connected. As well as being widely used in analyzing complex simulation models, this technique is widely used in many applications from image analysis to power network fragmentation studies. Our unique new algorithms, optimized for data parallelism available on graphical processing units, were rigorously tested using regular, small-world and other irregular data sets and their performance compared with CPUs and various GPU models. High citations indicate others have found the explicit program source codes we included for our eight algorithms useful.