SIAM Conference on Parallel Processing for Scientific Computing location:Seattle, Washington, U.S. date:24-26 February 2010
Graphic cards provides enormous amounts of parallel computing power. However, actually using it is sometimes rather technical. We consider the Kullback-Leibler divergence, a measure for the similarity between to sample sets, that is studied in the context of particle filters. Using the OpenCL standard, we were able to add a GPU accelerated implementation to our open source Bayesian Filtering Library (BFL) in a portable fashion and observed speedups of up to 40 (relative to C++).