GPU-PRISM: An Extension of PRISM for General Purpose Graphics Processing Units

Abstract

We present an extension of the model checker PRISM for (general purpose) graphics processing units (GPUs). The extension is based on parallel algorithms for probabilistic model checking which are tuned for GPUs. In particular, we parallelize the parts of the algorithms that boil down to linear algebraic operations, like solving systems of linear equations and matrix vector multiplication. These computations are performed very efficiently on GPGPUs which results inconsiderable runtime improvements compared to the standard versions of PRISM. We evaluated the extension of PRISM on several case studies in which we observed significant speedup over the standard CPU implementation of the tool.