Solving Mean-Payoff Games on the GPU
General purpose computation on graphics processing units (GPGPU) is a recent trend in areas which heavily depend on linear algebra, in particular solving large systems of linear equations. Many games, both qualitative (e.g. parity games) and quantitative (e.g. mean-payoff games) can be seen as systems of linear equations, too, albeit on more general algebraic structures. Building up on our GPU-based implementation of several solvers for parity games , we present in this paper a solver for mean-payoff games. Our implementation uses OpenCL which allows us to execute it without any changes on both the CPU and on the GPU allowing for direct comparison.
We evaluate our implementation on several benchmarks (obtained via reduction from parity games and optimization of controllers for hybrid systems ) where we obtain a speedup of up to 10 on the GPU in cases of MPGs with \(20\cdot 10^6\) nodes and \(60\cdot 10^6\) edges.
KeywordsModel Check Graphic Processing Unit Strategy Iteration Benchmark Suite Model Check Problem
We thank Majid Zamani and Matthias Rungger for kindly providing the example for the hybrid controller for our experimental evaluation.
- 3.Björklund, H., Sandberg, S., Vorobyov, S.: A combinatorial strongly subexponential strategy improvement algorithm for mean payoff games. In: Fiala, J., Koubek, V., Kratochvíl, J. (eds.) MFCS 2004. LNCS, vol. 3153, pp. 673–685. Springer, Heidelberg (2004). doi: 10.1007/978-3-540-28629-5_52 CrossRefGoogle Scholar
- 6.Friedmann, O., Lange, M.: The PGSolver collection of parity game solvers. University of Munich (2009). http://www2.tcs.ifi.lmu.de/pgsolver/
- 10.Rungger, M., Zamani, M.: SCOTS: a tool for the synthesis of symbolic controllers. In: Proceedings of the 19th International Conference on Hybrid Systems: Computation and Control, HSCC 2016, pp. 99–104. ACM, New York (2016)Google Scholar