Accelerating Energy Games Solvers on Modern Architectures

Abstract

Quantitative games, where quantitative objectives are defined on weighted game arenas, provide natural tools for designing faithful models of embedded controllers. Instances of these games that recently gained interest are the so called Energy Games. The fast-known algorithm solves Energy Games in O(EV W ) where W is the maximum weight. Starting from a sequential baseline implementation, we investigate the use of massively data computation capabilities supported by modern Graphics Processing Units to solve the initial credit problem for Energy Games. We present four different parallel implementations on multi-core CPU and GPU systems. Our solution outperforms the baseline implementation by up to 36x speedup and obtains a faster convergence time on real-world graphs.

GPU-based parallelism for ASP-solving

“…Second, because of intrinsic data-dependencies, procedures like propagation or learning often require to access large parts of the data/graph, sequentially. Similarly to what experienced in other complex graph-based problems [9], the kind of computation involved differs from that of traversal-like algorithms (such as, Breadth-First Search) which process a subset of the graph in iterative/incremental manners and for which advanced GPU-solutions exist. Furthermore, aspect specific to the underlying architecture enters into play, such as coalesced memory access and CUDA-thread balancing, which are major objectives in parallel algorithm design.…”