Massive Parallel Max-SAT Solver Based on Speculative Computation

This paper proposes a massive parallel Max-Sat solver based on speculative computation using GPUs. Max-SAT is a combinatorial optimization problem that maximizes the true clauses of the SAT clauses, and many Max-SAT solvers have been proposed. Parallelization is expected as an efficiency improvement method when solving problems with a large amount of computation such as Max-SAT. However, many of the Max-SAT solvers are designed as sequential programs, and the effect of parallelization is limited. In this study, we developed a massive parallel search algorithm by designing a parallel stochastic search algorithm and combining it with parallel speculative computation. From the comparison with existing solvers using the Max-SAT problem, we examine the feasibility of this method.