Boosting the Performance of SLS and CDCL Solvers by Preprocessor Tuning

Preprocessing techniques are crucial for SAT solvers when it comes to reaching state-of-the-art performance as it was shown by the results of the last SAT Competitions. The usefulness of a preprocessing technique depends highly on its own parameters, on the instances on which it is applied and on the used solver. In this paper we first give an extended analysis of the performance gain reached by using different preprocessing techniques individually in combination with CDCL solvers on application instances and SLS solvers on crafted instances. Further, we provide an analysis of combinations of preprocessing techniques by means of automated algorithm configuration, where we search for optimal preprocessor configurations for different scenarios. Our results show that the performance of CDCL and especially of SLS solvers can be further improved when using appropriate preprocessor configurations. The solvers augmented with the best found preprocessing configurations outperform the original solvers on the instances from the SAT Challenge 2012, achieving new state-of-the-art results.