SWORD – Module-based SAT Solving

Abstract

In the paper, SWORD is described – a decision procedure for bit-vector logic that uses SAT techniques and exploits word level information. The main idea of SWORD is based on the following observation: While current SAT solvers perform very well on instances with a large number of logic operations, their performance on arithmetic operations degrades with increasing data-path width. In contrast, pure word-level approaches are able to handle arithmetic operations very fast, but suffer from irregularities in the word-level structure (e.g. bit slicing). SWORD tries to combine the best of both worlds: On the one hand, it includes fast propagation, sophisticated data structures, as well as advanced techniques like non-chronological backtracking and learning from modern SAT solvers. On the other hand word-level information is exploited in the decision heuristic and during propagation.