SWORD v0.2 – Module-based SAT Solving

In this paper, we present SWORD – a SAT like solver that facilitates word level information. The main idea behind SWORD is based on the following observation: Current SAT solvers perform very well on instances with a large number of logic operations. But when more complex functions like arithmetic units are considered, the performance degrades with increasing data-path width. In contrast, pure word level approaches handle e.g. arithmetic operations very fast but suffer from complexity problems when irregularities in the word level structure (e.g. bit slicing) occur. SWORD tries to combine the best of both worlds: Logic operations like (bv)and, (bv)or, and (bv)xor are represented in terms of clauses while more complex functions like arithmetic operations or shifts are represented by so called modules. These modules inherit a problem specific decision as well as a problem specific propagation strategy, which is exploited during the search. Thus, SWORD combines the advantages of a Boolean proof procedure with the power of word level knowledge. Moreover, SWORD is not limited to pre-defined encodings as CNF or QF BV logic. Problem specific modules for respective domains can be developed.