Towards an Efficient Prolog System by Code Introspection

ABSTRACT. Several Prolog interpreters are based on the Warren Abstract Machine (WAM), an elegant model to compile Prolog programs. In order to make this model simpler and hence improve the performance several strategies have been proposed, such as: optimize the selection of clauses, specialize the unification, global analysis, native code generation and tabling. This paper proposes a different strategy to implement an efficient Prolog System, the creation of specialized emulators on the fly. The proposed strategy was implemented and evaluated at YAP Prolog System, and the results show an average performance improvement up to 30%.