Beyond depth-first strategies: improving tabled logic programs through alternative scheduling

Tabled evaluation ensures termination for programs with finite models by keeping track of which subgoals have been called. Given several variant subgoals in an evaluation, only the fi rst one encountered will use program-clause resolution; the rest will resolve with the answers generated by the first subgoal. This use of answer resolution prevents infi nite looping that sometimes happens in SLD. Because answers that are produced in one path of the computation may be consumed, asynchronously, in others, tabling systems face an important scheduling choice not present in traditional top-down evaluation: when to schedule answer resolution. This paper investigates alternate scheduling strategies for tabling in a WAM implementation, the SLG-WAM. The original SLG-WAM had a simple mechanism for scheduling answer resolution that was expensive in terms of trailing and choice-point creation. We propose here a more sophisticated scheduling strategy, batched scheduling, which reduces the overheads of these operations and provides dramatic space reduction as well as speedups for many programs. We also propose a second strategy, local scheduling, which has applications to nonmonotonic reasoning, and when combined with answer subsumption, can arbitrarily improve the performance of some programs.