Improving the Performance of FD Constraint Solving in a CFLP System

Abstract

Constraint Functional Logic Programming (CFLP) integrates lazy narrowing with constraint solving. It provides a high modeling abstraction, but its solving performance can be penalized by lazy narrowing and solver interface surcharges. As for real-world problems most of the solving time is carried out by solver computations, the system performance can be improved by interfacing state-of-the-art external solvers with proven performance. In this work we depart from the CFLP system $\mathcal{TOY(FD})$ , implemented in SICStus Prolog and supporting Finite Domain ( $\mathcal{FD}$ ) constraints by using its underlying Prolog $\mathcal{FD}$ solver. We present a scheme describing how to interface an external CP( $\mathcal{FD}$ ) solver to $\mathcal{TOY(FD})$ , and easily adaptable to other Prolog CLP or CFLP systems. We prove the scheme to be generic enough by interfacing Gecode and ILOG solvers, and we analyze the new performance achieved.