Optimization of linear logic programs using counting methods

Abstract

We present a general solution to the problem of optimized execution of logic programs containing linear recursive rules. Our solution is based on extensions of the classical counting method, which is known to be efficient but of limited applicability. In fact, the range of applicability of the counting method, and its variants proposed by previous researchers, suffer from one or more of the following limitations: the method can be applied only when (1) the adorned program contains one recursive rule, (2) the ‘left part’ and the ‘right part’ of the recursive rule do not have any common variable and (3) the relation associated with the left part of the recursive rule is ‘acyclic’. In this paper, a simple and unified framework is presented, where those limitations are removed, and the counting method thus become applicable to all programs with linear rules. This framework also allows a simple treatment of programs factorizable into segments, which can be computed separately yielding a much faster execution. A simple factorization technique based on argument reduction is presented that produces optimizations similar to those defined in the literature for RLC-linear programs (i.e., programs with right- linear rules, left-linear rules and a combination of the two).

Work done while visiting MCC, Austin, Texas, and supported by the project “Sistemi Informatici e Calcolo Parallelo” obiettivo ”Logidata+” of C.N.R. Italy.