Optimizations of CLP ( < Lin )

This paper describes the design and implementation of an optimizing compiler for CLP(< Lin), a constraint logic programming language over linear real constraints. The compiler performs a number of source to source optimizations which aim at replacing constraint solving, the basic operation of the language, by assignments (when, at runtime, all but one variables have a xed value) and tests (when, at runtime, all variables have a xed value) These optimizations follow the 3R's methodology which consists of reening, reordering, and removing constraints. The compiler uses abstract interpretation to decide when and how to perform the optimizations. The resulting system (40,000 lines of C, 10,000 of which concern the optimizations) has been evaluated experimentally. Our preliminary results indicate that substantial (often asymptotic) speedups can be obtained. Most of these speedups can be made arbitrary large, since they depend on the input size. This is, to our knowledge, the rst operational compiler for CLP(< Lin) with these functionalities and the rst implementation of reordering and constraint removal. It is also a convincing demonstration that abstract interpretation can be used to produce dramatic speedups for CLP(< Lin) programs.