Binding-time analysis applied to mathematical algorithms
Robert Glück, Ryo Nakashige, Robert Zöchling · 1996
Our goal is to incorporate state-of-the-art partial evaluation in a library of general-purpose algorithms — in particular, mathematical algorithms — in order to allow the automatic creation of efficient, special-purpose programs. The main goal is efficiency: a specialized program often runs significantly faster than its generic version. This paper shows how a binding-time analysis can be used to identify potential sources for specialization in mathematical algorithms. The method is surprisingly simple and effective. To demonstrate the effectiveness of this approach we used an automatic partial evaluator for Fortran that we developed. Results for five well-known algorithms show that some remarkable speedup factors can be obtained on a uniprocessor architecture.