Finding relevant templates via the principal component analysis

Abstract

The polyhedral model is widely used for the static analysis of programs, thanks to its expressiveness but it is also time consuming. To cope with this problem, weak-polyhedral analysis have been developed which offer a good trade off between expressiveness and efficiency. Some of these analysis are based on templates which fixed the form of the program’s invariant. These templates are defined statically at the beginning of the analysis, without taking into account the dynamic of programs. Finding good templates is a difficult problem. In this article, we present a method that uses the Principal Component analysis to compute an interesting template. We demonstrate the relevancy of the obtained templates on several benchmarks.