Precise Interprocedural Analysis through Linear Algebra

Markus Mueller-Olm, Helmut Seidl

To appear at Principles of Programming Languages (POPL04), Venice, Italy, January 14-16, 2004


Abstract

We apply linear algebra techniques to precise interprocedural dataflow analysis. Specifically, we describe analyses that determine for each program point identities that are valid among the program variables whenever control reaches that program point. Our analyses fully interpret assignment statements with affine expressions on the right hand side while considering other assignments as non-deterministic and ignoring conditions at branches. Under this abstraction, the analysis computes the set of all AFFINE IDENTITIES and, more generally, all POLYNOMIAL IDENTITIES OF BOUNDED DEGREE precisely. The running time of our algorithm is linear in the program size and polynomial in the number of occurring variables. We also show how to deal with local variables and indicate how to handle parameters and return values of procedures.


Server START Conference Manager
Update Time 19 Sep 2003 at 17:40:43
Maintainer Xavier.Leroy@inria.fr.
Start Conference Manager
Conference Systems