Context-sensitive interprocedural analysis in the presence of dynamic aliasing

Abstract

Context-sensitive interprocedural data-flow analysis, for problems where the flow value is a mapping from symbols to values from an abstract domain, requires solutions to be computed for differing aliasing conditions existing at distinct calling contexts. This paper presents an approach for computing context-sensitive solutions to forward, monotone data-flow problems for statically allocated scalar variables that does not require reanalysis of procedures. The algorithm handles dynamic aliasing, due to non-recursive pointer types, and recursion. This paper applies the technique to constant propagation for statically allocated scalars.

We propose an elimination-style approach that computes a single canonical transfer function for a procedure, under the assumption that no aliases hold between its arguments (including both explicitly and implicitly passed globals) on entry. The canonical transfer functions are expressed as a set of parameterised data flow mappings, augmented with sequence tokens and alias assertions. The sequence tokens and alias assertions succinctly capture sufficient control-flow and alias conditions, respectively, so that accurate solutions in the presence of aliasing can be computed from the canonical one. The information represented by the sequence tokens allows approximations that yield conservative solutions, with varying levels of flow and context sensitivity.

This research was supported in part by the Air Force Material Command and DARPA under the contract F30602-95-C-0098 and an NSF Young Investigator award.