A Note on "A polynomial-time algorithm for global value numbering"

TL;DR

This paper discusses limitations of a polynomial-time global value numbering algorithm and proposes improvements to enhance its redundancy detection capabilities in compiler optimization.

Contribution

It identifies specific limitations of Gulwani and Necula's GVN algorithm and suggests modifications to improve its effectiveness in detecting certain redundancies.

Findings

Identifies limitations in detecting classical global common subexpressions

Highlights issues in redundancy detection compared to local value numbering

Proposes improvements to extend the algorithm's detection capabilities

Abstract

Global Value Numbering(GVN) is a popular method for detecting redundant computations. A polynomial time algorithm for GVN is presented by Gulwani and Necula(2006). Here we present two limitations of this GVN algorithm due to which detection of certain kinds of redundancies can not be done using this algorithm. The first one is concerning the use of this algorithm in detecting some instances of the classical global common subexpressions, and the second is concerning its use in the detection of some redundancies that a local value numbering algorithm will detect. We suggest improvements that enable the algorithm to detect these kinds of redundancies as well.