Probabilistic Alias Analysis for Parallel Programming in SSA Forms

TL;DR

This paper introduces a probabilistic alias analysis method for parallel programs in SSA form, enhancing precision for compiler optimizations by associating alias results with occurrence probabilities.

Contribution

It presents a novel probabilistic alias analysis technique tailored for SPMD parallel programs in SSA form, using a rule-based inference system.

Findings

Enables more precise alias analysis for parallel programs.

Supports applications like Proof-Carrying Code.

Improves compiler optimization techniques.

Abstract

Static alias analysis of different type of programming languages has been drawing researcher attention. However most of the results of existing techniques for alias analysis are not precise enough compared to needs of modern compilers. Probabilistic versions of these results, in which result elements are associated with occurrence probabilities, are required in optimizations techniques of modern compilers. This paper presents a new probabilistic approach for alias analysis of parallel programs. The treated parallelism model is that of SPMD where in SPMD, a program is executed using a fixed number of program threads running on distributed machines on different data. The analyzed programs are assumed to be in the static single assignment (SSA) form which is a program representation form facilitating program analysis. The proposed technique has the form of simply-strictured system of inference rules. This enables using the system in applications like Proof-Carrying Code (PPC) which is a general technique for proving the safety characteristics of modern programs.