A semi-algebraic model for automatic loop parallelization

TL;DR

This paper presents a semi-algebraic model that extends the polyhedral approach for automatic parallelization of nested polynomial loops, enabling dependence analysis, schedule computation, and program transformation.

Contribution

It introduces a novel semi-algebraic model that generalizes the polyhedral model for better handling of polynomial loops in parallelization.

Findings

Supports dependence analysis for polynomial loops

Enables computation of valid parallel schedules

Facilitates loop program transformation

Abstract

In this work, we introduce a semi-algebraic model for automatic parallelization of perfectly nested polynomial loops, which generalizes the classical polyhedral model. This model supports the basic tasks for automatic loop parallelization, such as the representation of the nested loop, the dependence analysis, the computation of valid schedules, as well as the transformation of the loop program with a valid schedule.