Improvements to the Psi-SSA representation

TL;DR

This paper introduces the Psi-SSA representation, an extension of SSA tailored for architectures with predicated instruction sets, enabling easier adaptation of SSA algorithms to such architectures.

Contribution

The paper proposes the Psi-SSA representation and a new pseudo operation, the Psi operation, to better support predicated architectures in compiler optimizations.

Findings

Psi-SSA facilitates adaptation of SSA algorithms to predicated architectures

Introduction of the Psi operation for merging conditional definitions

Enhances compiler efficiency on architectures with predicated instructions

Abstract

Modern compiler implementations use the Static Single Assignment representation as a way to efficiently implement optimizing algorithms. However this representation is not well adapted to architectures with a predicated instruction set. The Psi-SSA representation extends the SSA representation such that standard SSA algorithms can be easily adapted to an architecture with a fully predicated instruction set. A new pseudo operation, the Psi operation, is introduced to merge several conditional definitions into a unique definition.