High-Performance Code Generation though Fusion and Vectorization

TL;DR

This paper introduces HFAV, a technique that automatically fuses and vectorizes nested loop kernels to reduce storage and enhance performance on modern hardware, using a declarative transformation approach.

Contribution

It presents a novel method for automatic kernel transformation involving fusion and vectorization, with a prototype implementation that improves HPC code performance.

Findings

Reduced intermediate storage in transformed kernels

Improved performance on contemporary hardware

Effective automatic transformation for nested loops

Abstract

We present a technique for automatically transforming kernel-based computations in disparate, nested loops into a fused, vectorized form that can reduce intermediate storage needs and lead to improved performance on contemporary hardware. We introduce representations for the abstract relationships and data dependencies of kernels in loop nests and algorithms for manipulating them into more efficient form; we similarly introduce techniques for determining data access patterns for stencil-like array accesses and show how this can be used to elide storage and improve vectorization. We discuss our prototype implementation of these ideas---named HFAV---and its use of a declarative, inference-based front-end to drive transformations, and we present results for some prominent codes in HPC.