Integrating Performance Tools in Model Reasoning for GPU Kernel Optimization

TL;DR

This paper introduces a methodology for training language models to incorporate performance tools during reasoning, enhancing GPU kernel optimization by considering environment and hardware factors.

Contribution

It presents a novel training approach enabling language models to interact with performance tools, improving their ability to optimize GPU kernels.

Findings

Enhanced model understanding of environment-hardware interactions

Improved GPU kernel optimization performance

Demonstrated state-of-the-art results in kernel optimization

Abstract

Language models are now prevalent in software engineering with many developers using them to automate tasks and accelerate their development. While language models have been tremendous at accomplishing complex software engineering tasks, there are still many areas where they fail to deliver desirable results, for instance code performance related tasks. Tasks like optimization depend on many complex data from the environment, hardware, etc. that are not directly represented in source code. Recent efforts have seen large improvements in general code modeling tasks using chain-of-thought style reasoning, but these models still fail to comprehend how the environment interacts with code performance. In this paper we propose a methodology to train language models that can interact with performance tools during their reasoning process. We then demonstrate how this methodology can be used to train a state-of-the-art GPU kernel optimization model.