Generating GPU Compiler Heuristics using Reinforcement Learning

TL;DR

This paper introduces a reinforcement learning-based framework for automatically generating GPU compiler heuristics, significantly improving graphics application frame rates and demonstrating stability over time without retraining.

Contribution

It presents a novel off-policy deep reinforcement learning approach for GPU compiler autotuning, reducing reliance on manual heuristic design and maintaining effectiveness over frequent compiler updates.

Findings

Achieves up to 15.8% frame rate improvement.

Matches or surpasses manual heuristics for 98% of benchmarks.

Demonstrates heuristic stability over a year of code changes.

Abstract

GPU compilers are complex software programs with many optimizations specific to target hardware. These optimizations are often controlled by heuristics hand-designed by compiler experts using time- and resource-intensive processes. In this paper, we developed a GPU compiler autotuning framework that uses off-policy deep reinforcement learning to generate heuristics that improve the frame rates of graphics applications. Furthermore, we demonstrate the resilience of these learned heuristics to frequent compiler updates by analyzing their stability across a year of code check-ins without retraining. We show that our machine learning-based compiler autotuning framework matches or surpasses the frame rates for 98% of graphics benchmarks with an average uplift of 1.6% up to 15.8%.