Data-driven Test Generation for Fuzzing AI Compiler

TL;DR

This paper introduces a comprehensive, data-driven testing framework for AI compilers that improves bug detection across different compilation stages, enhancing reliability and correctness.

Contribution

It presents a unified, stage-aware testing approach with three novel techniques for systematically detecting bugs in AI compiler stages.

Findings

Detected 266 previously unknown bugs in four AI compilers.

Enhanced testing coverage across compiler stages.

Improved reliability and correctness of AI compiler deployments.

Abstract

Artificial Intelligence (AI) compilers are critical for efficiently deploying AI models across diverse hardware platforms. However, they remain prone to bugs that can compromise both compiler reliability and model correctness. Thus, ensuring the quality of AI compilers is crucial. In this work, we present a unified data-driven testing framework that systematically addresses stage-specific challenges in AI compilers. Specifically, OPERA migrates tests for AI libraries to test various operator conversion logic in the model loading stage. OATest synthesizes diverse optimization-aware computational graphs for testing high-level optimizations. HARMONY generates and mutates diverse low-level IR seeds to generate hardware-optimization-aware tests for testing low-level optimizations. Together, these techniques provide a comprehensive, stage-aware framework that enhances testing coverage and effectiveness, detecting 266 previously unknown bugs in four widely used AI compilers.