一句话：提出 Opera，将 DL 库测试知识迁移到 DL 编译器模型加载阶段进行缺陷检测。
现有 DL 编译器测试主要关注模型优化阶段，较少检测模型加载阶段的缺陷。有效测试加载阶段需要覆盖各 DL 库中 DL 算子的多样用法，这与 DL 库测试目标一致，因此库测试中的嵌入知识可用于编译器加载阶段测试。本文提出 Opera，从 DL 库中迁移测试，包括文档测试与近期 fuzzer 生成的测试，共考虑三类测试来源，并引入基于多样性的测试优先级排序，优先迁移和执行更可能发现多样缺陷的测试。作者在 TVM、TensorRT 和 OpenVINO 三个 DL 编译器的八个前端上评估，共发现 170 个先前未知缺陷，其中 90 个已被确认或修复；其优先级排序策略相比通用策略平均提升测试效率 11.9%~47.4%。
Deep Learning (DL) compilers typically load a DL model and optimize it with intermediate representation. Existing DL compiler testing techniques mainly focus on model optimization stages, but rarely explore bug detection at the model loading stage. Effectively testing the model loading stage requires covering diverse usages of each DL operator from various DL libraries, which shares a common objective with DL library testing, indicating that the embedded knowledge in DL library tests is beneficial for testing the model loading stage of DL compilers. With this idea, we propose Opera to migrate the knowledge embedded in DL library tests to test the model loading stage. Opera constructs diverse tests from various tests for DL libraries (including the tests documented in DL libraries and those generated by recent fuzzers). In total, we considered three sources of tests in DL libraries for migration. In addition, it incorporates a diversity-based test prioritization strategy to migrate and execute those tests that are more likely to detect diverse bugs earlier. We then used eight frontends from three DL compilers (e.g., TVM, TensorRT, and OpenVINO) for evaluation. OPERA detected 170 previously unknown bugs in total, 90 of which have been confirmed/fixed by developers, demonstrating the effectiveness of such the migration-based idea. The test prioritization strategy in OPERA improves testing efficiency with migrated tests by$11.9 \% \sim 47.4 \%$on average compared to general test prioritization strategies.
DOI 原文 ·
@inproceedings{paperbot8826,
title = {A Tale of Two DL Cities: When Library Tests Meet Compiler},
author = {Qingchao Shen and Yongqiang Tian and Haoyang Ma and Junjie Chen and Lili Huang and Ruifeng Fu and Shing-Chi Cheung and Zan Wang},
booktitle = {ICSE},
year = {2025},
doi = {10.1109/icse55347.2025.00025}
}