Abstract
背景：深度学习编译器处理的深度学习模型与命令式程序有根本的不同，因为深度学习模型中的程序逻辑是隐式的。（the DL models processed by DL compilers differ fundamentally from imperative programs in that the program logic in DL models is implicit）
本文：TVMFuzz
Github: https://github.com/ShenQingchao/DLCstudy
Task: 1. Review Deep Learning Compiler Bugs, including root causes, symptoms, occurred stages 2. TVMFuzz: Test TVM DL Compiler based on TVM's original test suite
Method:
- 分析TVM, Glow和nGraph中的603个错误
- Test TVM DL Compiler based on TVM's original test suite
Discussions:
- 12项发现，例如很大一部分（近 20%）的 DL 编译器错误与类型有关，尤其是张量类型。
效果：
- +8 TVM bugs
- Comprehensive CDeepFuzz Compiler Learning Readingcomprehensive cdeepfuzz compiler learning comprehensive compiler learning survey combinatorial cdeepfuzz learning reading pre-trained cdeepfuzz natural reading differentiation cdeepfuzz automatic reading state-of-the-art cdeepfuzz the reading cdeepfuzz networks reading testing deepmutation cdeepfuzz mutation learning large-scale tensorflow cdeepfuzz learning metamorphic cdeepfuzz compilers learning