Please use this identifier to cite or link to this item:
https://rd.uffs.edu.br/handle/prefix/9707| Type: | Monografia |
| Title: | Differential testing SSA optimization passes in Go compilers using LLM generated code mutators |
| Author: | Terres, Luiz Paulo Grafetti |
| First advisor: | Feitosa, Samuel da Silva |
| Resume: | Compilers are fundamental tools for software infrastructure, yet their complex implementation makes them prone to bugs and hard to formally verify. The literature presents a range of techniques that aim at random testing (fuzzing) compilers for maintaining acceptable levels of correctness. Among traditional tools for compiler fuzzing, the complex implementation and maintenance of code generators and mutators remain an open challenge. Recent efforts aim to address this challenge by employing Large Language Models (LLMs) for test case generation. This paper presents Goopfuzz, a fuzzer that leverages Gemini 3.1 Pro and targets SSA passes in Go compilers in a differential testing, mutation-based, and coverageguided approach. By leveraging the Gemini model to generate code mutators and removing the LLM from the fuzzing loop, the Goopfuzz hybrid approach mitigates the complex implementation challenge associated with traditional methods and the low throughput of LLM-pure solutions. During Goopfuzz experiments, Gemini 3.1 Pro synthesized 76.25% of all 157 working mutators on the first try. Synthesized mutators produced a total of 165,321 unique Go programs to serve as test cases during the fuzzing campaign, exercising 15.75% of the SSA implementation portion of the Go compiler. The differential testing oracle evaluated the compilation results for those programs against the last two stable releases of the official Go compiler (1.25.10 and 1.26.3) in different optimization levels. |
| Keywords: | Montadores e compiladores Linguagem de programação Geração de código Testes Inteligência artificial Otimização |
| Language: | por |
| Country: | Brasil |
| Publisher: | Universidade Federal da Fronteira Sul |
| Acronym of the institution: | UFFS |
| College, Institute or Department: | Campus Chapecó |
| Type of Access: | Acesso Aberto |
| URI: | https://rd.uffs.edu.br/handle/prefix/9707 |
| Issue Date: | 2026 |
| Appears in Collections: | Ciência da Computação |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| TERRES.pdf | 663.06 kB | Adobe PDF | View/Open |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.