Replication Data for: LLM-Guided Differential Fuzzing for Detecting Platform-Specific Bugs in Scientific Applications

This artifact contains the anonymized code and data to replicate the results described in the paper titled `LLM-Guided Differential Fuzzing for Detecting Platform-Specific Bugs in Scientific Applications`, published in IEEE Artificial Intelligence x Software Engineering Conference (AIxSoftware), 2025. The artifact contains the following: The archive "source.tar.gz" contains the source code of the three main components of the HeteroBugDetect: `1.SIG (Seed Input Generator)`, `2. Grammar-based Kernel-Sensitive Fuzzer` and `3.Differential Driver` under respective directories and the main script that executes HeteroBugDetect technique. The archive "dataset.tar.gz" contains the `HeteroBug` dataset consisting of 20 known heterogeneous bugs in LAMMPS categorized across 7 different types and natural language descriptions of LAMMPS examples used by HeteroBugDetect to test LAMMPS. The archive "results.tar.gz" contains results organized in terms of the research questions. The `test_suite_part_*` files inside the `test_suite` directory contain the test suites generated using different fuzzing strategies. Note that these files are larger in size because they contain the tests generated by fuzzer using 10 random seeds and 3 fuzzing strategies for 40 example simulation. As the size of these files is large (1.9GB), please download all the files before they can be merged and extracted. Please use the following command to merge the split archives: cat `test_suite_part_*` > test_suite_combined.tar.gz` Please use the following command to decompress the archives: `tar -xvzf .tar.gz`