Compiler Autotuning Data via Bayesian Networks

COBAYN is a framework for automatically tuning compiler optimization parameters using Bayesian networks. The project was created by amirjamez and last updated in May 2022. It was developed as part of the European Union's Antarex project.

Use Cases

• Training Bayesian networks on historical compiler flag and performance metric pairs to predict optimal configurations.
• Analyzing the relationship between specific compiler optimization parameters and resulting execution time or code size.
• Building a recommender system that suggests compiler flags for new code based on features from previously tuned programs.

Strengths

• Framework is associated with the EU-funded Antarex project, suggesting formal research backing.
• Focuses on a specialized, high-impact application of machine learning for system performance.

Limitations

• The dataset size, specific features, and number of records are unknown.
• The last update was in 2022, so the included compiler versions and benchmarks may be outdated.

Provenance

Source

GitHub repository by author amirjamez, part of the Antarex EU project.

Collection Method

Likely involves benchmarking programs with different compiler optimization parameters.

Freshness

Last updated in May 2022.