Program autotuning has been demonstrated in many domains to achieve better or more portable performance. However, autotuners themselves are often not very portable between projects because using a domain informed search space representation is critical to achieving good results and because no single search technique performs best for all problems.

OpenTuner is a new framework for building domain-specific multi-objective program autotuners. OpenTuner supports fully customizable configuration representations, an extensible technique representation to allow for domain-specific techniques, and an easy to use interface for communicating with the tuned program. A key capability inside OpenTuner is the use of ensembles of disparate search techniques simultaneously, techniques which perform well will receive larger testing budgets and techniques which perform poorly will be disabled.

OpenTuner requires python 3.7+ and sqlite3 (or your supported database backend of choice). Install with:

sudo pip install opentuner

or

pip install --user opentuner

For development or running examples out of a git checkout, we recommend using miniconda3.

conda create --name=opentuner python=3.8
conda activate opentuner
pip install -r requirements.txt -r optional-requirements.txt
python setup.py develop

To check an installation you can run tests:

pytest tests/*

Or run an example program:

./examples/rosenbrock/rosenbrock.py

• Optimizing Block Matrix Multiplication
• Creating OpenTuner Techniques.

• OpenTuner: An Extensible Framework for Program Autotuning.
  Jason Ansel, Shoaib Kamil, Kalyan Veeramachaneni, Jonathan Ragan-Kelley, Jeffrey Bosboom, Una-May O'Reilly, Saman Amarasinghe.
  International Conference on Parallel Architectures and Compilation Techniques.
  Edmonton, Canada. August, 2014. Slides. Bibtex.

The preferred way to contribute code to OpenTuner is to fork the project on github and submit a pull request.

OpenTuner is supported in part by the United States Department of Energy X-Stack program as part of D-TEC.