From cTuning.org
NEWS: Since 2015, we moved all related developments to our new Collective Knowledge Framework! Demo of a new live repository is available here.
 |
MILEPOST project |
| machine learning for embedded programs optimization |
Web shortcut: http://cTuning.org/project-milepost
NOTE:
- Since 2014, we continue this R&D with the help of the non-profit cTuning foundation.
- MILEPOST GCC is now a part of cTuning CC and Collective Mind framework.
- Website milepost.eu was overtaken and has now nothing to do with the MILEPOST GCC project.
- Project reference: 035307 (Specific Targeted Research Project, funded by EU FP6 program)
- Official dates: 2006-07-01 - 2009-06-30
- Project coordinator: Prof. Michael O'Boyle, University of Edinburgh, UK
- Technical coordinator/cTuning founder: Dr. Grigori Fursin, INRIA, France
- Acknowledgments: Milepost colleagues
- Reference publications: MILEPOST GCC: machine learning enabled self-tuning compiler; Collective Optimization: A Practical Collaborative Approach; Practical Aggregation of Semantical Program Properties for Machine Learning Based Optimization.
Contents |
Official partners
Objectives
The overall objective of this project is to develop compiler technology that can automatically learn how to best optimise programs for reconfigurable heterogeneous embedded processors. If successful we will be able to dramatically reduce the time to market of reconfigurable systems. Rather than developing a specialised compiler by hand for each configuration, our project will produce optimising compilers automatically.
Current hand-crafted approaches to compiler development are no longer sustainable. With each generation of reconfigurable architecture, the compiler development time increases and the performance improvement achieved decreases. As high performance embedded systems move from application specific ASICs to programmable heterogeneous processors, this problem is becoming critical.
This project explores an emerging alternative approach where we use machine learning techniques, developed in the artificial intelligence arena, to learn how to generate compilers automatically. Such an approach, if successful, will have a dramatic impact on reconfigurable systems. This means that for a fixed amount of design time. We can evaluate many more configurations leading to better and more cost-effective performance. If successful, this will enable Europe to increase its dominance in this critical emerging market.
Software releases
- Milepost GCC - first public machine learning-enabled, self-tuning, adaptive compiler that correlates program features and optimizations during empirical learning to predict good optimization for unseen programs.
- Milepost Optimization Framework - infrastructure that combines MILEPOST GCC, CCC Framework, Collective Optimization Database and UNIDAPT Framework to find "good" program optimizations or architectural configurations for reconfigurable processors entirely automatically using statistical and machine learning techniques. After the end of the MILEPOST project in October, 2009, the MILEPOST framework has been fully integrated with cTools. Note: this framework is now fully integrated with the cTuning infrastructure, tools and repository so it is not used/extended anymore on its own.
Press
- IBM press-release about MILEPOST GCC (2009.06.30, 2008.06.30).
- InfoWorld: Rethinking code optimization for mobile and multicore
- Slashdot.org
- Dr Dobb
- HPCWire
- Machine-learning revolutionizes software development
Some industrial usages of the MILEPOST technology
- Since 2010, ICI-compatible plugin interface is available in mainline GCC, meaning that anyone using current GCC can take advantage of or extend MILEPOST and cTuning technology.
- IBM
- CAPS Entreprise
- Dr. Zbigniew Chamski joined Grigori Fursin's team at INRIA for 6 months to considerably extend, document and move ICI to the mainline GCC while using it in his "Infrasoft IT Solutions" company
Further work
- After the end of the MILEPOST project in October, 2009, Grigori integrated the MILEPOST framework with the cTuning framework to continue collaborative R&D on self-tuning computing systems together with the community.
- Grigori Fursin and Zbigniew Chamski collaborated with Google and Mozilla (see public GCC discussions on the GCC mailing list and wiki) to develop ICI-compatible plugin framework for GCC that has been eventually integrated to the mainline GCC 4.5+
- We are encouraging public contributions to cTuning framework particularly to add support for more adaptive scheduling, parallelization, fine-grain tuning, polyhedral optimizations while supporting other compilers such as LLVM, ROSE, etc.
Contacts
- cTuning discussions mailing list
- General questions about MILEPOST technology, vision and further R&D: Grigori Fursin