News

Revision as of 14:31, 26 February 2009 by Gfursin (Talk | contribs)

2009.February:

2008.November:

2008.March-July:

Finished developing the first version of the MILEPOST GCC, Collective Compilation Framework and Collective Optimization Database.
Prepared the prototype of the cTuning website for the MILEPOST project
Collected data from about 4,000,000 experiments by MILEPOST partners and prepare machine learning to predict good program optimizations to reduce execution time and code size or tune default GCC optimization heuristic on a number of platforms.
Presented MILEPOST GCC at the GCC Summit'08.

2005-2007:

Prepared multiple datasets (MiDataSets) to enable realistic program and architecture optimizations, run-time adaptation and performance evaluation.
Added performance counters support to the Continuous Collective Compilation Framework to predict good optimizations based on run-time program features.
Added support for architectural design space exploration in the Continious Collective Compilation Framework.
Started MILEPOST project.

2004:

Had initial discussions about continuous collective optimizations with colleagues from INRIA and the University of Edinburgh.
Started developing Interactive Compilation Interface to enable collaborative research using production compilers (based on Open64/PathScale compilers and later GCC).
Started developing UNIDAPT framework - a new hybrid static/dynamic approach to create self-tuning binaries based on static code multiversioning and run-time hardware counters monitoring routines to improve performance, power, fault-tolerance, etc. (and to enable run-time adaptation for statically compiled programs on heterogeneous multi-core architectures).
Started discussing statistical and machine learning techniques with colleagues from the University of Edinburgh to enable optimization knowledge reuse for program optimizations.