Machine Learning for Application Optimization with cTuning

07/26/2011 - 1:00pm to 1:25pm
Main Seminar Room - ML
William Petzke

William Petzke, University of Colorado, Denver

Abstract:  In some cases application execution time on high performance computing systems can be reduced by finding superior combinations of compiler optimization flags. Machine learning methods can be used to predict good combinations of compiler optimizations for previously unseen programs using the best optimizations found for similar programs whose search space has been explored with random search.  This discussion will describe experience gained setting up and extending the open source cTuning framework and results produced by using the framework to perform application optimization with iterative compilation and machine learning.

 

Video replay of the presentation

Please install the Flash Plug-in or click here for non-flash supported devices.

Download the entire presentation (right-click/save link as)


Presented on July 26, 2011 at NCAR