The Community for Technology Leaders
2012 21st International Conference on Parallel Architectures and Compilation Techniques (PACT) (2012)
Minneapolis, MN, USA
Sept. 19, 2012 to Sept. 23, 2012
ISBN: 978-1-5090-6609-4
pp: 437-438
Alexander Collins , University of Edinburgh, School of Informatics, Scotland
Christian Fensch , University of Edinburgh, School of Informatics, Scotland
Hugh Leather , University of Edinburgh, School of Informatics, Scotland
ABSTRACT
We present MaSiF, a novel tool to auto-tune parallelization parameters of skeleton parallel programs. It reduces the cost of searching the optimization space using a combination of machine learning and linear dimensionality reduction. To auto-tune a new program, a set of program features is determined statically and used to compute k nearest neighbors from a set of training programs. Previously collected performance data for the nearest neighbors is used to reduce the size of the search space using Principal Components Analysis. This results in a set of eigenvectors that are used to search the reduced space. MaSiF achieves 88% of the performance of the oracle, which searches a random set of 10,000 parameter values. MaSiF searches just 45 points, or 0.45% of the optimization space, to achieve this performance. MaSiF provides an average speedup of 1.18x over parallelization parameters chosen by a human expert.
INDEX TERMS
Skeleton, Optimization, Training, Feature extraction, Principal component analysis, Programming, Tuning,Parallel Skeletons, Auto-tuning, FastFlow, Machine Learning, Multi-core
CITATION
Alexander Collins, Christian Fensch, Hugh Leather, "MaSiF: Machine learning guided auto-tuning of Parallel Skeletons", 2012 21st International Conference on Parallel Architectures and Compilation Techniques (PACT), vol. 00, no. , pp. 437-438, 2012, doi:
99 ms
(Ver 3.3 (11022016))