Mapping parallelism to multi-cores: a machine learning based approach Export

@inproceedings{citeulike:4108189, abstract = {The efficient mapping of program parallelism to multi-core processors is highly dependent on the underlying architecture. This paper proposes a portable and automatic compiler-based approach to mapping such parallelism using machine learning. It develops two predictors: a data sensitive and a data insensitive predictor to select the best mapping for parallel programs. They predict the number of threads and the scheduling policy for any given program using a model learnt off-line. By using low-cost profiling runs, they predict the mapping for a new unseen program across multiple input data sets. We evaluate our approach by selecting parallelism mapping configurations for OpenMP programs on two representative but different multi-core platforms (the Intel Xeon and the Cell processors). Performance of our technique is stable across programs and architectures. On average, it delivers above 96\% performance of the maximum available on both platforms. It achieve, on average, a 37\% (up to 17.5 times ) performance improvement over the OpenMP runtime default scheme on the Cell platform. Compared to two recent prediction models, our predictors achieve better performance with a significant lower profiling cost.}, address = {New York, NY, USA}, author = {Wang, Zheng and O'Boyle, Michael F. P.}, booktitle = {PPoPP '09: Proceedings of the 14th ACM SIGPLAN symposium on Principles and practice of parallel programming}, citeulike-article-id = {4108189}, citeulike-linkout-0 = {http://portal.acm.org/citation.cfm?id=1504176.1504189}, citeulike-linkout-1 = {http://dx.doi.org/10.1145/1504176.1504189}, doi = {10.1145/1504176.1504189}, isbn = {978-1-60558-397-6}, location = {Raleigh, NC, USA}, pages = {75--84}, posted-at = {2009-04-21 02:22:17}, priority = {2}, publisher = {ACM}, title = {Mapping parallelism to multi-cores: a machine learning based approach}, url = {http://dx.doi.org/10.1145/1504176.1504189}, year = {2008} }

TY - CONF ID - citeulike:4108189 L3 - citeulike-article-id:4108189 N2 - The efficient mapping of program parallelism to multi-core processors is highly dependent on the underlying architecture. This paper proposes a portable and automatic compiler-based approach to mapping such parallelism using machine learning. It develops two predictors: a data sensitive and a data insensitive predictor to select the best mapping for parallel programs. They predict the number of threads and the scheduling policy for any given program using a model learnt off-line. By using low-cost profiling runs, they predict the mapping for a new unseen program across multiple input data sets. We evaluate our approach by selecting parallelism mapping configurations for OpenMP programs on two representative but different multi-core platforms (the Intel Xeon and the Cell processors). Performance of our technique is stable across programs and architectures. On average, it delivers above 96% performance of the maximum available on both platforms. It achieve, on average, a 37% (up to 17.5 times ) performance improvement over the OpenMP runtime default scheme on the Cell platform. Compared to two recent prediction models, our predictors achieve better performance with a significant lower profiling cost. CY - Raleigh, NC, USA SN - 978-1-60558-397-6 AD - New York, NY, USA EP - 84 TI - Mapping parallelism to multi-cores: a machine learning based approach PB - ACM T2 - PPoPP '09: Proceedings of the 14th ACM SIGPLAN symposium on Principles and practice of parallel programming SP - 75 AU - Wang, Zheng AU - O'Boyle, Michael PY - 2008/// UR - http://dx.doi.org/10.1145/1504176.1504189 ER -