TY - GEN

T1 - Mapping parallelism to multi-cores: a machine learning based approach

AU - Wang, Zheng

AU - O'Boyle, Michael F.P.

PY - 2009

Y1 - 2009

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.

AB - 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.

KW - artificial neural networks

KW - compiler optimization

KW - machine learning

KW - performance modeling

KW - support vector machine

UR - http://www.scopus.com/inward/record.url?scp=67650088253&partnerID=8YFLogxK

U2 - 10.1145/1504176.1504189

DO - 10.1145/1504176.1504189

M3 - Conference contribution/Paper

SN - 978-1-60558-397-6

SP - 75

EP - 84

BT - Proceedings of the 14th ACM SIGPLAN symposium on Principles and Practice of Parallel Programming (PPoPP '09)

PB - ACM

CY - New York, NY, USA

T2 - PPoPP 2009 14th ACM SIGPLAN symposium on Principles and practice of parallel programming

Y2 - 14 February 2009 through 18 February 2009

ER -