TY - GEN
T1 - Machine learning approach for the predicting performance of SpMV on GPU
AU - Benatia, Akrem
AU - Ji, Weixing
AU - Wang, Yizhuo
AU - Shi, Feng
N1 - Publisher Copyright:
© 2016 IEEE.
PY - 2016/7/2
Y1 - 2016/7/2
N2 - Sparse Matrix-Vector Multiplication (SpMV) kernel dominates the computing cost in numerous scientific applications. Many implementations based on different sparse formats were proposed recently for optimizing this kernel on the GPU side. Since the performance of the SpMV varies significantly according to the sparsity characteristics of the input matrix and the hardware features, developing an accurate performance model for this kernel is a challenging task. The traditional approach of building such models by analytical modeling is difficult in practice and requires a thorough understanding of the interaction between the GPU hardware and the sparse code. In this paper, we propose to use a machine learning approach to predict the performance of the SpMV kernel using several sparse formats (COO, CSR, ELL, and HYB) on GPU. We used two popular machine learning algorithms, Support Vector Regression (SVR) and Multilayer Perceptron neural network (MLP). Our experimental results on two different GPUs (Fermi GTX 512 and Maxwell GTX 980 Ti) show that the SVR models deliver the best accuracy with average prediction error ranging between 7% and 14%.
AB - Sparse Matrix-Vector Multiplication (SpMV) kernel dominates the computing cost in numerous scientific applications. Many implementations based on different sparse formats were proposed recently for optimizing this kernel on the GPU side. Since the performance of the SpMV varies significantly according to the sparsity characteristics of the input matrix and the hardware features, developing an accurate performance model for this kernel is a challenging task. The traditional approach of building such models by analytical modeling is difficult in practice and requires a thorough understanding of the interaction between the GPU hardware and the sparse code. In this paper, we propose to use a machine learning approach to predict the performance of the SpMV kernel using several sparse formats (COO, CSR, ELL, and HYB) on GPU. We used two popular machine learning algorithms, Support Vector Regression (SVR) and Multilayer Perceptron neural network (MLP). Our experimental results on two different GPUs (Fermi GTX 512 and Maxwell GTX 980 Ti) show that the SVR models deliver the best accuracy with average prediction error ranging between 7% and 14%.
KW - GPU computing
KW - Multilayer Perceptron (MLP)
KW - Performance modeling
KW - Sparse Matrix-Vector multiplication (SpMV)
KW - Support Vector Regression (SVR)
UR - http://www.scopus.com/inward/record.url?scp=85018508629&partnerID=8YFLogxK
U2 - 10.1109/ICPADS.2016.0120
DO - 10.1109/ICPADS.2016.0120
M3 - Conference contribution
AN - SCOPUS:85018508629
T3 - Proceedings of the International Conference on Parallel and Distributed Systems - ICPADS
SP - 894
EP - 901
BT - Proceedings - 22nd IEEE International Conference on Parallel and Distributed Systems, ICPADS 2016
A2 - Liao, Xiaofei
A2 - Lovas, Robert
A2 - Shen, Xipeng
A2 - Zheng, Ran
PB - IEEE Computer Society
T2 - 22nd IEEE International Conference on Parallel and Distributed Systems, ICPADS 2016
Y2 - 13 December 2016 through 16 December 2016
ER -