The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. ex. Some numerals are expressed as "XNUMX".
Copyrights notice
The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. Copyrights notice
A multiplicação esparsa de matrizes-vetores (SpMV) é um kernel computacional amplamente utilizado em muitas aplicações. Devido à importância, muitas implementações diferentes foram propostas para acelerar este kernel computacional. As características de desempenho dessas implementações de SpMV são bastante diferentes e é basicamente difícil selecionar a implementação que tem o melhor desempenho para uma determinada matriz esparsa sem perfil de desempenho. Uma abordagem existente para o problema de seleção do melhor código SpMV é usar recursos predefinidos manualmente e um modelo de aprendizado de máquina para a seleção. No entanto, geralmente é difícil definir manualmente características que possam expressar perfeitamente as características da matriz esparsa original necessária para a seleção do código. Além disso, alguma perda de informação aconteceria com o uso dessa abordagem. Portanto, este artigo apresenta um mecanismo eficaz de aprendizado profundo para seleção de código SpMV mais adequado para uma determinada matriz esparsa. Em vez de usar recursos predefinidos manualmente de uma matriz esparsa, uma imagem de recurso e uma rede de aprendizado profundo são usadas para mapear cada matriz esparsa para a implementação, que deverá ter o melhor desempenho, antes da execução. Os benefícios da utilização do mecanismo proposto são discutidos através do cálculo da precisão da previsão e do desempenho. De acordo com a avaliação, o mecanismo proposto pode selecionar uma implementação ideal ou subótima para uma matriz esparsa não vista no conjunto de dados de teste na maioria dos casos. Esses resultados demonstram que, ao usar o aprendizado profundo, toda uma matriz esparsa pode ser usada para fazer a melhor previsão de implementação, e a precisão da previsão alcançada pelo mecanismo proposto é maior do que a do uso de recursos predefinidos.
Hang CUI
Tohoku University
Shoichi HIRASAWA
National Institute of Informatics
Hiroaki KOBAYASHI
Tohoku University
Hiroyuki TAKIZAWA
Tohoku University
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Copiar
Hang CUI, Shoichi HIRASAWA, Hiroaki KOBAYASHI, Hiroyuki TAKIZAWA, "A Machine Learning-Based Approach for Selecting SpMV Kernels and Matrix Storage Formats" in IEICE TRANSACTIONS on Information,
vol. E101-D, no. 9, pp. 2307-2314, September 2018, doi: 10.1587/transinf.2017EDP7176.
Abstract: Sparse Matrix-Vector multiplication (SpMV) is a computational kernel widely used in many applications. Because of the importance, many different implementations have been proposed to accelerate this computational kernel. The performance characteristics of those SpMV implementations are quite different, and it is basically difficult to select the implementation that has the best performance for a given sparse matrix without performance profiling. One existing approach to the SpMV best-code selection problem is by using manually-predefined features and a machine learning model for the selection. However, it is generally hard to manually define features that can perfectly express the characteristics of the original sparse matrix necessary for the code selection. Besides, some information loss would happen by using this approach. This paper hence presents an effective deep learning mechanism for SpMV code selection best suited for a given sparse matrix. Instead of using manually-predefined features of a sparse matrix, a feature image and a deep learning network are used to map each sparse matrix to the implementation, which is expected to have the best performance, in advance of the execution. The benefits of using the proposed mechanism are discussed by calculating the prediction accuracy and the performance. According to the evaluation, the proposed mechanism can select an optimal or suboptimal implementation for an unseen sparse matrix in the test data set in most cases. These results demonstrate that, by using deep learning, a whole sparse matrix can be used to do the best implementation prediction, and the prediction accuracy achieved by the proposed mechanism is higher than that of using predefined features.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2017EDP7176/_p
Copiar
@ARTICLE{e101-d_9_2307,
author={Hang CUI, Shoichi HIRASAWA, Hiroaki KOBAYASHI, Hiroyuki TAKIZAWA, },
journal={IEICE TRANSACTIONS on Information},
title={A Machine Learning-Based Approach for Selecting SpMV Kernels and Matrix Storage Formats},
year={2018},
volume={E101-D},
number={9},
pages={2307-2314},
abstract={Sparse Matrix-Vector multiplication (SpMV) is a computational kernel widely used in many applications. Because of the importance, many different implementations have been proposed to accelerate this computational kernel. The performance characteristics of those SpMV implementations are quite different, and it is basically difficult to select the implementation that has the best performance for a given sparse matrix without performance profiling. One existing approach to the SpMV best-code selection problem is by using manually-predefined features and a machine learning model for the selection. However, it is generally hard to manually define features that can perfectly express the characteristics of the original sparse matrix necessary for the code selection. Besides, some information loss would happen by using this approach. This paper hence presents an effective deep learning mechanism for SpMV code selection best suited for a given sparse matrix. Instead of using manually-predefined features of a sparse matrix, a feature image and a deep learning network are used to map each sparse matrix to the implementation, which is expected to have the best performance, in advance of the execution. The benefits of using the proposed mechanism are discussed by calculating the prediction accuracy and the performance. According to the evaluation, the proposed mechanism can select an optimal or suboptimal implementation for an unseen sparse matrix in the test data set in most cases. These results demonstrate that, by using deep learning, a whole sparse matrix can be used to do the best implementation prediction, and the prediction accuracy achieved by the proposed mechanism is higher than that of using predefined features.},
keywords={},
doi={10.1587/transinf.2017EDP7176},
ISSN={1745-1361},
month={September},}
Copiar
TY - JOUR
TI - A Machine Learning-Based Approach for Selecting SpMV Kernels and Matrix Storage Formats
T2 - IEICE TRANSACTIONS on Information
SP - 2307
EP - 2314
AU - Hang CUI
AU - Shoichi HIRASAWA
AU - Hiroaki KOBAYASHI
AU - Hiroyuki TAKIZAWA
PY - 2018
DO - 10.1587/transinf.2017EDP7176
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E101-D
IS - 9
JA - IEICE TRANSACTIONS on Information
Y1 - September 2018
AB - Sparse Matrix-Vector multiplication (SpMV) is a computational kernel widely used in many applications. Because of the importance, many different implementations have been proposed to accelerate this computational kernel. The performance characteristics of those SpMV implementations are quite different, and it is basically difficult to select the implementation that has the best performance for a given sparse matrix without performance profiling. One existing approach to the SpMV best-code selection problem is by using manually-predefined features and a machine learning model for the selection. However, it is generally hard to manually define features that can perfectly express the characteristics of the original sparse matrix necessary for the code selection. Besides, some information loss would happen by using this approach. This paper hence presents an effective deep learning mechanism for SpMV code selection best suited for a given sparse matrix. Instead of using manually-predefined features of a sparse matrix, a feature image and a deep learning network are used to map each sparse matrix to the implementation, which is expected to have the best performance, in advance of the execution. The benefits of using the proposed mechanism are discussed by calculating the prediction accuracy and the performance. According to the evaluation, the proposed mechanism can select an optimal or suboptimal implementation for an unseen sparse matrix in the test data set in most cases. These results demonstrate that, by using deep learning, a whole sparse matrix can be used to do the best implementation prediction, and the prediction accuracy achieved by the proposed mechanism is higher than that of using predefined features.
ER -