Towards Dynamic Autotuning of SpMV in CUSP Library

D 2024

Towards Dynamic Autotuning of SpMV in CUSP Library

DEMEK, Miroslav and Jiří FILIPOVIČ

Basic information

Original name

Towards Dynamic Autotuning of SpMV in CUSP Library

Authors

DEMEK, Miroslav (203 Czech Republic, belonging to the institution) and Jiří FILIPOVIČ (203 Czech Republic, guarantor, belonging to the institution)

Edition

San Francisco, USA, IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW), p. 14-22, 9 pp. 2024

Publisher

IEEE

Other information

Language

English

Type of outcome

Proceedings paper

Field of Study

10201 Computer sciences, information science, bioinformatics

Country of publisher

United States of America

Confidentiality degree

is not subject to a state or trade secret

Publication form

electronic version available online

RIV identification code

RIV/00216224:14610/24:00137324

Organization unit

Institute of Computer Science

ISBN

979-8-3503-6461-3

DOI

http://dx.doi.org/10.1109/IPDPSW63119.2024.00012

UT WoS

001284697300084

EID Scopus

2-s2.0-85200751879

Keywords in English

dynamic autotuning; SpMV; CUPS; Kernel Tuning Toolkit

Abstract

In the original language

Sparse matrix-vector product (SpMV) is a central operation in many iterative methods for solving linear systems and, as such, is an attractive candidate for acceleration on the GPU. However, the performance of the SpMV kernel can vary depending both on the target architecture as well as on the sparsity pattern of the matrix. Thus, to achieve optimal performance, the implementation might need to be adjusted for each particular matrix and architecture. This can be achieved through dynamic autotuning, a method that can optimize a source code during program runtime. In this paper, we present a dynamic autotuning of SpMV kernel included in a production-quality CUSP library. We identify and implement tuning parameters and use the Kernel Tuning Toolkit framework for autotuning of SpMV working with the DIA and ELL sparse matrix formats. The dynamic autotuning integration is fully transparent to the library users - it can be activated just by re-compiling software using our tunable version of the CUSP. The proposed autotuned library is evaluated by comparing it with the original CUSP kernels on a set of representative matrices and by examining the contribution of autotuning. The results show that the autotuned kernels can reach up to about 16.9 × speedup compared to a fixed implementation.

Links

LM2023054, research and development project

Name: e-Infrastruktura CZ

Investor: Ministry of Education, Youth and Sports of the CR

Cite

DEMEK, Miroslav and Jiří FILIPOVIČ. Towards Dynamic Autotuning of SpMV in CUSP Library. Online. In IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW). San Francisco, USA: IEEE, 2024, p. 14-22. ISBN 979-8-3503-6461-3. Available from: https://dx.doi.org/10.1109/IPDPSW63119.2024.00012.

@inproceedings{2442222,
   author = {Demek, Miroslav and Filipovič, Jiří},
   address = {San Francisco, USA},
   booktitle = {IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW)},
   doi = {http://dx.doi.org/10.1109/IPDPSW63119.2024.00012},
   keywords = {dynamic autotuning; SpMV; CUPS; Kernel Tuning Toolkit},
   howpublished = {elektronická verze "online"},
   language = {eng},
   location = {San Francisco, USA},
   isbn = {979-8-3503-6461-3},
   pages = {14-22},
   publisher = {IEEE},
   title = {Towards Dynamic Autotuning of SpMV in CUSP Library},
   year = {2024}
}

TY  - CONF
ID  - 2442222
AU  - Demek, Miroslav - Filipovič, Jiří
PY  - 2024
TI  - Towards Dynamic Autotuning of SpMV in CUSP Library
PB  - IEEE
CY  - San Francisco, USA
SN  - 9798350364613
KW  - dynamic autotuning
KW  - SpMV
KW  - CUPS
KW  - Kernel Tuning Toolkit
N2  - Sparse matrix-vector product (SpMV) is a central operation in many iterative methods for solving linear systems and, as such, is an attractive candidate for acceleration on the GPU. However, the performance of the SpMV kernel can vary depending both on the target architecture as well as on the sparsity pattern of the matrix. Thus, to achieve optimal performance, the implementation might need to be adjusted for each particular matrix and architecture. This can be achieved through dynamic autotuning, a method that can optimize a source code during program runtime. In this paper, we present a dynamic autotuning of SpMV kernel included in a production-quality CUSP library. We identify and implement tuning parameters and use the Kernel Tuning Toolkit framework for autotuning of SpMV working with the DIA and ELL sparse matrix formats. The dynamic autotuning integration is fully transparent to the library users - it can be activated just by re-compiling software using our tunable version of the CUSP. The proposed autotuned library is evaluated by comparing it with the original CUSP kernels on a set of representative matrices and by examining the contribution of autotuning. The results show that the autotuned kernels can reach up to about 16.9 × speedup compared to a fixed implementation.
ER  -

DEMEK, Miroslav and Jiří FILIPOVIČ. Towards Dynamic Autotuning of SpMV in CUSP Library. Online. In \textit{IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW)}. San Francisco, USA: IEEE, 2024, p.~14-22. ISBN~979-8-3503-6461-3. Available from: https://dx.doi.org/10.1109/IPDPSW63119.2024.00012.

Přehled o publikaci