Auto-tuning Dense Vector and Matrix-vector Operations for Fermi GPUs

Hans Henrik Brandenborg Sørensen

doi:10.1007/978-3-642-31464-3_63

Auto-tuning Dense Vector and Matrix-vector Operations for Fermi GPUs

Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review

Abstract

In this paper, we consider the automatic performance tuning of dense vector and matrix-vector operations on GPUs. Such operations form the backbone of level 1 and level 2 routines in the Basic Linear Algebra Subroutines (BLAS) library and are therefore of great importance in many scientific applications. As examples, we develop single-precision CUDA kernels for the Euclidian norm (SNRM2) and the matrix-vector multiplication (SGEMV). The target hardware is the most recent Nvidia Tesla 20-series (Fermi architecture). We show that auto-tuning can be successfully applied to achieve high performance for dense vector and matrix-vector operations by appropriately utilizing the fine-grained parallelism of the GPU. Our tuned kernels display between 25-100% better performance than the current CUBLAS 3.2 library.

Original language	English
Title of host publication	Parallel Processing and Applied Mathematics : 9th International Conference, PPAM 2011
Editors	Roman Wyrzykowski, Jack Dongarra, Konrad Karczewski, Jerzy Wasniewski
Publisher	Springer
Publication date	2012
Pages	619-629
DOIs	https://doi.org/10.1007/978-3-642-31464-3_63
Publication status	Published - 2012
Event	Parallel Processing and Applied Mathematics. 9th International Conference, PPAM 2011 - Torun, Poland Duration: 11 Sept 2011 → 14 Sept 2011 http://ppam.pl/

Conference

Conference	Parallel Processing and Applied Mathematics. 9th International Conference, PPAM 2011
Country/Territory	Poland
City	Torun
Period	11/09/2011 → 14/09/2011
Internet address	http://ppam.pl/

Series	Lecture Notes in Computer Science
Volume	7203
ISSN	0302-9743

Keywords

GPU
BLAS
Dense linear algebra
Parallel algorithms

Access to Document

10.1007/978-3-642-31464-3_63

OpenUrl availability

Full text

Cite this

@inproceedings{a2ae08a60eae4bc9b2f98abdaa6d7973,

title = "Auto-tuning Dense Vector and Matrix-vector Operations for Fermi GPUs",

abstract = "In this paper, we consider the automatic performance tuning of dense vector and matrix-vector operations on GPUs. Such operations form the backbone of level 1 and level 2 routines in the Basic Linear Algebra Subroutines (BLAS) library and are therefore of great importance in many scientific applications. As examples, we develop single-precision CUDA kernels for the Euclidian norm (SNRM2) and the matrix-vector multiplication (SGEMV). The target hardware is the most recent Nvidia Tesla 20-series (Fermi architecture). We show that auto-tuning can be successfully applied to achieve high performance for dense vector and matrix-vector operations by appropriately utilizing the fine-grained parallelism of the GPU. Our tuned kernels display between 25-100% better performance than the current CUBLAS 3.2 library.",

keywords = "GPU, BLAS, Dense linear algebra, Parallel algorithms",

author = "S{\o}rensen, {Hans Henrik Brandenborg}",

year = "2012",

doi = "10.1007/978-3-642-31464-3_63",

language = "English",

series = "Lecture Notes in Computer Science",

publisher = "Springer",

pages = "619--629",

editor = "Wyrzykowski, {Roman } and Dongarra, {Jack } and Konrad Karczewski and Wasniewski, {Jerzy }",

booktitle = "Parallel Processing and Applied Mathematics",

note = "Parallel Processing and Applied Mathematics. 9th International Conference, PPAM 2011 ; Conference date: 11-09-2011 Through 14-09-2011",

url = "http://ppam.pl/",

}

Sørensen, HHB 2012, Auto-tuning Dense Vector and Matrix-vector Operations for Fermi GPUs. in R Wyrzykowski, J Dongarra, K Karczewski & J Wasniewski (eds), Parallel Processing and Applied Mathematics: 9th International Conference, PPAM 2011. Springer, Lecture Notes in Computer Science, vol. 7203, pp. 619-629, Parallel Processing and Applied Mathematics. 9th International Conference, PPAM 2011, Torun, Poland, 11/09/2011. https://doi.org/10.1007/978-3-642-31464-3_63

Auto-tuning Dense Vector and Matrix-vector Operations for Fermi GPUs. / Sørensen, Hans Henrik Brandenborg.
Parallel Processing and Applied Mathematics: 9th International Conference, PPAM 2011. ed. / Roman Wyrzykowski; Jack Dongarra; Konrad Karczewski; Jerzy Wasniewski. Springer, 2012. p. 619-629 (Lecture Notes in Computer Science, Vol. 7203).

Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review

TY - GEN

T1 - Auto-tuning Dense Vector and Matrix-vector Operations for Fermi GPUs

AU - Sørensen, Hans Henrik Brandenborg

PY - 2012

Y1 - 2012

N2 - In this paper, we consider the automatic performance tuning of dense vector and matrix-vector operations on GPUs. Such operations form the backbone of level 1 and level 2 routines in the Basic Linear Algebra Subroutines (BLAS) library and are therefore of great importance in many scientific applications. As examples, we develop single-precision CUDA kernels for the Euclidian norm (SNRM2) and the matrix-vector multiplication (SGEMV). The target hardware is the most recent Nvidia Tesla 20-series (Fermi architecture). We show that auto-tuning can be successfully applied to achieve high performance for dense vector and matrix-vector operations by appropriately utilizing the fine-grained parallelism of the GPU. Our tuned kernels display between 25-100% better performance than the current CUBLAS 3.2 library.

AB - In this paper, we consider the automatic performance tuning of dense vector and matrix-vector operations on GPUs. Such operations form the backbone of level 1 and level 2 routines in the Basic Linear Algebra Subroutines (BLAS) library and are therefore of great importance in many scientific applications. As examples, we develop single-precision CUDA kernels for the Euclidian norm (SNRM2) and the matrix-vector multiplication (SGEMV). The target hardware is the most recent Nvidia Tesla 20-series (Fermi architecture). We show that auto-tuning can be successfully applied to achieve high performance for dense vector and matrix-vector operations by appropriately utilizing the fine-grained parallelism of the GPU. Our tuned kernels display between 25-100% better performance than the current CUBLAS 3.2 library.

KW - GPU

KW - BLAS

KW - Dense linear algebra

KW - Parallel algorithms

U2 - 10.1007/978-3-642-31464-3_63

DO - 10.1007/978-3-642-31464-3_63

M3 - Article in proceedings

T3 - Lecture Notes in Computer Science

SP - 619

EP - 629

BT - Parallel Processing and Applied Mathematics

A2 - Wyrzykowski, Roman

A2 - Dongarra, Jack

A2 - Karczewski, Konrad

A2 - Wasniewski, Jerzy

PB - Springer

T2 - Parallel Processing and Applied Mathematics. 9th International Conference, PPAM 2011

Y2 - 11 September 2011 through 14 September 2011

ER -

Auto-tuning Dense Vector and Matrix-vector Operations for Fermi GPUs

Abstract

Conference

Keywords

Access to Document

OpenUrl availability

Fingerprint

Cite this