Generating optimal CUDA sparse matrix-vector product implementations for evolving GPU hardware

Ahmed H. El Zein; Alistair P. Rendell

doi:10.1002/cpe.1732

Generating optimal CUDA sparse matrix-vector product implementations for evolving GPU hardware

Ahmed H. El Zein^*, Alistair P. Rendell

^*Corresponding author for this work

Research output: Contribution to journal › Article › peer-review

23 Citations (Scopus)

Abstract

The CUDA model for graphics processing units (GPUs) presents the programmer with a plethora of different programming options. These includes different memory types, different memory access methods and different data types. Identifying which options to use and when is a non-trivial exercise. This paper explores the effect of these different options on the performance of a routine that evaluates sparse matrix-vector products (SpMV) across three different generations of NVIDIA GPU hardware. A process for analysing performance and selecting the subset of implementations that perform best is proposed. The potential for mapping sparse matrix attributes to optimal CUDA SpMV implementations is discussed.

Original language	English
Pages (from-to)	3-13
Number of pages	11
Journal	Concurrency Computation Practice and Experience
Volume	24
Issue number	1
DOIs	https://doi.org/10.1002/cpe.1732
Publication status	Published - Jan 2012

Access to Document

10.1002/cpe.1732

Cite this

@article{1d03bae303e14ee0af5dabdefe14ecf2,

title = "Generating optimal CUDA sparse matrix-vector product implementations for evolving GPU hardware",

abstract = "The CUDA model for graphics processing units (GPUs) presents the programmer with a plethora of different programming options. These includes different memory types, different memory access methods and different data types. Identifying which options to use and when is a non-trivial exercise. This paper explores the effect of these different options on the performance of a routine that evaluates sparse matrix-vector products (SpMV) across three different generations of NVIDIA GPU hardware. A process for analysing performance and selecting the subset of implementations that perform best is proposed. The potential for mapping sparse matrix attributes to optimal CUDA SpMV implementations is discussed.",

keywords = "CUDA, Fermi, GPU, NVIDIA, S2050, matrix-vector, sparse",

author = "{El Zein}, {Ahmed H.} and Rendell, {Alistair P.}",

year = "2012",

month = jan,

doi = "10.1002/cpe.1732",

language = "English",

volume = "24",

pages = "3--13",

journal = "Concurrency Computation Practice and Experience",

issn = "1532-0626",

publisher = "John Wiley & Sons Ltd.",

number = "1",

}

TY - JOUR

T1 - Generating optimal CUDA sparse matrix-vector product implementations for evolving GPU hardware

AU - El Zein, Ahmed H.

AU - Rendell, Alistair P.

PY - 2012/1

Y1 - 2012/1

N2 - The CUDA model for graphics processing units (GPUs) presents the programmer with a plethora of different programming options. These includes different memory types, different memory access methods and different data types. Identifying which options to use and when is a non-trivial exercise. This paper explores the effect of these different options on the performance of a routine that evaluates sparse matrix-vector products (SpMV) across three different generations of NVIDIA GPU hardware. A process for analysing performance and selecting the subset of implementations that perform best is proposed. The potential for mapping sparse matrix attributes to optimal CUDA SpMV implementations is discussed.

AB - The CUDA model for graphics processing units (GPUs) presents the programmer with a plethora of different programming options. These includes different memory types, different memory access methods and different data types. Identifying which options to use and when is a non-trivial exercise. This paper explores the effect of these different options on the performance of a routine that evaluates sparse matrix-vector products (SpMV) across three different generations of NVIDIA GPU hardware. A process for analysing performance and selecting the subset of implementations that perform best is proposed. The potential for mapping sparse matrix attributes to optimal CUDA SpMV implementations is discussed.

KW - CUDA

KW - Fermi

KW - GPU

KW - NVIDIA

KW - S2050

KW - matrix-vector

KW - sparse

UR - http://www.scopus.com/inward/record.url?scp=84855223315&partnerID=8YFLogxK

U2 - 10.1002/cpe.1732

DO - 10.1002/cpe.1732

M3 - Article

SN - 1532-0626

VL - 24

SP - 3

EP - 13

JO - Concurrency Computation Practice and Experience

JF - Concurrency Computation Practice and Experience

IS - 1

ER -

Generating optimal CUDA sparse matrix-vector product implementations for evolving GPU hardware

Abstract

Access to Document

Other files and links

Fingerprint

Cite this