A Set of Batched Basic Linear Algebra Subprograms and LAPACK Routines

Ahmad Abdelfattah; Timothy Costa; Jack Dongarra; Mark Gates; Azzam Haidar; Sven Hammarling; Nicholas J. Higham; Jakub Kurzak; Piotr Luszczek; Stanimire Tomov; Mawussi Zounon

doi:10.1145/3431921

A Set of Batched Basic Linear Algebra Subprograms and LAPACK Routines

Ahmad Abdelfattah, Timothy Costa, Jack Dongarra, Mark Gates, Azzam Haidar, Sven Hammarling, Nicholas J. Higham, Jakub Kurzak, Piotr Luszczek, Stanimire Tomov, Mawussi Zounon

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

23 Scopus citations

Abstract

This article describes a standard API for a set of Batched Basic Linear Algebra Subprograms (Batched BLAS or BBLAS). The focus is on many independent BLAS operations on small matrices that are grouped together and processed by a single routine, called a Batched BLAS routine. The matrices are grouped together in uniformly sized groups, with just one group if all the matrices are of equal size. The aim is to provide more efficient, but portable, implementations of algorithms on high-performance many-core platforms. These include multicore and many-core CPU processors, GPUs and coprocessors, and other hardware accelerators with floating-point compute facility. As well as the standard types of single and double precision, we also include half and quadruple precision in the standard. In particular, half precision is used in many very large scale applications, such as those associated with machine learning.

Original language	English
Article number	21
Journal	ACM Transactions on Mathematical Software
Volume	47
Issue number	3
DOIs	https://doi.org/10.1145/3431921
State	Published - Jun 2021
Externally published	Yes

Funding

This material is based upon work supported in part by the National Science Foundation under Grants No. OAC 1740250 and CSR 1514286 and OAC 2004850, NVIDIA, the Department of Energy, and in part by the Russian Scientific Foundation, Agreement N14-11-00190. This project was also funded in part from the European Union’s Horizon 2020 research and innovation programme under the NLAFET grant agreement No 671633. Authors’ addresses: A. Abdelfattah, M. Gates, P. Luszczek, and S. Tomov, University of Tennessee, 1122 Volunteer Blvd., Suite 203, Knoxville, TN 37996-3450, USA; emails: [email protected], [email protected], [email protected], [email protected]; T. Costa, NVIDIA, Santa Clara; J. Dongarra, University of Tennessee, Oak Ridge National Laboratory, and University of Manchester, Knoxville; email: [email protected]; A. Haidar, NVIDIA, Knoxville; email: [email protected]; S. Hammarling and N. J. Higham, University of Manchester, Manchester, UK; emails: [email protected], [email protected]; J. Kurzak, AMD, Knoxville; M. Zounon, NAG Ltd., Manchester, UK; email: [email protected]. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Request permissions from [email protected]. © 2021 Association for Computing Machinery. 0098-3500/2021/06-ART21 $15.00 https://doi.org/10.1145/3431921

Funders	Funder number
National Science Foundation	CSR 1514286, OAC 2004850, OAC 1740250
U.S. Department of Energy
NVIDIA
Horizon 2020 Framework Programme	671633
Russian Science Foundation	N14-11-00190

Keywords

BLAS
batched BLAS

Access to Document

10.1145/3431921

Cite this

@article{08873e54589045648cbb09e6c2ae67a3,

title = "A Set of Batched Basic Linear Algebra Subprograms and LAPACK Routines",

abstract = "This article describes a standard API for a set of Batched Basic Linear Algebra Subprograms (Batched BLAS or BBLAS). The focus is on many independent BLAS operations on small matrices that are grouped together and processed by a single routine, called a Batched BLAS routine. The matrices are grouped together in uniformly sized groups, with just one group if all the matrices are of equal size. The aim is to provide more efficient, but portable, implementations of algorithms on high-performance many-core platforms. These include multicore and many-core CPU processors, GPUs and coprocessors, and other hardware accelerators with floating-point compute facility. As well as the standard types of single and double precision, we also include half and quadruple precision in the standard. In particular, half precision is used in many very large scale applications, such as those associated with machine learning.",

keywords = "BLAS, batched BLAS",

author = "Ahmad Abdelfattah and Timothy Costa and Jack Dongarra and Mark Gates and Azzam Haidar and Sven Hammarling and Higham, {Nicholas J.} and Jakub Kurzak and Piotr Luszczek and Stanimire Tomov and Mawussi Zounon",

note = "Publisher Copyright: {\textcopyright} 2021 ACM.",

year = "2021",

month = jun,

doi = "10.1145/3431921",

language = "English",

volume = "47",

journal = "ACM Transactions on Mathematical Software",

issn = "0098-3500",

publisher = "Association for Computing Machinery (ACM)",

number = "3",

}

TY - JOUR

T1 - A Set of Batched Basic Linear Algebra Subprograms and LAPACK Routines

AU - Abdelfattah, Ahmad

AU - Costa, Timothy

AU - Dongarra, Jack

AU - Gates, Mark

AU - Haidar, Azzam

AU - Hammarling, Sven

AU - Higham, Nicholas J.

AU - Kurzak, Jakub

AU - Luszczek, Piotr

AU - Tomov, Stanimire

AU - Zounon, Mawussi

PY - 2021/6

Y1 - 2021/6

N2 - This article describes a standard API for a set of Batched Basic Linear Algebra Subprograms (Batched BLAS or BBLAS). The focus is on many independent BLAS operations on small matrices that are grouped together and processed by a single routine, called a Batched BLAS routine. The matrices are grouped together in uniformly sized groups, with just one group if all the matrices are of equal size. The aim is to provide more efficient, but portable, implementations of algorithms on high-performance many-core platforms. These include multicore and many-core CPU processors, GPUs and coprocessors, and other hardware accelerators with floating-point compute facility. As well as the standard types of single and double precision, we also include half and quadruple precision in the standard. In particular, half precision is used in many very large scale applications, such as those associated with machine learning.

AB - This article describes a standard API for a set of Batched Basic Linear Algebra Subprograms (Batched BLAS or BBLAS). The focus is on many independent BLAS operations on small matrices that are grouped together and processed by a single routine, called a Batched BLAS routine. The matrices are grouped together in uniformly sized groups, with just one group if all the matrices are of equal size. The aim is to provide more efficient, but portable, implementations of algorithms on high-performance many-core platforms. These include multicore and many-core CPU processors, GPUs and coprocessors, and other hardware accelerators with floating-point compute facility. As well as the standard types of single and double precision, we also include half and quadruple precision in the standard. In particular, half precision is used in many very large scale applications, such as those associated with machine learning.

KW - BLAS

KW - batched BLAS

UR - http://www.scopus.com/inward/record.url?scp=85108906597&partnerID=8YFLogxK

U2 - 10.1145/3431921

DO - 10.1145/3431921

M3 - Article

AN - SCOPUS:85108906597

SN - 0098-3500

VL - 47

JO - ACM Transactions on Mathematical Software

JF - ACM Transactions on Mathematical Software

IS - 3

M1 - 21

ER -

A Set of Batched Basic Linear Algebra Subprograms and LAPACK Routines

Abstract

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Cite this