High Level Transforms for SIMD and Low-Level Computer Vision Algorithms
Identifieur interne : 000101 ( Hal/Corpus ); précédent : 000100; suivant : 000102High Level Transforms for SIMD and Low-Level Computer Vision Algorithms
Auteurs : Lionel Lacassagne ; Daniel Etiemble ; Hassan Zahraee ; Alain Dominguez ; Pascal VezolleSource :
English descriptors
- mix :
Abstract
This paper presents a review of algorithmic transforms called High Level Transforms for IBM, Intel and ARM SIMD multi-core pro-cessors to accelerate the implementation of low level image pro-cessing algorithms. We show that these optimizations provide a significant acceleration. A first evaluation of 512-bit SIMD Xeon-Phi is also presented. We focus on the point that the combination of optimizations leading to the best execution time cannot be pre-dicted, and thus, systematic benchmarking is mandatory. Once the best configuration is found for each architecture, a comparison of these performances is presented. The Harris points detection opera-tor is selected as being representative of low level image processing and computer vision algorithms. Being composed of five convolu-tions, it is more complex than a simple filter and enables more op-portunities to combine optimizations. The presented work can scale across a wide range of codes using 2D stencils and convolutions.
Url:
Links to Exploration step
Hal:hal-01094906Le document en format XML
<record><TEI><teiHeader><fileDesc><titleStmt><title xml:lang="en">High Level Transforms for SIMD and Low-Level Computer Vision Algorithms</title><author><name sortKey="Lacassagne, Lionel" sort="Lacassagne, Lionel" uniqKey="Lacassagne L" first="Lionel" last="Lacassagne">Lionel Lacassagne</name><affiliation><hal:affiliation type="laboratory" xml:id="struct-2544" status="VALID"><orgName>Laboratoire de Recherche en Informatique</orgName><orgName type="acronym">LRI</orgName><desc><address><addrLine>LRI - Bâtiments 650-660 Université Paris-Sud 91405 Orsay Cedex</addrLine><country key="FR"></country></address><ref type="url">http://www.lri.fr/</ref></desc><listRelation><relation active="#struct-92966" type="direct"></relation><relation name="UMR8623" active="#struct-441569" type="direct"></relation></listRelation><tutelles><tutelle active="#struct-92966" type="direct"><org type="institution" xml:id="struct-92966" status="VALID"><orgName>Université Paris-Sud - Paris 11</orgName><orgName type="acronym">UP11</orgName><desc><address><addrLine>Bâtiment 300 - 91405 Orsay cedex</addrLine><country key="FR"></country></address><ref type="url">http://www.u-psud.fr/</ref></desc></org></tutelle><tutelle name="UMR8623" active="#struct-441569" type="direct"><org type="institution" xml:id="struct-441569" status="VALID"><idno type="IdRef">02636817X</idno><idno type="ISNI">0000000122597504</idno><orgName>Centre National de la Recherche Scientifique</orgName><orgName type="acronym">CNRS</orgName><date type="start">1939-10-19</date><desc><address><country key="FR"></country></address><ref type="url">http://www.cnrs.fr/</ref></desc></org></tutelle></tutelles></hal:affiliation></affiliation></author><author><name sortKey="Etiemble, Daniel" sort="Etiemble, Daniel" uniqKey="Etiemble D" first="Daniel" last="Etiemble">Daniel Etiemble</name><affiliation><hal:affiliation type="laboratory" xml:id="struct-2544" status="VALID"><orgName>Laboratoire de Recherche en Informatique</orgName><orgName type="acronym">LRI</orgName><desc><address><addrLine>LRI - Bâtiments 650-660 Université Paris-Sud 91405 Orsay Cedex</addrLine><country key="FR"></country></address><ref type="url">http://www.lri.fr/</ref></desc><listRelation><relation active="#struct-92966" type="direct"></relation><relation name="UMR8623" active="#struct-441569" type="direct"></relation></listRelation><tutelles><tutelle active="#struct-92966" type="direct"><org type="institution" xml:id="struct-92966" status="VALID"><orgName>Université Paris-Sud - Paris 11</orgName><orgName type="acronym">UP11</orgName><desc><address><addrLine>Bâtiment 300 - 91405 Orsay cedex</addrLine><country key="FR"></country></address><ref type="url">http://www.u-psud.fr/</ref></desc></org></tutelle><tutelle name="UMR8623" active="#struct-441569" type="direct"><org type="institution" xml:id="struct-441569" status="VALID"><idno type="IdRef">02636817X</idno><idno type="ISNI">0000000122597504</idno><orgName>Centre National de la Recherche Scientifique</orgName><orgName type="acronym">CNRS</orgName><date type="start">1939-10-19</date><desc><address><country key="FR"></country></address><ref type="url">http://www.cnrs.fr/</ref></desc></org></tutelle></tutelles></hal:affiliation></affiliation></author><author><name sortKey="Zahraee, Hassan" sort="Zahraee, Hassan" uniqKey="Zahraee H" first="Hassan" last="Zahraee">Hassan Zahraee</name><affiliation><hal:affiliation type="laboratory" xml:id="struct-227498" status="INCOMING"><orgName>Institut d′Electronique Fondamentale</orgName><desc><address><addrLine>Orsay</addrLine><country key="FR"></country></address></desc><listRelation><relation active="#struct-92966" type="direct"></relation></listRelation><tutelles><tutelle active="#struct-92966" type="direct"><org type="institution" xml:id="struct-92966" status="VALID"><orgName>Université Paris-Sud - Paris 11</orgName><orgName type="acronym">UP11</orgName><desc><address><addrLine>Bâtiment 300 - 91405 Orsay cedex</addrLine><country key="FR"></country></address><ref type="url">http://www.u-psud.fr/</ref></desc></org></tutelle></tutelles></hal:affiliation></affiliation></author><author><name sortKey="Dominguez, Alain" sort="Dominguez, Alain" uniqKey="Dominguez A" first="Alain" last="Dominguez">Alain Dominguez</name><affiliation><hal:affiliation type="laboratory" xml:id="struct-118323" status="VALID"><orgName>Intel Corporation [USA]</orgName><desc><address><country key="US"></country></address></desc><listRelation><relation active="#struct-310927" type="direct"></relation></listRelation><tutelles><tutelle active="#struct-310927" type="direct"><org type="institution" xml:id="struct-310927" status="INCOMING"><orgName>Intel Corporation</orgName><desc><address><country key="FR"></country></address></desc></org></tutelle></tutelles></hal:affiliation></affiliation></author><author><name sortKey="Vezolle, Pascal" sort="Vezolle, Pascal" uniqKey="Vezolle P" first="Pascal" last="Vezolle">Pascal Vezolle</name><affiliation><hal:affiliation type="laboratory" xml:id="struct-95040" status="INCOMING"><orgName>IBM Deep Computing Europe</orgName><desc><address><addrLine>34060 Montpellier</addrLine><country key="FR"></country></address></desc><listRelation><relation active="#struct-300665" type="direct"></relation></listRelation><tutelles><tutelle active="#struct-300665" type="direct"><org type="institution" xml:id="struct-300665" status="VALID"><orgName>IBM</orgName><desc><address><country key="US"></country></address></desc></org></tutelle></tutelles></hal:affiliation></affiliation></author></titleStmt><publicationStmt><idno type="wicri:source">HAL</idno><idno type="RBID">Hal:hal-01094906</idno><idno type="halId">hal-01094906</idno><idno type="halUri">https://hal.inria.fr/hal-01094906</idno><idno type="url">https://hal.inria.fr/hal-01094906</idno><date when="2014-02-15">2014-02-15</date><idno type="wicri:Area/Hal/Corpus">000101</idno></publicationStmt><sourceDesc><biblStruct><analytic><title xml:lang="en">High Level Transforms for SIMD and Low-Level Computer Vision Algorithms</title><author><name sortKey="Lacassagne, Lionel" sort="Lacassagne, Lionel" uniqKey="Lacassagne L" first="Lionel" last="Lacassagne">Lionel Lacassagne</name><affiliation><hal:affiliation type="laboratory" xml:id="struct-2544" status="VALID"><orgName>Laboratoire de Recherche en Informatique</orgName><orgName type="acronym">LRI</orgName><desc><address><addrLine>LRI - Bâtiments 650-660 Université Paris-Sud 91405 Orsay Cedex</addrLine><country key="FR"></country></address><ref type="url">http://www.lri.fr/</ref></desc><listRelation><relation active="#struct-92966" type="direct"></relation><relation name="UMR8623" active="#struct-441569" type="direct"></relation></listRelation><tutelles><tutelle active="#struct-92966" type="direct"><org type="institution" xml:id="struct-92966" status="VALID"><orgName>Université Paris-Sud - Paris 11</orgName><orgName type="acronym">UP11</orgName><desc><address><addrLine>Bâtiment 300 - 91405 Orsay cedex</addrLine><country key="FR"></country></address><ref type="url">http://www.u-psud.fr/</ref></desc></org></tutelle><tutelle name="UMR8623" active="#struct-441569" type="direct"><org type="institution" xml:id="struct-441569" status="VALID"><idno type="IdRef">02636817X</idno><idno type="ISNI">0000000122597504</idno><orgName>Centre National de la Recherche Scientifique</orgName><orgName type="acronym">CNRS</orgName><date type="start">1939-10-19</date><desc><address><country key="FR"></country></address><ref type="url">http://www.cnrs.fr/</ref></desc></org></tutelle></tutelles></hal:affiliation></affiliation></author><author><name sortKey="Etiemble, Daniel" sort="Etiemble, Daniel" uniqKey="Etiemble D" first="Daniel" last="Etiemble">Daniel Etiemble</name><affiliation><hal:affiliation type="laboratory" xml:id="struct-2544" status="VALID"><orgName>Laboratoire de Recherche en Informatique</orgName><orgName type="acronym">LRI</orgName><desc><address><addrLine>LRI - Bâtiments 650-660 Université Paris-Sud 91405 Orsay Cedex</addrLine><country key="FR"></country></address><ref type="url">http://www.lri.fr/</ref></desc><listRelation><relation active="#struct-92966" type="direct"></relation><relation name="UMR8623" active="#struct-441569" type="direct"></relation></listRelation><tutelles><tutelle active="#struct-92966" type="direct"><org type="institution" xml:id="struct-92966" status="VALID"><orgName>Université Paris-Sud - Paris 11</orgName><orgName type="acronym">UP11</orgName><desc><address><addrLine>Bâtiment 300 - 91405 Orsay cedex</addrLine><country key="FR"></country></address><ref type="url">http://www.u-psud.fr/</ref></desc></org></tutelle><tutelle name="UMR8623" active="#struct-441569" type="direct"><org type="institution" xml:id="struct-441569" status="VALID"><idno type="IdRef">02636817X</idno><idno type="ISNI">0000000122597504</idno><orgName>Centre National de la Recherche Scientifique</orgName><orgName type="acronym">CNRS</orgName><date type="start">1939-10-19</date><desc><address><country key="FR"></country></address><ref type="url">http://www.cnrs.fr/</ref></desc></org></tutelle></tutelles></hal:affiliation></affiliation></author><author><name sortKey="Zahraee, Hassan" sort="Zahraee, Hassan" uniqKey="Zahraee H" first="Hassan" last="Zahraee">Hassan Zahraee</name><affiliation><hal:affiliation type="laboratory" xml:id="struct-227498" status="INCOMING"><orgName>Institut d′Electronique Fondamentale</orgName><desc><address><addrLine>Orsay</addrLine><country key="FR"></country></address></desc><listRelation><relation active="#struct-92966" type="direct"></relation></listRelation><tutelles><tutelle active="#struct-92966" type="direct"><org type="institution" xml:id="struct-92966" status="VALID"><orgName>Université Paris-Sud - Paris 11</orgName><orgName type="acronym">UP11</orgName><desc><address><addrLine>Bâtiment 300 - 91405 Orsay cedex</addrLine><country key="FR"></country></address><ref type="url">http://www.u-psud.fr/</ref></desc></org></tutelle></tutelles></hal:affiliation></affiliation></author><author><name sortKey="Dominguez, Alain" sort="Dominguez, Alain" uniqKey="Dominguez A" first="Alain" last="Dominguez">Alain Dominguez</name><affiliation><hal:affiliation type="laboratory" xml:id="struct-118323" status="VALID"><orgName>Intel Corporation [USA]</orgName><desc><address><country key="US"></country></address></desc><listRelation><relation active="#struct-310927" type="direct"></relation></listRelation><tutelles><tutelle active="#struct-310927" type="direct"><org type="institution" xml:id="struct-310927" status="INCOMING"><orgName>Intel Corporation</orgName><desc><address><country key="FR"></country></address></desc></org></tutelle></tutelles></hal:affiliation></affiliation></author><author><name sortKey="Vezolle, Pascal" sort="Vezolle, Pascal" uniqKey="Vezolle P" first="Pascal" last="Vezolle">Pascal Vezolle</name><affiliation><hal:affiliation type="laboratory" xml:id="struct-95040" status="INCOMING"><orgName>IBM Deep Computing Europe</orgName><desc><address><addrLine>34060 Montpellier</addrLine><country key="FR"></country></address></desc><listRelation><relation active="#struct-300665" type="direct"></relation></listRelation><tutelles><tutelle active="#struct-300665" type="direct"><org type="institution" xml:id="struct-300665" status="VALID"><orgName>IBM</orgName><desc><address><country key="US"></country></address></desc></org></tutelle></tutelles></hal:affiliation></affiliation></author></analytic></biblStruct></sourceDesc></fileDesc><profileDesc><textClass><keywords scheme="mix" xml:lang="en"><term>2D stencil</term><term>ARM Neon</term><term>High Level Transforms</term><term>IBM Altivec</term><term>Intel SSE & XeonPhi</term><term>SIMD</term><term>code optimization</term><term>low-level computer vision and image processing algorithms</term></keywords></textClass></profileDesc></teiHeader><front><div type="abstract" xml:lang="en">This paper presents a review of algorithmic transforms called High Level Transforms for IBM, Intel and ARM SIMD multi-core pro-cessors to accelerate the implementation of low level image pro-cessing algorithms. We show that these optimizations provide a significant acceleration. A first evaluation of 512-bit SIMD Xeon-Phi is also presented. We focus on the point that the combination of optimizations leading to the best execution time cannot be pre-dicted, and thus, systematic benchmarking is mandatory. Once the best configuration is found for each architecture, a comparison of these performances is presented. The Harris points detection opera-tor is selected as being representative of low level image processing and computer vision algorithms. Being composed of five convolu-tions, it is more complex than a simple filter and enables more op-portunities to combine optimizations. The presented work can scale across a wide range of codes using 2D stencils and convolutions.</div></front></TEI><hal api="V3"><titleStmt><title xml:lang="en">High Level Transforms for SIMD and Low-Level Computer Vision Algorithms</title><author role="aut"><persName><forename type="first">Lionel</forename><surname>Lacassagne</surname></persName><email></email><idno type="halauthor">1095834</idno><affiliation ref="#struct-2544"></affiliation><affiliation ref="#struct-244753"></affiliation></author><author role="aut"><persName><forename type="first">Daniel</forename><surname>Etiemble</surname></persName><email></email><ptr type="url" target="https://www.lri.fr/~de"></ptr><idno type="halauthor">1112753</idno><affiliation ref="#struct-2544"></affiliation><affiliation ref="#struct-244753"></affiliation></author><author role="aut"><persName><forename type="first">Hassan</forename><surname>Zahraee</surname></persName><email></email><idno type="halauthor">1107824</idno><affiliation ref="#struct-227498"></affiliation></author><author role="aut"><persName><forename type="first">Alain</forename><surname>Dominguez</surname></persName><email>alain.dominguez@intel.com</email><idno type="halauthor">1107825</idno><affiliation ref="#struct-118323"></affiliation></author><author role="aut"><persName><forename type="first">Pascal</forename><surname>Vezolle</surname></persName><email>pascal.vezolle@fr.ibm.com</email><idno type="halauthor">1107826</idno><affiliation ref="#struct-95040"></affiliation></author><editor role="depositor"><persName><forename>Lionel</forename><surname>Lacassagne</surname></persName><email>lionel.lacassagne@lri.fr</email></editor></titleStmt><editionStmt><edition n="v1" type="current"><date type="whenSubmitted">2015-01-06 16:41:42</date><date type="whenModified">2016-01-27 11:41:33</date><date type="whenReleased">2015-01-06 16:41:42</date><date type="whenProduced">2014-02-15</date><date type="whenEndEmbargoed">2015-01-06</date><ref type="file" target="https://hal.inria.fr/hal-01094906/document"><date notBefore="2015-01-06"></date></ref><ref type="file" subtype="author" n="1" target="https://hal.inria.fr/hal-01094906/file/simd-140122.pdf"><date notBefore="2015-01-06"></date></ref></edition><respStmt><resp>contributor</resp><name key="300884"><persName><forename>Lionel</forename><surname>Lacassagne</surname></persName><email>lionel.lacassagne@lri.fr</email></name></respStmt></editionStmt><publicationStmt><distributor>CCSD</distributor><idno type="halId">hal-01094906</idno><idno type="halUri">https://hal.inria.fr/hal-01094906</idno><idno type="halBibtex">lacassagne:hal-01094906</idno><idno type="halRefHtml">Symposium on Principles and Practice of Parallel Programming / WPMVP, Feb 2014, Orlando, Florida, United States. pp.8, <https://sites.google.com/site/ppopp2014/>. <10.1145/2568058.2568067></idno><idno type="halRef">Symposium on Principles and Practice of Parallel Programming / WPMVP, Feb 2014, Orlando, Florida, United States. pp.8, <https://sites.google.com/site/ppopp2014/>. <10.1145/2568058.2568067></idno></publicationStmt><seriesStmt><idno type="stamp" n="INRIA">INRIA - Institut National de Recherche en Informatique et en Automatique</idno><idno type="stamp" n="CNRS">CNRS - Centre national de la recherche scientifique</idno><idno type="stamp" n="INRIA-SACLAY">INRIA Saclay - Ile de France</idno><idno type="stamp" n="UMR8623">Laboratoire de Recherche en Informatique</idno><idno type="stamp" n="INRIA2">INRIA 2</idno><idno type="stamp" n="INRIA_TEST">INRIA - Institut National de Recherche en Informatique et en Automatique</idno><idno type="stamp" n="LRI-PARSYS" p="UMR8623">Laboratoire de recherche en informatique. Équipe: Systèmes Parallèles</idno></seriesStmt><notesStmt><note type="audience" n="2">International</note><note type="invited" n="0">No</note><note type="popular" n="0">No</note><note type="peer" n="1">Yes</note><note type="proceedings" n="1">Yes</note></notesStmt><sourceDesc><biblStruct><analytic><title xml:lang="en">High Level Transforms for SIMD and Low-Level Computer Vision Algorithms</title><author role="aut"><persName><forename type="first">Lionel</forename><surname>Lacassagne</surname></persName><idno type="halAuthorId">1095834</idno><affiliation ref="#struct-2544"></affiliation><affiliation ref="#struct-244753"></affiliation></author><author role="aut"><persName><forename type="first">Daniel</forename><surname>Etiemble</surname></persName><ptr type="url" target="https://www.lri.fr/~de"></ptr><idno type="halAuthorId">1112753</idno><affiliation ref="#struct-2544"></affiliation><affiliation ref="#struct-244753"></affiliation></author><author role="aut"><persName><forename type="first">Hassan</forename><surname>Zahraee</surname></persName><idno type="halAuthorId">1107824</idno><affiliation ref="#struct-227498"></affiliation></author><author role="aut"><persName><forename type="first">Alain</forename><surname>Dominguez</surname></persName><email>alain.dominguez@intel.com</email><idno type="halAuthorId">1107825</idno><affiliation ref="#struct-118323"></affiliation></author><author role="aut"><persName><forename type="first">Pascal</forename><surname>Vezolle</surname></persName><email>pascal.vezolle@fr.ibm.com</email><idno type="halAuthorId">1107826</idno><affiliation ref="#struct-95040"></affiliation></author></analytic><monogr><meeting><title>Symposium on Principles and Practice of Parallel Programming / WPMVP</title><date type="start">2014-02-15</date><date type="end">2014-02-19</date><settlement>Orlando, Florida</settlement><country key="US">United States</country></meeting><imprint><biblScope unit="pp">8</biblScope></imprint></monogr><idno type="doi">10.1145/2568058.2568067</idno><ref type="publisher">https://sites.google.com/site/ppopp2014/</ref></biblStruct></sourceDesc><profileDesc><langUsage><language ident="en">English</language></langUsage><textClass><keywords scheme="author"><term xml:lang="en">High Level Transforms</term><term xml:lang="en">ARM Neon</term><term xml:lang="en">IBM Altivec</term><term xml:lang="en">code optimization</term><term xml:lang="en">low-level computer vision and image processing algorithms</term><term xml:lang="en">2D stencil</term><term xml:lang="en">SIMD</term><term xml:lang="en">Intel SSE & XeonPhi</term></keywords><classCode scheme="acm" n="C.1.1"></classCode><classCode scheme="acm" n="C.1.4"></classCode><classCode scheme="acm" n="D.1.3.1"></classCode><classCode scheme="acm" n="D.3.4.1"></classCode><classCode scheme="acm" n="D.3.4.6"></classCode><classCode scheme="acm" n="I.4.7"></classCode><classCode scheme="halDomain" n="info.info-ds">Computer Science [cs]/Data Structures and Algorithms [cs.DS]</classCode><classCode scheme="halDomain" n="info.info-ar">Computer Science [cs]/Hardware Architecture [cs.AR]</classCode><classCode scheme="halDomain" n="info.info-se">Computer Science [cs]/Software Engineering [cs.SE]</classCode><classCode scheme="halDomain" n="info.info-dm">Computer Science [cs]/Discrete Mathematics [cs.DM]</classCode><classCode scheme="halDomain" n="info.info-rb">Computer Science [cs]/Robotics [cs.RO]</classCode><classCode scheme="halDomain" n="info.info-ti">Computer Science [cs]/Image Processing</classCode><classCode scheme="halDomain" n="info.info-ts">Computer Science [cs]/Signal and Image Processing</classCode><classCode scheme="halDomain" n="info.info-cv">Computer Science [cs]/Computer Vision and Pattern Recognition [cs.CV]</classCode><classCode scheme="halDomain" n="spi.auto">Engineering Sciences [physics]/Automatic</classCode><classCode scheme="halDomain" n="spi.signal">Engineering Sciences [physics]/Signal and Image processing</classCode><classCode scheme="halDomain" n="info.info-ao">Computer Science [cs]/Computer Arithmetic</classCode><classCode scheme="halDomain" n="info.info-dc">Computer Science [cs]/Distributed, Parallel, and Cluster Computing [cs.DC]</classCode><classCode scheme="halTypology" n="COMM">Conference papers</classCode></textClass><abstract xml:lang="en">This paper presents a review of algorithmic transforms called High Level Transforms for IBM, Intel and ARM SIMD multi-core pro-cessors to accelerate the implementation of low level image pro-cessing algorithms. We show that these optimizations provide a significant acceleration. A first evaluation of 512-bit SIMD Xeon-Phi is also presented. We focus on the point that the combination of optimizations leading to the best execution time cannot be pre-dicted, and thus, systematic benchmarking is mandatory. Once the best configuration is found for each architecture, a comparison of these performances is presented. The Harris points detection opera-tor is selected as being representative of low level image processing and computer vision algorithms. Being composed of five convolu-tions, it is more complex than a simple filter and enables more op-portunities to combine optimizations. The presented work can scale across a wide range of codes using 2D stencils and convolutions.</abstract></profileDesc></hal></record>Pour manipuler ce document sous Unix (Dilib)
EXPLOR_STEP=$WICRI_ROOT/Wicri/Musique/explor/OperaV1/Data/Hal/Corpus
HfdSelect -h $EXPLOR_STEP/biblio.hfd -nk 000101 | SxmlIndent | more
Ou
HfdSelect -h $EXPLOR_AREA/Data/Hal/Corpus/biblio.hfd -nk 000101 | SxmlIndent | more
Pour mettre un lien sur cette page dans le réseau Wicri
{{Explor lien
|wiki= Wicri/Musique
|area= OperaV1
|flux= Hal
|étape= Corpus
|type= RBID
|clé= Hal:hal-01094906
|texte= High Level Transforms for SIMD and Low-Level Computer Vision Algorithms
}}
| This area was generated with Dilib version V0.6.21. |  |