{"id":"https://openalex.org/W3009762362","doi":"https://doi.org/10.1137/19m1289546","title":"Mixed Precision Block Fused Multiply-Add: Error Analysis and Application to GPU Tensor Cores","display_name":"Mixed Precision Block Fused Multiply-Add: Error Analysis and Application to GPU Tensor Cores","publication_year":2020,"publication_date":"2020-01-01","ids":{"openalex":"https://openalex.org/W3009762362","doi":"https://doi.org/10.1137/19m1289546","mag":"3009762362"},"language":"en","primary_location":{"id":"doi:10.1137/19m1289546","is_oa":true,"landing_page_url":"https://doi.org/10.1137/19m1289546","pdf_url":null,"source":{"id":"https://openalex.org/S165512578","display_name":"SIAM Journal on Scientific Computing","issn_l":"1064-8275","issn":["1064-8275","1095-7197"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310320508","host_organization_name":"Society for Industrial and Applied Mathematics","host_organization_lineage":["https://openalex.org/P4310320508"],"host_organization_lineage_names":["Society for Industrial and Applied Mathematics"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"SIAM Journal on Scientific Computing","raw_type":"journal-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"hybrid","oa_url":"https://doi.org/10.1137/19m1289546","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5011183093","display_name":"Pierre Blanchard","orcid":"https://orcid.org/0000-0003-4785-3409"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Pierre Blanchard","raw_affiliation_strings":["Department of Mathematics [Manchester]"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Mathematics [Manchester]","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5069339254","display_name":"Nicholas J. Higham","orcid":"https://orcid.org/0000-0001-5956-4976"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Nicholas J. Higham","raw_affiliation_strings":["Department of Mathematics [Manchester]"],"raw_orcid":"https://orcid.org/0000-0001-5956-4976","affiliations":[{"raw_affiliation_string":"Department of Mathematics [Manchester]","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5014227566","display_name":"Florent Lopez","orcid":"https://orcid.org/0000-0002-8479-5749"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Florent Lopez","raw_affiliation_strings":["Innovative Computing Laboratory [Knoxville]"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Innovative Computing Laboratory [Knoxville]","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5065253192","display_name":"Th\u00e9o Mary","orcid":"https://orcid.org/0000-0001-9949-4634"},"institutions":[{"id":"https://openalex.org/I1294671590","display_name":"Centre National de la Recherche Scientifique","ror":"https://ror.org/02feahw73","country_code":"FR","type":"government","lineage":["https://openalex.org/I1294671590"]}],"countries":["FR"],"is_corresponding":false,"raw_author_name":"Theo Mary","raw_affiliation_strings":["Performance et Qualit\u00e9 des Algorithmes Num\u00e9riques","Centre National de la Recherche Scientifique"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Performance et Qualit\u00e9 des Algorithmes Num\u00e9riques","institution_ids":[]},{"raw_affiliation_string":"Centre National de la Recherche Scientifique","institution_ids":["https://openalex.org/I1294671590"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5070702896","display_name":"Srikara Pranesh","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Srikara Pranesh","raw_affiliation_strings":["Department of Mathematics [Manchester]"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Mathematics [Manchester]","institution_ids":[]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":5,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":10.1496,"has_fulltext":false,"cited_by_count":64,"citation_normalized_percentile":{"value":0.98873081,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":95,"max":99},"biblio":{"volume":"42","issue":"3","first_page":"C124","last_page":"C141"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.9997000098228455,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.9997000098228455,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10792","display_name":"Matrix Theory and Algorithms","score":0.9890000224113464,"subfield":{"id":"https://openalex.org/subfields/1703","display_name":"Computational Theory and Mathematics"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10363","display_name":"Low-power high-performance VLSI design","score":0.9883000254631042,"subfield":{"id":"https://openalex.org/subfields/2208","display_name":"Electrical and Electronic Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/rounding","display_name":"Rounding","score":0.7709071636199951},{"id":"https://openalex.org/keywords/tensor","display_name":"Tensor (intrinsic definition)","score":0.6433868408203125},{"id":"https://openalex.org/keywords/matrix-multiplication","display_name":"Matrix multiplication","score":0.5738157629966736},{"id":"https://openalex.org/keywords/block","display_name":"Block (permutation group theory)","score":0.525212824344635},{"id":"https://openalex.org/keywords/mathematics","display_name":"Mathematics","score":0.4828127324581146},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.47256606817245483},{"id":"https://openalex.org/keywords/generalized-minimal-residual-method","display_name":"Generalized minimal residual method","score":0.45329564809799194},{"id":"https://openalex.org/keywords/iterative-refinement","display_name":"Iterative refinement","score":0.45210641622543335},{"id":"https://openalex.org/keywords/multiplication","display_name":"Multiplication (music)","score":0.44780078530311584},{"id":"https://openalex.org/keywords/matrix","display_name":"Matrix (chemical analysis)","score":0.4376150667667389},{"id":"https://openalex.org/keywords/algorithm","display_name":"Algorithm","score":0.43638724088668823},{"id":"https://openalex.org/keywords/factorization","display_name":"Factorization","score":0.43339404463768005},{"id":"https://openalex.org/keywords/linear-algebra","display_name":"Linear algebra","score":0.41197559237480164},{"id":"https://openalex.org/keywords/computational-science","display_name":"Computational science","score":0.4058823883533478},{"id":"https://openalex.org/keywords/iterative-method","display_name":"Iterative method","score":0.2965012788772583},{"id":"https://openalex.org/keywords/pure-mathematics","display_name":"Pure mathematics","score":0.08638438582420349},{"id":"https://openalex.org/keywords/geometry","display_name":"Geometry","score":0.0766899585723877}],"concepts":[{"id":"https://openalex.org/C136625980","wikidata":"https://www.wikidata.org/wiki/Q663208","display_name":"Rounding","level":2,"score":0.7709071636199951},{"id":"https://openalex.org/C155281189","wikidata":"https://www.wikidata.org/wiki/Q3518150","display_name":"Tensor (intrinsic definition)","level":2,"score":0.6433868408203125},{"id":"https://openalex.org/C17349429","wikidata":"https://www.wikidata.org/wiki/Q1049914","display_name":"Matrix multiplication","level":3,"score":0.5738157629966736},{"id":"https://openalex.org/C2777210771","wikidata":"https://www.wikidata.org/wiki/Q4927124","display_name":"Block (permutation group theory)","level":2,"score":0.525212824344635},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.4828127324581146},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.47256606817245483},{"id":"https://openalex.org/C155332342","wikidata":"https://www.wikidata.org/wiki/Q1432976","display_name":"Generalized minimal residual method","level":3,"score":0.45329564809799194},{"id":"https://openalex.org/C2779982483","wikidata":"https://www.wikidata.org/wiki/Q6094420","display_name":"Iterative refinement","level":2,"score":0.45210641622543335},{"id":"https://openalex.org/C2780595030","wikidata":"https://www.wikidata.org/wiki/Q3860309","display_name":"Multiplication (music)","level":2,"score":0.44780078530311584},{"id":"https://openalex.org/C106487976","wikidata":"https://www.wikidata.org/wiki/Q685816","display_name":"Matrix (chemical analysis)","level":2,"score":0.4376150667667389},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.43638724088668823},{"id":"https://openalex.org/C187834632","wikidata":"https://www.wikidata.org/wiki/Q188804","display_name":"Factorization","level":2,"score":0.43339404463768005},{"id":"https://openalex.org/C139352143","wikidata":"https://www.wikidata.org/wiki/Q82571","display_name":"Linear algebra","level":2,"score":0.41197559237480164},{"id":"https://openalex.org/C459310","wikidata":"https://www.wikidata.org/wiki/Q117801","display_name":"Computational science","level":1,"score":0.4058823883533478},{"id":"https://openalex.org/C159694833","wikidata":"https://www.wikidata.org/wiki/Q2321565","display_name":"Iterative method","level":2,"score":0.2965012788772583},{"id":"https://openalex.org/C202444582","wikidata":"https://www.wikidata.org/wiki/Q837863","display_name":"Pure mathematics","level":1,"score":0.08638438582420349},{"id":"https://openalex.org/C2524010","wikidata":"https://www.wikidata.org/wiki/Q8087","display_name":"Geometry","level":1,"score":0.0766899585723877},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.0},{"id":"https://openalex.org/C114614502","wikidata":"https://www.wikidata.org/wiki/Q76592","display_name":"Combinatorics","level":1,"score":0.0},{"id":"https://openalex.org/C159985019","wikidata":"https://www.wikidata.org/wiki/Q181790","display_name":"Composite material","level":1,"score":0.0},{"id":"https://openalex.org/C121332964","wikidata":"https://www.wikidata.org/wiki/Q413","display_name":"Physics","level":0,"score":0.0},{"id":"https://openalex.org/C62520636","wikidata":"https://www.wikidata.org/wiki/Q944","display_name":"Quantum mechanics","level":1,"score":0.0},{"id":"https://openalex.org/C84114770","wikidata":"https://www.wikidata.org/wiki/Q46344","display_name":"Quantum","level":2,"score":0.0},{"id":"https://openalex.org/C192562407","wikidata":"https://www.wikidata.org/wiki/Q228736","display_name":"Materials science","level":0,"score":0.0}],"mesh":[],"locations_count":5,"locations":[{"id":"doi:10.1137/19m1289546","is_oa":true,"landing_page_url":"https://doi.org/10.1137/19m1289546","pdf_url":null,"source":{"id":"https://openalex.org/S165512578","display_name":"SIAM Journal on Scientific Computing","issn_l":"1064-8275","issn":["1064-8275","1095-7197"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310320508","host_organization_name":"Society for Industrial and Applied Mathematics","host_organization_lineage":["https://openalex.org/P4310320508"],"host_organization_lineage_names":["Society for Industrial and Applied Mathematics"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"SIAM Journal on Scientific Computing","raw_type":"journal-article"},{"id":"pmh:oai:pure.atira.dk:openaire_cris_publications/04093107-013b-45d7-994c-ed72ca13b725","is_oa":false,"landing_page_url":"https://research.manchester.ac.uk/en/publications/04093107-013b-45d7-994c-ed72ca13b725","pdf_url":null,"source":{"id":"https://openalex.org/S4306400662","display_name":"Research Explorer (The University of Manchester)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I28407311","host_organization_name":"University of Manchester","host_organization_lineage":["https://openalex.org/I28407311"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Blanchard, P, Higham, N, Lopez, F, Mary, T & Pranesh, S 2020, 'Mixed precision block fused multiply-add: error analysis and application to GPU tensor cores', S I A M Journal on Scientific Computing.","raw_type":"info:eu-repo/semantics/publishedVersion"},{"id":"pmh:oai:eprints.maths.manchester.ac.uk.MIMS.EPrints:2733","is_oa":false,"landing_page_url":null,"pdf_url":null,"source":{"id":"https://openalex.org/S4306400452","display_name":"MIMS EPrints (University of Southampton)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I43439940","host_organization_name":"University of Southampton","host_organization_lineage":["https://openalex.org/I43439940"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"acceptedVersion","is_accepted":true,"is_published":false,"raw_source_name":"","raw_type":"MIMS Preprint"},{"id":"pmh:oai:HAL:hal-02491076v1","is_oa":true,"landing_page_url":"https://hal.archives-ouvertes.fr/hal-02491076","pdf_url":null,"source":{"id":"https://openalex.org/S4306402512","display_name":"HAL (Le Centre pour la Communication Scientifique Directe)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I1294671590","host_organization_name":"Centre National de la Recherche Scientifique","host_organization_lineage":["https://openalex.org/I1294671590"],"host_organization_lineage_names":[],"type":"repository"},"license":"other-oa","license_id":"https://openalex.org/licenses/other-oa","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"ISSN: 1064-8275","raw_type":"info:eu-repo/semantics/article"},{"id":"pmh:oai:HAL:hal-02491076v2","is_oa":true,"landing_page_url":"https://hal.science/hal-02491076","pdf_url":null,"source":{"id":"https://openalex.org/S4306402512","display_name":"HAL (Le Centre pour la Communication Scientifique Directe)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I1294671590","host_organization_name":"Centre National de la Recherche Scientifique","host_organization_lineage":["https://openalex.org/I1294671590"],"host_organization_lineage_names":[],"type":"repository"},"license":"other-oa","license_id":"https://openalex.org/licenses/other-oa","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"SIAM Journal on Scientific Computing, 2020, 42 (3), pp.C124-C141. ⟨10.1137/19M1289546⟩","raw_type":"Journal articles"}],"best_oa_location":{"id":"doi:10.1137/19m1289546","is_oa":true,"landing_page_url":"https://doi.org/10.1137/19m1289546","pdf_url":null,"source":{"id":"https://openalex.org/S165512578","display_name":"SIAM Journal on Scientific Computing","issn_l":"1064-8275","issn":["1064-8275","1095-7197"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310320508","host_organization_name":"Society for Industrial and Applied Mathematics","host_organization_lineage":["https://openalex.org/P4310320508"],"host_organization_lineage_names":["Society for Industrial and Applied Mathematics"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"SIAM Journal on Scientific Computing","raw_type":"journal-article"},"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G1322834978","display_name":"Inference, COmputation and Numerics for Insights into Cities (ICONIC)","funder_award_id":"EP/P020720/1","funder_id":"https://openalex.org/F4320334627","funder_display_name":"Engineering and Physical Sciences Research Council"},{"id":"https://openalex.org/G1935751110","display_name":null,"funder_award_id":"EP/P020720/1","funder_id":"https://openalex.org/F4320334627","funder_display_name":"Engineering and Physical Sciences Research Council"}],"funders":[{"id":"https://openalex.org/F4320315144","display_name":"MathWorks","ror":"https://ror.org/01n8qtk87"},{"id":"https://openalex.org/F4320320006","display_name":"Royal Society","ror":"https://ror.org/03wnrjx87"},{"id":"https://openalex.org/F4320334627","display_name":"Engineering and Physical Sciences Research Council","ror":"https://ror.org/0439y7842"}],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":17,"referenced_works":["https://openalex.org/W1540050069","https://openalex.org/W2026891778","https://openalex.org/W2080094880","https://openalex.org/W2170345902","https://openalex.org/W2560012549","https://openalex.org/W2599973081","https://openalex.org/W2752879223","https://openalex.org/W2765439756","https://openalex.org/W2808102735","https://openalex.org/W2895305554","https://openalex.org/W2903589674","https://openalex.org/W2910374893","https://openalex.org/W2945097505","https://openalex.org/W2999942548","https://openalex.org/W3021869609","https://openalex.org/W3022909171","https://openalex.org/W3101543398"],"related_works":["https://openalex.org/W3121314575","https://openalex.org/W3147497457","https://openalex.org/W4221154943","https://openalex.org/W2599973081","https://openalex.org/W4225552076","https://openalex.org/W4206336957","https://openalex.org/W3014877903","https://openalex.org/W4366778549","https://openalex.org/W3004823601","https://openalex.org/W4379985703"],"abstract_inverted_index":{"Computing":[0],"units":[1,17,31,119],"that":[2,52,169,184],"carry":[3],"out":[4],"a":[5,47,69],"fused":[6],"multiply-add":[7],"(FMA)":[8],"operation":[9],"with":[10,101,136,190],"matrix":[11,79,131],"arguments,":[12],"referred":[13],"to":[14,66,97,115],"as":[15],"tensor":[16,61,117,154],"by":[18],"some":[19],"vendors,":[20],"have":[21],"great":[22],"potential":[23],"for":[24,149],"use":[25],"in":[26,72,120,145],"scientific":[27],"computing.":[28],"However,":[29],"these":[30,128],"are":[32],"inherently":[33],"mixed":[34,48],"precision,":[35],"and":[36,59,81,84,124,133,138,181],"existing":[37,60],"rounding":[38,87],"error":[39,88],"analyses":[40,89],"do":[41],"not":[42],"support":[43],"them.":[44],"We":[45,63],"consider":[46],"precision":[49,147],"block":[50,70,102],"FMA":[51,58,71],"generalizes":[53],"both":[54,91],"the":[55,73,116,121,146,150,153,164,167,170,178,185],"usual":[56],"scalar":[57],"units.":[62],"describe":[64],"how":[65],"exploit":[67],"such":[68],"numerical":[74],"linear":[75],"algebra":[76],"kernels":[77],"of":[78,90,141,152,166],"multiplication":[80,132],"LU":[82,134],"factorization":[83,135],"give":[85],"detailed":[86],"kernels.":[92],"An":[93],"important":[94],"application":[95],"is":[96,113,173,188],"GMRES-based":[98],"iterative":[99],"refinement":[100],"FMAs,":[103],"about":[104],"which":[105,143],"our":[106],"analysis":[107,168],"provides":[108],"new":[109],"insight.":[110],"Our":[111,156],"framework":[112],"applicable":[114],"core":[118],"NVIDIA":[122],"Volta":[123],"Turing":[125],"GPUs.":[126],"For":[127],"we":[129],"compare":[130],"TC16":[137,179],"TC32":[139,171],"forms":[140],"FMA,":[142],"differ":[144],"used":[148],"output":[151],"cores.":[155],"experiments":[157],"on":[158],"an":[159],"NVDIA":[160],"V100":[161],"GPU":[162],"confirm":[163],"predictions":[165],"variant":[172],"much":[174],"more":[175],"accurate":[176],"than":[177],"one,":[180],"they":[182],"show":[183],"accuracy":[186],"boost":[187],"obtained":[189],"almost":[191],"no":[192],"performance":[193],"loss.":[194]},"counts_by_year":[{"year":2026,"cited_by_count":1},{"year":2025,"cited_by_count":10},{"year":2024,"cited_by_count":10},{"year":2023,"cited_by_count":9},{"year":2022,"cited_by_count":8},{"year":2021,"cited_by_count":15},{"year":2020,"cited_by_count":11}],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2025-10-10T00:00:00"}