{"id":"https://openalex.org/W7124146610","doi":"https://doi.org/10.1109/icpads67057.2025.11322894","title":"Breaking the Fusion Barrier: An Online Algorithm for Fused Normalization and Linear Layers","display_name":"Breaking the Fusion Barrier: An Online Algorithm for Fused Normalization and Linear Layers","publication_year":2025,"publication_date":"2025-12-14","ids":{"openalex":"https://openalex.org/W7124146610","doi":"https://doi.org/10.1109/icpads67057.2025.11322894"},"language":null,"primary_location":{"id":"doi:10.1109/icpads67057.2025.11322894","is_oa":false,"landing_page_url":"https://doi.org/10.1109/icpads67057.2025.11322894","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 IEEE 31th International Conference on Parallel and Distributed Systems (ICPADS)","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":false,"oa_status":"closed","oa_url":null,"any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Hanghang Cao","orcid":null},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"government","lineage":["https://openalex.org/I19820366"]},{"id":"https://openalex.org/I4210128818","display_name":"Institute of Software","ror":"https://ror.org/033dfsn42","country_code":"CN","type":"facility","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210128818"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Hanghang Cao","raw_affiliation_strings":["Institute of Software Chinese Academy of Sciences,Beijing,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Institute of Software Chinese Academy of Sciences,Beijing,China","institution_ids":["https://openalex.org/I4210128818","https://openalex.org/I19820366"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5032839961","display_name":"Shihao Gao","orcid":"https://orcid.org/0000-0002-5804-8758"},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"government","lineage":["https://openalex.org/I19820366"]},{"id":"https://openalex.org/I4210128818","display_name":"Institute of Software","ror":"https://ror.org/033dfsn42","country_code":"CN","type":"facility","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210128818"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Shihao Gao","raw_affiliation_strings":["Institute of Software Chinese Academy of Sciences,Beijing,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Institute of Software Chinese Academy of Sciences,Beijing,China","institution_ids":["https://openalex.org/I4210128818","https://openalex.org/I19820366"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123011915","display_name":"Quanyi Li","orcid":null},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"government","lineage":["https://openalex.org/I19820366"]},{"id":"https://openalex.org/I4210128818","display_name":"Institute of Software","ror":"https://ror.org/033dfsn42","country_code":"CN","type":"facility","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210128818"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Quanyi Li","raw_affiliation_strings":["Institute of Software Chinese Academy of Sciences,Beijing,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Institute of Software Chinese Academy of Sciences,Beijing,China","institution_ids":["https://openalex.org/I4210128818","https://openalex.org/I19820366"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123036980","display_name":"Hongbin Zhang","orcid":null},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"government","lineage":["https://openalex.org/I19820366"]},{"id":"https://openalex.org/I4210128818","display_name":"Institute of Software","ror":"https://ror.org/033dfsn42","country_code":"CN","type":"facility","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210128818"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Hongbin Zhang","raw_affiliation_strings":["Institute of Software Chinese Academy of Sciences,Beijing,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Institute of Software Chinese Academy of Sciences,Beijing,China","institution_ids":["https://openalex.org/I4210128818","https://openalex.org/I19820366"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123045427","display_name":"Mingjie Xing","orcid":null},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"government","lineage":["https://openalex.org/I19820366"]},{"id":"https://openalex.org/I4210128818","display_name":"Institute of Software","ror":"https://ror.org/033dfsn42","country_code":"CN","type":"facility","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210128818"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Mingjie Xing","raw_affiliation_strings":["Institute of Software Chinese Academy of Sciences,Beijing,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Institute of Software Chinese Academy of Sciences,Beijing,China","institution_ids":["https://openalex.org/I4210128818","https://openalex.org/I19820366"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5121131266","display_name":"Yanjun Wu","orcid":null},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"government","lineage":["https://openalex.org/I19820366"]},{"id":"https://openalex.org/I4210128818","display_name":"Institute of Software","ror":"https://ror.org/033dfsn42","country_code":"CN","type":"facility","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210128818"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yanjun Wu","raw_affiliation_strings":["Institute of Software Chinese Academy of Sciences,Beijing,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Institute of Software Chinese Academy of Sciences,Beijing,China","institution_ids":["https://openalex.org/I4210128818","https://openalex.org/I19820366"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5123042589","display_name":"Chen Zhao","orcid":null},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"government","lineage":["https://openalex.org/I19820366"]},{"id":"https://openalex.org/I4210128818","display_name":"Institute of Software","ror":"https://ror.org/033dfsn42","country_code":"CN","type":"facility","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210128818"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Chen Zhao","raw_affiliation_strings":["Institute of Software Chinese Academy of Sciences,Beijing,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Institute of Software Chinese Academy of Sciences,Beijing,China","institution_ids":["https://openalex.org/I4210128818","https://openalex.org/I19820366"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":7,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.60686748,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"1","last_page":"8"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.3361000120639801,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/T10036","display_name":"Advanced Neural Network Applications","score":0.3361000120639801,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.19789999723434448,"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/T11948","display_name":"Machine Learning in Materials Science","score":0.0658000037074089,"subfield":{"id":"https://openalex.org/subfields/2505","display_name":"Materials Chemistry"},"field":{"id":"https://openalex.org/fields/25","display_name":"Materials Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/normalization","display_name":"Normalization (sociology)","score":0.8112000226974487},{"id":"https://openalex.org/keywords/computation","display_name":"Computation","score":0.5135999917984009},{"id":"https://openalex.org/keywords/compiler","display_name":"Compiler","score":0.47999998927116394},{"id":"https://openalex.org/keywords/database-normalization","display_name":"Database normalization","score":0.4659000039100647},{"id":"https://openalex.org/keywords/decoding-methods","display_name":"Decoding methods","score":0.3910999894142151},{"id":"https://openalex.org/keywords/kernel","display_name":"Kernel (algebra)","score":0.37700000405311584},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.3702999949455261},{"id":"https://openalex.org/keywords/overhead","display_name":"Overhead (engineering)","score":0.3659000098705292},{"id":"https://openalex.org/keywords/data-set","display_name":"Data set","score":0.32739999890327454}],"concepts":[{"id":"https://openalex.org/C136886441","wikidata":"https://www.wikidata.org/wiki/Q926129","display_name":"Normalization (sociology)","level":2,"score":0.8112000226974487},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7597000002861023},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.6488000154495239},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.5135999917984009},{"id":"https://openalex.org/C169590947","wikidata":"https://www.wikidata.org/wiki/Q47506","display_name":"Compiler","level":2,"score":0.47999998927116394},{"id":"https://openalex.org/C162984825","wikidata":"https://www.wikidata.org/wiki/Q339072","display_name":"Database normalization","level":3,"score":0.4659000039100647},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.3910999894142151},{"id":"https://openalex.org/C74193536","wikidata":"https://www.wikidata.org/wiki/Q574844","display_name":"Kernel (algebra)","level":2,"score":0.37700000405311584},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.3702999949455261},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.3659000098705292},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.35030001401901245},{"id":"https://openalex.org/C58489278","wikidata":"https://www.wikidata.org/wiki/Q1172284","display_name":"Data set","level":2,"score":0.32739999890327454},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.31029999256134033},{"id":"https://openalex.org/C122280245","wikidata":"https://www.wikidata.org/wiki/Q620622","display_name":"Kernel method","level":3,"score":0.3061000108718872},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.30329999327659607},{"id":"https://openalex.org/C78766204","wikidata":"https://www.wikidata.org/wiki/Q555032","display_name":"Multi-core processor","level":2,"score":0.29510000348091125},{"id":"https://openalex.org/C162319229","wikidata":"https://www.wikidata.org/wiki/Q175263","display_name":"Data structure","level":2,"score":0.2833000123500824},{"id":"https://openalex.org/C139352143","wikidata":"https://www.wikidata.org/wiki/Q82571","display_name":"Linear algebra","level":2,"score":0.2816999852657318},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.28110000491142273},{"id":"https://openalex.org/C70437156","wikidata":"https://www.wikidata.org/wiki/Q7228652","display_name":"Pooling","level":2,"score":0.27649998664855957},{"id":"https://openalex.org/C158525013","wikidata":"https://www.wikidata.org/wiki/Q2593739","display_name":"Fusion","level":2,"score":0.2655999958515167},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.2639000117778778},{"id":"https://openalex.org/C33954974","wikidata":"https://www.wikidata.org/wiki/Q486494","display_name":"Sensor fusion","level":2,"score":0.2621000111103058},{"id":"https://openalex.org/C141353440","wikidata":"https://www.wikidata.org/wiki/Q182221","display_name":"Fuse (electrical)","level":2,"score":0.26010000705718994},{"id":"https://openalex.org/C21080849","wikidata":"https://www.wikidata.org/wiki/Q13611879","display_name":"Data point","level":2,"score":0.25940001010894775},{"id":"https://openalex.org/C75684735","wikidata":"https://www.wikidata.org/wiki/Q858810","display_name":"Big data","level":2,"score":0.2590999901294708},{"id":"https://openalex.org/C2989134064","wikidata":"https://www.wikidata.org/wiki/Q288510","display_name":"Execution time","level":2,"score":0.25870001316070557},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.25679999589920044}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/icpads67057.2025.11322894","is_oa":false,"landing_page_url":"https://doi.org/10.1109/icpads67057.2025.11322894","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 IEEE 31th International Conference on Parallel and Distributed Systems (ICPADS)","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":3,"referenced_works":["https://openalex.org/W1987034518","https://openalex.org/W4394998532","https://openalex.org/W4415797154"],"related_works":[],"abstract_inverted_index":{"The":[0],"performance":[1],"of":[2,71,129,174,193],"Large":[3],"Language":[4],"Model":[5],"(LLM)":[6],"inference":[7],"is":[8,121],"critically":[9],"hindered":[10],"by":[11,94],"memory-bound":[12],"operations,":[13],"among":[14],"which":[15],"normalization":[16,33,135,215],"layers":[17],"are":[18],"a":[19,36,39,64,97,115,127,153,167],"primary":[20],"bottleneck,":[21],"especially":[22],"during":[23],"the":[24,47,69,103,124,134,142,147,172,178,201,214],"latency-sensitive":[25],"decoding":[26,210],"phase.":[27],"While":[28],"deep":[29],"learning":[30],"compilers":[31,185],"fuse":[32],"operations":[34],"into":[35,126],"single":[37],"kernel,":[38],"fundamental":[40],"fusion":[41],"barrier":[42,93],"prevents":[43],"them":[44],"from":[45],"merging":[46],"ubiquitous":[48],"Normalization":[49],"and":[50,80,177,189,211],"subsequent":[51],"Linear":[52],"layer":[53],"(<tex":[54],"xmlns:mml=\"http://www.w3.org/1998/Math/MathML\"":[55,105,197,203],"xmlns:xlink=\"http://www.w3.org/1999/xlink\">$N":[56,106,204],"\\&amp;":[57,107,205],"L$</tex>)":[58],"pattern.":[59],"This":[60],"barrier,":[61],"rooted":[62],"in":[63,208],"core":[65,148],"data":[66,82,149],"dependency,":[67],"forces":[68],"execution":[70],"two":[72],"separate":[73],"kernels,":[74],"incurring":[75],"prohibitive":[76],"kernel":[77,157],"launch":[78],"overhead":[79],"costly":[81],"round-trips":[83],"to":[84,110,114,122,137,163,170,195],"global":[85],"memory.":[86],"In":[87],"this":[88,92,161],"paper,":[89],"we":[90],"break":[91],"introducing":[95],"FlashFusion,":[96],"novel":[98],"online":[99],"algorithm":[100,162],"that":[101,158],"reformulates":[102],"<tex":[104,196,202],"L$</tex>":[108,206],"pattern":[109,207],"be":[111,138],"mathematically":[112],"equivalent":[113],"single-pass":[116],"computation.":[117],"Our":[118],"key":[119],"insight":[120],"decompose":[123],"computation":[125],"set":[128],"independent":[130],"parallel":[131],"sums,":[132],"allowing":[133],"statistics":[136],"calculated":[139],"concurrently":[140],"with":[141],"matrix":[143],"multiplication,":[144],"thus":[145],"eliminating":[146,213],"dependency.":[150],"We":[151],"co-design":[152],"high-performance,":[154],"hardware-aware":[155],"GPU":[156],"efficiently":[159],"maps":[160],"modern":[164],"architectures,":[165],"leveraging":[166],"tiling":[168],"strategy":[169],"maximize":[171],"utilization":[173],"Tensor":[175],"Cores":[176],"memory":[179],"hierarchy.":[180],"FlashFusion":[181],"significantly":[182],"outperforms":[183],"state-of-the-art":[184],"like":[186],"PyTorch":[187],"Inductor":[188],"TensorRT,":[190],"achieving":[191],"speedups":[192],"up":[194],"xmlns:xlink=\"http://www.w3.org/1999/xlink\">$3.03":[198],"\\times$</tex>":[199],"for":[200],"LLM":[209],"effectively":[212],"bottleneck.":[216]},"counts_by_year":[],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2026-01-15T00:00:00"}