{"id":"https://openalex.org/W7167088804","doi":"https://doi.org/10.1145/3797905.3800548","title":"DynSpAttn: Efficient Attention via Dual-Side Dynamic Sparsity on Sparse Tensor Cores","display_name":"DynSpAttn: Efficient Attention via Dual-Side Dynamic Sparsity on Sparse Tensor Cores","publication_year":2026,"publication_date":"2026-07-02","ids":{"openalex":"https://openalex.org/W7167088804","doi":"https://doi.org/10.1145/3797905.3800548"},"language":null,"primary_location":{"id":"doi:10.1145/3797905.3800548","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3797905.3800548","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 40th ACM International Conference on Supercomputing","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://doi.org/10.1145/3797905.3800548","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5101297034","display_name":"Xiangrui Yu","orcid":"https://orcid.org/0009-0005-2478-1512"},"institutions":[{"id":"https://openalex.org/I889458895","display_name":"University of Hong Kong","ror":"https://ror.org/02zhqgq86","country_code":"HK","type":"education","lineage":["https://openalex.org/I889458895"]},{"id":"https://openalex.org/I90610280","display_name":"South China University of Technology","ror":"https://ror.org/0530pts50","country_code":"CN","type":"education","lineage":["https://openalex.org/I90610280"]}],"countries":["CN","HK"],"is_corresponding":false,"raw_author_name":"Xiangrui Yu","raw_affiliation_strings":["Baidu Inc., Beijing, China and The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China"],"raw_orcid":"https://orcid.org/0009-0005-2478-1512","affiliations":[{"raw_affiliation_string":"Baidu Inc., Beijing, China and The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China","institution_ids":["https://openalex.org/I90610280","https://openalex.org/I889458895"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5054056254","display_name":"Ruibo Fan","orcid":"https://orcid.org/0009-0006-7492-2069"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ruibo Fan","raw_affiliation_strings":["The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China"],"raw_orcid":"https://orcid.org/0009-0006-7492-2069","affiliations":[{"raw_affiliation_string":"The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100455077","display_name":"Zeyu Li","orcid":"https://orcid.org/0009-0008-4381-0544"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zeyu Li","raw_affiliation_strings":["The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China"],"raw_orcid":"https://orcid.org/0009-0008-4381-0544","affiliations":[{"raw_affiliation_string":"The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5104247871","display_name":"Weile Luo","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Weile Luo","raw_affiliation_strings":["The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China"],"raw_orcid":"https://orcid.org/0009-0007-2875-0056","affiliations":[{"raw_affiliation_string":"The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139936034","display_name":"Gu Gong","orcid":"https://orcid.org/0009-0008-2269-2358"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gu Gong","raw_affiliation_strings":["The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China"],"raw_orcid":"https://orcid.org/0009-0008-2269-2358","affiliations":[{"raw_affiliation_string":"The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China","institution_ids":[]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5100730785","display_name":"Xiaowen Chu","orcid":"https://orcid.org/0000-0001-9745-4372"},"institutions":[{"id":"https://openalex.org/I200769079","display_name":"Hong Kong University of Science and Technology","ror":"https://ror.org/00q4vv597","country_code":"HK","type":"education","lineage":["https://openalex.org/I200769079"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Xiaowen Chu","raw_affiliation_strings":["The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China and The Hong Kong University of Science and Technology, Hong Kong, Hong Kong"],"raw_orcid":"https://orcid.org/0000-0001-9745-4372","affiliations":[{"raw_affiliation_string":"The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China and The Hong Kong University of Science and Technology, Hong Kong, Hong Kong","institution_ids":["https://openalex.org/I200769079"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":3,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"1053","last_page":"1064"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.20559999346733093,"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":0.20559999346733093,"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/T12303","display_name":"Tensor decomposition and applications","score":0.1379999965429306,"subfield":{"id":"https://openalex.org/subfields/2605","display_name":"Computational Mathematics"},"field":{"id":"https://openalex.org/fields/26","display_name":"Mathematics"},"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":0.11079999804496765,"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/sparse-matrix","display_name":"Sparse matrix","score":0.644599974155426},{"id":"https://openalex.org/keywords/pipeline","display_name":"Pipeline (software)","score":0.6229000091552734},{"id":"https://openalex.org/keywords/bottleneck","display_name":"Bottleneck","score":0.6022999882698059},{"id":"https://openalex.org/keywords/operand","display_name":"Operand","score":0.5468999743461609},{"id":"https://openalex.org/keywords/tensor","display_name":"Tensor (intrinsic definition)","score":0.5361999869346619},{"id":"https://openalex.org/keywords/block","display_name":"Block (permutation group theory)","score":0.45980000495910645},{"id":"https://openalex.org/keywords/scheme","display_name":"Scheme (mathematics)","score":0.44749999046325684},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.44339999556541443},{"id":"https://openalex.org/keywords/executable","display_name":"Executable","score":0.4408999979496002}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7226999998092651},{"id":"https://openalex.org/C56372850","wikidata":"https://www.wikidata.org/wiki/Q1050404","display_name":"Sparse matrix","level":3,"score":0.644599974155426},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.6229000091552734},{"id":"https://openalex.org/C2780513914","wikidata":"https://www.wikidata.org/wiki/Q18210350","display_name":"Bottleneck","level":2,"score":0.6022999882698059},{"id":"https://openalex.org/C55526617","wikidata":"https://www.wikidata.org/wiki/Q719375","display_name":"Operand","level":2,"score":0.5468999743461609},{"id":"https://openalex.org/C155281189","wikidata":"https://www.wikidata.org/wiki/Q3518150","display_name":"Tensor (intrinsic definition)","level":2,"score":0.5361999869346619},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.47279998660087585},{"id":"https://openalex.org/C2777210771","wikidata":"https://www.wikidata.org/wiki/Q4927124","display_name":"Block (permutation group theory)","level":2,"score":0.45980000495910645},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.44749999046325684},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.44339999556541443},{"id":"https://openalex.org/C160145156","wikidata":"https://www.wikidata.org/wiki/Q778586","display_name":"Executable","level":2,"score":0.4408999979496002},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.4133000075817108},{"id":"https://openalex.org/C2778119891","wikidata":"https://www.wikidata.org/wiki/Q477690","display_name":"CUDA","level":2,"score":0.4088999927043915},{"id":"https://openalex.org/C106487976","wikidata":"https://www.wikidata.org/wiki/Q685816","display_name":"Matrix (chemical analysis)","level":2,"score":0.40860000252723694},{"id":"https://openalex.org/C124066611","wikidata":"https://www.wikidata.org/wiki/Q28684319","display_name":"Sparse approximation","level":2,"score":0.39820000529289246},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.37299999594688416},{"id":"https://openalex.org/C179799912","wikidata":"https://www.wikidata.org/wiki/Q205084","display_name":"Computational complexity theory","level":2,"score":0.3346000015735626},{"id":"https://openalex.org/C54219892","wikidata":"https://www.wikidata.org/wiki/Q4080929","display_name":"Matrix-free methods","level":4,"score":0.3343999981880188},{"id":"https://openalex.org/C459310","wikidata":"https://www.wikidata.org/wiki/Q117801","display_name":"Computational science","level":1,"score":0.32670000195503235},{"id":"https://openalex.org/C2777472644","wikidata":"https://www.wikidata.org/wiki/Q16968992","display_name":"Approximate inference","level":3,"score":0.311599999666214},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.28999999165534973},{"id":"https://openalex.org/C99138194","wikidata":"https://www.wikidata.org/wiki/Q183427","display_name":"Hash function","level":2,"score":0.2865000069141388},{"id":"https://openalex.org/C78766204","wikidata":"https://www.wikidata.org/wiki/Q555032","display_name":"Multi-core processor","level":2,"score":0.28139999508857727},{"id":"https://openalex.org/C17349429","wikidata":"https://www.wikidata.org/wiki/Q1049914","display_name":"Matrix multiplication","level":3,"score":0.28110000491142273},{"id":"https://openalex.org/C2778770139","wikidata":"https://www.wikidata.org/wiki/Q1966904","display_name":"Solver","level":2,"score":0.28060001134872437},{"id":"https://openalex.org/C77637269","wikidata":"https://www.wikidata.org/wiki/Q7002051","display_name":"Neural coding","level":2,"score":0.27300000190734863},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.2669000029563904},{"id":"https://openalex.org/C2778915421","wikidata":"https://www.wikidata.org/wiki/Q3643177","display_name":"Performance improvement","level":2,"score":0.26600000262260437},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.2606000006198883},{"id":"https://openalex.org/C89611455","wikidata":"https://www.wikidata.org/wiki/Q6804646","display_name":"Mechanism (biology)","level":2,"score":0.2581999897956848},{"id":"https://openalex.org/C2777904410","wikidata":"https://www.wikidata.org/wiki/Q7397","display_name":"Software","level":2,"score":0.2565999925136566}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3797905.3800548","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3797905.3800548","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 40th ACM International Conference on Supercomputing","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.1145/3797905.3800548","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3797905.3800548","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 40th ACM International Conference on Supercomputing","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":26,"referenced_works":["https://openalex.org/W2736953746","https://openalex.org/W2808128431","https://openalex.org/W2914631005","https://openalex.org/W2946609015","https://openalex.org/W2964337156","https://openalex.org/W2998617917","https://openalex.org/W3132695675","https://openalex.org/W3208099998","https://openalex.org/W4293024985","https://openalex.org/W4309672181","https://openalex.org/W4320063610","https://openalex.org/W4321466385","https://openalex.org/W4387321091","https://openalex.org/W4388667495","https://openalex.org/W4389576338","https://openalex.org/W4391986945","https://openalex.org/W4395106472","https://openalex.org/W4408843954","https://openalex.org/W4408891420","https://openalex.org/W4408894144","https://openalex.org/W4409248726","https://openalex.org/W4409362760","https://openalex.org/W4415797154","https://openalex.org/W4415797534","https://openalex.org/W4415799245","https://openalex.org/W7133216168"],"related_works":[],"abstract_inverted_index":{"The":[0],"high":[1],"computational":[2],"complexity":[3],"of":[4,39],"the":[5,37,76,83,156],"self-attention":[6],"mechanism":[7,58],"constitutes":[8],"a":[9,54,67,107,113,161],"primary":[10],"performance":[11],"bottleneck":[12],"in":[13,87],"LLM":[14,167],"inference.":[15,168],"Existing":[16],"sparse":[17,56,90],"attention":[18,57,88,158],"mechanisms":[19],"commonly":[20],"adopt":[21],"coarse-grained":[22],"block":[23],"sparsity":[24,48,71,151],"to":[25,131],"align":[26],"with":[27,60,152],"FlashAttention\u2019s":[28],"tiling":[29],"and":[30,46,78,106,123,139,163],"rely":[31],"on":[32,95,120],"dense":[33],"Tensor":[34,43,63],"Cores,":[35],"leaving":[36],"potential":[38],"emerging":[40],"hardware":[41,153],"Sparse":[42,62],"Cores":[44],"(SpTCs)":[45],"semi-structured":[47,150],"largely":[49],"untapped.":[50],"We":[51],"present":[52],"DynSpAttn,":[53],"dynamic":[55],"co-designed":[59],"NVIDIA":[61],"Cores.":[64],"DynSpAttn":[65,101,128],"introduces":[66],"dual-side":[68],"2:4":[69],"structured":[70],"strategy":[72],"that":[73,127,148],"prunes":[74],"both":[75],"Query":[77],"Score":[79],"matrices,":[80],"thereby":[81],"transforming":[82],"dominant":[84],"matrix":[85,91],"multiplications":[86,92],"into":[89],"(SpMMs)":[93],"executable":[94],"SpTCs.":[96],"To":[97],"realize":[98],"this":[99],"transformation,":[100],"incorporates":[102],"lightweight":[103],"in-register":[104],"pruners":[105],"shuffle-free":[108],"operand":[109],"remapping":[110],"scheme":[111],"within":[112],"fully":[114],"fused,":[115],"I/O-aware":[116],"CUDA":[117],"kernel.":[118],"Evaluations":[119],"RTX":[121],"4090":[122],"L20":[124],"GPUs":[125],"show":[126],"achieves":[129],"up":[130],"1.70":[132],"\u00d7":[133,141],"kernel-level":[134],"preformance":[135],"improvement":[136],"over":[137],"FlashAttention":[138],"1.58":[140],"end-to-end":[142],"inference":[143],"speedup.":[144],"These":[145],"results":[146],"demonstrate":[147],"co-designing":[149],"support":[154],"across":[155],"full":[157],"pipeline":[159],"provides":[160],"practical":[162],"efficient":[164],"solution":[165],"for":[166]},"counts_by_year":[],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2026-07-03T00:00:00"}