{"id":"https://openalex.org/W4414198815","doi":"https://doi.org/10.1109/dac63849.2025.11132101","title":"PacQ: A SIMT Microarchitecture for Efficient Dataflow in Hyper-asymmetric GEMMs","display_name":"PacQ: A SIMT Microarchitecture for Efficient Dataflow in Hyper-asymmetric GEMMs","publication_year":2025,"publication_date":"2025-06-22","ids":{"openalex":"https://openalex.org/W4414198815","doi":"https://doi.org/10.1109/dac63849.2025.11132101"},"language":"en","primary_location":{"id":"doi:10.1109/dac63849.2025.11132101","is_oa":false,"landing_page_url":"https://doi.org/10.1109/dac63849.2025.11132101","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 62nd ACM/IEEE Design Automation Conference (DAC)","raw_type":"proceedings-article"},"type":"conference-paper","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":"https://openalex.org/A5057610010","display_name":"Ruokai Yin","orcid":"https://orcid.org/0000-0002-7550-0638"},"institutions":[{"id":"https://openalex.org/I32971472","display_name":"Yale University","ror":"https://ror.org/03v76x132","country_code":"US","type":"education","lineage":["https://openalex.org/I32971472"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Ruokai Yin","raw_affiliation_strings":["Yale University,New Haven,USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Yale University,New Haven,USA","institution_ids":["https://openalex.org/I32971472"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100443277","display_name":"Yuhang Li","orcid":"https://orcid.org/0000-0002-9364-4125"},"institutions":[{"id":"https://openalex.org/I32971472","display_name":"Yale University","ror":"https://ror.org/03v76x132","country_code":"US","type":"education","lineage":["https://openalex.org/I32971472"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yuhang Li","raw_affiliation_strings":["Yale University,New Haven,USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Yale University,New Haven,USA","institution_ids":["https://openalex.org/I32971472"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5050310538","display_name":"Priyadarshini Panda","orcid":"https://orcid.org/0000-0002-4167-6782"},"institutions":[{"id":"https://openalex.org/I32971472","display_name":"Yale University","ror":"https://ror.org/03v76x132","country_code":"US","type":"education","lineage":["https://openalex.org/I32971472"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Priyadarshini Panda","raw_affiliation_strings":["Yale University,New Haven,USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Yale University,New Haven,USA","institution_ids":["https://openalex.org/I32971472"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I32971472"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.24065978,"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":"7"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.9929999709129333,"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":0.9929999709129333,"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/T11181","display_name":"Advanced Data Storage Technologies","score":0.991599977016449,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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/T10481","display_name":"Computer Graphics and Visualization Techniques","score":0.9598000049591064,"subfield":{"id":"https://openalex.org/subfields/1704","display_name":"Computer Graphics and Computer-Aided Design"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/dataflow","display_name":"Dataflow","score":0.9309999942779541},{"id":"https://openalex.org/keywords/memory-footprint","display_name":"Memory footprint","score":0.6065999865531921},{"id":"https://openalex.org/keywords/microarchitecture","display_name":"Microarchitecture","score":0.5285999774932861},{"id":"https://openalex.org/keywords/computation","display_name":"Computation","score":0.5019999742507935},{"id":"https://openalex.org/keywords/speedup","display_name":"Speedup","score":0.46700000762939453},{"id":"https://openalex.org/keywords/multiplier","display_name":"Multiplier (economics)","score":0.4106999933719635},{"id":"https://openalex.org/keywords/quantization","display_name":"Quantization (signal processing)","score":0.38260000944137573},{"id":"https://openalex.org/keywords/unpacking","display_name":"Unpacking","score":0.37279999256134033},{"id":"https://openalex.org/keywords/operand","display_name":"Operand","score":0.36410000920295715},{"id":"https://openalex.org/keywords/dataflow-architecture","display_name":"Dataflow architecture","score":0.3628999888896942}],"concepts":[{"id":"https://openalex.org/C96324660","wikidata":"https://www.wikidata.org/wiki/Q205446","display_name":"Dataflow","level":2,"score":0.9309999942779541},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8274000287055969},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.7408000230789185},{"id":"https://openalex.org/C74912251","wikidata":"https://www.wikidata.org/wiki/Q6815727","display_name":"Memory footprint","level":2,"score":0.6065999865531921},{"id":"https://openalex.org/C107598950","wikidata":"https://www.wikidata.org/wiki/Q259864","display_name":"Microarchitecture","level":2,"score":0.5285999774932861},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.5019999742507935},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.46700000762939453},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.42260000109672546},{"id":"https://openalex.org/C124584101","wikidata":"https://www.wikidata.org/wiki/Q1053266","display_name":"Multiplier (economics)","level":2,"score":0.4106999933719635},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.38260000944137573},{"id":"https://openalex.org/C94375191","wikidata":"https://www.wikidata.org/wiki/Q11205","display_name":"Arithmetic","level":1,"score":0.38260000944137573},{"id":"https://openalex.org/C2777256151","wikidata":"https://www.wikidata.org/wiki/Q7897273","display_name":"Unpacking","level":2,"score":0.37279999256134033},{"id":"https://openalex.org/C55526617","wikidata":"https://www.wikidata.org/wiki/Q719375","display_name":"Operand","level":2,"score":0.36410000920295715},{"id":"https://openalex.org/C176727019","wikidata":"https://www.wikidata.org/wiki/Q1172415","display_name":"Dataflow architecture","level":3,"score":0.3628999888896942},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3544999957084656},{"id":"https://openalex.org/C2776834041","wikidata":"https://www.wikidata.org/wiki/Q25346349","display_name":"Execution model","level":2,"score":0.3431999981403351},{"id":"https://openalex.org/C111335779","wikidata":"https://www.wikidata.org/wiki/Q3454686","display_name":"Reduction (mathematics)","level":2,"score":0.32199999690055847},{"id":"https://openalex.org/C206729178","wikidata":"https://www.wikidata.org/wiki/Q2271896","display_name":"Scheduling (production processes)","level":2,"score":0.3203999996185303},{"id":"https://openalex.org/C2777575374","wikidata":"https://www.wikidata.org/wiki/Q1644704","display_name":"MicroBlaze","level":3,"score":0.30559998750686646},{"id":"https://openalex.org/C162319229","wikidata":"https://www.wikidata.org/wiki/Q175263","display_name":"Data structure","level":2,"score":0.303600013256073},{"id":"https://openalex.org/C105339364","wikidata":"https://www.wikidata.org/wiki/Q2297740","display_name":"Software deployment","level":2,"score":0.3010999858379364},{"id":"https://openalex.org/C56086750","wikidata":"https://www.wikidata.org/wiki/Q6042592","display_name":"Integer programming","level":2,"score":0.3005000054836273},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.2994999885559082},{"id":"https://openalex.org/C200833197","wikidata":"https://www.wikidata.org/wiki/Q333707","display_name":"Compile time","level":3,"score":0.29420000314712524},{"id":"https://openalex.org/C134750763","wikidata":"https://www.wikidata.org/wiki/Q625277","display_name":"Unit load","level":2,"score":0.28949999809265137},{"id":"https://openalex.org/C97137487","wikidata":"https://www.wikidata.org/wiki/Q729138","display_name":"Integer (computer science)","level":2,"score":0.2849999964237213},{"id":"https://openalex.org/C82687282","wikidata":"https://www.wikidata.org/wiki/Q66221","display_name":"Auxiliary memory","level":2,"score":0.28299999237060547},{"id":"https://openalex.org/C169590947","wikidata":"https://www.wikidata.org/wiki/Q47506","display_name":"Compiler","level":2,"score":0.28299999237060547},{"id":"https://openalex.org/C75684735","wikidata":"https://www.wikidata.org/wiki/Q858810","display_name":"Big data","level":2,"score":0.27390000224113464},{"id":"https://openalex.org/C176649486","wikidata":"https://www.wikidata.org/wiki/Q2308807","display_name":"Memory management","level":3,"score":0.271699994802475},{"id":"https://openalex.org/C164620267","wikidata":"https://www.wikidata.org/wiki/Q376953","display_name":"Adder","level":3,"score":0.2709999978542328},{"id":"https://openalex.org/C35555965","wikidata":"https://www.wikidata.org/wiki/Q189057","display_name":"Merge sort","level":4,"score":0.2685999870300293},{"id":"https://openalex.org/C197129107","wikidata":"https://www.wikidata.org/wiki/Q1921621","display_name":"Merge (version control)","level":2,"score":0.26409998536109924}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/dac63849.2025.11132101","is_oa":false,"landing_page_url":"https://doi.org/10.1109/dac63849.2025.11132101","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 62nd ACM/IEEE Design Automation Conference (DAC)","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320306076","display_name":"National Science Foundation","ror":"https://ror.org/021nxhr62"}],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":5,"referenced_works":["https://openalex.org/W2963122961","https://openalex.org/W2963989532","https://openalex.org/W4360831846","https://openalex.org/W4385573671","https://openalex.org/W4393407021"],"related_works":[],"abstract_inverted_index":{"Weight-only":[0],"quantization":[1],"has":[2],"been":[3],"widely":[4],"explored":[5],"in":[6,28,36,175],"large":[7],"language":[8],"models":[9],"(LLMs)":[10],"to":[11,41,63,66,70,86,127,155,167,178],"reduce":[12],"memory":[13,46],"storage":[14],"and":[15,48,75,95,111,130,146,171],"data":[16,49],"loading":[17,50],"overhead.":[18],"During":[19],"deployment":[20],"on":[21,184],"single-instruction-multiple-threads":[22],"(SIMT)":[23],"architectures,":[24],"weights":[25,68,94],"are":[26,55],"stored":[27],"low-precision":[29,92],"integer":[30],"(INT)":[31],"format,":[32],"while":[33],"activations":[34],"remain":[35,60],"full-precision":[37],"floating-point":[38],"(FP)":[39],"format":[40,72],"preserve":[42],"inference":[43],"accuracy.":[44],"Although":[45],"footprint":[47],"requirements":[51],"for":[52,114],"weight":[53,116],"matrices":[54],"reduced,":[56],"computation":[57],"performance":[58],"gains":[59],"limited":[61],"due":[62],"the":[64,102,143],"need":[65],"convert":[67],"back":[69],"FP":[71,97],"through":[73],"unpacking":[74],"dequantization":[76],"before":[77],"GEMM":[78,88,104],"operations.":[79],"In":[80],"this":[81,100],"work,":[82],"we":[83,141],"investigate":[84],"methods":[85],"accelerate":[87,157],"operations":[89],"involving":[90],"packed":[91],"INT":[93,115,138],"high-precision":[96],"activations,":[98],"defining":[99],"as":[101],"hyper-asymmetric":[103,158],"problem.":[105],"Our":[106],"approach":[107],"co-optimizes":[108],"tile-level":[109],"packing":[110,129],"dataflow":[112,131],"strategies":[113],"matrices.":[117],"We":[118,160],"further":[119],"design":[120],"a":[121,151],"specialized":[122],"FP-INT":[123],"multiplier":[124,147],"unit":[125,148],"tailored":[126],"our":[128],"strategies,":[132],"enabling":[133],"parallel":[134],"processing":[135],"of":[136],"multiple":[137],"weights.":[139],"Finally,":[140],"integrate":[142],"packing,":[144],"dataflow,":[145],"into":[149],"PacQ,":[150],"SIMT":[152,186],"microarchitecture":[153],"designed":[154],"efficiently":[156],"GEMMs.":[159],"show":[161],"that":[162],"PacQ":[163],"can":[164],"achieve":[165],"up":[166],"$1.99":[168],"\\times$":[169],"speedup":[170],"$81.4":[172],"\\%$":[173],"reduction":[174],"EDP":[176],"compared":[177],"weight-only":[179],"quantized":[180],"LLM":[181],"workloads":[182],"running":[183],"conventional":[185],"baselines.":[187]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2025-10-10T00:00:00"}