{"id":"https://openalex.org/W4407857898","doi":"https://doi.org/10.1145/3696443.3708953","title":"Accelerating LLMs using an Efficient GEMM Library and Target-Aware Optimizations on Real-World PIM Devices","display_name":"Accelerating LLMs using an Efficient GEMM Library and Target-Aware Optimizations on Real-World PIM Devices","publication_year":2025,"publication_date":"2025-02-22","ids":{"openalex":"https://openalex.org/W4407857898","doi":"https://doi.org/10.1145/3696443.3708953"},"language":"en","primary_location":{"id":"doi:10.1145/3696443.3708953","is_oa":false,"landing_page_url":"https://doi.org/10.1145/3696443.3708953","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 23rd ACM/IEEE International Symposium on Code Generation and Optimization","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":null,"display_name":"Hyeoncheol Kim","orcid":"https://orcid.org/0009-0009-6819-2591"},"institutions":[{"id":"https://openalex.org/I193775966","display_name":"Yonsei University","ror":"https://ror.org/01wjejq96","country_code":"KR","type":"education","lineage":["https://openalex.org/I193775966"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Hyeoncheol Kim","raw_affiliation_strings":["Yonsei University, Seoul, Republic of Korea"],"raw_orcid":"https://orcid.org/0009-0009-6819-2591","affiliations":[{"raw_affiliation_string":"Yonsei University, Seoul, Republic of Korea","institution_ids":["https://openalex.org/I193775966"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Taehoon Kim","orcid":"https://orcid.org/0009-0002-2380-8160"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Taehoon Kim","raw_affiliation_strings":["Rebellions, Seongnam, Republic of Korea"],"raw_orcid":"https://orcid.org/0009-0002-2380-8160","affiliations":[{"raw_affiliation_string":"Rebellions, Seongnam, Republic of Korea","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5068076002","display_name":"Taehyeong Park","orcid":"https://orcid.org/0009-0004-2703-5024"},"institutions":[{"id":"https://openalex.org/I193775966","display_name":"Yonsei University","ror":"https://ror.org/01wjejq96","country_code":"KR","type":"education","lineage":["https://openalex.org/I193775966"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Taehyeong Park","raw_affiliation_strings":["Yonsei University, Seoul, Republic of Korea"],"raw_orcid":"https://orcid.org/0009-0004-2703-5024","affiliations":[{"raw_affiliation_string":"Yonsei University, Seoul, Republic of Korea","institution_ids":["https://openalex.org/I193775966"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100334088","display_name":"Donghyeon Kim","orcid":"https://orcid.org/0000-0001-5019-8199"},"institutions":[{"id":"https://openalex.org/I4575257","display_name":"Hanyang University","ror":"https://ror.org/046865y68","country_code":"KR","type":"education","lineage":["https://openalex.org/I4575257"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Donghyeon Kim","raw_affiliation_strings":["Hanyang University, Seoul, Republic of Korea"],"raw_orcid":"https://orcid.org/0000-0001-5019-8199","affiliations":[{"raw_affiliation_string":"Hanyang University, Seoul, Republic of Korea","institution_ids":["https://openalex.org/I4575257"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5085503041","display_name":"Yongseung Yu","orcid":"https://orcid.org/0000-0001-9838-4310"},"institutions":[{"id":"https://openalex.org/I193775966","display_name":"Yonsei University","ror":"https://ror.org/01wjejq96","country_code":"KR","type":"education","lineage":["https://openalex.org/I193775966"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Yongseung Yu","raw_affiliation_strings":["Yonsei University, Seoul, Republic of Korea"],"raw_orcid":"https://orcid.org/0000-0001-9838-4310","affiliations":[{"raw_affiliation_string":"Yonsei University, Seoul, Republic of Korea","institution_ids":["https://openalex.org/I193775966"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100722345","display_name":"Hanjun Kim","orcid":"https://orcid.org/0000-0002-0762-7901"},"institutions":[{"id":"https://openalex.org/I193775966","display_name":"Yonsei University","ror":"https://ror.org/01wjejq96","country_code":"KR","type":"education","lineage":["https://openalex.org/I193775966"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Hanjun Kim","raw_affiliation_strings":["Yonsei University, Seoul, Republic of Korea"],"raw_orcid":"https://orcid.org/0000-0002-0762-7901","affiliations":[{"raw_affiliation_string":"Yonsei University, Seoul, Republic of Korea","institution_ids":["https://openalex.org/I193775966"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5086980677","display_name":"Yongjun Park","orcid":"https://orcid.org/0000-0003-3725-0380"},"institutions":[{"id":"https://openalex.org/I193775966","display_name":"Yonsei University","ror":"https://ror.org/01wjejq96","country_code":"KR","type":"education","lineage":["https://openalex.org/I193775966"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Yongjun Park","raw_affiliation_strings":["Yonsei University, Seoul, Republic of Korea"],"raw_orcid":"https://orcid.org/0000-0003-3725-0380","affiliations":[{"raw_affiliation_string":"Yonsei University, Seoul, Republic of Korea","institution_ids":["https://openalex.org/I193775966"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":2,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"225","last_page":"240"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T12808","display_name":"Ferroelectric and Negative Capacitance Devices","score":0.9848999977111816,"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"}},"topics":[{"id":"https://openalex.org/T12808","display_name":"Ferroelectric and Negative Capacitance Devices","score":0.9848999977111816,"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"}},{"id":"https://openalex.org/T12784","display_name":"Modular Robots and Swarm Intelligence","score":0.9308000206947327,"subfield":{"id":"https://openalex.org/subfields/2210","display_name":"Mechanical Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11181","display_name":"Advanced Data Storage Technologies","score":0.9222999811172485,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.703603982925415},{"id":"https://openalex.org/keywords/computer-architecture","display_name":"Computer architecture","score":0.4777999520301819},{"id":"https://openalex.org/keywords/parallel-computing","display_name":"Parallel computing","score":0.3765946924686432}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.703603982925415},{"id":"https://openalex.org/C118524514","wikidata":"https://www.wikidata.org/wiki/Q173212","display_name":"Computer architecture","level":1,"score":0.4777999520301819},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.3765946924686432}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3696443.3708953","is_oa":false,"landing_page_url":"https://doi.org/10.1145/3696443.3708953","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 23rd ACM/IEEE International Symposium on Code Generation and Optimization","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":45,"referenced_works":["https://openalex.org/W1988888548","https://openalex.org/W2014277916","https://openalex.org/W2030407693","https://openalex.org/W2149381887","https://openalex.org/W2613989746","https://openalex.org/W2802367674","https://openalex.org/W2805566098","https://openalex.org/W2920081941","https://openalex.org/W2965373594","https://openalex.org/W2971801673","https://openalex.org/W2980282514","https://openalex.org/W2981852735","https://openalex.org/W2982008795","https://openalex.org/W2985229340","https://openalex.org/W2990138404","https://openalex.org/W2996428491","https://openalex.org/W3033187248","https://openalex.org/W3098220359","https://openalex.org/W3100710793","https://openalex.org/W3134274954","https://openalex.org/W3158634533","https://openalex.org/W3189166979","https://openalex.org/W3206328251","https://openalex.org/W3210580311","https://openalex.org/W4211105506","https://openalex.org/W4214734582","https://openalex.org/W4221001402","https://openalex.org/W4226126604","https://openalex.org/W4246166885","https://openalex.org/W4249322926","https://openalex.org/W4280496502","https://openalex.org/W4282551050","https://openalex.org/W4285121610","https://openalex.org/W4288089799","https://openalex.org/W4291002173","https://openalex.org/W4295036115","https://openalex.org/W4302296459","https://openalex.org/W4304140739","https://openalex.org/W4321636575","https://openalex.org/W4322718191","https://openalex.org/W4385438314","https://openalex.org/W4387490777","https://openalex.org/W4392427708","https://openalex.org/W4393949386","https://openalex.org/W4400985603"],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W2390279801","https://openalex.org/W4391913857","https://openalex.org/W2358668433","https://openalex.org/W4396701345","https://openalex.org/W2376932109","https://openalex.org/W2001405890","https://openalex.org/W4396696052"],"abstract_inverted_index":{"Real-time":[0],"processing":[1],"of":[2,32,171,226],"deep":[3,189],"learning":[4,190],"models":[5],"on":[6,102,128,251],"conventional":[7],"systems,":[8],"such":[9],"as":[10],"CPUs":[11],"and":[12,81,94,123,153,162,168,214],"GPUs,":[13],"is":[14,22,84,96],"highly":[15],"challenging":[16],"due":[17],"to":[18,147,228],"memory":[19,83,205],"bottlenecks.":[20],"This":[21],"exacerbated":[23],"in":[24],"Large":[25],"Language":[26],"Models":[27],"(LLMs),":[28],"where":[29],"the":[30,79,88,92,103,187,196,231],"majority":[31],"executions":[33],"are":[34,43],"dominated":[35],"by":[36,199],"General":[37],"Matrix":[38],"Multiplication":[39],"(GEMM)":[40],"operations,":[41],"which":[42,52,136],"relatively":[44],"more":[45],"memory-intensive":[46],"than":[47],"convolution":[48],"operations.":[49],"Processing-in-Memory":[50],"(PIM),":[51],"provides":[53,137,244],"high":[54],"internal":[55],"bandwidth,":[56],"can":[57],"be":[58],"a":[59],"promising":[60],"alternative":[61],"for":[62,98,141,164,234,247],"LLM":[63,197,249],"serving.":[64],"However,":[65],"since":[66],"current":[67],"PIM":[68,95,130,172,208,253],"systems":[69,173],"do":[70],"not":[71],"fully":[72],"replace":[73],"traditional":[74],"memory,":[75],"data":[76,150],"transfer":[77,89,151],"between":[78,91],"host":[80,93],"PIM-side":[82],"essential.":[85],"Therefore,":[86],"minimizing":[87],"cost":[90],"crucial":[97],"serving":[99,250],"LLMs":[100,116,180],"efficiently":[101],"PIM.":[104],"In":[105,156],"this":[106,242],"paper,":[107],"we":[108,183],"propose":[109,134],"PIM-LLM,":[110],"an":[111,118,175],"end-to-end":[112],"framework":[113],"that":[114,220,241],"accelerates":[115],"using":[117,174,181],"efficient":[119,248],"tiled":[120],"GEMM":[121,166],"library":[122],"several":[124,201,235],"key":[125,202,245],"target-aware":[126],"optimizations":[127],"real-world":[129],"systems.":[131],"We":[132,193,238],"first":[133],"PGEMMlib,":[135,182],"optimized":[138,160],"tiling":[139],"techniques":[140,163],"PIM,":[142],"considering":[143],"architecture":[144],"specific":[145],"characteristics":[146],"minimize":[148],"unnecessary":[149],"overhead":[152],"maximize":[154],"parallelism.":[155],"addition,":[157],"Tile-Selector":[158],"explores":[159],"parameters":[161],"different":[165],"shapes":[167],"available":[169],"resources":[170],"analytical":[176],"model.":[177],"To":[178],"accelerate":[179],"integrate":[184],"it":[185],"into":[186],"TVM":[188,232],"compiler":[191],"framework.":[192],"further":[194],"optimize":[195],"execution":[198],"applying":[200],"optimizations:":[203],"Build-time":[204],"layout":[206],"adjustment,":[207],"resource":[209],"pooling,":[210],"CPU/PIM":[211],"cooperation":[212],"support,":[213],"QKV":[215],"generation":[216],"fusion.":[217],"Evaluation":[218],"shows":[219],"PIM-LLM":[221],"achieves":[222],"significant":[223],"performance":[224],"gains":[225],"up":[227],"45.75x":[229],"over":[230],"baseline":[233],"well-known":[236],"LLMs.":[237],"strongly":[239],"believe":[240],"work":[243],"insights":[246],"real":[252],"devices.":[254]},"counts_by_year":[{"year":2026,"cited_by_count":1},{"year":2025,"cited_by_count":1}],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2025-10-10T00:00:00"}