Article(id=1327012611629101484, tenantId=1146029695717560320, journalId=1324759914146529478, issueId=1327012311115608215, articleNumber=null, orderNo=null, doi=10.7544/issn1000-1239.202330864, pmid=null, cstr=null, oa=null, hot=null, price=null, onlineType=0, articleFormat=0, articleType=null, articleTypeStr=research-article, receivedDate=1698681600000, receivedDateStr=2023-10-31, revisedDate=1708963200000, revisedDateStr=2024-02-27, acceptedDate=null, acceptedDateStr=null, onlineDate=1794314061314, onlineDateStr=2026-11-10, pubDate=1738339200000, pubDateStr=2025-02-01, doiRegisterDate=null, doiRegisterDateStr=null, onlineIssueDate=1794314061314, onlineIssueDateStr=2026-11-10, onlineJustAcceptDate=null, onlineJustAcceptDateStr=null, onlineFirstDate=null, onlineFirstDateStr=null, sourceXml=null, magXml=null, createTime=1794314061314, creator=13701087609, updateTime=1794314061314, updator=13701087609, issue=Issue{id=1327012311115608215, tenantId=1146029695717560320, journalId=1324759914146529478, year='2025', volume='62', issue='2', pageStart='285', pageEnd='541', issueExtLink='null', onlineDate='null', pubDate='1738339200000', pubDateStr='2025-02-01', beforeIssueId=null, nextIssueId=null, price=null, status=1, issueComplete=1, articleOrder=1, issueType=-1, specialIssue=null, createTime=1794313989665, creator='13701087609', updateTime=1794376696965, updator='13701087609', preIssue=null, nextIssue=null, articleTotal=null, ext={EN=IssueExt(id=1327275325060395748, tenantId=1146029695717560320, journalId=1324759914146529478, issueId=1327012311115608215, language=EN, specialIssueTitle=, coverIllustrator=null, specialIssueEditor=, specialIssueAbout=), CN=IssueExt(id=1327275325060395749, tenantId=1146029695717560320, journalId=1324759914146529478, issueId=1327012311115608215, language=CN, specialIssueTitle=, coverIllustrator=null, specialIssueEditor=, specialIssueAbout=)}, issueFiles=null, downloadFileDto=null}, startPage=517, endPage=531, ext={EN=ArticleExt(id=1327012613113885102, articleId=1327012611629101484, tenantId=1146029695717560320, journalId=1324759914146529478, language=EN, title=Small-Scale Irregular TRSM Implementation and Optimization, columnId=1325493593498059289, journalTitle=Journal of Computer Research and Development, columnName=High Performance Computing, runingTitle=null, highlight=null, articleAbstract=

TRSM (triangular matrix equation solver) is a commonly used algorithm for solving systems of linear equations, and is the core algorithm of various scientific computing libraries and mathematical software, which is widely used in the fields of scientific computing, engineering computing and machine learning. The small-scale irregular TRSM algorithm limits the scope of problem-solving and is an algorithm for efficiently handling smaller-scale, irregular data inputs. With the development of personalization and refinement in the field of high-performance computing, the demand for small-scale irregular TRSM computation in the scientific and industrial communities is becoming more and more obvious. While traditional algorithms are better suited for large-scale and regular TRSM computation, there is still room for improvement in the computational efficiency of small-scale and irregular TRSM. In this paper, we propose a small-scale irregular TRSM optimization scheme by combining hardware architecture and application scenario characteristics, designing a high-performance kernel from the perspectives of register chunking, boundary processing, and vectorization computation, and constructing an algorithmic library of small-scale irregular SI_TRSM (small-scale irregular TRSM) covering double-precision real numbers and double-precision complex numbers based on which the performance of this algorithm is greatly improved. Based on experimental results, the double-precision small-scale irregular TRSM algorithm library developed in this paper has shown to enhance the average performance of double-precision small-scale irregular real numbers by 29.4 times, and double-precision small-scale irregular complex numbers by 24.6 times in comparison with similar algorithms available in the MKL (Intel math kernel library).

, authors=Rongyuan Guo1, 2, Haipeng Jia2, *, Yunquan Zhang2, Cunyang Wei2, Mingsen Deng1, Jingrui Chen3, Zhenya Zhou3, authorsList=Rongyuan Guo, Haipeng Jia, Yunquan Zhang, Cunyang Wei, Mingsen Deng, Jingrui Chen, Zhenya Zhou, authorCompany=null, correspAuthors=Haipeng Jia, authorNote=

born in 1999. Master candidate. His main research interests include parallel computing and high performance computing

born in 1983. PhD, senior engineer. His main research interests include parallel computing and heterogeneous computing

born in 1973. PhD, professor. His main research interests include parallel algorithm and parallel software

born in 1999. Master. His main research interests include parallel computing and high performance computing

born in 1981. PhD, professor. His main research interests include distributed parallel computing and feature extraction of big data

born in 1987. Master, R&D manager. Her main research interests include circuit simulation and linear equation solvers

born in 1980. Master, senior engineer. His main research interest includes algorithms and software for transistor-level circuit simulation

, correspAuthorsNote=null, copyrightStatement=Copyright © 2025 Journal of Computer Research and Development. All rights reserved., copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=null, magXml=null, pdfUrl=null, pdf=null, pdfFileSize=null, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=null, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=null, mapNumber=null, fund=null), CN=ArticleExt(id=1327012617928946134, articleId=1327012611629101484, tenantId=1146029695717560320, journalId=1324759914146529478, language=CN, title=小规模非规则TRSM实现与优化, columnId=1325493595259666971, journalTitle=计算机研究与发展, columnName=高性能计算, runingTitle=null, highlight=null, articleAbstract=

TRSM(triangular matrix equation solver)是线性方程组求解的常用算法,是各种科学计算库和数学软件的核心算法,广泛应用于科学计算、工程计算、机器学习等领域. 小规模非规则TRSM算法限定解决问题范围,是高效处理较小规模、非规则数据输入的算法. 随着高性能计算领域个性化、精细化发展,科学界、工业界对小规模非规则TRSM计算的需求愈加明显. 传统算法更偏重于大规模、规则TRSM的计算,在小规模非规则TRSM计算上效率不佳. 结合硬件体系结构、应用场景特征提出小规模非规则TRSM优化方案,从寄存器分块、边界处理、向量化计算角度设计高性能内核,在此基础上构建覆盖双精度实数、双精度复数的小规模非规则算法库SI_TRSM (small-scale irregular TRSM),大幅度提升该算法性能. 实验结果表明,构建的双精度小规模非规则TRSM算法库,较MKL(Intel math kernel library)同类算法,在双精度小规模非规则实数上平均性能提高29.4倍,在双精度小规模非规则复数上平均性能提高24.6倍.

, authors=郭容园1, 2, 贾海鹏2, *, 张云泉2, 韦存阳2, 邓明森1, 陈婧蕊3, 周振亚3, authorsList=郭容园, 贾海鹏, 张云泉, 韦存阳, 邓明森, 陈婧蕊, 周振亚, authorCompany=null, correspAuthors=贾海鹏, authorNote=

1999年生. 硕士研究生. 主要研究方向为并行计算、高性能计算

1983年生. 博士,高级工程师. 主要研究方向为并行计算、异构计算

1973年生. 博士,研究员. 主要研究方向为并行算法、并行软件

1999年生. 硕士. 主要研究方向为并行计算、高性能计算

1981年生. 博士,教授. 主要研究方向为分布式并行计算、大数据特征提取

1987年生. 硕士,研发经理. 主要研究方向为电路仿真、线性方程求解器

1980年生. 硕士,高级工程师. 主要研究方向为晶体管级电路仿真的算法和软件

, correspAuthorsNote=
贾海鹏()
, copyrightStatement=版权所有 © 2025 计算机研究与发展编辑部, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=nOGfrNUOFSfPl1BoHZlOrQ==, magXml=S6ZNIKkltnKBDrglzrKNxw==, pdfUrl=null, pdf=Puwez2VmzLCPbhJQ1oRvsw==, pdfFileSize=2544958, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=ehGPWnu+E9I2N4IzdNnn/Q==, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=ldHw8C6Joz6vZtqdaWvSTw==, mapNumber=null, fund=null)}, authors=[Author(id=1327012618470011362, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=guoryuangradu@163.com, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1327012618587451877, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012618470011362, language=EN, stringName=Rongyuan Guo, firstName=Rongyuan, middleName=null, lastName=Guo, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, address=1School of Information, Guizhou University of Finance and Economics, Guiyang 550025
2Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190, bio={"content":"

born in 1999. Master candidate. His main research interests include parallel computing and high performance computing

"}, bioImg=null, bioContent=

born in 1999. Master candidate. His main research interests include parallel computing and high performance computing

, aboutCorrespAuthor=null), CN=AuthorExt(id=1327012618671337959, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012618470011362, language=CN, stringName=郭容园, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, address=1贵州财经大学信息学院 贵阳 550025
2中国科学院计算技术研究所 北京 100190, bio={"img":"iWqAG/ctkkR/2rAJ/wmKTQ==","content":"

1999年生. 硕士研究生. 主要研究方向为并行计算、高性能计算

"}, bioImg=iWqAG/ctkkR/2rAJ/wmKTQ==, bioContent=

1999年生. 硕士研究生. 主要研究方向为并行计算、高性能计算

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1327012618147049943, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, xref=1, ext=[AuthorCompanyExt(id=1327012618155438552, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618147049943, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1School of Information, Guizhou University of Finance and Economics, Guiyang 550025), AuthorCompanyExt(id=1327012618168021465, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618147049943, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1贵州财经大学信息学院 贵阳 550025)]), AuthorCompany(id=1327012618235130330, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, xref=2, ext=[AuthorCompanyExt(id=1327012618243518939, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618235130330, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190), AuthorCompanyExt(id=1327012618247713244, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618235130330, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2中国科学院计算技术研究所 北京 100190)])]), Author(id=1327012618767806953, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=jiahaipeng@ict.ac.cn, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1327012619027853803, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012618767806953, language=EN, stringName=Haipeng Jia, firstName=Haipeng, middleName=null, lastName=Jia, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, *, address=2Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190, bio={"content":"

born in 1983. PhD, senior engineer. His main research interests include parallel computing and heterogeneous computing

"}, bioImg=null, bioContent=

born in 1983. PhD, senior engineer. His main research interests include parallel computing and heterogeneous computing

, aboutCorrespAuthor=null), CN=AuthorExt(id=1327012619136905708, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012618767806953, language=CN, stringName=贾海鹏, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, *, address=2中国科学院计算技术研究所 北京 100190, bio={"img":"rIPqZ1zmlW7/6TUyOpUtMg==","content":"

1983年生. 博士,高级工程师. 主要研究方向为并行计算、异构计算

"}, bioImg=rIPqZ1zmlW7/6TUyOpUtMg==, bioContent=

1983年生. 博士,高级工程师. 主要研究方向为并行计算、异构计算

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1327012618235130330, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, xref=2, ext=[AuthorCompanyExt(id=1327012618243518939, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618235130330, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190), AuthorCompanyExt(id=1327012618247713244, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618235130330, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2中国科学院计算技术研究所 北京 100190)])]), Author(id=1327012619258540526, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, orderNo=2, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1327012619329843696, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012619258540526, language=EN, stringName=Yunquan Zhang, firstName=Yunquan, middleName=null, lastName=Zhang, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190, bio={"content":"

born in 1973. PhD, professor. His main research interests include parallel algorithm and parallel software

"}, bioImg=null, bioContent=

born in 1973. PhD, professor. His main research interests include parallel algorithm and parallel software

, aboutCorrespAuthor=null), CN=AuthorExt(id=1327012619409535473, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012619258540526, language=CN, stringName=张云泉, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2中国科学院计算技术研究所 北京 100190, bio={"img":"RM5ub5FSk9Vs0a7bIbvcFA==","content":"

1973年生. 博士,研究员. 主要研究方向为并行算法、并行软件

"}, bioImg=RM5ub5FSk9Vs0a7bIbvcFA==, bioContent=

1973年生. 博士,研究员. 主要研究方向为并行算法、并行软件

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1327012618235130330, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, xref=2, ext=[AuthorCompanyExt(id=1327012618243518939, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618235130330, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190), AuthorCompanyExt(id=1327012618247713244, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618235130330, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2中国科学院计算技术研究所 北京 100190)])]), Author(id=1327012619480838643, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, orderNo=3, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1327012619560530421, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012619480838643, language=EN, stringName=Cunyang Wei, firstName=Cunyang, middleName=null, lastName=Wei, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190, bio={"content":"

born in 1999. Master. His main research interests include parallel computing and high performance computing

"}, bioImg=null, bioContent=

born in 1999. Master. His main research interests include parallel computing and high performance computing

, aboutCorrespAuthor=null), CN=AuthorExt(id=1327012619623444982, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012619480838643, language=CN, stringName=韦存阳, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2中国科学院计算技术研究所 北京 100190, bio={"img":"b7v0Y/ziU31+DuIZ3dLVog==","content":"

1999年生. 硕士. 主要研究方向为并行计算、高性能计算

"}, bioImg=b7v0Y/ziU31+DuIZ3dLVog==, bioContent=

1999年生. 硕士. 主要研究方向为并行计算、高性能计算

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1327012618235130330, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, xref=2, ext=[AuthorCompanyExt(id=1327012618243518939, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618235130330, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190), AuthorCompanyExt(id=1327012618247713244, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618235130330, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2中国科学院计算技术研究所 北京 100190)])]), Author(id=1327012619698942456, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, orderNo=4, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1327012619766051322, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012619698942456, language=EN, stringName=Mingsen Deng, firstName=Mingsen, middleName=null, lastName=Deng, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1School of Information, Guizhou University of Finance and Economics, Guiyang 550025, bio={"content":"

born in 1981. PhD, professor. His main research interests include distributed parallel computing and feature extraction of big data

"}, bioImg=null, bioContent=

born in 1981. PhD, professor. His main research interests include distributed parallel computing and feature extraction of big data

, aboutCorrespAuthor=null), CN=AuthorExt(id=1327012619841548795, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012619698942456, language=CN, stringName=邓明森, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1贵州财经大学信息学院 贵阳 550025, bio={"img":"eCSaQkOG4UP4sc3szoNOJQ==","content":"

1981年生. 博士,教授. 主要研究方向为分布式并行计算、大数据特征提取

"}, bioImg=eCSaQkOG4UP4sc3szoNOJQ==, bioContent=

1981年生. 博士,教授. 主要研究方向为分布式并行计算、大数据特征提取

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1327012618147049943, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, xref=1, ext=[AuthorCompanyExt(id=1327012618155438552, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618147049943, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1School of Information, Guizhou University of Finance and Economics, Guiyang 550025), AuthorCompanyExt(id=1327012618168021465, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618147049943, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1贵州财经大学信息学院 贵阳 550025)])]), Author(id=1327012619925434877, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, orderNo=5, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1327012620009320959, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012619925434877, language=EN, stringName=Jingrui Chen, firstName=Jingrui, middleName=null, lastName=Chen, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3Empyrean Technology Co., Ltd., Beijing 100102, bio={"content":"

born in 1987. Master, R&D manager. Her main research interests include circuit simulation and linear equation solvers

"}, bioImg=null, bioContent=

born in 1987. Master, R&D manager. Her main research interests include circuit simulation and linear equation solvers

, aboutCorrespAuthor=null), CN=AuthorExt(id=1327012620084818432, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012619925434877, language=CN, stringName=陈婧蕊, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3北京华大九天科技股份有限公司 北京 100102, bio={"img":"2GedsPjM2M+ZnLG7jl6pqA==","content":"

1987年生. 硕士,研发经理. 主要研究方向为电路仿真、线性方程求解器

"}, bioImg=2GedsPjM2M+ZnLG7jl6pqA==, bioContent=

1987年生. 硕士,研发经理. 主要研究方向为电路仿真、线性方程求解器

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1327012618319016413, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, xref=3, ext=[AuthorCompanyExt(id=1327012618331599326, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618319016413, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3Empyrean Technology Co., Ltd., Beijing 100102), AuthorCompanyExt(id=1327012618356765151, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618319016413, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3北京华大九天科技股份有限公司 北京 100102)])]), Author(id=1327012620219036162, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, orderNo=6, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1327012620286145028, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012620219036162, language=EN, stringName=Zhenya Zhou, firstName=Zhenya, middleName=null, lastName=Zhou, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3Empyrean Technology Co., Ltd., Beijing 100102, bio={"content":"

born in 1980. Master, senior engineer. His main research interest includes algorithms and software for transistor-level circuit simulation

"}, bioImg=null, bioContent=

born in 1980. Master, senior engineer. His main research interest includes algorithms and software for transistor-level circuit simulation

, aboutCorrespAuthor=null), CN=AuthorExt(id=1327012621951283717, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012620219036162, language=CN, stringName=周振亚, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3北京华大九天科技股份有限公司 北京 100102, bio={"img":"+uf3tSWTa8p1WUJxm5pIew==","content":"

1980年生. 硕士,高级工程师. 主要研究方向为晶体管级电路仿真的算法和软件

"}, bioImg=+uf3tSWTa8p1WUJxm5pIew==, bioContent=

1980年生. 硕士,高级工程师. 主要研究方向为晶体管级电路仿真的算法和软件

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1327012618319016413, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, xref=3, ext=[AuthorCompanyExt(id=1327012618331599326, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618319016413, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3Empyrean Technology Co., Ltd., Beijing 100102), AuthorCompanyExt(id=1327012618356765151, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618319016413, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3北京华大九天科技股份有限公司 北京 100102)])])], keywords=[Keyword(id=1327012622064529926, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=EN, orderNo=1, keyword=TRSM algorithm), Keyword(id=1327012622131638791, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=EN, orderNo=2, keyword=BLAS), Keyword(id=1327012622194553352, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=EN, orderNo=3, keyword=small-scale irregular), Keyword(id=1327012622257467913, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=EN, orderNo=4, keyword=SIMD), Keyword(id=1327012622324576778, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=EN, orderNo=5, keyword=assembly optimization), Keyword(id=1327012622387491339, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=CN, orderNo=1, keyword=TRSM算法), Keyword(id=1327012622450405900, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=CN, orderNo=2, keyword=BLAS), Keyword(id=1327012622504931853, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=CN, orderNo=3, keyword=小规模非规则), Keyword(id=1327012622588817934, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=CN, orderNo=4, keyword=SIMD), Keyword(id=1327012622664315407, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=CN, orderNo=5, keyword=汇编优化)], refs=[Reference(id=1327012627148026416, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=1, rfOrder=0, authorNames=null, journalName=null, refType=null, unstructuredReference=BLAS Forum. BLAS: Basic linear algebra subprograms[EB/OL]. [2023-05-28]. http://netlib.org/blas/, articleTitle=null, refAbstract=null), Reference(id=1327012627210940977, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=2, rfOrder=1, authorNames=null, journalName=null, refType=null, unstructuredReference=Choi J, Dongarra J J, Walker D W. PB-BLAS: A set of parallel block basic linear algebra subprograms[J]. Concurrency: Practice and Experience, 1996, 8(7): 517−535, articleTitle=null, refAbstract=null), Reference(id=1327012627269661234, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=3, rfOrder=2, authorNames=null, journalName=null, refType=null, unstructuredReference=Intel. Intel oneAPI math kernel library[EB/OL]. [2023-05-21]. http://www.intel.com/content/www/us/en/developer/tools/oneapi/onemkl.html, articleTitle=null, refAbstract=null), Reference(id=1327012627349353011, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=4, rfOrder=3, authorNames=null, journalName=null, refType=null, unstructuredReference=Zhang Xianyi. OpenBLAS: An optimized BLAS library[EB/OL]. [2023-06-13]. http://www.openblas.net/, articleTitle=null, refAbstract=null), Reference(id=1327012627420656180, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=5, rfOrder=4, authorNames=null, journalName=null, refType=null, unstructuredReference=AMD. AMDLIB (AMD math library): A high-performance mathematical function library[EB/OL]. [2023-06-29]. https://www.amd.com/en/developer/aocl/libm.html, articleTitle=null, refAbstract=null), Reference(id=1327012627487765045, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=6, rfOrder=5, authorNames=null, journalName=null, refType=null, unstructuredReference=McNamee J M. Algorithm 408: A sparse matrix package (part I)[J]. Communications of the ACM, 1971, 14(4): 265−273, articleTitle=null, refAbstract=null), Reference(id=1327012627567456822, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=7, rfOrder=6, authorNames=null, journalName=null, refType=null, unstructuredReference=Dongarra J J, Du C J, Hammarling S, et al. Algorithm 679: A set of level 3 basic linear algebra subprograms: Model implementation and test programs[J]. ACM Transactions on Mathematical Software, 1990, 16(1): 18−28, articleTitle=null, refAbstract=null), Reference(id=1327012627638759991, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=8, rfOrder=7, authorNames=null, journalName=null, refType=null, unstructuredReference=Hanson R. PC basic linear algebra subroutines[EB/OL]. [2023-06-15]. https://www.osti.gov/biblio/1230151, articleTitle=null, refAbstract=null), Reference(id=1327012627701674552, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=9, rfOrder=8, authorNames=null, journalName=null, refType=null, unstructuredReference=Choi J, Walker D W, Dongarra J J. PUMMA: Parallel universal matrix multiplication algorithms on distributed memory concurrent computers.[J]. Concurrency: Practice and Experience, 1994, 6(7): 543−570, articleTitle=null, refAbstract=null), Reference(id=1327012627768783417, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=10, rfOrder=9, authorNames=null, journalName=null, refType=null, unstructuredReference=Dongarra J J. Preface: Basic linear algebra subprograms technical (blast) forum standard[J]. The International Journal of High Performance Computing Applications, 2002, 16(1): 1−1, articleTitle=null, refAbstract=null), Reference(id=1327012627835892282, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=11, rfOrder=10, authorNames=null, journalName=null, refType=null, unstructuredReference=Chan E, Van Zee F G, Bientinesi P, et al. Supermatrix: A multithreaded runtime scheduling system for algorithms-by-blocks[C]//Proc of the 13th ACM SIGPLAN Symp on Principles and Practice of Parallel Programming. New York: ACM, 2008: 123−132, articleTitle=null, refAbstract=null), Reference(id=1327012627915584059, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=12, rfOrder=11, authorNames=null, journalName=null, refType=null, unstructuredReference=Volkov V, Demmel J W. Benchmarking GPUs to tune dense linear algebra[C/OL]//Proc of the 2008 ACM/IEEE Conf on Supercomputing(SC’08). Piscataway, NJ: IEEE, 2008[2023-06-20]. https://ieeexplore.ieee.org/document/5214359, articleTitle=null, refAbstract=null), Reference(id=1327012628016247356, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=13, rfOrder=12, authorNames=null, journalName=null, refType=null, unstructuredReference=Soorishetty A. Accelerating Linear Algebra and Machine Learning Kernels on a Massively Parallel Reconfigurable Architecture[M]. Arizona: Arizona State University, 2019, articleTitle=null, refAbstract=null), Reference(id=1327012628184019517, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=14, rfOrder=13, authorNames=null, journalName=null, refType=null, unstructuredReference=Wei Cunyang, Jia Haipeng, Zhang Yunquan, et al. IATF: An input-aware tuning framework for compact BLAS based on ARMv8 CPUs[C/OL]//Proc of the 51st Int Conf on Parallel Processing. New York: ACM, 2022[2023-06-25].https://dl.acm.org/doi/10.1145/3545008.3545032, articleTitle=null, refAbstract=null), Reference(id=1327012628330820158, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=15, rfOrder=14, authorNames=null, journalName=null, refType=null, unstructuredReference=Valero-Lara P, Greenwalt C, Vetter J S. SparseLU, a novel algorithm and math library for sparse LU factorization[C]//Proc of the 2022 IEEE/ACM Workshop on Irregular Applications: Architectures and Algorithms (IA3). Piscataway, NJ: IEEE, 2022: 25−31, articleTitle=null, refAbstract=null), Reference(id=1327012628397929023, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=16, rfOrder=15, authorNames=null, journalName=null, refType=null, unstructuredReference=胡怡, 陈道琨, 杨超, 等. 国产SW26010-Pro处理器上3级BLAS函数众核并行优化[J]. 软件学报,2023,35(3):1569−1584, articleTitle=null, refAbstract=null), Reference(id=1327012628465037888, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=16, rfOrder=16, authorNames=null, journalName=null, refType=null, unstructuredReference=Hu Yi, Chen Daokun, Yang Chao, et al. Many-core parallel optimization of level–3 BLAS function on domestic SW26010-Pro processor[J]. Journal of Software, 2023, 35(3): 1569−1584 (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1327012628532146753, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=17, rfOrder=17, authorNames=null, journalName=null, refType=null, unstructuredReference=Whaley R C, Petitet A, Dongarra J J. Automated empirical optimizations of software and the ATLAS project[J]. Parallel Computing, 2001, 27(1/2): 3−25, articleTitle=null, refAbstract=null), Reference(id=1327012628590867010, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=18, rfOrder=18, authorNames=null, journalName=null, refType=null, unstructuredReference=Lu H, Matz M, Girkar M, et al. System V application binary interface AMD64 architecture processor supplement (with LP64 and ILP32 programming models) version 1.0[EB/OL]. [2023-06-28]. https://cs61.seas.harvard.edu/site/pdf/X86-64-abi-20210928.pdf, articleTitle=null, refAbstract=null), Reference(id=1327012628662170179, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=19, rfOrder=19, authorNames=null, journalName=null, refType=null, unstructuredReference=Amiri H, Shahbahrami A. SIMD programming using Intel vector extensions[J]. Journal of Parallel and Distributed Computing, 2020, 135: 83−100, articleTitle=null, refAbstract=null), Reference(id=1327012628733473348, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=20, rfOrder=20, authorNames=null, journalName=null, refType=null, unstructuredReference=Kusswurm D. Modern X86 Assembly Language Programming: Covers X86 64-bit, AVX, AVX2, and AVX−512[M]. Berkeley, CA: Apress, 2018: 421−432, articleTitle=null, refAbstract=null), Reference(id=1327012628859302469, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=21, rfOrder=21, authorNames=null, journalName=null, refType=null, unstructuredReference=Li Chendi, Jia Haipeng, Cao hang, et al. Autotsmm: An auto-tuning framework for building high-performance tall-and-skinny matrix-matrix multiplication on CPUs[C]//Proc of the 2021 IEEE Intel Conf on Parallel & Distributed Processing with Applications, Big Data & Cloud Computing, Sustainable Computing & Communications, Social Computing & Networking (ISPA/BDCloud/SocialCom/SustainCom). Piscataway, NJ: IEEE, 2021: 159−166, articleTitle=null, refAbstract=null), Reference(id=1327012628951577158, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=22, rfOrder=22, authorNames=null, journalName=null, refType=null, unstructuredReference=Yang Weiling, Fang Jianbin, Dong Dezun, et al. LIBSHALOM: Optimizing small and irregular-shaped matrix multiplications on ARMv8 multi-cores[C/OL]//Proc of the Int Conf for High Performance Computing, Networking, Storage and Analysis. New York: ACM, 2021[2023-06-26].https://dl.acm.org/doi/10.1145/3458817.3476217, articleTitle=null, refAbstract=null), Reference(id=1327012629018686023, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=23, rfOrder=23, authorNames=null, journalName=null, refType=null, unstructuredReference=Sorokin A, Malkovsky S, Tsoy G. Comparing the performance of general matrix multiplication routine on heterogeneous computing systems[J]. Journal of Parallel and Distributed Computing, 2022, 160: 39−48, articleTitle=null, refAbstract=null)], funds=null, companyList=[AuthorCompany(id=1327012618147049943, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, xref=1, ext=[AuthorCompanyExt(id=1327012618155438552, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618147049943, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1School of Information, Guizhou University of Finance and Economics, Guiyang 550025), AuthorCompanyExt(id=1327012618168021465, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618147049943, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1贵州财经大学信息学院 贵阳 550025)]), AuthorCompany(id=1327012618235130330, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, xref=2, ext=[AuthorCompanyExt(id=1327012618243518939, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618235130330, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190), AuthorCompanyExt(id=1327012618247713244, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618235130330, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2中国科学院计算技术研究所 北京 100190)]), AuthorCompany(id=1327012618319016413, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, xref=3, ext=[AuthorCompanyExt(id=1327012618331599326, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618319016413, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3Empyrean Technology Co., Ltd., Beijing 100102), AuthorCompanyExt(id=1327012618356765151, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618319016413, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3北京华大九天科技股份有限公司 北京 100102)])], figs=[ArticleFig(id=1327012622802727440, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=EN, label=Figure 1, caption=TRSM demand hotspot size, figureFileSmall=ZmpS0i7fwS9HzF/rkPFPcA==, figureFileBig=ehGPWnu+E9I2N4IzdNnn/Q==, tableContent=null), ArticleFig(id=1327012622895002129, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=CN, label=图1, caption=TRSM需求热点规模, figureFileSmall=ZmpS0i7fwS9HzF/rkPFPcA==, figureFileBig=ehGPWnu+E9I2N4IzdNnn/Q==, tableContent=null), ArticleFig(id=1327012623134077458, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=EN, label=Figure 2, caption=Small-scale irregular features, figureFileSmall=SKhNa2Dt/DkwcZIN93qpcg==, figureFileBig=Gkc1NKKSmBPslA8mCE6vmQ==, tableContent=null), ArticleFig(id=1327012623201186323, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=CN, label=图2, caption=小规模非规则特征, figureFileSmall=SKhNa2Dt/DkwcZIN93qpcg==, figureFileBig=Gkc1NKKSmBPslA8mCE6vmQ==, tableContent=null), ArticleFig(id=1327012623289266708, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=EN, label=Figure 3, caption=Classification of small-scale irregular problems, figureFileSmall=JBfwabG08anKFIHoE7LaTQ==, figureFileBig=CyLZvvnKxubNiAjfQy9hOw==, tableContent=null), ArticleFig(id=1327012623352181269, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=CN, label=图3, caption=小规模非规则问题分类, figureFileSmall=JBfwabG08anKFIHoE7LaTQ==, figureFileBig=CyLZvvnKxubNiAjfQy9hOw==, tableContent=null), ArticleFig(id=1327012623427678742, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=EN, label=Figure 4, caption=Double-precision real LNLN kernel chunking, figureFileSmall=vvS4/OR6H7rT2tExu9KifA==, figureFileBig=UYRzSibgZvLkRSsbqTjxJg==, tableContent=null), ArticleFig(id=1327012623490593303, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=CN, label=图4, caption=双精度实数LNLN内核分块, figureFileSmall=vvS4/OR6H7rT2tExu9KifA==, figureFileBig=UYRzSibgZvLkRSsbqTjxJg==, tableContent=null), ArticleFig(id=1327012623549313560, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=EN, label=Figure 5, caption=Double-precision complex LNLN kernel chunking, figureFileSmall=7zZ50GuE3TwGAwqV6rcLqg==, figureFileBig=2D8qXgv85K9XWu/wsmEZLQ==, tableContent=null), ArticleFig(id=1327012623649976857, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=CN, label=图5, caption=双精度复数 LNLN内核分块, figureFileSmall=7zZ50GuE3TwGAwqV6rcLqg==, figureFileBig=2D8qXgv85K9XWu/wsmEZLQ==, tableContent=null), ArticleFig(id=1327012623717085722, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=EN, label=Figure 6, caption=LNLN boundary processing, figureFileSmall=C7QAE134G8j+WG4CbP710Q==, figureFileBig=zRhuqYpJwh5K/6G9ock1YQ==, tableContent=null), ArticleFig(id=1327012623796777499, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=CN, label=图6, caption=LNLN边界处理, figureFileSmall=C7QAE134G8j+WG4CbP710Q==, figureFileBig=zRhuqYpJwh5K/6G9ock1YQ==, tableContent=null), ArticleFig(id=1327012623889052188, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=EN, label=Figure 7, caption=Handling method of transposed kernels, figureFileSmall=6WXlt1f2LUjAaPGJJeKZBg==, figureFileBig=XYXQvr+rFUlVVursn62LUA==, tableContent=null), ArticleFig(id=1327012623956161053, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=CN, label=图7, caption=转置内核的处理方法, figureFileSmall=6WXlt1f2LUjAaPGJJeKZBg==, figureFileBig=XYXQvr+rFUlVVursn62LUA==, tableContent=null), ArticleFig(id=1327012624031658526, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=EN, label=Figure 8, caption=SIMD multiplication, figureFileSmall=vwnFExo8j+42p4hjEnah2w==, figureFileBig=sD1E9Ry6/xAHh7goX+zyhg==, tableContent=null), ArticleFig(id=1327012624107155999, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=CN, label=图8, caption=SIMD乘法, figureFileSmall=vwnFExo8j+42p4hjEnah2w==, figureFileBig=sD1E9Ry6/xAHh7goX+zyhg==, tableContent=null), ArticleFig(id=1327012624291705376, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=EN, label=Figure 9, caption=Data distribution, figureFileSmall=BHH8zSDNBEMICFGLn7v7mg==, figureFileBig=Dd1QtaUCrlmqgbNXwV4ufw==, tableContent=null), ArticleFig(id=1327012624375591457, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=CN, label=图9, caption=数据分布, figureFileSmall=BHH8zSDNBEMICFGLn7v7mg==, figureFileBig=Dd1QtaUCrlmqgbNXwV4ufw==, tableContent=null), ArticleFig(id=1327012624442700322, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=EN, label=Figure 10, caption=Real LNLN flow control, figureFileSmall=gUHtXt/Yh5z6LmjwqYvEBA==, figureFileBig=oQYQzZ2jmVuoTaO6VlRupA==, tableContent=null), ArticleFig(id=1327012624509809187, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=CN, label=图10, caption=实数LNLN流控制, figureFileSmall=gUHtXt/Yh5z6LmjwqYvEBA==, figureFileBig=oQYQzZ2jmVuoTaO6VlRupA==, tableContent=null), ArticleFig(id=1327012624644026916, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=EN, label=Figure 11, caption=Comparison of the performance of each type of double-precision real SI_TRSM and MKL, figureFileSmall=hBlL5Pbw2curyPpsz0clrA==, figureFileBig=LgaqQQ7CeY0QMnb502Jwgg==, tableContent=null), ArticleFig(id=1327012624715330085, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=CN, label=图11, caption=双精度实数SI_TRSM与MKL各类型性能对比, figureFileSmall=hBlL5Pbw2curyPpsz0clrA==, figureFileBig=LgaqQQ7CeY0QMnb502Jwgg==, tableContent=null), ArticleFig(id=1327012624782438950, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=EN, label=Figure 12, caption=Comparison of the performance of each type of double-precision complex SI_TRSM and MKL, figureFileSmall=BFYv+Rsn48zvQgO0ppTDmg==, figureFileBig=Xc40HUR8sItlXS2DfIiVPg==, tableContent=null), ArticleFig(id=1327012626460160552, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=CN, label=图12, caption=双精度复数SI_TRSM与MKL各类型性能对比, figureFileSmall=BFYv+Rsn48zvQgO0ppTDmg==, figureFileBig=Xc40HUR8sItlXS2DfIiVPg==, tableContent=null), ArticleFig(id=1327012626544046634, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=EN, label=Table 1, caption=Kernel Names, figureFileSmall=null, figureFileBig=null, tableContent=
计算类型DTRSMZTRSM
LNLNKernel=D_m_LNLNKernel=Z_m_LNLN
m=1, 2, …, 8, nm=1, 2, 3, n
LTLNKernel=D_m_LTLNKernel=Z_m_LTLN
m=1, 2, …, 8, nm=1, 2, 3, n
LCLN Kernel=Z_m_LCLN
 m=1, 2, 3, n
LNUNKernel=D_m_LNUNKernel=Z_m_LNUN
m=1, 2, …, 8, nm=1, 2, 3, n
LTUNKernel=D_m_LTUNKernel=Z_m_LTUN
m=1, 2, …, 8, nm=1, 2, 3, n
LCUN Kernel=Z_m_LCUN
 m=1, 2, 3, n
RNLNKernel=D_n_RNLNKernel=Z_n_RNLN
n=1, 2, …, 8, nn=1, 2, 3, n
RTLNKernel=D_n_RTLNKernel=Z_n_RTLN
n=1, 2, …, 8, nn=1, 2, 3, n
RCLN Kernel=Z_n_RCLN
 n=1, 2, 3, n
RNUNKernel=D_n_RNUNKernel=Z_n_RNUN
n=1, 2, …, 8, nn=1, 2, 3, n
RTUNKernel=D_n_RTUNKernel=Z_n_RTUN
n=1, 2, …, 8, nn=1, 2, 3, n
RCUN Kernel=Z_n_RCUN
 n=1, 2, 3, n
), ArticleFig(id=1327012626636321323, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=CN, label=表1, caption=

内核名称

, figureFileSmall=null, figureFileBig=null, tableContent=
计算类型DTRSMZTRSM
LNLNKernel=D_m_LNLNKernel=Z_m_LNLN
m=1, 2, …, 8, nm=1, 2, 3, n
LTLNKernel=D_m_LTLNKernel=Z_m_LTLN
m=1, 2, …, 8, nm=1, 2, 3, n
LCLN Kernel=Z_m_LCLN
 m=1, 2, 3, n
LNUNKernel=D_m_LNUNKernel=Z_m_LNUN
m=1, 2, …, 8, nm=1, 2, 3, n
LTUNKernel=D_m_LTUNKernel=Z_m_LTUN
m=1, 2, …, 8, nm=1, 2, 3, n
LCUN Kernel=Z_m_LCUN
 m=1, 2, 3, n
RNLNKernel=D_n_RNLNKernel=Z_n_RNLN
n=1, 2, …, 8, nn=1, 2, 3, n
RTLNKernel=D_n_RTLNKernel=Z_n_RTLN
n=1, 2, …, 8, nn=1, 2, 3, n
RCLN Kernel=Z_n_RCLN
 n=1, 2, 3, n
RNUNKernel=D_n_RNUNKernel=Z_n_RNUN
n=1, 2, …, 8, nn=1, 2, 3, n
RTUNKernel=D_n_RTUNKernel=Z_n_RTUN
n=1, 2, …, 8, nn=1, 2, 3, n
RCUN Kernel=Z_n_RCUN
 n=1, 2, 3, n
), ArticleFig(id=1327012626766344748, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=EN, label=Table 2, caption=Experimental Environments, figureFileSmall=null, figureFileBig=null, tableContent=
不同环境环境参数参数指标
硬件环境处理器Intel Xeon Gold 6240
架构x86_64
主频/GHz2.6
SIMD/b512
L1d cache/KB32
L1i cache/KB32
L2 cache/KB1024
软件环境编译器GCC7.5
MKL2023.0.0
), ArticleFig(id=1327012626833453613, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=CN, label=表2, caption=

实验环境

, figureFileSmall=null, figureFileBig=null, tableContent=
不同环境环境参数参数指标
硬件环境处理器Intel Xeon Gold 6240
架构x86_64
主频/GHz2.6
SIMD/b512
L1d cache/KB32
L1i cache/KB32
L2 cache/KB1024
软件环境编译器GCC7.5
MKL2023.0.0
), ArticleFig(id=1327012626934116910, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=EN, label=Table 3, caption=Average Acceleration Statistics Relative to MKL, figureFileSmall=null, figureFileBig=null, tableContent=
类型平均加速 类型平均加速
DTRSMLNLN34.7倍 ZTRSMLCLN16.8倍
DTRSMLTLN30.1倍 ZTRSMLNUN22.1倍
DTRSMLNUN32.8倍 ZTRSMLTUN28.9倍
DTRSMLTUN26.1倍 ZTRSMLCUN18.3倍
DTRSMRNLN19.8倍 ZTRSMRNLN26.5倍
DTRSMRTLN37.5倍 ZTRSMRTLN25.1倍
DTRSMRNUN31.6倍 ZTRSMRCLN32.1倍
DTRSMRTUN23.2倍 ZTRSMRNUN30.1倍
ZTRSMLNLN24.7倍 ZTRSMRTUN27.1倍
ZTRSMLTLN17.1倍 ZTRSMRCUN26.5倍
), ArticleFig(id=1327012627026391599, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, language=CN, label=表3, caption=

相对于MKL的平均加速统计

, figureFileSmall=null, figureFileBig=null, tableContent=
类型平均加速 类型平均加速
DTRSMLNLN34.7倍 ZTRSMLCLN16.8倍
DTRSMLTLN30.1倍 ZTRSMLNUN22.1倍
DTRSMLNUN32.8倍 ZTRSMLTUN28.9倍
DTRSMLTUN26.1倍 ZTRSMLCUN18.3倍
DTRSMRNLN19.8倍 ZTRSMRNLN26.5倍
DTRSMRTLN37.5倍 ZTRSMRTLN25.1倍
DTRSMRNUN31.6倍 ZTRSMRCLN32.1倍
DTRSMRTUN23.2倍 ZTRSMRNUN30.1倍
ZTRSMLNLN24.7倍 ZTRSMRTUN27.1倍
ZTRSMLTLN17.1倍 ZTRSMRCUN26.5倍
)], attaches=null, journal=Journal(id=1324759606947315908, delFlag=0, nameCn=计算机研究与发展, nameEn=Journal of Computer Research and Development, nameHistory1=null, nameHistory2=null, issn=1000-1239, eissn=null, cn=11-1777/TP, coden=null, periodic=0, language=CN, oaType=null, ccby=null, superviseOffice=null, ownerOffice=null, pubOffice=null, editorOffice=null, officeType=null, aims=null, clcCode=null, officeProv=null, officeCity=null, officeAddr=null, officeZip=null, officeEmail=null, officePhone=null, editDirector=null, officeDirector=null, officeDirectorPhone=null, officeStaffNum=null, officeEmpNum=null, coverPicUrl=k2l8zgqwstuxxevFX2qxVA==, journalPrice=null, startedYear=null, abbrevIsoEn=Journal of Computer Research and Development, journalRemark=null, publicationField=null, createdTime=1793776903141, updatedTime=1793850698612, createdBy=18614031015, updatedBy=13701087609, firstLetterCn=J, firstLetterEn=J, subjectCode=Electronic Information, subjectName=null, subjectCodeEn=Electronic Information, subjectNameEn=null, picCn=k2l8zgqwstuxxevFX2qxVA==, picEn=r6k0Ry+RD+c/VXj5QarmDw==, jcr=null, cjcr=null, exts=[JournalExt(id=1325069127710286688, language=CN, name=计算机研究与发展, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=, createdTime=1793850698640, updatedTime=1793850698640, createdBy=13701087609, updatedBy=13701087609, submissionGuidelinesUrl=, submissionAuthorUrl=https://crad.ict.ac.cn/journalx/authorLogOn.action, submissionEditorUrl=https://crad.ict.ac.cn/journalx/editorLogOn.action, submissionReviewUrl=https://crad.ict.ac.cn/journalx/expertLogOn.action, submissionCeEditorUrl=https://crad.ict.ac.cn/journalx/editorInChiefLogOn.action, submissionAeEditorUrl=, option={"copyright":""}), JournalExt(id=1325069127777395553, language=EN, name=Journal of Computer Research and Development, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=, createdTime=1793850698656, updatedTime=1793850698656, createdBy=13701087609, updatedBy=13701087609, submissionGuidelinesUrl=, submissionAuthorUrl=https://crad.ict.ac.cn/journalx/authorLogOn.action, submissionEditorUrl=https://crad.ict.ac.cn/journalx/editorLogOn.action, submissionReviewUrl=https://crad.ict.ac.cn/journalx/expertLogOn.action, submissionCeEditorUrl=https://crad.ict.ac.cn/journalx/editorInChiefLogOn.action, submissionAeEditorUrl=, option={"copyright":""})], databaseList=null, tenantJournalId=1324759914146529478, websiteList=[Website(id=1325070300139892771, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1324759914146529478, journalNameCn=null, journalNameEn=null, grayFlag=null, tenantId=1146029695717560320, platformId=null, journalGroupId=null, journalGroupNameCn=null, journalGroupNameEn=null, type=1, domain=https://castjournals.cast.org.cn/joweb/jsjyjyfz/CN, language=CN, createTime=1793850978169, createBy=18614031015, updateTime=1793851040259, updateBy=18614031015, name=计算机研究与发展-中文, tplId=1146099689490845704, title=计算机研究与发展, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1325070893751349360, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=articleTextType, value=kx, createTime=1793851119697, updateTime=1793851119697, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893730377837, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=banner, value=null, createTime=1793851119692, updateTime=1793851119692, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893772320883, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=grayFlag, value=0, createTime=1793851119702, updateTime=1793851119702, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893721989228, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=logo, value=https://castjournals.cast.org.cn/joweb/jsjyjyfz/CN/file/pic?fileId=U27tDsy717YF7XSPA+Vwdw==, createTime=1793851119690, updateTime=1793851119690, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893784903797, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=minRunFlag, value=0, createTime=1793851119705, updateTime=1793851119705, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893742960751, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/jsjyjyfz/CN/file/pic, createTime=1793851119695, updateTime=1793851119695, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893780709492, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=silenceFlag, value=0, createTime=1793851119704, updateTime=1793851119704, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893738766446, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_cn_619/, createTime=1793851119694, updateTime=1793851119694, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893755543665, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=themeColor, value=null, createTime=1793851119698, updateTime=1793851119698, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893763932274, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=themeStyle, value=null, createTime=1793851119700, updateTime=1793851119700, creator=18614031015, updator=18614031015)]), Website(id=1325070300580294698, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1324759914146529478, journalNameCn=null, journalNameEn=null, grayFlag=null, tenantId=1146029695717560320, platformId=null, journalGroupId=null, journalGroupNameCn=null, journalGroupNameEn=null, type=1, domain=https://castjournals.cast.org.cn/joweb/jsjyjyfz/EN, language=EN, createTime=1793850978274, createBy=18614031015, updateTime=1793851036638, updateBy=18614031015, name=计算机研究与发展-英文, tplId=1146101810881728533, title=Journal of Computer Research and Development, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1325070925288321148, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=articleTextType, value=kx, createTime=1793851127216, updateTime=1793851127216, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925263155321, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=banner, value=null, createTime=1793851127210, updateTime=1793851127210, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925309292671, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=grayFlag, value=0, createTime=1793851127221, updateTime=1793851127221, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925254766712, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=logo, value=https://castjournals.cast.org.cn/joweb/jsjyjyfz/EN/file/pic?fileId=U27tDsy717YF7XSPA+Vwdw==, createTime=1793851127208, updateTime=1793851127208, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925326069889, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=minRunFlag, value=0, createTime=1793851127225, updateTime=1793851127225, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925279932539, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/jsjyjyfz/EN/file/pic, createTime=1793851127214, updateTime=1793851127214, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925317681280, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=silenceFlag, value=0, createTime=1793851127223, updateTime=1793851127223, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925271543930, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_en_623/, createTime=1793851127212, updateTime=1793851127212, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925292515453, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=themeColor, value=null, createTime=1793851127217, updateTime=1793851127217, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925300904062, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=themeStyle, value=null, createTime=1793851127219, updateTime=1793851127219, creator=18614031015, updator=18614031015)])], journalTitle=计算机研究与发展, weixinUrl=null, journalUrl=https://crad.ict.ac.cn/, iacademicId=null, status=1, seqNo=null, journalTitleEn=Journal of Computer Research and Development, journalPhotoCn=k2l8zgqwstuxxevFX2qxVA==, journalPhotoEn=r6k0Ry+RD+c/VXj5QarmDw==, journalFirstLetter=J, journalRecommend=null, journalNew=null, journalCollection=null, jcrJf=null, cjcrJf=null, jcrJfStr=null, cjcrJfStr=null, submissionFirstDecision=null, sciSubjectClassification=null, casSubjectClassification=null, citeScore=null, totalCitationFrequency=null, icpCode=null, psCode=null, advertisingLicenseCode=null, copyrightInformation=null, country=null, option=, provinceCode=null, provinceName=null, collectFlag=false, interPubPlatform=, interPubPlatformUrl=null), detailUrlCn=https://castjournals.cast.org.cn/joweb/jsjyjyfz/CN/10.7544/issn1000-1239.202330864, detailUrlEn=https://castjournals.cast.org.cn/joweb/jsjyjyfz/EN/10.7544/issn1000-1239.202330864, pdfUrlCn=https://castjournals.cast.org.cn/joweb/jsjyjyfz/CN/PDF/10.7544/issn1000-1239.202330864, pdfUrlEn=https://castjournals.cast.org.cn/joweb/jsjyjyfz/EN/PDF/10.7544/issn1000-1239.202330864, aliStartDate=0, aliEndDate=0, collectionFlag=false, citedCount=null, citedUrl=null, previewStatus=0, delFlag=0, hasFullText=1, orderTime=1738339200000, fullTextJson=null, articleText=null, reference=null)
收藏切换
小规模非规则TRSM实现与优化
郭容园 1, 2 , 贾海鹏 2, * , 张云泉 2 , 韦存阳 2 , 邓明森 1 , 陈婧蕊 3 , 周振亚 3
计算机研究与发展 | 高性能计算 2025,62(2): 517-531
收起
收藏切换
计算机研究与发展 |高性能计算 2025 , 62 (2) : 517 -531
小规模非规则TRSM实现与优化
全屏
[Author(id=1327012618470011362, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=guoryuangradu@163.com, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1327012618587451877, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012618470011362, language=EN, stringName=Rongyuan Guo, firstName=Rongyuan, middleName=null, lastName=Guo, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, address=1School of Information, Guizhou University of Finance and Economics, Guiyang 550025
2Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190, bio={"content":"

born in 1999. Master candidate. His main research interests include parallel computing and high performance computing

"}, bioImg=null, bioContent=

born in 1999. Master candidate. His main research interests include parallel computing and high performance computing

, aboutCorrespAuthor=null), CN=AuthorExt(id=1327012618671337959, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012618470011362, language=CN, stringName=郭容园, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, address=1贵州财经大学信息学院 贵阳 550025
2中国科学院计算技术研究所 北京 100190, bio={"img":"iWqAG/ctkkR/2rAJ/wmKTQ==","content":"

1999年生. 硕士研究生. 主要研究方向为并行计算、高性能计算

"}, bioImg=iWqAG/ctkkR/2rAJ/wmKTQ==, bioContent=

1999年生. 硕士研究生. 主要研究方向为并行计算、高性能计算

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1327012618147049943, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, xref=1, ext=[AuthorCompanyExt(id=1327012618155438552, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618147049943, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1School of Information, Guizhou University of Finance and Economics, Guiyang 550025), AuthorCompanyExt(id=1327012618168021465, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618147049943, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1贵州财经大学信息学院 贵阳 550025)]), AuthorCompany(id=1327012618235130330, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, xref=2, ext=[AuthorCompanyExt(id=1327012618243518939, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618235130330, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190), AuthorCompanyExt(id=1327012618247713244, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618235130330, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2中国科学院计算技术研究所 北京 100190)])]), Author(id=1327012618767806953, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=jiahaipeng@ict.ac.cn, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1327012619027853803, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012618767806953, language=EN, stringName=Haipeng Jia, firstName=Haipeng, middleName=null, lastName=Jia, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, *, address=2Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190, bio={"content":"

born in 1983. PhD, senior engineer. His main research interests include parallel computing and heterogeneous computing

"}, bioImg=null, bioContent=

born in 1983. PhD, senior engineer. His main research interests include parallel computing and heterogeneous computing

, aboutCorrespAuthor=null), CN=AuthorExt(id=1327012619136905708, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012618767806953, language=CN, stringName=贾海鹏, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, *, address=2中国科学院计算技术研究所 北京 100190, bio={"img":"rIPqZ1zmlW7/6TUyOpUtMg==","content":"

1983年生. 博士,高级工程师. 主要研究方向为并行计算、异构计算

"}, bioImg=rIPqZ1zmlW7/6TUyOpUtMg==, bioContent=

1983年生. 博士,高级工程师. 主要研究方向为并行计算、异构计算

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1327012618235130330, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, xref=2, ext=[AuthorCompanyExt(id=1327012618243518939, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618235130330, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190), AuthorCompanyExt(id=1327012618247713244, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618235130330, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2中国科学院计算技术研究所 北京 100190)])]), Author(id=1327012619258540526, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, orderNo=2, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1327012619329843696, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012619258540526, language=EN, stringName=Yunquan Zhang, firstName=Yunquan, middleName=null, lastName=Zhang, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190, bio={"content":"

born in 1973. PhD, professor. His main research interests include parallel algorithm and parallel software

"}, bioImg=null, bioContent=

born in 1973. PhD, professor. His main research interests include parallel algorithm and parallel software

, aboutCorrespAuthor=null), CN=AuthorExt(id=1327012619409535473, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012619258540526, language=CN, stringName=张云泉, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2中国科学院计算技术研究所 北京 100190, bio={"img":"RM5ub5FSk9Vs0a7bIbvcFA==","content":"

1973年生. 博士,研究员. 主要研究方向为并行算法、并行软件

"}, bioImg=RM5ub5FSk9Vs0a7bIbvcFA==, bioContent=

1973年生. 博士,研究员. 主要研究方向为并行算法、并行软件

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1327012618235130330, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, xref=2, ext=[AuthorCompanyExt(id=1327012618243518939, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618235130330, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190), AuthorCompanyExt(id=1327012618247713244, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618235130330, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2中国科学院计算技术研究所 北京 100190)])]), Author(id=1327012619480838643, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, orderNo=3, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1327012619560530421, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012619480838643, language=EN, stringName=Cunyang Wei, firstName=Cunyang, middleName=null, lastName=Wei, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190, bio={"content":"

born in 1999. Master. His main research interests include parallel computing and high performance computing

"}, bioImg=null, bioContent=

born in 1999. Master. His main research interests include parallel computing and high performance computing

, aboutCorrespAuthor=null), CN=AuthorExt(id=1327012619623444982, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012619480838643, language=CN, stringName=韦存阳, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2中国科学院计算技术研究所 北京 100190, bio={"img":"b7v0Y/ziU31+DuIZ3dLVog==","content":"

1999年生. 硕士. 主要研究方向为并行计算、高性能计算

"}, bioImg=b7v0Y/ziU31+DuIZ3dLVog==, bioContent=

1999年生. 硕士. 主要研究方向为并行计算、高性能计算

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1327012618235130330, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, xref=2, ext=[AuthorCompanyExt(id=1327012618243518939, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618235130330, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190), AuthorCompanyExt(id=1327012618247713244, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618235130330, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2中国科学院计算技术研究所 北京 100190)])]), Author(id=1327012619698942456, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, orderNo=4, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1327012619766051322, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012619698942456, language=EN, stringName=Mingsen Deng, firstName=Mingsen, middleName=null, lastName=Deng, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1School of Information, Guizhou University of Finance and Economics, Guiyang 550025, bio={"content":"

born in 1981. PhD, professor. His main research interests include distributed parallel computing and feature extraction of big data

"}, bioImg=null, bioContent=

born in 1981. PhD, professor. His main research interests include distributed parallel computing and feature extraction of big data

, aboutCorrespAuthor=null), CN=AuthorExt(id=1327012619841548795, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012619698942456, language=CN, stringName=邓明森, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1贵州财经大学信息学院 贵阳 550025, bio={"img":"eCSaQkOG4UP4sc3szoNOJQ==","content":"

1981年生. 博士,教授. 主要研究方向为分布式并行计算、大数据特征提取

"}, bioImg=eCSaQkOG4UP4sc3szoNOJQ==, bioContent=

1981年生. 博士,教授. 主要研究方向为分布式并行计算、大数据特征提取

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1327012618147049943, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, xref=1, ext=[AuthorCompanyExt(id=1327012618155438552, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618147049943, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1School of Information, Guizhou University of Finance and Economics, Guiyang 550025), AuthorCompanyExt(id=1327012618168021465, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618147049943, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1贵州财经大学信息学院 贵阳 550025)])]), Author(id=1327012619925434877, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, orderNo=5, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1327012620009320959, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012619925434877, language=EN, stringName=Jingrui Chen, firstName=Jingrui, middleName=null, lastName=Chen, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3Empyrean Technology Co., Ltd., Beijing 100102, bio={"content":"

born in 1987. Master, R&D manager. Her main research interests include circuit simulation and linear equation solvers

"}, bioImg=null, bioContent=

born in 1987. Master, R&D manager. Her main research interests include circuit simulation and linear equation solvers

, aboutCorrespAuthor=null), CN=AuthorExt(id=1327012620084818432, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012619925434877, language=CN, stringName=陈婧蕊, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3北京华大九天科技股份有限公司 北京 100102, bio={"img":"2GedsPjM2M+ZnLG7jl6pqA==","content":"

1987年生. 硕士,研发经理. 主要研究方向为电路仿真、线性方程求解器

"}, bioImg=2GedsPjM2M+ZnLG7jl6pqA==, bioContent=

1987年生. 硕士,研发经理. 主要研究方向为电路仿真、线性方程求解器

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1327012618319016413, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, xref=3, ext=[AuthorCompanyExt(id=1327012618331599326, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618319016413, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3Empyrean Technology Co., Ltd., Beijing 100102), AuthorCompanyExt(id=1327012618356765151, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618319016413, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3北京华大九天科技股份有限公司 北京 100102)])]), Author(id=1327012620219036162, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, orderNo=6, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1327012620286145028, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012620219036162, language=EN, stringName=Zhenya Zhou, firstName=Zhenya, middleName=null, lastName=Zhou, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3Empyrean Technology Co., Ltd., Beijing 100102, bio={"content":"

born in 1980. Master, senior engineer. His main research interest includes algorithms and software for transistor-level circuit simulation

"}, bioImg=null, bioContent=

born in 1980. Master, senior engineer. His main research interest includes algorithms and software for transistor-level circuit simulation

, aboutCorrespAuthor=null), CN=AuthorExt(id=1327012621951283717, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, authorId=1327012620219036162, language=CN, stringName=周振亚, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3北京华大九天科技股份有限公司 北京 100102, bio={"img":"+uf3tSWTa8p1WUJxm5pIew==","content":"

1980年生. 硕士,高级工程师. 主要研究方向为晶体管级电路仿真的算法和软件

"}, bioImg=+uf3tSWTa8p1WUJxm5pIew==, bioContent=

1980年生. 硕士,高级工程师. 主要研究方向为晶体管级电路仿真的算法和软件

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1327012618319016413, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, xref=3, ext=[AuthorCompanyExt(id=1327012618331599326, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618319016413, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3Empyrean Technology Co., Ltd., Beijing 100102), AuthorCompanyExt(id=1327012618356765151, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1327012611629101484, companyId=1327012618319016413, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3北京华大九天科技股份有限公司 北京 100102)])])]
郭容园1, 2 , 贾海鹏2, * , 张云泉2, 韦存阳2, 邓明森1, 陈婧蕊3, 周振亚3
作者信息
  • 1贵州财经大学信息学院 贵阳 550025
  • 2中国科学院计算技术研究所 北京 100190
  • 3北京华大九天科技股份有限公司 北京 100102
通讯作者:
贾海鹏()
作者简介:

1999年生. 硕士研究生. 主要研究方向为并行计算、高性能计算

1983年生. 博士,高级工程师. 主要研究方向为并行计算、异构计算

1973年生. 博士,研究员. 主要研究方向为并行算法、并行软件

1999年生. 硕士. 主要研究方向为并行计算、高性能计算

1981年生. 博士,教授. 主要研究方向为分布式并行计算、大数据特征提取

1987年生. 硕士,研发经理. 主要研究方向为电路仿真、线性方程求解器

1980年生. 硕士,高级工程师. 主要研究方向为晶体管级电路仿真的算法和软件

Small-Scale Irregular TRSM Implementation and Optimization
Rongyuan Guo1, 2 , Haipeng Jia2, * , Yunquan Zhang2, Cunyang Wei2, Mingsen Deng1, Jingrui Chen3, Zhenya Zhou3
Affiliations
  • 1School of Information, Guizhou University of Finance and Economics, Guiyang 550025
  • 2Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190
  • 3Empyrean Technology Co., Ltd., Beijing 100102
  • born in 1999. Master candidate. His main research interests include parallel computing and high performance computing

    born in 1983. PhD, senior engineer. His main research interests include parallel computing and heterogeneous computing

    born in 1973. PhD, professor. His main research interests include parallel algorithm and parallel software

    born in 1999. Master. His main research interests include parallel computing and high performance computing

    born in 1981. PhD, professor. His main research interests include distributed parallel computing and feature extraction of big data

    born in 1987. Master, R&D manager. Her main research interests include circuit simulation and linear equation solvers

    born in 1980. Master, senior engineer. His main research interest includes algorithms and software for transistor-level circuit simulation

出版时间: 2025-02-01 doi: 10.7544/issn1000-1239.202330864
文章导航
收藏切换

TRSM(triangular matrix equation solver)是线性方程组求解的常用算法,是各种科学计算库和数学软件的核心算法,广泛应用于科学计算、工程计算、机器学习等领域. 小规模非规则TRSM算法限定解决问题范围,是高效处理较小规模、非规则数据输入的算法. 随着高性能计算领域个性化、精细化发展,科学界、工业界对小规模非规则TRSM计算的需求愈加明显. 传统算法更偏重于大规模、规则TRSM的计算,在小规模非规则TRSM计算上效率不佳. 结合硬件体系结构、应用场景特征提出小规模非规则TRSM优化方案,从寄存器分块、边界处理、向量化计算角度设计高性能内核,在此基础上构建覆盖双精度实数、双精度复数的小规模非规则算法库SI_TRSM (small-scale irregular TRSM),大幅度提升该算法性能. 实验结果表明,构建的双精度小规模非规则TRSM算法库,较MKL(Intel math kernel library)同类算法,在双精度小规模非规则实数上平均性能提高29.4倍,在双精度小规模非规则复数上平均性能提高24.6倍.

TRSM算法  /  BLAS  /  小规模非规则  /  SIMD  /  汇编优化

TRSM (triangular matrix equation solver) is a commonly used algorithm for solving systems of linear equations, and is the core algorithm of various scientific computing libraries and mathematical software, which is widely used in the fields of scientific computing, engineering computing and machine learning. The small-scale irregular TRSM algorithm limits the scope of problem-solving and is an algorithm for efficiently handling smaller-scale, irregular data inputs. With the development of personalization and refinement in the field of high-performance computing, the demand for small-scale irregular TRSM computation in the scientific and industrial communities is becoming more and more obvious. While traditional algorithms are better suited for large-scale and regular TRSM computation, there is still room for improvement in the computational efficiency of small-scale and irregular TRSM. In this paper, we propose a small-scale irregular TRSM optimization scheme by combining hardware architecture and application scenario characteristics, designing a high-performance kernel from the perspectives of register chunking, boundary processing, and vectorization computation, and constructing an algorithmic library of small-scale irregular SI_TRSM (small-scale irregular TRSM) covering double-precision real numbers and double-precision complex numbers based on which the performance of this algorithm is greatly improved. Based on experimental results, the double-precision small-scale irregular TRSM algorithm library developed in this paper has shown to enhance the average performance of double-precision small-scale irregular real numbers by 29.4 times, and double-precision small-scale irregular complex numbers by 24.6 times in comparison with similar algorithms available in the MKL (Intel math kernel library).

TRSM algorithm  /  BLAS  /  small-scale irregular  /  SIMD  /  assembly optimization
郭容园, 贾海鹏, 张云泉, 韦存阳, 邓明森, 陈婧蕊, 周振亚. 小规模非规则TRSM实现与优化. 计算机研究与发展, 2025 , 62 (2) : 517 -531 . DOI: 10.7544/issn1000-1239.202330864
Rongyuan Guo, Haipeng Jia, Yunquan Zhang, Cunyang Wei, Mingsen Deng, Jingrui Chen, Zhenya Zhou. Small-Scale Irregular TRSM Implementation and Optimization[J]. Journal of Computer Research and Development, 2025 , 62 (2) : 517 -531 . DOI: 10.7544/issn1000-1239.202330864
TRSM(triangular matrix equation solver)算法[1-2]是矩阵计算核心操作之一,也是基本线性代数子程序(BLAS)的重要组成部分. TRSM是针对A×X=α×B或X×A=α×B进行求解的算法,其中A是一个上三角或下三角矩阵,X和B是矩阵,α是一个标量. 该算法应用广泛,是高斯消元、QR分解和LU分解等算法的基础操作,对电磁场分析、声学模拟、计算流体力学有支撑作用,也可被用于实现矩阵求逆操作和解线性方程组问题.
TRSM算法在许多软件的底层具有丰富的应用. 例如材料科学和生物分子模拟DFT软件包、芯片电路设计EDA软件等在底层对TRSM算法需求颇丰. 图1展示某EDA软件部分模块在底层对TRSM算法7194396次调用需求的规模分布,占比86.3%的调用规模分布在图1中SI区域(m≤100,n≤100). SI区域以外的规模,业界标准库MKL[3],OpenBLAS[4],AMDLIB(AMD math library)[5]等可取得良好性能效果,甚至可达到硬件峰值的90%以上,但在SI区域上,目前已有算法的性能表现并不理想.
SI区域TRSM计算具备“小规模非规则”特征. 该区域包含计算规模“m≤100,n≤100”,定义“q=m/n”,根据q值的变化,矩阵会呈现长条(q<1)、瘦高(q>1)、小方块(q=1)等小规模、非规则情况. 另外,SI区域TRSM计算包含的数据量基本可一次性放入一级缓存(主流处理器32~256 KB),符合“规模较小”特征.
传统TRSM算法在SI区域性能略显不足. 传统TRSM算法通常采用分块、Pack的方法来优化性能. Pack操作保证计算的连续性,分块操作充分利用缓存,这些方法在较大规模、方阵或趋近方阵的数据上将获得较好的性能优势. 但SI区域的数据规模较小,基本可一次性放入一级缓存,已具备缓存优势,继续采用基于缓存的分块、Pack操作对性能提升没有意义;另外,“边界处理”在SI小规模非规则区域TRSM计算时不可忽视. 传统TRSM算法在内核设计时,对边界数据并不关注,较大规模TRSM计算中边界数据计算量占比较小,忽视边界处理对整体性能的影响较小. 但在SI区域的TRSM计算中,边界数据计算量占比较高,忽视边界处理严重影响整体性能.
本文研究的小规模非规则TRSM算法充分考虑SI区域TRSM计算特征,摒弃传统算法的优化思路,结合硬件体系结构和应用场景,提出小规模非规则TRSM优化方案,从寄存器分块、边界处理、向量化计算角度,重新设计了高性能内核,最后遵循BLAS标准完成算法库,定义为SI_TRSM(small-scale irregular TRSM)库. 实验表明,在x86平台上,本文TRSM算法较业界领先且在x86平台适配性最佳的MKL同类算法,有大幅度的性能提升.
本文的贡献有3点:
1)提出了基于应用场景、关注小规模非规则数据特征的优化方案. 该方案为BLAS相关算法的个性化需求提供参考.
2)设计了有效利用硬件特征进行分块、关注双精度小规模非规则数据边界处理问题的高性能内核. 与传统内核思路不同,本文内核设计大幅度提高了运行效率.
3)构建了覆盖双精度实数与复数的SI_TRSM算法库. 该算法库接口符合BLAS标准,与x86平台适配性最佳的MKL同类算法相比,表现出较好的性能优势.
TRSM算法是建立在广泛研究和深入理论探索基础上的经典算法,但针对本文所述小规模非规则TRSM的研究缺乏相关的成果. 早期的TRSM算法基于高斯消元法,难以处理大规模矩阵. 1971年,McNamee[6]提出用于求解稀疏上三角线性方程组的算法,这是率先用于解决TRSM问题的子程序. 1979年BLAS作为标准正式发布,TRSM算法随之得到规范,逐步发展出高效的双精度DTRSM算法,可处理一般密集矩阵. 1990年,Dongarra等人[7]提出level-3 BLAS的模型实现与测试方法,单精度STRSM得到详细测试,为该算法的优化奠定基础. 1992年,Hanson[8]发布PC-BLAS,在精度、实数、复数方面对BLAS进行高度优化,推动TRSM算法在内的BLAS库等走向应用. 1994年,Choi等人[9]在分布式系统上研究BLAS计算,BLAS算法的应用场景进一步扩大. 2002年,Dongarra[10]主导推进BLAS标准更新,对密集和稀疏矩阵的计算都提出并行要求,TRSM算法也受到规范. 2008年,Chan等人[11]对多线程BLAS深入研究,其中Cholesky因式分解过程时通过分块大量运用TRSM算法,TRSM算法在多线程的计算中得到发展. 2010年,GPU硬件广泛应用,Volkov等人[12]提出一种基于GPU的相关算法,该算法利用GPU的高并行性来加速矩阵计算. 近年来随着深度学习、机器学习的发展,研究人员开始研究将TRSM算法与其相结合[13],用于直接并行化加速训练学习.
学者们对TRSM算法的其他研究[14-16]在不断深入,以满足越来越复杂的矩阵计算需求. 但针对本文所述小规模非规则TRSM实现与优化的研究,目前没有针对性较强的成果.
另外,TRSM算法也在众多公司和组织的BLAS商业库及解决方案中得到了发展. MKL,OpenBLAS,ATLAS[17],BLIS等算法库在业界处于领先地位,但也存在缺陷. 针对本文研究的小规模非规则TRSM算法,MKL对于自身的体系架构充分优化适配,但缺乏小规模非规则TRSM的专门优化;OpenBLAS会因默认使用标准分块而对小规模非规则TRSM数据过度分块;ATLAS自动参数优化在小规模计算时管理开销太大;BLIS模块化的设计带来的函数调用开销影响较大.
综上,传统算法对小规模非规则TRSM问题尚未进行较好解决,因此本文算法具有一定的研究价值和改进空间,有较强的工业实践意义和学术理论意义.
本文提出适合SI_TRSM的优化方案,该方案充分适应双精度小规模非规则数据. 如图2所示,我们梳理SI_TRSM遇到的基本情形,包括“长条” “瘦高”“小方块”等特征的数据,图2中三角形代表矩阵A,矩形代表矩阵B.
对于三角矩阵位于左侧的TRSM计算,需要将更多注意力放在“长条”数据特征上. 在TRSM计算中,规模参数m和n决定矩阵A和B的形状. 当三角矩阵A在左侧时,矩阵B的列长度受限于参数m,需同列长度、行长度皆为m的矩阵A保持一致. 此时矩阵B的行长度n不受限制,在小规模的限定下,n值或大或小,本质是“长条”数据特征问题,故对于左侧为三角时TRSM计算,需更注重“长条”形状数据的处理.
对于三角矩阵位于右侧的TRSM计算,需将更多注意力放在“瘦高”数据特征上. 当三角矩阵A位于右侧时,矩阵B的行长度受限于参数n,需同列长度、行长度皆为n的矩阵A保持一致. 此时矩阵B的列长度m不受限制,在小规模的限定下,m值或大或小,本质上是“瘦高”数据特征问题. 故对于右侧三角的TRSM计算,需更注重“瘦高”形状数据的处理.
对于“小方块”数据特征,其本质考验SI_TRSM的边界处理能力. 该特征包含于n较小的“长条”问题中,或包含于m较小的“瘦高”问题中. 区别是相对于“长条”“瘦高”特征,“小方块”数据的边界计算占比升高. 本文SI_TRSM关注“边界处理”工作,故“小方块”无需单独进行考虑.
综上,SI_TRSM需做到对“长条”“瘦高”特征自适应选择特定处理内核. 这2类特征的共同点是在规模较小时,矩阵A可完全以广播形式放入向量寄存器,此时“长条”或“瘦高”形状的矩阵B可在计算上采用分块的方式复用矩阵A,这样既降低矩阵A在整个计算中的访存量,又能极大降低TRSM边界处理在整个计算中的占比. 基于这个发现,我们给出了SI_TRSM覆盖双精度实数、双精度复数的优化方案.
为方便下文阐述,这里定义参数side,transa,uplo,diag 依次为“R(或L)、N(或T,C)、L(或U)、N(或U)”,例如“LNLN”表示“A为左侧下三角非单位矩阵且无需转置操作”,“LCLN”表示“A为左侧下三角非单位矩阵且需进行共轭转置操作”.
图3左侧对双精度实数的8种类型进行分类,LNLN,LTLN,LNUN,LTUN这4种类型属于“长条”问题,RNLN,RTLN,RNUN,RTUN这4种类型属于“瘦高”问题.
矩阵A位于左侧时,双精度实数包含LNLN,LTLN,LNUN,LTUN这4种类型,此时关注于解决“长条”问题. 于是,每种类型下参数m=1,2,3,4,5,6,7,8和m>8这9种情况,SI_TRSM自适应选择特定内核,共计36个. 对于参数m=1,2,3,4,5,6,7,8时的特定内核,我们已知矩阵A大小,其不需要分块. 矩阵B的参数n是不固定的,会呈现“长条”特征,这种情况下,需处理n维度上矩阵B分块、边界处理问题;对于参数m>8时的特定内核,参数m,n皆未知,故矩阵A,B都需要进行分块,此时涉及矩阵A,B分块和A,B边界处理问题.
矩阵A位于右侧时,双精度实数包含RNLN,RTLN,RNUN,RTUN这4种类型,此时关注于解决“瘦高”问题. 于是,每种类型下参数n=1,2,3,4,5,6,7,8和n>8这9种情况,SI_TRSM自适应选择特定内核,共计36个. 对于参数n=1,2,3,4,5,6,7,8时的特定内核,我们已知矩阵A大小,其不需要分块. 矩阵B的参数m是未知的,会呈现“瘦高”特征,这种情况下,需处理m维度上矩阵B分块、边界处理问题;对于参数n>8时的特定内核,参数m,n皆未知,故矩阵A,B都需要进行分块,此时涉及A,B分块和A,B边界处理问题.
图3中右侧对双精度复数的12种类型进行分类,LNLN,LTLN,LCLN,LNUN,LTUN,LCUN这6种类型属于“长条”问题,RNLN,RTLN,RCLN,RNUN,RTUN,RCUN这6种类型属于“瘦高”问题.
矩阵A位于左侧时,双精度复数包含LNLN,LTLN,LCLN,LNUN,LTUN,LCUN这6种类型,此时面临的是“长条”问题,于是,每种类型下参数m=1,2,3和m>3这4种情况,SI_TRSM自适应选择特定内核,共计24个. 对于参数m=1,2,3时的特定内核,我们已知矩阵A大小,其不需要分块. 矩阵B的参数n是不固定的,所以呈现“长条”特征,这种情况下,需处理n维度上矩阵B分块、边界处理问题;对于参数m>3时的特定内核,参数m,n皆未知,故矩阵A,B都需要进行分块,此时涉及A,B分块和A,B边界处理问题.
矩阵A位于右侧时,双精度复数包含RNLN,RTLN,RCLN,RNUN,RTUN,RCUN这6种类型,此时面临的是“瘦高”问题,于是,每种类型下参数m=1,2,3和m>3这4种情况,SI_TRSM自适应选择特定内核,共计24个. 对于参数n=1,2,3时的特定内核,我们已知矩阵A大小,其不需要分块. 矩阵B的参数m是未知的,所以呈现“瘦高”特征,这种情况下,需要处理m维度上矩阵B分块、边界处理问题;对于参数n>3时的特定内核,参数m,n皆未知,故矩阵A,B都需要进行分块,此时涉及A,B分块及A,B边界处理问题.
在本节中给出了小规模非规则TRSM优化方案,下文介绍高性能内核实现,重点讲解分块和边界处理.
SI_TRSM优化方案中20种内核扩展为120个实际内核,第2节介绍了这些内核如何自适应“长条”“瘦高”等特征. 本节将具体介绍内核实现部分的寄存器分块与边界处理策略.
表1中给出所有内核的名称. 本文SI_TRSM算法去掉Pack步骤,内核设计需考虑转置、实数、复数等问题,其处理双精度实数和双精度复数需20种不同类型内核,在第3节优化方案里拓展为120个实际内核. 如表1所示,对于双精度实数DTRSM,每种类型下包含“参数m=1, 2, …, 8, n”,或“参数n=1, 2, …, 8, n”分别9个内核. 对于双精度复数ZTRSM,每种类型下包含“参数m=1, 2, 3, n”,或“参数n=1, 2, 3, n”分别4个内核.
本节重点介绍基于寄存器的分块及边界处理. 边界处理也是一种特殊分块,本文关注边界处理,故将其从整体分块中取出单独命名为“边界处理”. 根据第3节优化方案,当side=L时,以参数m作为参照数. 在双精度实数side=L时m≤8的内核、双精度复数side=L时m≤3的内核中,对矩阵A尽可能放入向量寄存器复用,对矩阵B进行“长条”或“瘦高”方向上分块和边界处理. 在双精度实数中side=L时m>8的内核、双精度复数中side=L时m>3的内核中,需对矩阵A、矩阵B分别进行分块和边界处理. 同理,当side=R时,以参数n作为参照数. 本文分块和边界处理归纳如下:
1)分块. 作为参照数的m或n为已知常数时,此时矩阵A尽可能被放入寄存器进行复用,涉及到矩阵B的分块. 作为参照数的m或n为未知数(整型正数)时,此时涉及矩阵A、矩阵B分块.
2)边界处理. 作为参照数的m或n为已知常数时,涉及矩阵B的边界处理. 作为参照数的m或n为未知数(整型正数)时,此时涉及矩阵A、矩阵B两者的边界处理.
存储、计算、通信是高性能计算系统的3个基本组成部分,本文不涉及分布式系统和多核处理,故本文为内核设计只确立关于存储和计算的2条原则,这是本文高性能内核设计的基石. 存储层面,充分发挥向量寄存器在小规模非规则数据特征上的存取优势. 本文去掉传统算法中基于缓存的Pack操作和分块方法,由于SI_TRSM规模较小,本身已经具备在缓存方面的优势,现需将注意力放在主动利用向量化寄存器的优势上,获得在寄存器级别的优势. 计算层面,充分关注小规模非规则数据向量化并行计算. 无论非边界处理部分和边界处理部分,我们都将尽可能使用向量化并行计算方式,关注边界处理本质上是关注边界处理上的并行计算.
本文研究中内核设计与传统算法的内核设计有较大区别,在实现中比传统算法的内核数量更多、复杂度也更高,但在小规模非规则TRSM各类型求解上获得了较好的计算效率.
本文分块方式紧密联系硬件体系结构特征,传统分块在这方面更关注缓存层面,本文分块更关注寄存器层面. 除去边界块处理,寄存器分块的2种情况如下:
1)作为参照数的m或n为已知常数(实数类型下参照数为1, 2, …, 8;复数类型下参照数为1,2,3)时,此时矩阵A尽可能放入寄存器进行复用,涉及到矩阵B的分块.
2)作为参照数的m或n为未知数(实数类型下参照数大于8,复数类型下参照数大于3)时,此时涉及矩阵A、矩阵B分块.
分块大小和原理与计算机体系结构和应用场景密切相关.
在计算机体系结构上,由于面向支持512 b向量运算的处理器进行优化,其典型特点之一是具有位宽达到512 b的向量寄存器,支持AVX-512向量化指令. 这意味着1个ZMM寄存器可同时处理8个双精度浮点数,可极大限度地利用SIMD技术. 故分块的某一维度应该以8为参照,这样才能充分利用512 b向量寄存器,避免向量寄存器资源浪费.
在应用场景上,本文将目前小规模非规则TRSM需求引申为SI_TRSM问题,优化与实现重点集中在“参数m≤100,n≤100”的“长条”“瘦高”问题上. “长条”的长度方向和“瘦高”的高度方向显然是本文向量化的方向,这2个方向上的分块以填满512 b向量寄存器为准,即8个双精度浮点数据. “长条”的宽度方向或“瘦高”的瘦度方向等于矩阵A的边长,即决定矩阵A的大小,在规模较小时我们尽可能要求矩阵A以广播形式放入向量寄存器复用,限于硬件体系结构32个向量寄存器数量,设矩阵A边长为K,2条边的边长均至少应该满足公式$ (K\times \left(K+1\right))/2\le 32 $,故对于双精度实数,其中一个块边长不应该超过8,对于双精度复数,边长均不应该超过3.
基于以上对计算机体系结构和应用场景的分析研究,本文对于分块大小概括如下:
对于情况1,矩阵B分块大小是“已知常数×8”. 这样便于在“长条”“瘦高”上对矩阵A进行复用,在“长条”的长度方向、“瘦高”的高度方向做充分的向量化并行计算,这2个方向上的剩余计算交给高效的边界处理予以解决.
对于情况2,矩阵A,B分块大小依次是“K×K”和“K×8”(实数时K=8,复数时K=3). 这种情况下,矩阵A因为元素较多,已经无法全部保留在寄存器中进行复用. 此时我们可以通过分块,使得当前计算的矩阵A的分块一直保留在向量寄存器中,然后在矩阵B分块后进行复用. 概括来说,就是把当前规模稍大(实数时K>8,复数时K>3)的计算转化为高效的“长条计算”或“高瘦计算”,最后剩下的部分交给边界处理.
图4展示情况2中双精度实数LNLN类型下的内核分块,当K=m=n=8时,矩阵A包含36个双精度浮点数,矩阵B包含8行8列双精度浮点数据,即内核实现上对A,B均采用8×8分块. 这里需要说明的是,矩阵A元素是通过广播的形式在向量寄存器中存放,当K=6,7,8时,将依次占用21,28,36个向量寄存器,K=6,7时显然满足硬件特征,K=8显然不满足硬件特征. 通过反复实验,对8×8的矩阵A采取个别元素嵌入式广播,依然比对矩阵A采用6×6,7×7分块高效.
图5展示了情况2下双精度复数LNLN的内核分块. 双精度复数原始数据的实部、虚部紧邻排列,在实际运算中实现向量化并行,需继续借助向量寄存器将实部和虚部分离,为了留出足够的向量寄存器进行辅助计算,情况2下双精度复数的分块对矩阵A按照3×3分块,对矩阵B按照3×8分块. 另外,图5也展示了双精度复数LNLN分块的数据分布,在列长度上1个实部和1个虚部构成1个复数,对于3×3复数矩阵A,其实际包含12个元素;对于3×8复数矩阵B,其实际包含6行48个元素.
边界处理在SI_TRSM内核设计中不可忽视. SI_TRSM规模较小,边界处理在整个计算中的占比较高,SI_TRSM形状非规则,边界处理难以向量化,需单独考虑. 涉及边界处理的情况有2个:
1)作为参照数的m或n为已知常数(实数类型下参照数为1, 2, …, 8;复数类型下参照数为1,2,3)时,此时涉及矩阵B的边界处理.
2)作为参照数的m,n为未知数(实数类型下参照数大于8,复数类型下参照数大于3)时,此时涉及矩阵A和矩阵B的边界处理.
对于边界处理情况1,只涉及矩阵B的边界处理. TRSM求解时,矩阵A规模大小已知且理论上尽可能保留在向量寄存器中,在矩阵B“长条”或“瘦高”方向分块计算时进行复用,以及在矩阵B “长条”或“瘦高”方向上最后不满足完整分块的地方进行边界处理. 这种情况下边界部分的特点是,其中一个维度的大小等于前面完整分块的一个维度,另一个维度取决于“长条”或“瘦高”方向上分块的剩余. 此时采用寄存器掩码的方式解决,例如“长条”或“瘦高”方向上余量为7时,使用掩码方式只启用向量寄存器前448位,存放7个双精度浮点数,然后在向量化计算上保持和前文所述完整分块相同.
对于边界处理情况2,涉及矩阵A和矩阵B的边界处理. 本文定义一系列边界块,有可变边界块B、受限边界块A、边界三角块. 图6展示LNLN边界部分的分块,其中矩阵A规模是27×27,矩阵B规模是27×23,矩阵A和B分别进行8×8分块后,出现Ⅶ,Ⅷ,Ⅸ,Ⅹ和c,f,i,j,k,l这些边界块.
本文对边界块进行分类,其中c,f,i,l的维度不受矩阵计算限制,在LNLN类型下,B列长度需与A列长度保持一致,在行长度上是不受限制的,本文把这种边界形成的块定义为B可变块;Ⅶ,Ⅷ,Ⅸ与j,k边界块大小受限于矩阵A列长度上分块剩余,定义为A受限边界块,规模为n×8(n<8);Ⅹ自成一类,即边界三角块,规模包含于n×n(n<8).
对于B可变块和A受限边界块有4种情况:
1)B可变块需同类似图6中Ⅰ这样的三角块进行矩阵计算. Ⅰ块采用广播形式取入ZMM寄存器,矩阵B采用的按行分散取值(gather指令)的方式取入ZMM寄存器,此时的边界块采用寄存器掩码操作即可,本质上是行长度上取几个数据的问题,对取几个问题进行判别,给予自适应的掩码解决边界问题.
2)B可变块需同类似图6中Ⅱ这样的8×8常规分块进行矩阵计算. 在介绍这种情况如何处理之前,先探讨图6中矩阵B的f块结果如何得到,首先需要完成“f–Ⅱ×c”来更新f值,然后求解f就变成“Ⅲ×X=new_f ”问题(这里X表示解矩阵),而“Ⅲ×X=new_f ”是情况1中所讲的可变块遇到三角块的处理方式(这里X表示解矩阵),在情况1中已经给出处理办法.
因此B可变块主要是解决“f–Ⅱ×c”边界问题,针对此边界问题的处理,我们设计了微内核B_8x7_kernel, B_8x6_kernel, B_8x5_kernel, B_8x4_kernel, …, B_8x1_kernel,依次处理“f–Ⅱ×c”中f规模为8×7,8×6,8×5,8×4,8×3,8×2,8×1的情况,具体的实现是采用带减法的矩阵乘. 这里给出B_8x7_kernel的设计思想如下:第1步,加载矩阵A中Ⅱ块数据到ZMM2, ZMM3, …, ZMM9中;第2步,加载矩阵B中f块数据到ZMM10, ZMM11, …, ZMM16等7个向量寄存器;第3步,采用融合的乘减指令完成ZMM10, ZMM11, …, ZMM16数据减去“ZMM2, ZMM3, …, ZMM9数据与矩阵B中c块数据相乘结果”. 具体实现可参考GEMM算法.
3)B可变块需同A受限边界块进行矩阵计算. 针对这种情况,我们根据A受限程度(可能有7,6,5,4,3,2,1等列长度剩余),仿照B_8x7_kernel设计了B_Nx7_kernel, B_Nx6_kernel, B_Nx5_kernel, B_Nx4_kernel, …, B_Nx1_kernel(N=7,6,5,4,3,2,1)等49种微内核.
4)A受限边界块需同B中a这样8×8正常分块进行矩阵计算,我们设计了KERNEL_Nx8_M(N=7,6,5,4,3,2,1)微内核,实现上是矩阵乘的变体.
对于图6中Ⅹ边界三角块,本文设计覆盖所有情况的微内核. 我们按照图4中分块8×8思想,完成了7×7,6×6,5×5,4×4,3×3,2×2,1×1等规模的边界三角块微内核.
双精度复数边界处理原理与双精度实数边界处理原理相同,只是规模大小发生了变化. 对双精度复数LNLN类型下矩阵A,按照3×3(复数包含12个双精度元素)分块,对双精度复数矩阵B按照3×8(复数是6行8列共48个双精度元素)分块. 对应图6中B可变边界块、A受限边界块、边界三角块依旧存在,只是规模发生相应变化,例如对于边界三角,双精度实数是N×N(N=1, 2, …, 7)规模,双精度复数是N×N(N=1,2)规模,故这里不再详述.
内核之间存在高度关联性,下面通过分类描述其关联,打通前文分块、边界处理等策略在各种类型内核上的应用阻碍.
处理双精度实数及复数共需20种不同类型内核,本文将LNLN,LNUN,RNLN,RNUN类型归为普通内核,特点是不涉及转置,双精度实数、复数都有这些类型,所以共8种;将LTLN,LTUN,RTLN,RTUN类型归为转置内核设计,特点是需要对三角矩阵A进行转置操作,双精度实数、复数都有这些类型,共8种;将LCLN,LCUN,RCLN,RCUN类型归为共轭转置内核设计,特点是需要完成数据的共轭转置操作,只有双精度复数有这些类型,共4种;以上包含SI_TRSM所有不同类型的内核.
对于转置内核设计,其相比于普通内核设计,需要对矩阵A进行转置操作. 例如对于LTUN类型的数据进行操作时,上三角矩阵转置后是下三角矩阵,可以采用LNLN计算形式,但需要计算的数据分布于上三角部分,如图7所示,需先按照转置的方式取值,再按照LNLN形式计算.
共轭转置内核设计是在转置内核的基础上增加复数矩阵的共轭操作. 复数可以表示为 a+bi 的形式,其中 a 和 b 分别是实部和虚部,共轭转置操作将复数的虚部取负数,即 b 变为 “–b”. 共轭转置内核设计需在转置内核的基础上对复数矩阵A增加取共轭步骤.
以上第2和3节依次介绍小规模非规则TRSM优化方案、高性能内核的设计,注重从思路、研究的角度讲解本文的工作,并从实现与优化方法的角度对本文工作进行介绍.
本文SI_TRSM在汇编级别实现TRSM任务的存取、计算等全部任务,汇编实现提供高效、精确的硬件控制能力,同时也为优化奠定了基础.
汇编实现为优化奠定基础. 首先,按照System V AMD64[18]调用约定,我们在汇编层得到TRSM全参数,从而获得对数据操作较为直接的权限,可以直接访问和控制内存,精确地管理数据的存储和访问方式,减少内存访问的延迟和带宽消耗. 其次,汇编实现不会引入高级语言中可能存在的不必要抽象和额外代码,我们可以直接控制生成的指令,减少代码的冗余和不必要的计算,从而提高执行效率. 最后,汇编实现支持数据重排、指令融合、循环展开、手动分配寄存器、优化数据结构的存储布局、使用相关SIMD指令集等,为性能优化提供了更多思路.
SIMD(single instruction multiple data)[19]是一种计算机指令级并行处理的技术,旨在同时处理多个数据元素,提高计算性能和效率.
AVX-512指令集[20]的使用极大增强并行化程度. 该指令集是Intel公司推出的一套扩展指令集,是SIMD技术的一种具体实现,它引入了32个512 b的向量寄存器(ZMM0-ZMM31),用于存储和处理更多的数据元素,执行并行程度更高的向量化操作. 其次提供掩码操作,这为边界数据的向量化操作提供了帮助. SI_TRSM在计算时,可充分采用SIMD技术进行优化,本文研究的TRSM内核计算时涉及到的基本数据类型都是双精度浮点数,1个ZMM寄存器可容纳8个双精度浮点数,意味着我们可以同时对8个双精度浮点数据执行SIMD操作,在边界处理时不能填满向量寄存器时,可采用掩码操作尽最大程度实现向量化.
如图4所示,对于双精度实数TRSM,记$ \boldsymbol{X}1 $为矩阵X第1行的结果,记 A1为矩阵A中标号1的元素,记$ \boldsymbol{B}1 $ 为矩阵B标号1的行元素. 在不考虑标量α时,图8中展示简单的$ \boldsymbol{X}1=\boldsymbol{B}1/A1 $计算,使用ZMM寄存器执行SIMD的乘法操作. 对于数据较多的公式$ \boldsymbol{X}8=(\boldsymbol{B}8\mathrm{ }-A29\times \boldsymbol{X}1-A30\times \boldsymbol{X}2-A31\times \boldsymbol{X}3-A32\times \boldsymbol{X}4- A33\times \boldsymbol{X}5-A34\times \boldsymbol{X}6-A35\times \boldsymbol{X}7)/A36 $,包含乘法和减法操作,依旧可采用图8的思路,可使用“vfnmadd231pd”指令实现减法和乘法的指令融合.
寄存器是位于处理器内部的存储器单元,相比硬盘、主存、缓存,其具有最快的存取速度. 向量寄存器是一种特殊类型的寄存器,用于并行处理数据,以提高数据处理的并行性和效率.
本文追求对向量寄存器的“贪心”使用,这体现在对全部向量寄存器的分配和单个寄存器的使用上.
向量寄存器的分配使用与SI_TRSM的特征密切相关. SI_TRSM“小规模非规则”的特征让我们摒弃传统算法对于缓存命中、数据局部性的追求,本文工作更加追求对于向量寄存器的合理分配与使用. “AX=αB或XA=αB”中A为三角方阵,在小规模的限定下,“AX=αB”容易呈现“长条”特征,“XA=αB”容易呈现“瘦高”特征,“长条”形状的宽度和“瘦高”形状的瘦度本质上是矩阵A的边长,故“长条”形状的宽度或“瘦高”形状的瘦度决定矩阵A元素数量. 增强矩阵A在“长条”“瘦高”特征方向的复用性对于处理非规则问题是关键,我们期望矩阵A(或A的分块)能尽可能保留在向量寄存器,这样可在“长条”“瘦高”方向上分块计算时进行复用,因此矩阵A元素数量是本文分配向量寄存器的首要标准,在向量寄存器分配上较为“贪心”,矩阵A能放则放,不能存放时,分块存放.
单个向量寄存器的充分利用需尽可能填满数据. 从单个向量寄存器的角度看,我们期望在“长条”长度方向、“瘦高”高度方向进行向量化操作,这样除边界处理外,向量寄存器的使用基本可被填满,进而执行并行计算操作.
反复实验对比,以上对于全部向量寄存器、单个向量寄存器的“贪心”使用方法,对于性能的提高有显著作用.
结合汇编实现、SIMD向量化、向量寄存器使用等方法,本文工作在计算上尽最大可能实现向量化并行计算.
对于双精度实数TRSM,在不考虑标量α时,获得解矩阵X第1行的结果(记为$ \boldsymbol{X}1 $),需图4中A1(矩阵A中标号1的元素)取倒数后与$ \boldsymbol{B}1 $ (矩阵B标号1的行)元素对应相乘,公式为$ \boldsymbol{X}1=\boldsymbol{B}1/A1 $. 同理,X第2行的结果$ \boldsymbol{X}2=(\boldsymbol{B}2-A2\times \boldsymbol{X}1)/A3 $,X第3行的结果$ \boldsymbol{X}3=(\boldsymbol{B}3-A4\times \boldsymbol{X}1-A5\times \boldsymbol{X}2)/A6 $, …, X第8行的结果$ \boldsymbol{X}8=(\boldsymbol{B}8\mathrm{ }-A29\times \boldsymbol{X}1-A30\times \boldsymbol{X}2-A31\times\boldsymbol{X}3-A32\times \boldsymbol{X}4- $$ A33\times \boldsymbol{X}5-A34\times \boldsymbol{X}6-A35\times \boldsymbol{X}7)/A36 $.
根据上述公式,我们在算法1中给出矩阵$ \boldsymbol{X}1,\boldsymbol{X}2, \boldsymbol{X}3 $的求解伪代码. 行①~③对3×3三角矩阵主对角线元素进行取倒数操作,行④~⑨将矩阵A广播进向量寄存器,行⑩~⑫将规模3×8的矩阵B分散取值到向量化寄存器,行⑬~⑱是求解三角矩阵的四则运算,行⑲⑳将计算结果存入矩阵B,其完成了一个简易的TRSM计算(这里以LNLN模式下3×3的矩阵A和3×8的矩阵B中TRSM求解问题为例).
算法1. 双精度实数LNLN类型向量化计算.
输入:标量A1,A2,A3,A4,A5,A6,向量B1,B2,B3;
输出:结果向量B1(X1),B2(X2),B3(X3).
① XMM1 ← 1.0/A1;
② XMM3 ← 1.0/A3;
③ XMM6 ← 1.0/A6;
④ ZMM1 ← broadcast(XMM1);
⑤ ZMM3 ← broadcast(XMM3);
⑥ ZMM6 ← broadcast(XMM6);
⑦ ZMM2 ← broadcast(A2);
⑧ ZMM4 ← broadcast(A4);
⑨ ZMM5 ← broadcast(A5);
⑩ ZMM22 ← gather(B1);
⑪ ZMM23 ← gather(B2);
⑫ ZMM24 ← gather(B3);
⑬ ZMM22 ← ZMM22 × ZMM1;
⑭ ZMM23 ← ZMM23–ZMM2 × ZMM22;
⑮ ZMM23 ← ZMM23 × ZMM3;
⑯ ZMM24 ← ZMM24–ZMM4 × ZMM22;
⑰ ZMM24 ← ZMM24–ZMM5 × ZMM23;
⑱ ZMM24 ← ZMM24 × ZMM6;
⑲ scatter(B1,ZMM22);
⑳ scatter(B2,ZMM23);
㉑ scatter(B3,ZMM24).
双精度复数与双精度实数差异不大. 在图5中,记A1为1个实部和1个虚部构成的复数,记$ \boldsymbol{B}1 $是1行实部和1行虚部组成的1行复数,定义解矩阵$ \boldsymbol{X}1 $的数据分布形式与$ \boldsymbol{B}1 $相同. 在不考虑α时, $ \boldsymbol{X}1=\boldsymbol{B}1/ A1 $,以此类推,$ \boldsymbol{X}2=(\boldsymbol{B}2-A2\times \boldsymbol{X}1)/A3 $,$ \boldsymbol{X}3=(\boldsymbol{B}3-A4\times \boldsymbol{X}1-A5\times \boldsymbol{X}2)/A6 $. 以上是复数矩阵计算涉及到的公式,其形式、原理与双精度实数相同.
双精度复数计算通过“实部和虚部分离”最大程度对数据进行向量化操作. 算法2给出了复数矩阵A规模“3×3”、矩阵B规模“3×8”的主要计算步骤. 输入A1, A2, …, A6的实部、虚部(其中A1,A3,A6等主对角线元素已完成取倒数操作),依次按照双精度64位广播到ZMM2,ZMM3,ZMM4,ZMM5,ZMM8,ZMM9,ZMM6,ZMM7,ZMM10,ZMM11,ZMM12,ZMM13中. $ \boldsymbol{B}1,\boldsymbol{B}2,\boldsymbol{B}3 $的实部、虚部依次取入ZMM22,ZMM23,ZMM24,ZMM25,ZMM26,ZMM27中,其他计算步骤如算法2所示,行①②计算出$ \boldsymbol{X}1 $,行⑤~⑩计算出$ \boldsymbol{X}2 $,行⑬~㉒计算出$ \boldsymbol{X}3 $,其余伪代码负责存储.
算法2. 双精度复数LNLN类型向量化计算.
输入:复数标量A1,A2,A3,A4,A5,A6, 向量 B1,B2,B3,B4,B5,B6;
输出:向量B1,B2,B3,B4,B5,B6(依次存储X1,X2,X3,X4,X5,X6).
① ZMM14=ZMM2×ZMM22+ZMM3×ZMM23;
② ZMM15=ZMM3×ZMM22+ZMM2×ZMM23;
③ scatter([rdx],ZMM22 ← ZMM14);
④ scatter([rdx+0x8],ZMM23 ← ZMM15);
⑤ZMM14=ZMM4×ZMM22–ZMM5×ZMM23;
⑥ ZMM15=ZMM5×ZMM22+ZMM4×ZMM23;
⑦ ZMM24=ZMM24–ZMM14;
⑧ ZMM25=ZMM25–ZMM15;
⑨ ZMM14=ZMM8×ZMM24–ZMM9×ZMM25;
⑩ ZMM15=ZMM9×ZMM24+ZMM8×ZMM25;
⑪ scatter ([rdx+0x10],ZMM24 ← ZMM14);
⑫ scatter ([rdx+0x18],ZMM25 ← ZMM15);
⑬ ZMM14=ZMM6×ZMM22–ZMM7×ZMM23;
⑭ ZMM15=ZMM7×ZMM22+ZMM6×ZMM23;
⑮ ZMM26=ZMM26–ZMM14;
⑯ ZMM27=ZMM27–ZMM15;
⑰ ZMM14=ZMM10×ZMM2–ZMM11×ZMM25;
⑱ ZMM15=ZMM11×ZMM24+ZMM10×ZMM25;
⑲ ZMM26=ZMM26–ZMM14;
⑳ ZMM27=ZMM27–ZMM15;
㉑ZMM14=ZMM12×ZMM26–ZMM13×ZMM27;
㉒ZMM15=ZMM13×ZMM26+ZMM12×ZMM27;
㉓ scatter ([rdx+0x20], ZMM26 ← ZMM14);
㉔ scatter ([rdx+0x28],ZMM27 ← ZMM15).
向量化计算方式极大限度地发挥硬件体系结构优势. 算法1伪代码展示了矩阵B规模为“3×8”,倘或矩阵B是规模“3×96”的“长条”矩阵,矩阵B第1个分块后面11个“3×8”分块计算时,矩阵A无需再次加载. 这种硬件优势在算法2上表现也是一致的.
除4.4节所述向量化计算指令外,本文TRSM还包含取值、流控制、存储等指令操作.
取值阶段负责将数据加载至特定向量寄存器. 对于三角矩阵A,其取值方式为广播操作,每一个数据填充1个寄存器,在寄存器不够用时,少许数据使用嵌入式广播操作. 对于矩阵B,如图9所示,若计算机系统是列主序存储,涉及转置、非转置问题,至少有1个方向是不连续的,对于图9中非连续方向上的数据,我们采用分散取值的指令“vgatherqpd”收集到寄存器中,对于连续方向上的数据,可以采用“vmovupd/vmovapd”等指令连续取值到寄存器. 对于连续、非连续边界,都有特定的微内核,其取值可采用K寄存器掩码取值方式或者直接改变计算方式适应边界. 存储阶段是取值阶段的反向操作,使用指令的思路是一致的,存取方向是相反的.
流控制主要完成条件判断、跳转和循环控制. 由上文可知,我们设计了仅需矩阵B分块的内核和矩阵A,B皆需分块的内核. 如图10中①所示LNLN类型下,对于仅需矩阵B分块的内核,其流控制只涉及规模参数n维度,m维度对于我们来说是已知的,此时仅需要条件判断和跳转指令完成规模n维度循环,在循环过程中,矩阵A保留在向量寄存器中进行复用. 图10中②展示了矩阵A,B皆需分块的内核,其规模m,n对于内核都是未知的,其流控制需要负责2个维度上的条件判断和跳转,指令的使用变得更加精细和复杂,但本质并没有改变,只是将其通过分块转化成“长条”(或“瘦高”)形状计算,此时矩阵A已无法一次性放入向量寄存器,但其三角分块可保留在寄存器中,依然可在矩阵B通过分块形成“长条”(或“瘦高”)形状计算时进行复用.
第2~4节对小规模非规则TRSM的工作已进行介绍,经过参考其他的评估方法[21-23],下面我们给出针对本文工作的评估方案.
本文在x86架构平台测试,软硬件具体指标如表2所示,另外,选取了x86架构上适配性和运行效果最佳的MKL商业库进行对比.
本文对双精度实数TRSM的评估思路如下:对TRSM规模参数m≤100(n=96)的样本点测试作为“小规模非规则长条数据特征”的评估;对TRSM规模参数n≤100(m=96)的样本点测试,作为“小规模非规则瘦高数据特征”的评估;对TRSM规模参数m=n≤100的样本点测试,作为“小规模数据特征”的评估.
本文对双精度复数TRSM的评估思路如下:复数包含实部和虚部(等同2个实数),因此本文在测试规模上进行收缩. 对TRSM规模参数m≤50(n=48)的样本点进行测试,作为“小规模非规则长条数据特征”的评估;对TRSM规模参数n≤50(m=48)的样本点进行测试,作为“小规模非规则瘦高数据特征”的评估;对TRSM规模参数m=n≤50的样本点进行测试,作为“小规模数据特征”的评估.
本文处理的数据都是双精度浮点数,采用每秒执行10亿次浮点运算数(GFLOPS)衡量性能. 为避免偶然性,本文实验数据随机生成,衡量时采用均值.
按照5.1节中的评估方案,实验结果如图11、图12所示. 图11包含双精度实数小规模非规则TRSM和MKL各类型性能,图12包含双精度复数小规模非规则TRSM和MKL的各类型性能.
图11展示双精度实数小规模非规则TRSM和MKL的各类型性能. 图11(a)点线图中,规模参数n=96,规模参数m是自变量,该图展示LNLN,LTLN,LNUN,LTUN类型在“小规模非规则长条形状”数据上的浮点性能;图11(b)点线图中,规模参数m=n≤100,规模参数m=n是自变量,该图展示LNLN,LTLN,LNUN,LTUN类型在“小规模方块形状”数据上的浮点性能;图11(c)点线图中,规模参数m=96, 规模参数n是自变量,该图展示RNLN,RTLN,RNUN,RTUN类型在“小规模非规则高瘦形状”数据上的浮点性能;图11(d)点线图中,规模参数m=n≤100,规模参数m=n是自变量,该图展示RNLN,RTLN,RNUN,RTUN类型在“小规模方块形状”数据上的浮点性能.
图12展示双精度复数小规模非规则TRSM和MKL的各类型性能. 图12(a)点线图中,规模参数n=48,规模参数m是自变量,该图展示LNLN,LTLN,LCLN,LNUN,LTUN,LCUN类型在“小规模非规则长条形状”数据上的浮点性能;图12(b)点线图中,规模参数m=n≤50,规模参数m=n是自变量,该图展示LNLN,LTLN,LCLN,LNUN,LTUN,LCUN类型在“小规模方块形状”数据上的浮点性能;图12(c)点线图中,规模参数m=48,规模参数n是自变量,该图展示RNLN,RTLN,RCLN,RNUN,RTUN,RCUN类型在“小规模非规则高瘦形状”数据上的浮点性能;图12(d)点线图中,规模参数m=n≤50,规模参数m=n是自变量,该图展示RNLN,RTLN,RCLN,RNUN,RTUN,RCUN类型在“小规模方块形状”数据上的浮点性能.
图11(a)点线图中,相应类型TRSM在“小规模非规则长条形状”数据上相对MKL平均每个样本点加速13.5倍. 图11(b)点线图中,相应类型TRSM在“小规模方块形状”数据上相对MKL平均每个样本点加速46.6倍. 图11(c)点线图中,相应类型TRSM在“小规模非规则高瘦形状”数据上相对MKL平均每个样本点加速16.1倍. 图11(d)点线图中,相应类型TRSM在“小规模方块形状”数据上相对MKL平均每个样本点加速33.4倍.
图12(a)点线图中,相应类型TRSM在“小规模非规则长条形状”数据上相对MKL平均每个样本点加速19.9倍. 图12(b)点线图中,相应类型TRSM在“小规模方块形状”数据上相对MKL平均每个样本点加速50.9倍. 图12(c)点线图中,相应类型TRSM在“小规模非规则高瘦形状”数据上相对MKL平均每个样本点加速33.6倍. 图12(d)点线图中,相应类型TRSM在“小规模方块形状”数据上相对MKL平均每个样本点加速74.5倍.
根据以上评估数据,本文TRSM工作在SI区域(具备小规模非规则特征)相对于x86平台适配性最佳的MKL各类型算法取得了较大性能提升. 另外,根据图11、图12可知,MKL各类型算法性能在SI区域表现不佳,但在规模m或n趋近SI区域边界时逐渐获得性能优势,根据其上升趋势,随着规模继续增大取得较好效果. 故本文工作旨在解决小规模非规则TRSM问题,对于规模较大的TRSM计算,本文对业界已有BLAS库的卓越成果表示肯定.
上面已经针对双精度实数、复数TRSM各类型适配小规模非规则特征进行样本评估. 为保证我们工作成果的通用性,在各类型下,让规模参数m取值1~100,参数n取值1~100,每种类型m,n皆排列组合10000个样本,基本完全覆盖各种可能的小规模非规则情况,相对于MKL同类算法性能,我们完成平均加速统计如表3所示. 在双精度小规模非规则实数上平均性能提高29.4倍,在双精度小规模非规则复数上平均性能提高24.6倍.
本文旨在解决传统TRSM算法在处理小规模非规则数据时性能不佳的问题. 我们结合硬件特征、应用场景,分析研究了传统算法在该问题上的不足,提出了适合小规模非规则TRSM的自适应优化方案,从寄存器分块、边界处理、向量化计算角度实现了汇编级别的高性能内核,介绍了实现与优化方法,最终构建了完整规范的双精度小规模非规则SI_TRSM算法库.
未来,我们将研究实现单精度小规模非规则TRSM算法,进一步研究和优化TRSM算法在不同体系结构、应用场景和数据特征下的性能. 同时,我们还将关注BLAS领域中其他算法的个性化需求,以实现算法思路的迁移、扩展和创新.
作者贡献声明:郭容园完成代码实现、开展实验和撰写论文;贾海鹏指导研究方法,确立论文结构和协助论文修改;韦存阳指导代码的实现及评估;张云泉、邓明森审阅和指导论文;陈婧蕊、周振亚参与界定研究问题及审核实验结果.

参考文献 引证文献
排序方式:
1
BLAS Forum. BLAS: Basic linear algebra subprograms[EB/OL]. [2023-05-28]. http://netlib.org/blas/
2
Choi J, Dongarra J J, Walker D W. PB-BLAS: A set of parallel block basic linear algebra subprograms[J]. Concurrency: Practice and Experience, 1996, 8(7): 517−535
3
Intel. Intel oneAPI math kernel library[EB/OL]. [2023-05-21]. http://www.intel.com/content/www/us/en/developer/tools/oneapi/onemkl.html
4
Zhang Xianyi. OpenBLAS: An optimized BLAS library[EB/OL]. [2023-06-13]. http://www.openblas.net/
5
AMD. AMDLIB (AMD math library): A high-performance mathematical function library[EB/OL]. [2023-06-29]. https://www.amd.com/en/developer/aocl/libm.html
6
McNamee J M. Algorithm 408: A sparse matrix package (part I)[J]. Communications of the ACM, 1971, 14(4): 265−273
7
Dongarra J J, Du C J, Hammarling S, et al. Algorithm 679: A set of level 3 basic linear algebra subprograms: Model implementation and test programs[J]. ACM Transactions on Mathematical Software, 1990, 16(1): 18−28
8
Hanson R. PC basic linear algebra subroutines[EB/OL]. [2023-06-15]. https://www.osti.gov/biblio/1230151
9
Choi J, Walker D W, Dongarra J J. PUMMA: Parallel universal matrix multiplication algorithms on distributed memory concurrent computers.[J]. Concurrency: Practice and Experience, 1994, 6(7): 543−570
10
Dongarra J J. Preface: Basic linear algebra subprograms technical (blast) forum standard[J]. The International Journal of High Performance Computing Applications, 2002, 16(1): 1−1
11
Chan E, Van Zee F G, Bientinesi P, et al. Supermatrix: A multithreaded runtime scheduling system for algorithms-by-blocks[C]//Proc of the 13th ACM SIGPLAN Symp on Principles and Practice of Parallel Programming. New York: ACM, 2008: 123−132
12
Volkov V, Demmel J W. Benchmarking GPUs to tune dense linear algebra[C/OL]//Proc of the 2008 ACM/IEEE Conf on Supercomputing(SC’08). Piscataway, NJ: IEEE, 2008[2023-06-20]. https://ieeexplore.ieee.org/document/5214359
13
Soorishetty A. Accelerating Linear Algebra and Machine Learning Kernels on a Massively Parallel Reconfigurable Architecture[M]. Arizona: Arizona State University, 2019
14
Wei Cunyang, Jia Haipeng, Zhang Yunquan, et al. IATF: An input-aware tuning framework for compact BLAS based on ARMv8 CPUs[C/OL]//Proc of the 51st Int Conf on Parallel Processing. New York: ACM, 2022[2023-06-25].https://dl.acm.org/doi/10.1145/3545008.3545032
15
Valero-Lara P, Greenwalt C, Vetter J S. SparseLU, a novel algorithm and math library for sparse LU factorization[C]//Proc of the 2022 IEEE/ACM Workshop on Irregular Applications: Architectures and Algorithms (IA3). Piscataway, NJ: IEEE, 2022: 25−31
16
胡怡, 陈道琨, 杨超, 等. 国产SW26010-Pro处理器上3级BLAS函数众核并行优化[J]. 软件学报,2023,35(3):1569−1584
Hu Yi, Chen Daokun, Yang Chao, et al. Many-core parallel optimization of level–3 BLAS function on domestic SW26010-Pro processor[J]. Journal of Software, 2023, 35(3): 1569−1584 (in Chinese)
17
Whaley R C, Petitet A, Dongarra J J. Automated empirical optimizations of software and the ATLAS project[J]. Parallel Computing, 2001, 27(1/2): 3−25
18
Lu H, Matz M, Girkar M, et al. System V application binary interface AMD64 architecture processor supplement (with LP64 and ILP32 programming models) version 1.0[EB/OL]. [2023-06-28]. https://cs61.seas.harvard.edu/site/pdf/X86-64-abi-20210928.pdf
19
Amiri H, Shahbahrami A. SIMD programming using Intel vector extensions[J]. Journal of Parallel and Distributed Computing, 2020, 135: 83−100
20
Kusswurm D. Modern X86 Assembly Language Programming: Covers X86 64-bit, AVX, AVX2, and AVX−512[M]. Berkeley, CA: Apress, 2018: 421−432
21
Li Chendi, Jia Haipeng, Cao hang, et al. Autotsmm: An auto-tuning framework for building high-performance tall-and-skinny matrix-matrix multiplication on CPUs[C]//Proc of the 2021 IEEE Intel Conf on Parallel &amp; Distributed Processing with Applications, Big Data &amp; Cloud Computing, Sustainable Computing &amp; Communications, Social Computing &amp; Networking (ISPA/BDCloud/SocialCom/SustainCom). Piscataway, NJ: IEEE, 2021: 159−166
22
Yang Weiling, Fang Jianbin, Dong Dezun, et al. LIBSHALOM: Optimizing small and irregular-shaped matrix multiplications on ARMv8 multi-cores[C/OL]//Proc of the Int Conf for High Performance Computing, Networking, Storage and Analysis. New York: ACM, 2021[2023-06-26].https://dl.acm.org/doi/10.1145/3458817.3476217
23
Sorokin A, Malkovsky S, Tsoy G. Comparing the performance of general matrix multiplication routine on heterogeneous computing systems[J]. Journal of Parallel and Distributed Computing, 2022, 160: 39−48
2025年第62卷第2期
146
71
0
引用本文
BibTeX
文章信息
doi: 10.7544/issn1000-1239.202330864
  • 接收时间:2023-10-31
  • 首发时间:2026-11-10
  • 出版时间:2025-02-01
补充材料
相关文章
文章信息
作者
出版历史
  • 收稿日期:2023-10-31
  • 修回日期:2024-02-27
基金
作者信息
    1贵州财经大学信息学院 贵阳 550025
    2中国科学院计算技术研究所 北京 100190
    3北京华大九天科技股份有限公司 北京 100102

通讯作者:

贾海鹏()
参考文献
分享链接
https://castjournals.cast.org.cn/joweb/jsjyjyfz/CN/10.7544/issn1000-1239.202330864
分享至
全文二维码

扫描看全文

引用本文
BibTeX
本文的引用情况
2种不同金属材料的力学参数
科
Family
属数
Number of
genus
种数
Number of
species
占总种数比例
Percentage of
total species (%)
属
Genus
种数
Number of
species
占总种数比例
Percentage of total
species (%)
鹅膏菌科Amanitaceae 2 11 5.26 鹅膏菌属 Amanita 10 4.78
小菇科 Mycenaceae 2 12 5.74 丝盖伞属 Inocybe 5 2.39
多孔菌科 Polyporaceae 8 14 6.70 蜡蘑属 Laccaria 5 2.39
红菇科 Russulaceae 3 23 11.00 小皮伞属 Marasmius 6 2.87
小菇属 Mycena 11 5.26
光柄菇属 Pluteus 5 2.39
红菇属 Russula 17 8.13
栓菌属 Trametes 5 2.39
关闭全屏