• Search Research Projects
  • Search Researchers
  • How to Use
  1. Back to project page

2014 Fiscal Year Final Research Report

Research on Software Foundations for General Purpose Automatic Tuning Mechanisms

Research Project

  • PDF
Project/Area Number 23240005
Research Category

Grant-in-Aid for Scientific Research (A)

Allocation TypeSingle-year Grants
Section一般
Research Field Computer system/Network
Research InstitutionThe University of Tokyo

Principal Investigator

SUDA Reiji  東京大学, 情報理工学系研究科, 教授 (40251392)

Co-Investigator(Kenkyū-buntansha) SATO Hiroyuki  東京大学, 情報基盤センター, 准教授 (20225999)
YAMAMOTO Yusaku  電気通信大学, 情報理工学研究科, 教授 (20362288)
IMAMURA Toshiyuki  独立行政法人理化学研究所, 計算科学研究機構, チームリーダー (60361838)
YOSHIZOE Kazuki  東京工業大学, 情報理工学研究科, 研究員 (80449115)
Co-Investigator(Renkei-kenkyūsha) KATAGIRI Takahiro  東京大学, 情報基盤センター, 准教授 (40345434)
FUJII Akihiro  工学院大学, 情報学部, 講師 (10383986)
Research Collaborator YUBA Toshitsugu  
YASUGI Masahiro  
OSAWA Noritaka  
Project Period (FY) 2011-04-01 – 2015-03-31
Keywords自動チューニング / 性能モデリング / 逐次実験計画 / 自動チューニング記述言語 / コンパイラ技術 / 軽量化 / 行列計算 / 大規模探索
Outline of Final Research Achievements

Aiming to establish Automatic Tuning (AT) as a general technology, we researched in the following four domains. In Math Domain, we developed and released AT mathematical core library ATMathCoreLib, and developed mathematical methods for AT with performance correlation. In Programming Domain, we combined ATMathCoreLib with AT description language ppOpen-AT, and proposed a method to cancel performance variation from code placements. In System Domain, we proposed architecture of general purpose Auto-Tuner, and created light-weight performance data collection mechanism. In Application Domain, we apply AT to matrix computations and large scale search, and apply ATMathCoreLib to some applications. Thus we researched on four domains in focus and in collaboration, and achieved research results toward general purpose AT methodology.

Free Research Field

ハイパフォーマンスコンピューティング

URL: 

Published: 2016-06-03  

Information User Guide FAQ News Terms of Use Attribution of KAKENHI

Powered by NII kakenhi