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

2022 Fiscal Year Final Research Report

Development of efficient arbitrary precision numerical linear computation library optimized for multi-core CPUs

Research Project

  • PDF
Project/Area Number 20K11843
Research Category

Grant-in-Aid for Scientific Research (C)

Allocation TypeMulti-year Fund
Section一般
Review Section Basic Section 60090:High performance computing-related
Research InstitutionShizuoka Institute of Science and Technology

Principal Investigator

Kouya Tomonori  静岡理工科大学, 情報学部, 教授 (80319152)

Project Period (FY) 2020-04-01 – 2023-03-31
Keywords多倍長精度浮動小数点演算 / 基本線形計算 / 最適化 / SIMD / OpenMP
Outline of Final Research Achievements

We have developed our original fast basic linear computation library supporting multiple-precision floating-point arithmetic, and already published it as open-source project. That includes multi-component-way fixed-precision arithmetic and optimized routines using AVX2, and multi-digit-way arbitrary precision arithmetic using C-native function call which can reduce overhead due to C++ class. For matrix multiplication, the Ozaki scheme and the Strassen algorithm have been implemented, so we have observed that the Ozaki scheme can be more efficient than Strassen algorithm in middle-precision arithmetic, and already published it as the international proceeding paper.

Free Research Field

高性能計算

Academic Significance and Societal Importance of the Research Achievements

我々の開発した多倍長精度最適化基本線形計算ライブラリは,標準的な多倍長精度線形計算ライブラリよりも高速であり,OpenMP, AVX2, 尾崎スキーム,Strassenアルゴリズムといった,あらゆる最適化手法を取り入れている。よって,大規模な悪条件問題をより高速に解くことができ,スーパーコンピュータからコンシューマ向けのノートパソコンまで利用することができるものである。Pythonモジュールとしても実験的に実装しており,深層学習結果の検証や,可視化ツールとの連携も可能であり,敷居の高い多倍長精度計算環境をカジュアルに,高価な商用ソフトウェアを介することなく利用できる。

URL: 

Published: 2024-01-30  

Information User Guide FAQ News Terms of Use Attribution of KAKENHI

Powered by NII kakenhi