菱沼 利彰 (Toshiaki Hishinuma)

Current Status

  • SB Intuitions株式会社，シニアHPCエンジニア
    (Senior HPC Engineer, SB Intuitions Corp.)
  • 筑波大学 図書館情報メディア研究科 博士 (情報学)
    (Ph.D. in Informatics, Univ. of Tsukuba)
  • 情報処理学会 HPC研究会 運営委員
    (Steering committee member, SIG-HPC)
  • 情報処理学会 正会員
    (Member, Information Processing Society of Japan)

SUMMARY

I am an HPC software engineer with a Ph.D. in Informatics. I have 5+ years of experience developing linear algebra libraries for accelerating CAE simulation and machine learning. My target applications are the Krylov subspace method, which is one of solutions of the linear equations, and the graph convolutional network, one of the machine learning methods. I specialize in implementing and optimizing these kernels, sparse matrix operations, in high- and low-precision. I optimized these operations for various hardware using SIMD, CUDA, OpenCL, and OpenMP Offloading. I have not only implemented these functions but also considered easy-to-use APIs, developed them as libraries, and released them as OSS with a build script, test code, benchmark programs, container images, CI/CD settings, and documents.

Domain of expertise

  • High performance computing for Linear algebra
    • Sparse matrix operations
    • BLAS, LAPACK
    • Lis, Eigen, ARPACK, PETSc
    • Krylov subspace methods (for liner equations)
  • High precision arithmetics
    • Double-Double precision
    • GMP, MPFR
  • Optimize CAE Softwares
    • FrontISTR
    • OpenFOAM

Skills

  • Programming Languages
    • C/C++, x86 / 64 asm (well)
    • MATLAB, Scilab, C#, Vlang (previously used)
    • Rust, Fortran (normal)
    • Python, Ruby, Go (learning)
  • Parallelization
    • SIMD (AVX512, AVX2, AVX, ARM NEON, ARM SVE?)
    • Multi-threading (OpenMP, pthread)
    • Multi-Process (MPI)
  • Performance Tuning on Special Hardware
    • NVIDIA GPGPU (CUDA, OpenACC, OpenCL, OpenMP Offloading)
    • SX-Aurora TSUBASA
    • A64fx (ARM SVE)
    • PEZY-SC2 (PZCL)
    • SPARC64 VII
    • Xeon Phi
    • AMD APU (OpenCL)
  • etc:
    • zsh, tmux, vim, gnuplot, GNU Make, awk, sed…
    • Linux server (Ubuntu, RedHat)
    • Docker, KVM, Virtual Box
    • git, svn
    • github actions, gitlab-CI
    • LaTeX
    • HTML, Markdown, javascript (this blog is written in Markdown, and generated by Hugo)

Work Experience

  • 2024.10 - present SB Intuitions株式会社、プリンシパルHPCエンジニア、超並列計算基盤チーム リーダー
    (Principal HPC Engineer, Leader of HPC Team, SB Intuitions Corp.)

  • 2024.02 - 2024.09 SB Intuitions株式会社，シニアHPCエンジニア (出向比率100%)
    (Senior HPC Engineer, SB Intuitions Corp.)

  • 2024.02 - 2024.09 ソフトバンク株式会社，担当課長
    (Section manager, Softbank Corp.)

  • 2023.01 - 2024.01 Software Engineer, Braid Technologies Inc.
    (ブレイドテクノロジーズ株式会社，ソフトウェアエンジニア)

  • 2022.07 - 2022.12 天進技術株式会社 (OPPO日本研究所), 顧問
    (Software Engineering Adviser, OPPO Japan Research Center)

  • 2020.07 - 2022.06 株式会社 科学計算総合研究所, 主任研究員
    (Chief Researcher, Research Institute for Computational Science Co. Ltd. (RICOS Co. Ltd.))

  • 2019.05 - 2020.06 株式会社 科学計算総合研究所, 研究員
    (Researcher, Research Institute for Computational Science Co. Ltd. (RICOS Co. Ltd.))

  • 2017.04 - 2019.04 株式会社 PEZY Computing, ソフトウェアエンジニア
    (Software Engineer, PEZY Compuing K.K.)

  • 2016.04 - 2017.03 リサーチアシスタント (筑波大学) , 「数値計算ライブラリによる 超並列複合システムの階層的抽象化に関する研究
    (Research Assistant (U. of Tsukuba), “A study of hierarchical abstraction by numerical library for massively-parallel heterogeneous systems”)

  • 2015.06 - 2015.12 リサーチアシスタント (筑波大学) , 「大規模なユーザ生成コンテンツ集合からの知識発見
    (Research Assistant (U. of Tsukuba), “Knowledge Discovery from Large-scale User-generated Content Set”)

Education

  • 2020.03 筑波大学 図書館情報メディア研究科 図書館情報メディア専攻 修了 (情報学博士)
    (Doctor of philosophy in informatics, Univ. of Tsukuba)

  • 2015.04 筑波大学　図書館情報メディア研究科 図書館情報メディア専攻 入学 (Hasegawa Lab)
    (Doctoral Program, Library, Information and Media Studies, University of Tsukuba : matriculate)

  • 2015.04 - 2018.03 工学院大学 客員研究員
    (Visiting researcher, Kogakuin Univ.)

  • 2015.03 工学院大学 工学研究科 情報学専攻 修了(情報学修士)
    (Major of Informatics, Graduate School, Kogakuin Univ. : graduate, Master of Informatics)

  • 2013.04 工学院大学 工学研究科 情報学専攻 入学
    (Major of Informatics, Graduate School, Kogakuin Univ. : matriculate)

  • 2013.03 工学院大学情報学部情報デザイン学科 卒業
    (Department of Information Design, Faculty of informatics, Kogakuin Univ : graduate)

  • 2009.04 工学院大学情報学部情報デザイン学科 入学
    (Department of Information Design, Faculty of informatics, Kogakuin Univ : matriculate)

OSS Development

  • monolish: MONOlithic LInear equation Solvers for Highly-parallel architecture
    • Monolithically fuses data type, matrix structures, matrix data format, vendor-specific data transfer APIs, and vendor-specific linear algebra library APIs written in C++.
    • It is a monolithic C++ numerical library designed to bridge the gap between HPC engineers and application engineers. It provides an API that integrates the numerical algebra libraries of each vendor (MKL, CUDA Libraries, OpenBLAS, Etc.).
    • monolish solves cumbersome package management by Docker.
    • GitHub: https://github.com/ricosjp/monolish
  • DD-AVX: Library of High Precision Sparse Matrix Operations Accelerated by SIMD
    • SIMD accelerated simple interface high precision BLAS and Sparse BLAS library written in C++.
    • Multi-threading and SIMD AVX, AVX2, and AVX512 acceleration are available.
    • Provides an easy way to implement a fast and accurate Krylov subspace method.
    • GitHub: https://github.com/t-hishinuma/DD-AVX_v3

Other informations

  • 博士 (情報学)　(2020)
  • 修士 (情報学)　(2015)
  • 基本情報技術者 (2010)
  • 普通自動車免許 (2009)
  • 誕生 (birth) (1990/03/23)