大規模アプリケーションの高性能な実用的アクセラレータ対応手法
下川辺　隆史(東京大学　情報基盤センター)

This project focuses on porting CPU-based applications to accelerator-equipped supercomputers and developing efficient methodologies for this transition. Targeting the upcoming Miyabi-G supercomputer, equipped with NVIDIA GH200, the project aims to optimize the seismic wave simulation code OpenSWPC and other applications using standard methods, ensuring ease of use for non-HPC researchers. Key objectives include evaluating Kokkos, a performance portability library, against OpenMP, OpenACC, and SYCL, and exploring its use for both C++ and Fortran applications. By utilizing GH200’s tightly coupled GPU-CPU architecture, the project aims to enhance communication, IO, and overall application performance while maintaining productivity and portability.

研究紹介ポスター／最終報告書

(1) 学術論文 （査読あり）
[]Yuuichi Asahi, Thomas Padioleau, Paul Zehner, Julien Bigot, Damien Lebrun-Grandie, 2025, kokkos-fft: A shared-memory FFT for the Kokkos ecosystem, Journal of Open Source Software, 10 (111), 8391
(2) 国際会議プロシーディングス （査読あり）
[]Yuuichi Asahi, Trévis Morvany, Thomas Padioleau, Julien Bigot, 2025, Development of a performance portable distributed FFT interface on top of the Kokkos ecosystem, Proceedings of the SC '25 Workshops of the International Conference for High Performance Computing, Networking, Storage and Analysis, 1233-1242
[]Ziheng Yuan, Takashi Shimokawabe, 2025, Accelerating LBM with C++ STL Asynchronous Parallel Model, Lecture Notes in Computer Science Computational Science – ICCS 2025, 249-256
(3) 国際会議発表（査読なし）
[]Yohei Miki, 2026, Solomon: unified schemes for directive-based GPU offloading, OpenACC User Workshop at SCA2026,
[]Ziheng Yuan, Takashi Shimokawabe, Accelerating stencil computation by using Tensor Core, 4th International Conference on Computational Engineering and Science for Safety and Environmental Problems (COMPSAFE2025), Kobe, Japan, 2025.
[]Su Jipeng, Shimokawabe Takashi, Yuan Ziheng, Performance Comparison of Kokkos-Based and CUDA/OpenACC Lattice Boltzmann Solvers, SCA/HPC Asia 2026, Osaka, Japan, 2026 (poster).
[]Yize Yang, Takashi Shimokawabe, Offloading the IBM Workloads for Efficient LBM Fluid Simulations on Grace Hopper, SCA/HPC Asia 2026, Osaka, Japan, 2026 (poster).
[]Tao Wang, Takashi Shimokawabe, Empowering Ozaki Scheme with Hopper Architecture, SCA/HPC Asia 2026, Osaka, Japan, 2026 (poster, Best Student Poster Award ).
(4) 国内会議発表（査読なし）
[]三木 洋平, 2026, Pathways to vendor-neutral GPU computing for sustainable simulation code development, 「富岳成果創出加速プログラム」基礎科学合同シンポジウム 2025,
[]三木 洋平, 2026, 多様なプログラミング手法を用いたN体計算コードのGPU実装：NVIDIA GH200およびAMD MI300A上での性能比較, 日本天文学会 2026年春季年会
[]三木 洋平, 2026, GPU向け指示文統合マクロライブラリSolomon, ワークショップ「アプリ開発者のためのアクセラレータプログラミング最新情報」
[]三木 洋平, 塙 敏博, 2026, N体計算におけるGPUプログラミング手法比較：NVIDIA GH200/B200およびAMD MI300Aでの性能評価, 情報処理学会研究報告, Vol.2026-HPC-203 No.47 (10pp)
[]下川辺 隆史, スパコンMiyabiとGPUアプリケーション開発, 日本応用数理学会 2025年度 年会, 東京, 2025.
(5) 公開したライブラリなど
該当なし
(6) その他（特許，プレスリリース，著書等）
該当なし