Second ITC-NU High Performance Computing Division International Seminar

第2回 名古屋大学情報基盤センター 大規模計算支援環境研究部門 国際セミナー

  Date   July 25 (Thu), 2024
  2024年7月25日(木)
  Venue   2F Lecture Room, Information Technology Center, Nagoya University
  名古屋大学情報基盤センター 2階演習室
  Financial Support   Kayamori Foundation of Information Science Advancement
  栢森情報科学振興財団
  Program
  • 13:00 - 13:10    Opening
  • Prof. Takahiro Katagiri
    High Performance Computing Division, Information Technology, Center, Nagoya University

  • 13:10 - 14:10    Invited Talk
  • Dr. Keita Teranishi
    Group Leader, Senior Scientist, Programming Systems, Oak Ridge National Laboratory

    Title: Toward Automatic Test Synthesis for Kokkos Performance Portable Programming Models

    Abstract: Performance Portable Programming Models have significantly improved productivity when it comes to leveraging various heterogeneous node architectures in high-performance computing (HPC). Nevertheless, these portability layers still contain intricacies in how they interact with target node architectures and their low-level runtime systems. This often demands expert knowledge from application developers to write programs that are both portable and free of errors.
    To address this challenge, we propose the use of program analysis tools and methodologies for Kokkos, a prominent performance portable programming framework. This approach eliminates the need for platform-specific testing to ensure the correctness of applications developed using performance portable programming frameworks.
    Firstly, we will introduce an automated analysis tool designed for Kokkos parallel programs. This tool employs guided symbolic execution through an LLVM-based Klee plugin. This automated analysis serves as an initial pass over the compiled program, aiding in the detection of bugs and acting as a cost-effective precursor to dynamic analysis. Secondly, we will explore how a collection of parallel programming examples, gathered from the community and categorized as either correct or incorrect, can enhance the feasibility of concolic analysis for parallel programs.

  • 14:10 - 14:20    Closing
最終更新:2024年5月30日