Skip to content

Conference call notes 20210106

Davide Vanzo edited this page Jan 6, 2021 · 6 revisions

(back to Conference calls)

Notes on the 164th EasyBuild conference call, Wednesday January 6th 2021 (16:00 UTC - 17:00 CET)

Attendees

Alphabetical list of attendees (11):

  • Sebastian Achilles (Jülich Supercomputing Centre, Germany)
  • Simon Branford (University of Birmingham, UK)
  • Alex Domingo (Vrije Universiteit Brussel, Belgium)
  • Fotis Georgatos (SDSC, Switzerland)
  • Kenneth Hoste (HPC-UGent, Belgium)
  • Adam Huffman (Big Data Institute, Oxford, UK)
  • Terje Kvernes (University of Oslo, Norway)
  • Kurt Lust (Univ. of Antwerpen - LUMI)
  • Robert Mijakovic (LuxProvide)
  • Bart Oldeman (ComputeCanada)
  • Jörg Saßmannshausen (NIHR Biomedical Research Centre, UK)

Agenda

  • update on recent developments
  • Q&A

Recent developments

  • next release (v4.3.3): before EUM'21?
  • recent changes
    • framework
      • bug fixes
        • add --init and --recursive options to 'git submodule update' command that is used when creating source tarball for specific commit (PR #3537)
        • filter out duplicate paths in RPATH wrapper script (PR #3538)
      • enhancements
        • add support for --accept-eula configuration option + 'accept_eula' easyconfig parameter (PR #3535)
          • used for AOCC, will also be useful for Intel oneAPI (and maybe also NVHPC, see PR #11920)
    • easyblocks
      • bug fixes
        • only configure PETSc with --with-numpy or --with-mpi4py if those Python packages are available, and only if Python is not a build-only dependency (PR #2299)
        • fix handling of build deps in PETSc/SLEPc/Trilinos easyblocks (PR #2299, PR #2300, PR #2301)
      • enhancements
        • set $TF_GPU_COUNT and $TF_TESTS_PER_GPU for TensorFlow tests (PR #2292)
        • update BerkeleyGW (PR #2297) and QuantumESPRESSO (PR #2298) easyblocks to fix compilation issues with GCC 10.x
      • new software
        • custom easyblock for AOCC (PR #)
    • easyconfigs
      • bug fixes
        • fix source URL in expat easyconfigs (and consistently add custom sanity_check_paths) (PR #11960)
      • enhancements
      • new software
      • software updates
      • changes
        • (nothing special)
  • to merge/fix/tackle soon
    • framework (v4.3.3 milestone)
    • easyblocks (v4.3.3 milestone)
      • bug fixes
        • correctly determine path to active binutils in TensorFlow easyblock (PR #2218)
        • fix taking into account --sysroot when installing/using CMake
          • two options:
            • use toolchain file in CMakeMake (PR #2247)
            • patch CMake installation (PR #2248)
      • enhancements
        • enhance OpenBLAS easyblock to make it aware of optarch (PR #1946)
          • needs testing...
        • add support for statically linking Bazel (PR #2272)
        • set $PYTHONNOUSERSITE in PythonBundle.extensions_step (PR #2272)
        • improve Bazel easyblock: add support for running tests, enabl static linking, use build dir rather than tmpdir, verbose output (PR #2285)
        • add support for skipping steps in Extension PythonPackages (PR #2290)
      • changes
        • create less temporary directories for TensorFlow by (only) using --output_user_root (PR #2293)
      • new software
    • easyconfigs (v4.3.3 milestone)
      • bug fixes
        • added missing space in configopts in ParaView 5.8.0 easyconfigs using 2020a toolchain (PR #10989)
          • (cfr. discussion in previous conf calls)
      • enhancements
        • (nothing special)
      • new software
        • Intel oneAPI
          • there will be no further updates to Intel Parallel Studio (apparently)
      • software updates

Q&A

  • Sebastian: AMD toolchains
    • what should be in it as compiler: AOCC or Clang (+ flang) + AMD LibM
      • performance differences between AOCC and standard Clang seem small
      • Clang-based toolchain is more portable (and open source)
      • AOCC may be used for vendor support reasons
    • should we consult AMD or George @ LUMI?
      • cfr.
    • Kurt: similar question w.r.t. MPI library
      • we're currently focused a lot on OpenMPI
      • but we should consider MPICH too, for performance and compability with vendor MPIs (Cray MPI, Intel MPI, etc.)
    • Robert: ATOS seems to favor OpenMPI
    • Bart: AMD-optimized HPL is built on top of OpenMPI
    • do we need to reconsider the approach of common toolchains going forward?
      • increasing variety in CPUs (Intel, AMD, Arm, POWER, ...)
      • probably implies we need more diverse toolchains w.r.t. compiler & libraries
      • another Clang-based common toolchains?
        • since NVHPC, AOCC, Intel oneAPI, Cray are all based on Clang/LLVM...
        • MPI should be MPICH2 due to arguments above
          • what about MVAPICH2? (ping Xavier)
Clone this wiki locally