Go to the Indico Home Page Oldest event Older event Up to category Newer event Newest event Printable version
 
Focus on:  Hide Contributions
  • Compact style
  • Indico style
  • Indico style - inline minutes
  • Indico style - numbered
  • Indico style - numbered + minutes
  • Indico Weeks View
    • Switch to the management area of this event
    Login

    Madgraph5 GPU development - !! ATTENTION Tuesday 16:00 CEST !!

    → Europe/Zurich
    Virtual (Zoom)

    Virtual (Zoom)
    Zoom URL
    Stefan Roiser
    Hide

    # Madgraph meeting Tue 20.07.2021

    Present: Carl, Taylor, Andy, AV, SR, Walter (later), Olivier (later)
    Apologies: Josh

    ## Round table

    Carl: busy with CMS, ntr

    Taylor: not much, fixed json file for Andy's tool

    Andy: thanks to Taylor, the tool can now show results also for kokkos which is very nice!
    Unfortunaly cannot yet show them today.
    Thesis defense is next week on Wed, https://indico.cern.ch/event/1060793

    Stefan: not a lot on my side, helped Andy build Kokkos

    AV: showed some slides
    - SR: could we test Power9 also with Andy? AV: will contact Tommaso
    - TC: any effect on Cuda? AV: not sure, should check
    - TC/OM: some discussion of vectorization and event vs helicity parallelism
    - SR: should also try the compiler optimizations (auto-FDO) that Nathalie had tried
      https://indico.cern.ch/event/587970/contributions/2369824/attachments/1374948/2087355/slides.pdf
      (OM: feedback directed is mainly with a lot of branches, not our case)
    - OM: reweighting is not MadWeight, it is a reweighting package in Madgraph

    WH: ntr, was on holiday

    OM: ntr, also was on holiday or working on other projects
    Status of Fortran madevent: essentially ready for integration,
    but there is a bug to analyse in the rapidity cuts.
    The problem is that the cuts have moved.
    The code is not crashing, but it is producing wrong results.
    - SR: if it is just giving wrong results, maybe we can try an interface?
      OM: maybe yes, but need to clean up the API first
    - TC: would be interested in calling Kokkos kernels from MadEvent
      TC: can we have a meeting for that?
      OM: yes, can even be a mini workshop or tutorial
      AV: we could already decide an API, same for cuda/c++ or kokkos
      OM: actually two things, the API should be fast,
      the mini workshop is about automatic code generation

    Date for this workshop?
    AV: away Aug ~1st to ~10 September
    OM: away at a school September 6-15
    OM: maybe Thu 2nd September? SR: busy with C++ school
    TC: say Thu 16 September 3pm-5pm GVA time - OK approved

    ## Next meeting

    Friday 30 July 3pm!

    Dont forget Andy's thesis on Wed July 28!

     

    There are minutes attached to this event. Show them.