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 15:00 CEST !!

    → Europe/Zurich
    Virtual (Zoom)

    Virtual (Zoom)
    Stefan Roiser
    Hide

    # Madgraph dev meeting Tue 12 October 2021

    Present: SR, AV, OM, LaurenceF, DavidS, TaylorC, Carl Vuosalo, Walter Hopkins (partly), Nathan Nichols
    Excused: StephanH

    ## Round table

    1) About the backport - and abstraction layers

    AV: show slides about epochX (cudacpp code generation) and also about a documentation wiki
    (A few questions and answers)

    WH: ntr

    TC: progressing on kokkos code generation, see github jtchilders/madgraph_kokkos_plugin
    (A few questions and answers)

    DS: working on Alpaka code generation, ongoing though not producing yet code that compiles

    Discussion - TC "we keep it messy now and clean up later", general agreement

    NN: have been working on sycl to get it up and running (but no code generation yet)
    Some trouble getting sycl cuda to work for Madgraph (maybe due to nvidia ptx code?)
    Maybe some issues with the toolchain itself rather than with MG (eg linking on sycl-cuda or sycl-cudacl)
    LF: I had some similar issues but got it to work, let's discuss offline
    NN: also discussed some points about Intel hardware (under NDA!) with Taylor

    LF: was waiting for golden tag, will then resume work on the sycl port
    AV: golden tag is now there (a frozen version - but developments go on in master...)
    LF: was not having code generation on my todo list however

    2) Fortran and AOB

    OM: ntr, busy with other things
    Implemented an overload of + and - in the plugin (eg was interested in fortran vectorization)

    CV: ntr, busy with CMS

    SR: merged a PR for cuda/fortran bridging studies, first bridge is done
    Calling a cuda function via C from Fortran, first this does a transposition of the array
    Still some functionalities missing from the Fortran however, may need to discuss with Olivier
    Working on a proton-proton process however, adapting this to only look at gg collisions
    OM: can also just use random numbers to replace some missing info if needed

    ## AOB

    Next meeting? Agree on Mon 25 October

    NN: do we have a mattermost for this team?
    No better use emails and the mailing list (or github issues)

    There are minutes attached to this event. Show them.