Skip to content
Snippets Groups Projects

Extend tests for Allen

    • Open Issue created by Dorothea Vom Bruch @dovombru

    Physics performance: nightly as part of ci-test

    • Forward tracking (with UT), 100 events, minbias FTv6 sample
    • Seeding and matching, 100 events, minbias FTv6 sample
    • veloUT tracking, 100 events, minbias FTv6 sample
    • velo tracking, 100 events, minbias FTv6 sample
    • calo cluster efficiency, 1k events, Bd->K*gamma sample
    • calo cluster resolution, 1k events, Bd->K*gamma sample
    • switch existing track reconstruction tests to use BsPhiPhi sample, 1k events
    • Forward tracking w/o UT, 1k events, BsPhiPhi sample
    • MuonID efficiency for forward tracks, 1k events, ZMuMu sample
    • MuonID efficiency for seeding-matching tracks, 1k events, ZMuMu sample
    • PV efficiency on pp minbias and pHe-pp minbias samples
    • PV resolution on pp minbias and pHe-pp minbias samples

    Physics performance: PR dashboard nightly / weekly tests

    • Forward tracking with UT: efficiency versus various kinematic variables, track properties
    • Track reconstruction efficiency, with separate category for electrons
      • Forward tracking with and w/o UT, seeding and matching
        • long tracks versus eta, phi, p, pt
        • long from B versus eta, phi, p, pt
        • ghosts versus eta, nPVs, p, pt
      • veloUT tracks
        • veloUT tracks versus eta, phi, p, pt
        • long from B versus eta, phi, p, pt
        • ghosts versus eta, nPVs, p, pt
      • velo tracks
        • velo tracks versus eta, phi, p, pt
        • long from B versus eta, phi, p, pt
        • ghosts versus eta, nPVs, p, pt
    • momentum resolution (TrackResChecker) versus p
      • forward (with and w/o UT) tracks
      • seeding-matching tracks
    • IP resolution in x and y versus 1/pT
      • velo Kalman fitted tracks from forward (with and w/o UT)
      • velo Kalman fitted tracks from seeding-matching
    • Muon ID efficiency versus eta, p, phi for muons and pions
      • forward (with and w/o UT) tracks
      • seeding-matching tracks
    • electron ID efficiency versus eta, p, phi for muons and pions
      • forward (with and w/o UT) tracks
      • seeding-matching tracks
    • PV efficiency versus nTracks, r, z
    • PV resolution
      • mean x, y, z versus nTracks
      • mean x, y, z versus z
      • sigma x, y, z versus nTracks
      • sigma x, y, z versus z
    • calo cluster efficiency plots
    • calo cluster resolution plots
    • secondary vertex reconstruction efficiency plots
    • Brems recovery performance
    • charge asymmetry: #216

    Consistent results across architectures, input format, software framework

    Test ID Gaudi-Allen Allen event loop from MooreOnline Allen standalone events / slice architecture input type
    1 x 1 x86 DIGI
    2 x 1 x86 MDF
    3 x 1 x86 MDF
    4 x 1000 x86 MDF
    5 x 1000 cuda MDF
    6 x 1000 cuda MEP
    7 x 1000 cuda MDF
    8 x 1000 x86 MDF
    • compare velo, veloUT, forward, seeding-matching efficiency for 1k BsPhiPhi events
      • 1 versus 2: digi versus MDF in Gaudi-Allen -> expect same results
      • 2 versus 3: Gaudi-Allen versus Allen event loop from Gaudi -> expect same results
      • 3 versus 4: 1 event/slice versus 1000 events/slice -> expect same results
      • 5 versus 6: MDF versus MEP, Allen event loop from Gaudi -> expect same results
      • 5 versus 7: Allen event loop from Gaudi versus Allen standalone, 1000 events/slice -> expect same results, also checking efficiency checker implemented inside Allen
      • 7 versus 8: x86 versus cuda, 1000 events/slice -> expect differences, should monitor them; already tested in Allen CI test, but no direct comparison of x86 and cuda performance is done
    • Add test running hlt1_pp_validation sequence within Moore using MC information written to a RawEvent (dumped to MDF files for Allen standalone processing)
    • Standalone Allen physics performance checker versus Gaudi checker: compare the output of the two previous tests
    • Test throughput several times to check for variations
    • Add test with gpu-burn to check stability of GPUs
    • Add test with and w/o offloading algorithms to the host

    Stability, integration, commissioning

    • Add test for velo raw banks decoding
    • Add test for UT raw banks decoding
    • Add test for muon raw banks decoding
    • Add test for Calo raw banks decoding: Moore#393 (closed)
    • Add test for SciFi raw bank decoding
    • Add test for Plume raw bank decoding: Moore!2034 (merged)
    • Test for velo open, w/o UT, similar to Moore!1396 (merged)
    • Test that output file writing (MDF) works
    • Add test on MU samples
    • Add test on PbPb samples
    • Add test on compressed and uncompressed MDF input
    • MPI build and run
    • Test copying banks onto device

    Functionality needed for these tests that is not yet tested / existing:

    • a server equipped with a GPU (preferably RTX A5000), callable from the nightly tests
    • calling an Allen validator from Gaudi-Allen
    • read track and PV MC information raw banks within Gaudi-Allen
    • DIGI, MDF and MEP files from the same sample, including MC information raw banks
    • test Allen validator from MEP file
    • scripts to call IP resolution, muon ID efficiency, PV efficiency and resolution using Gaudi-Allen
    11 of 80 checklist items completed · Edited

    Designs

      Child items ...

      2. Activity

      Please register or sign in to reply