Seed Parameter Estimation Fix, main branch (2025.02.10.)

[krasznaa]

This is a follow-up about the issue discussed (amongst other places) in #842. That in RelWithDebInfo mode we would see the following output from traccc_seq_example_cuda:

[bash][Celeborn]:out > ./build/cuda-fp64/bin/traccc_seq_example_cuda --compare-with-cpu

Running Full Tracking Chain Using CUDA

Detector Options:
├ Detector file:                   geometries/odd/odd-detray_geometry_detray.json
├ Material file:                   geometries/odd/odd-detray_material_detray.json
├ Surface grid file:               geometries/odd/odd-detray_surface_grids_detray.json
├ Use detray detector:             true
└ Digitization file:               geometries/odd/odd-digi-geometric-config.json
...
WARNING: @traccc::io::csv::read_cells: 28 duplicate cells found in /home/krasznaa/ATLAS/projects/traccc/traccc/data/odd/geant4_10muon_10GeV/event000000000-cells.csv
===>>> Event 0 <<<===
Number of measurements: 631 (host), 631 (device)
  Matching rate(s):
    - 100% at 0.01% uncertainty
    - 100% at 0.1% uncertainty
    - 100% at 1% uncertainty
    - 100% at 5% uncertainty
Number of spacepoints: 451 (host), 451 (device)
  Matching rate(s):
    - 100% at 0.01% uncertainty
    - 100% at 0.1% uncertainty
    - 100% at 1% uncertainty
    - 100% at 5% uncertainty
Number of seeds: 181 (host), 181 (device)
  Matching rate(s):
    - 100% at 0.01% uncertainty
    - 100% at 0.1% uncertainty
    - 100% at 1% uncertainty
    - 100% at 5% uncertainty
Number of track parameters: 181 (host), 181 (device)
  Matching rate(s):
    - 0% at 0.01% uncertainty
    - 0% at 0.1% uncertainty
    - 0% at 1% uncertainty
    - 0% at 5% uncertainty
Number of track candidates (header): 183 (host), 183 (device)
  Matching rate(s):
    - 0% at 0.01% uncertainty
    - 0% at 0.1% uncertainty
    - 0% at 1% uncertainty
    - 0% at 5% uncertainty
  Track candidates (item) matching rate: 100%
Number of track states: 183 (host), 183 (device)
  Matching rate(s):
    - 98.9071% at 0.01% uncertainty
    - 98.9071% at 0.1% uncertainty
    - 98.9071% at 1% uncertainty
    - 98.9071% at 5% uncertainty
...

(Note that in these examples I built the code with FP64 precision, to make the point even more strongly.)

With this small update, this changes to:

[bash][Celeborn]:out > ./build/cuda-fp64/bin/traccc_seq_example_cuda --compare-with-cpu

Running Full Tracking Chain Using CUDA

Detector Options:
├ Detector file:                   geometries/odd/odd-detray_geometry_detray.json
├ Material file:                   geometries/odd/odd-detray_material_detray.json
├ Surface grid file:               geometries/odd/odd-detray_surface_grids_detray.json
├ Use detray detector:             true
└ Digitization file:               geometries/odd/odd-digi-geometric-config.json
...
WARNING: @traccc::io::csv::read_cells: 28 duplicate cells found in /home/krasznaa/ATLAS/projects/traccc/traccc/data/odd/geant4_10muon_10GeV/event000000000-cells.csv
===>>> Event 0 <<<===
Number of measurements: 631 (host), 631 (device)
  Matching rate(s):
    - 100% at 0.01% uncertainty
    - 100% at 0.1% uncertainty
    - 100% at 1% uncertainty
    - 100% at 5% uncertainty
Number of spacepoints: 451 (host), 451 (device)
  Matching rate(s):
    - 100% at 0.01% uncertainty
    - 100% at 0.1% uncertainty
    - 100% at 1% uncertainty
    - 100% at 5% uncertainty
Number of seeds: 181 (host), 181 (device)
  Matching rate(s):
    - 100% at 0.01% uncertainty
    - 100% at 0.1% uncertainty
    - 100% at 1% uncertainty
    - 100% at 5% uncertainty
Number of track parameters: 181 (host), 181 (device)
  Matching rate(s):
    - 100% at 0.01% uncertainty
    - 100% at 0.1% uncertainty
    - 100% at 1% uncertainty
    - 100% at 5% uncertainty
Number of track candidates (header): 183 (host), 183 (device)
  Matching rate(s):
    - 100% at 0.01% uncertainty
    - 100% at 0.1% uncertainty
    - 100% at 1% uncertainty
    - 100% at 5% uncertainty
  Track candidates (item) matching rate: 100%
Number of track states: 183 (host), 183 (device)
  Matching rate(s):
    - 100% at 0.01% uncertainty
    - 100% at 0.1% uncertainty
    - 100% at 1% uncertainty
    - 100% at 5% uncertainty
...

Though without backward filtering, the agreement is indeed not perfect. So that part is still as we discussed earlier. 🤔

[bash][Celeborn]:out > ./build/cuda-fp64/bin/traccc_seq_example_cuda --compare-with-cpu --fit-use-backward-filter=false

Running Full Tracking Chain Using CUDA

Detector Options:
├ Detector file:                   geometries/odd/odd-detray_geometry_detray.json
├ Material file:                   geometries/odd/odd-detray_material_detray.json
├ Surface grid file:               geometries/odd/odd-detray_surface_grids_detray.json
├ Use detray detector:             true
└ Digitization file:               geometries/odd/odd-digi-geometric-config.json
...
WARNING: @traccc::io::csv::read_cells: 28 duplicate cells found in /home/krasznaa/ATLAS/projects/traccc/traccc/data/odd/geant4_10muon_10GeV/event000000000-cells.csv
===>>> Event 0 <<<===
Number of measurements: 631 (host), 631 (device)
  Matching rate(s):
    - 100% at 0.01% uncertainty
    - 100% at 0.1% uncertainty
    - 100% at 1% uncertainty
    - 100% at 5% uncertainty
Number of spacepoints: 451 (host), 451 (device)
  Matching rate(s):
    - 100% at 0.01% uncertainty
    - 100% at 0.1% uncertainty
    - 100% at 1% uncertainty
    - 100% at 5% uncertainty
Number of seeds: 181 (host), 181 (device)
  Matching rate(s):
    - 100% at 0.01% uncertainty
    - 100% at 0.1% uncertainty
    - 100% at 1% uncertainty
    - 100% at 5% uncertainty
Number of track parameters: 181 (host), 181 (device)
  Matching rate(s):
    - 100% at 0.01% uncertainty
    - 100% at 0.1% uncertainty
    - 100% at 1% uncertainty
    - 100% at 5% uncertainty
Number of track candidates (header): 183 (host), 183 (device)
  Matching rate(s):
    - 100% at 0.01% uncertainty
    - 100% at 0.1% uncertainty
    - 100% at 1% uncertainty
    - 100% at 5% uncertainty
  Track candidates (item) matching rate: 100%
Number of track states: 183 (host), 183 (device)
  Matching rate(s):
    - 80.3279% at 0.01% uncertainty
    - 83.6066% at 0.1% uncertainty
    - 84.153% at 1% uncertainty
    - 84.153% at 5% uncertainty
...

It took me a little while to realize that we're not actually using detray::bound_parameters_vector in that part of the code, but just the underlying algebra-plugins "vector type" directly. So the params variable was in an undefined state over there. Leading to its time property ending up being undefined. (As it's not set to anything explicitly by either the host or the device code.) And this undefined time value made its way all the way down to the final fitted tracks. Leading to an imperfect (though not 0%) agreement even there. 🤔

[krasznaa]

Note that one could also just set the time parameter to 0.f explicitly, instead of using matrix::zero. Which may be an optimization for the future... 🤔

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud