Skip to content

Final(?) tweaks to ID track silicon seeding

Maximilian Emanuel Goblirsch-Kolb requested to merge goblirsc/athena:master-tuneTrackFinderFurther into master

This MR adds some small further performance tweaks to the seeding phase of ID track reconstruction, as part of the ATLIDTRKCP-260 optimization round:

  • The z0 interval for track reconstruction is reduced from 250 to 200mm
  • veto of some specific 'impossible' regions for the seed's central space point (for example, first/last layers or top of a disk)
  • increase min delta R between consecutive SP in a seed from 5 to 10mm
  • update of the space point map in the seed maker event data to allow it to grow in size if needed, allowing a smaller size in memory when starting out (200 rather than 5k slots)
  • Improve documentation of the logic being used in the seed maker

The changes should leave the tracking efficiency stable at sub-permille level, while the fake rate is reduced by around 10%. The CPU time for the track finding step should improve by slightly below 10 percent, with slight downstream profits to the ambiguity resolution and track extension, and a 3-4% improvement in overall tracking CPU. A detailed discussion of the changes is available in these slides.

Adding @norap and @sroe. Also including @amete for information in case this shows up in SPOT.

Edited by Maximilian Emanuel Goblirsch-Kolb

Merge request reports