Gabriele

Tested new parallel algorithm on CPU, similar to current implementation but with a parallel approach:

  1. For each track, find z component of DCA to beamline
  2. Sort tracks per z value
  3. For each track, count how many tracks fall into a window centered in z (and are time compatible) --> density
  4. For each track, check if it is a local maximum (look at neighbours)
  5. For each local maximum, fit tracks which falls into the windows
  6. For each seed, check if it is not a duplicate with neighboring seeds
  7. Surviving seeds promoted to vertices

 

Current version:

[1727121:its-tracker]: [17:12:06][INFO]  - Vertex finding: found 132 vertices (total 132) in 50.63 ms

Parallel version (20 threads):

[1725878:its-tracker]: [17:05:17][INFO]  - Vertex finding: found 132 vertices (total 132) in 14.35 ms

Preliminary results from 4 simulated Pb-Pb 50kHz TF:

 

Next steps