Add Allen configuration in Moore control flow + HLT1 PV finding with input HLT2 tracks to Moore configuration
DESCRIPTION
Trying to use the Pr::Velo::Tracks converter to the Allen objects needed to run the Allen TBLV algorithm.
Definition of allen_velo_track and make_hlt1_pvs functions to test the converter algorithm defined in Allen!1861.
These tracks are used as input for running the algorithm defined in a modified version of gaudi_allen_pv.py and hlt1_hlt2_vertexcompare.py, using the converted tracks as input of the HLT1 PV finding algorithm.
The qmt tests are added to the MR.
The HLT1 PV finding with converted track is added to the global reconstruction and set to False by default!
THROUGHPUT STUDIES
Without Allen!2285, running 10 mins the entire HLT2 sequence on 2025 data (link), using 56 threads (to emulate pit machine conditions):
| PV reconstruction | Throughput (Hz) |
|---|---|
| With Moore VeloTracks + Converter4Tracks + Allen TBLV (HLT1 PV reconstruction) + Converter4PVs | 260.6 (-1.8%) |
| NOMINAL (With Moore VeloTracks + Moore TBLV (HLT2 PV reconstruction) | 265.4 |
With Allen!2285, running 10 mins the HLT2 sequence on 2025 data using 56 threads shows a throughput decrease of -1.3 Events/s (corresponding to -0.4%)
Detailed comparison
| PV reconstruction | Avg. execution time (us) |
|---|---|
| HLT2 TBLV (nominal) | ~120 |
| HLT1 TBLV with converted tracks and PVs | ~280 (multi_fitter ~100, calculate_denom ~52, histo ~44, peak ~22, cleanup ~19, HLT2->HLT1 velo tracks converter ~40, HLT1->HLT2 PVs converter ~7) |