Improved u-v-hit search window calculation for the ForwardTracking
Adds a class which calculates a y shift of the predicted hit position in u-v layers
- overhead because of extra calculation might be mitigated by vectorisation
- allows to define smaller search window which speeds up the algorithm
- window parameters are tuned such that there is no loss of efficiency
Summary of efficiencies and throughput results
- ForwardFast: default parameters tolYCollectX=3.5 and tolYSlopeCollectX=0.001
- efficiencies (given in %) tested with Bs->PhiPhi
- Throughput tested with
NoCutNoFit
scenario on quanta machine (20 threads)
Branch | (TolYCollectX, TolYSlopeCollectX) | long_fromB_P>3GeV_Pt>0.5GeV | Throughput [Evts/s] | |
---|---|---|---|---|
0 | Master | default | 89.32 | 7671.25 |
1 | yCorr | default | 89.7 | 7670.26 |
2 | yCorr | (2.1875, 0.000625) | 89.71 | 8129.14 |
3 | yCorr | (1.75, 0.0004) | 89.39 | 8351.61 |
- ForwardBest: default parameters tolYCollectX=4.1 and tolYSlopeCollectX=0.0018
Branch | Parameters | 01_long | 02_long>5GeV | 05_long_B | 06_long_B>5Gev | Throughput [Evts/s] | |
---|---|---|---|---|---|---|---|
0 | Master | default | 80.18 | 89.04 | 87.41 | 92.54 | 1018.18 |
1 | yCorr | default | 80.26 | 89.04 | 87.38 | 92.53 | 993.54 |
2 | yCorr | (3.5, 0.001) | 80.23 | 89.13 | 87.41 | 92.51 | 1226.09 |
3 | yCorr | (3.0, 0.0009) | 80.17 | 89.19 | 87.34 | 92.47 | 1344.75 |
Edited by Marco Cattaneo