Skip to content

Double candidate SciFi

  • Change default number of candidates in SciFi to 16.
  • Use 32 threads in a block to process the 16 hits in the middle module. This produces a maximum of two candidates per middle hit, increasing both speed (with respect to vanilla 16 candidates) and slightly efficiency.

The performance improves significantly with this MR by about 16%.

Physics efficiency 1000 bsphiphi:

Forward tracks:
TrackChecker output                               :      2196/    28448   7.72% ghosts
for P>3GeV,Pt>0.5GeV                              :      1239/    18841   6.58% ghosts
Long                                              :     24928/    57988  42.99% ( 44.50%),       171 (  0.68%) clones, pur  96.15%, hit eff  90.52%
Long, p > 5 GeV                                   :     22137/    36811  60.14% ( 61.81%),       152 (  0.68%) clones, pur  96.36%, hit eff  91.65%
Long strange                                      :       691/     2542  27.18% ( 27.11%),         2 (  0.29%) clones, pur  95.39%, hit eff  89.36%
Long strange, p > 5 GeV                           :       565/     1197  47.20% ( 47.71%),         2 (  0.35%) clones, pur  95.76%, hit eff  91.10%
Long from B                                       :      2659/     3930  67.66% ( 69.51%),        19 (  0.71%) clones, pur  96.62%, hit eff  92.22%
Long from B, p > 5 GeV                            :      2539/     3219  78.88% ( 79.60%),        17 (  0.67%) clones, pur  96.72%, hit eff  92.81%
Long electrons                                    :       496/     4548  10.91% ( 11.46%),        15 (  2.94%) clones, pur  94.97%, hit eff  89.80%
Long electrons from B                             :        67/      202  33.17% ( 35.86%),         2 (  2.90%) clones, pur  95.26%, hit eff  90.69%
Long electrons from B, p > 5 GeV                  :        64/      130  49.23% ( 52.42%),         2 (  3.03%) clones, pur  95.42%, hit eff  91.41%
Long from D electrons, p > 3 GeV, pt > 0.5 GeV    :        27/       49  55.10% ( 56.67%),         0 (  0.00%) clones, pur  95.62%, hit eff  92.21%
Long from D, p > 3 GeV, pt > 0.5 GeV              :       548/      728  75.27% ( 72.99%),         2 (  0.36%) clones, pur  96.26%, hit eff  91.29%
Long from B electrons, p > 3 GeV, pt > 0.5 GeV    :        54/       82  65.85% ( 66.88%),         1 (  1.82%) clones, pur  95.42%, hit eff  91.35%
Long from B, p > 3 GeV, pt > 0.5 GeV              :      2420/     2978  81.26% ( 80.73%),        18 (  0.74%) clones, pur  96.72%, hit eff  92.62%
Long from B, p > 3 GeV, pt > 0.5 GeV, eta < 2.5, (phi-pi/2)<0.8:       123/      236  52.12% ( 51.26%),         0 (  0.00%) clones, pur  95.83%, hit eff  88.83%
Long from B, p > 3 GeV, pt > 0.5 GeV, eta > 2.5, (phi-pi/2)<0.8:      1137/     1342  84.72% ( 84.86%),         8 (  0.70%) clones, pur  96.88%, hit eff  93.07%
Long from B, p > 3 GeV, pt > 0.5 GeV, eta < 2.5, (phi-pi/2)>0.8:       155/      237  65.40% ( 62.74%),         0 (  0.00%) clones, pur  95.38%, hit eff  89.47%
Long from B, p > 3 GeV, pt > 0.5 GeV, eta > 2.5, (phi-pi/2)>0.8:      1005/     1163  86.41% ( 86.42%),        10 (  0.99%) clones, pur  96.85%, hit eff  93.05%

The original MR was https://gitlab.cern.ch/lhcb-parallelization/Allen/merge_requests/198 but I messed it up with other changes.

Merge request reports