As per title, this fixes a bug in the setting of hit selection dphi windows.
Additionally, it adds an upper bound on 1/pT for window parametrization, for both dq and dphi, to avoid that window size explodes if pT<0.1 GeV.
These issues seem to only have affected detachedTripletStep and pixelLessStep, in terms of timing.
Effects on physics performance are ~ invisible (see below).
examples of timing for a few events from /eos/cms/store/user/cmsbuild/profiling/data/CMSSW_12_2_0_pre2/slc7_amd64_gcc900/11834.21/step2.root.unused (for future reference)
PR description:
As per title, this fixes a bug in the setting of hit selection dphi windows. Additionally, it adds an upper bound on 1/pT for window parametrization, for both dq and dphi, to avoid that window size explodes if pT<0.1 GeV. These issues seem to only have affected detachedTripletStep and pixelLessStep, in terms of timing. Effects on physics performance are ~ invisible (see below).
PR validation:
TTbar=50: https://mmasciovecchio.web.cern.ch/MkFit_hitSelectionWindows_Nov21/MTV_fixHitSelWindows_Nov21_TTbarPU50_122X/
10muPt0p2to10: https://mmasciovecchio.web.cern.ch/MkFit_hitSelectionWindows_Nov21/MTV_fixHitSelWindows_Nov21_10muPt0p2to10_122X/
"Luckily", changes in physics performance are ~ invisible. On the other hand, timing for pixelLessStep and detachedTripletStep appears to be "fixed": https://mmasciovecchio.web.cern.ch/MkFit_hitSelectionWindows_Nov21/MTV_fixHitSelWindows_Nov21_TTbarPU50_122X/plots_timing/iterationsReal.pdf