More AMVF performance upgrades

[baschlag]

This PR removes the estimateDeltaZ function (which contributed 5.5% to the overall AMVF timing, see below) and replaces it with a simple z position diff check which is way less expensive and gives the same outputs.

[baschlag]

Somebody wants to review this small PR, @asalzburger, @robertlangenberg ?

[robertlangenberg]

Did you do a proper test with >100 vertices to see if the reduction in physics performance is negligible?

[baschlag]

Did you do a proper test with >100 vertices to see if the reduction in physics performance is negligible?

Yes, I checked on 300 events (i.e. 10536 vertices). The average time for running the AMVF decreased from 124ms to 120ms and the resulting vertex x,y,z positions, covariance matrices and number of tracks match 100%.
Also, the estimateDeltaZ method and the new calculation seem to give matching numbers for the z difference. And yes, I'm sure that I'm comparing the right things 😃 doing one more cross-check now... hang on

[codecov]

Codecov Report

Merging #277 into master will increase coverage by 3.54%.
The diff coverage is 52.46%.

@@            Coverage Diff             @@
##           master     #277      +/-   ##
==========================================
+ Coverage   44.84%   48.38%   +3.54%     
==========================================
  Files         376      318      -58     
  Lines       18866    16357    -2509     
  Branches     8970     7580    -1390     
==========================================
- Hits         8461     7915     -546     
+ Misses       4905     3180    -1725     
+ Partials     5500     5262     -238     

Impacted Files	Coverage Δ
...e/include/Acts/EventData/SingleTrackParameters.hpp	75.00% <ø> (+2.14%)	⬆️
...Acts/EventData/detail/initialize_parameter_set.hpp	81.81% <ø> (ø)
...include/Acts/TrackFinder/CKFSourceLinkSelector.hpp	46.37% <ø> (-1.45%)	⬇️
Core/include/Acts/Utilities/Helpers.hpp	58.90% <ø> (ø)
...clude/Acts/Vertexing/AdaptiveMultiVertexFinder.hpp	78.57% <ø> (ø)
...nclude/Acts/Vertexing/TrackDensityVertexFinder.hpp	75.00% <ø> (ø)
Core/src/Geometry/CutoutCylinderVolumeBounds.cpp	40.29% <0.00%> (ø)
Core/src/Material/SurfaceMaterialMapper.cpp	10.20% <0.00%> (ø)
Core/src/Material/VolumeMaterialMapper.cpp	11.69% <0.00%> (ø)
...lude/Acts/Visualization/EventDataVisualization.hpp	40.00% <30.00%> (-2.47%)	⬇️
... and 78 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 4acad15...54d4d12. Read the comment docs.

[baschlag]

Did you do a proper test with >100 vertices to see if the reduction in physics performance is negligible?

Just to clarify: The part

1. / std::tan(th) * (X * std::cos(phi) + Y * std::sin(phi))

is basically always 0 (order 1e-17 or smaller) and therefore negligible. That's why I'm seeing the same physics results with less CPU time.

[robertlangenberg]

As discussed, removed part never exceeds 10e-16 in test with 10000 vertices, shouldn't influence end result.