KalmanVertexUpdator performance tuning for use in AMVF

This MR significantly speeds up the KalmanVertexUpdator's update method by replacing a temporarily created and subsequently destroyed xAOD::Vertex object by a slim struct. This is also done in the IVertexUpdator and propagated to all clients. This, together with local caching of a matrix inversion result, yields a factor of up to 4 speed improvement for the Adaptive multi vertex fitter (see https://its.cern.ch/jira/browse/ATLIDTRKCP-245).

In addition to this performance tuning, some clean-up to the code is performed (removal of commented lines, attempt to clean up indentation to a consistent scheme, uniform use of messaging macros, use of std algorithms where possible, fix of potential memory leak).

The changes were tested to not change the reco output (except for CPU/timings) compared to running the AMVF in an unchanged master.

Adding @sroe and @npetters who are also working on improving the AMVF performance.

added Tracking label

marked as a Work In Progress

changed title from re-implement initial tuning step, trying to keep gitlab history happy to WIP: KalmanVertexUpdator performance tuning for use in AMVF

changed the description

added 1 commit

• c2327bd9 - remove commented couts

Compare with previous version

build area is cleaned as the previous build was for the MR labeled as full-build

This merge request affects 1 package:

• Tracking/TrkVertexFitter/TrkVertexFitterUtils

Adding @amorley as watcher

added master review-pending-level-1 labels

added 1 commit

• 95d489a7 - replace loop by std::find

Compare with previous version

removed review-pending-level-1 label

removing review-pending for now as still WIP

added 2 commits

• 8efaf082 - avoid unnecessary pointer juggling
• d9006186 - copyright...

Compare with previous version

CI Result SUCCESS (hash 41494daf)

	Athena	AthSimulation
externals
cmake
make
required tests
optional tests

Full details available on this CI monitor view
Athena: number of compilation errors 0, warnings 0
AthSimulation: number of compilation errors 0, warnings 0
For experts only: Jenkins output [CI-MERGE-REQUEST-CC7 10071]

added 2 commits

• e783b5ad - protect positionUpdate from invalid input
• 1f431c0f - add some protection to sign argument, add enum for sign in private methods

Compare with previous version

resolved all threads

I just noticed that the KalmanVertexTrackUpdator calls the (slow) positionUpdate method via the IVertexUpdator interface, resulting in another performance hit on the AMVF.

To solve this, we would need to move our struct out of this class into the interface and make the positionUpdateFast method public - or alternatively change positionUpdate itself and adapt all callers?

Any opinions? @sroe / @npetters in particular?

added 188 commits

• 1f431c0f...cd6feeb6 - 186 commits from branch atlas:master
• 62c70dda - Merge remote-tracking branch 'upstream/master' into KalmanVertexUpdatorTuning_cleanedUp
• 56f1a4a2 - Optimize IVertexUpdator interface and update clients to avoid temporary...

Compare with previous version

Now updated the IVertexUpdator to avoid the temporary xAOD::Vertex return object everywhere.