ATLASRECTS-7693: Keep one impl of the newCovariance, the one that should be ~ 10-15% faster

Review changes
Download
Patches
Plain diff

Merged ATLASRECTS-7693: Keep one impl of the newCovariance, the one that should be ~ 10-15% faster

ATLAS-EGamma/athena:ATLASRECTS-7693_single_new_covariance_faster_version into main

Overview 33
Commits 4
Pipelines 4
Changes 7

Merged Christos Anastopoulos requested to merge ATLAS-EGamma/athena:ATLASRECTS-7693_single_new_covariance_faster_version into main 1 year ago

Overview 9
Commits 4
Pipelines 4
Changes 7

ATLASRECTS-7693

This is also for @jdandoy @jchapman @jcatmore @nstyles to approve.

This came up during summer on the recurring question (that time by @zmarshal on if we can do things do faster here and there ala HLT).

The answer is kind of yes but since we deal with float if we alter a bit the order of operations it will alter the "binary" result.

In any case

We have practically 2 implementations of the same thing
One using Eigen and managing a bit better vectorization
Actually, up to summer I had assumed that what is tried here was done 2-3 years ago
@sroe, @goblirsc prb I mis-remember from the time we introduced the common JacobianHelper.h , why we did not also unified this?
The idea is to both de-duplicate code and retain only the faster of the 2.
From then one we will have to deal only with one code location which we could improve furhter.
The one we keep is located [https://gitlab.cern.ch/atlas/athena/-/blob/main/Tracking/TrkExtrapolation/TrkExUtils/src/RungeKuttaUtils.cxx#L1412]

So this gives something along the following magnitude of changes . It can be viewed as similar in concept to [!66067 (merged)] by @smh

Py:diff-root         INFO 000.InDetForwardTrackParticlesAux.definingParametersCovMatrixOffDiag.5.57 8.812105079414323e-05 -> 8.812105807010084e-05 => diff= [-0.00000206%]
Py:diff-root         INFO 000.InDetForwardTrackParticlesAux.2.standardDeviationOfChi2OS 'A' -> '!' => diff= [N/A]
Py:diff-root         INFO 000.InDetLargeD0TrackParticlesAux.330.qOverP -0.00038124711136333644 -> -0.0003812477516476065 => diff= [-0.00004199%]
Py:diff-root         INFO 000.InDetLargeD0TrackParticlesAux.definingParametersCovMatrixDiag.0.330 0.1922348141670227 -> 0.19223608076572418 => diff= [-0.00016472%]
Py:diff-root         INFO 000.InDetLargeD0TrackParticlesAux.definingParametersCovMatrixOffDiag.0.330 0.6220205426216125 -> 0.6220209002494812 => diff= [-0.00001437%]
Py:diff-root         INFO 000.InDetLargeD0TrackParticlesAux.330.chiSquared 89.23895263671875 -> 89.23896026611328 => diff= [-0.00000214%]
Py:diff-root         INFO 000.InDetTrackParticlesAux.22.d0 -0.29108744859695435 -> -0.29106953740119934 => diff= [0.00153835%]
Py:diff-root         INFO 000.InDetTrackParticlesAux.22.z0 -31.78347396850586 -> -31.783472061157227 => diff= [0.00000150%]
Py:diff-root         INFO 000.InDetTrackParticlesAux.22.phi -0.640583872795105 -> -0.6405847668647766 => diff= [-0.00003489%]
Py:diff-root         INFO 000.InDetTrackParticlesAux.206.theta 0.3397007882595062 -> 0.33970075845718384 => diff= [0.00000219%]
Py:diff-root         INFO 000.InDetTrackParticlesAux.22.qOverP 0.0004895639140158892 -> 0.0004895543679594994 => diff= [0.00048748%]
Py:diff-root         INFO 000.InDetTrackParticlesAux.definingParametersCovMatrixDiag.0.22 0.057841114699840546 -> 0.057839393615722656 => diff= [0.00074390%]
Py:diff-root         INFO 000.InDetTrackParticlesAux.definingParametersCovMatrixDiag.0.206 0.04129693657159805 -> 0.041296765208244324 => diff= [0.00010374%]
Py:diff-root         INFO 000.InDetTrackParticlesAux.definingParametersCovMatrixOffDiag.0.22 0.012589438818395138 -> 0.012590291909873486 => diff= [-0.00169400%]
Py:diff-root         INFO 000.InDetTrackParticlesAux.definingParametersCovMatrixOffDiag.0.206 -0.02126721665263176 -> -0.021267127245664597 => diff= [0.00010510%]
Py:diff-root         INFO 000.InDetTrackParticlesAux.definingParametersCovMatrixOffDiag.7.790 0.0005897537339478731 -> 0.000589753792155534 => diff= [-0.00000247%]
Py:diff-root         INFO 000.InDetTrackParticlesAux.22.chiSquared 22.065128326416016 -> 22.06508445739746 => diff= [0.00004970%]
Py:diff-root         INFO 000.InDetTrackParticlesAux.22.eProbabilityComb 0.08699578791856766 -> 0.08699649572372437 => diff= [-0.00020340%]
Py:diff-root         INFO 000.InDetTrackParticlesAux.22.eProbabilityHT 0.4067322015762329 -> 0.4067312777042389 => diff= [0.00005679%]
Py:diff-root         INFO 000.22.InDetTrackParticlesAuxDyn.TRTdEdx 0.829282283782959 -> 0.8292827010154724 => diff= [-0.00001258%]
Py:diff-root         INFO 000.PrimaryVerticesAux.0.chiSquared 237.71218872070312 -> 237.71214294433594 => diff= [0.00000481%]
Py:diff-root         INFO 000.PrimaryVerticesAux.5.numberDoF 44.76664733886719 -> 44.76664352416992 => diff= [0.00000213%]
Py:diff-root         INFO 000.PrimaryVerticesAux.0.x -0.4937134385108948 -> -0.49371346831321716 => diff= [-0.00000151%]
Py:diff-root         INFO 000.PrimaryVerticesAux.covariance.1.0 7.870347644711728e-07 -> 7.870345370974974e-07 => diff= [0.00000722%]
Py:diff-root         INFO 000.PrimaryVerticesAux.covariance.1.2 5.103515476889697e-08 -> 5.103516187432433e-08 => diff= [-0.00000348%]
Py:diff-root         INFO 000.PrimaryVerticesAux.covariance.2.5 8.140814315993339e-05 -> 8.140813588397577e-05 => diff= [0.00000223%]
Py:diff-root         INFO 000.PrimaryVerticesAux.covariance.1.23 -2.1415432911453536e-06 -> -2.141543518519029e-06 => diff= [-0.00000265%]
Py:diff-root         INFO 000.PrimaryVerticesAux.covariance.1.40 7.870347644711728e-07 -> 7.870345370974974e-07 => diff= [0.00000722%]
Py:diff-root         INFO 000.PrimaryVerticesAux.trackWeights.1.0 0.19020986557006836 -> 0.1902097910642624 => diff= [0.00000979%]
Py:diff-root         INFO 000.PrimaryVerticesAux.trackWeights.1.2 0.1038067564368248 -> 0.1038067489862442 => diff= [0.00000179%]
Py:diff-root         INFO 000.PrimaryVerticesAux.trackWeights.0.5 0.02166411466896534 -> 0.021664118394255638 => diff= [-0.00000430%]
Py:diff-root         INFO 000.PrimaryVerticesAux.trackWeights.22.11 0.24990658462047577 -> 0.24990659952163696 => diff= [-0.00000149%]
Py:diff-root         INFO 000.PrimaryVerticesAux.trackWeights.1.23 0.8795447945594788 -> 0.879544734954834 => diff= [0.00000169%]
Py:diff-root         INFO 000.PrimaryVerticesAux.trackWeights.9.27 0.002159126801416278 -> 0.0021591265685856342 => diff= [0.00000270%]
Py:diff-root         INFO 000.2.PrimaryVerticesAuxDyn.sumPt2 136.55674743652344 -> 136.5567626953125 => diff= [-0.00000279%]
....