sPHENIX tracking discussion

[Christof Roland <christof.roland AT cern.ch>]

Hi Everybody, 

following up on the discussion on timeing performance of our current code 
I ran a few benchmark jobs. 1000 jobs one event each hijing 0-20 + 50kHz of 
pileup.
These are jobs submitted from sphnx02 so I assume even if there are a few 
slow machines this will not dominate the results. 

Time per event is here:
InttClusterizer                                 1000 0.014891 sec
MvtxClusterizer                         1000 0.088210 sec
TpcClusterizer                                  1000 2.677374 sec
TpcClusterCleaner                               1000 0.045161 sec
PHActsSiliconSeeding                    1000 509.086792 sec
PHActsVertexPropagator          1000 0.044235 sec
PHCASeeding                                     1000 3.974943 sec
PHSimpleKFProp                          1000 3.371190 sec
PrePropagatorPHTpcTrackSeedCircleFit    1000 0.108033 sec
PHTpcTrackSeedCircleFit         1000 0.107930 sec
PHTrackCleaner                          1000 0.007340 sec
PHGhostRejection                                1000 0.211178 sec
PHSiliconTpcTrackMatching               1000 1.440864 sec
PHActsFirstTrkFitter                    1000 2.409158 sec
PHSimpleVertexFinder                    1000 0.044909 sec
PHRaveVertexing                         1000 0.409105 sec
PHGenFitTrackProjection         1000 0.000386 sec
SvtxEvaluator                                   1000 418.793579 sec

It looks like all modules touching the actual hits are horribly slow now. 
This looks like in our recent changes to the local coordinate storage we 
introduced 
an inefficient loop to look up things. 

I'll try to run a job through callgrind now to see if I can trace this down 
some more. 

cheers

   Christof