New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Helix #3295
Helix #3295
Conversation
… of iteration: never seen more than 7 if successful
A new Pull Request was created by @VinInn (Vincenzo Innocente) for CMSSW_7_1_X. Helix It involves the following packages: TrackingTools/GeomPropagators @nclopezo, @cmsbuild, @anton-a, @thspeer, @slava77, @Degano can you please review it and eventually sign? Thanks. |
@lgray @dbenedet @nancymarinelli
Daniele and Lindsey, could you please check the changes on your side, before we push this into the release. (maybe there's a better/cleaner sample to look at) Vincenzo, is there a place to fine-tune the changes so that the conversions are not lost at this level. |
Hi Slava, for the changes in question: |
I made some test with workflow 19 photon 35 GeV |
Hi Vincenzo, what was your "benchmark"? was it some TTbar PU or something realistically enhanced with real conversions? |
On 14 Apr, 2014, at 2:53 PM, Slava Krutelyov notifications@github.com wrote:
in any case It is clear that current Conversion code depends on those slowsly converging trajecttories. Either we leave this PR pending or we simply closing it as potentially impacting critical physics… v. |
I would keep it pending for now, waiting for some feedback from the egamma POG. |
Hi, For the conversionStepTracks this part of the code is technically maintained by the tracking POG rather than egamma. (And in practice may be maintained by no one.) I wouldn't be surprised to see similar numerical sensitivity in the conversion vertex fitting itself though. |
Hi Josh, My plots were taken directly from the default PhotonValidator (you can guess the meaning of the plots based on the directory and histogram names). @cerati @rovere |
Well, it's not clear from the plots that the difference is only coming from the convStepTracks, there may indeed be some contribution from the ConversionProducer as well (which is maintained by egamma). |
I don't find any discrepancies in conversionStepTracks in the gamma35 On 4/14/14, 8:52 AM, Josh Bendavid wrote:
Vyacheslav (Slava) Krutelyov |
On 14 Apr, 2014, at 3:56 PM, Slava Krutelyov notifications@github.com wrote:
|
No the InOut and OutIn ecal-seeded tracks are separate from the conversionStepTracks, they have their own collections. |
(InOut and OutIn are the ecal-seeded tracks, whereas the conversionStepTracks are the soft conversion partners seeded from existing standard tracks in the tracker) |
Sorry my conclusions were wrong about the number of iterations (they are counted backward) |
I actually discovered that it is pretty common for conversion seed |
regressions gone |
Speed ups and cleanup in Helix computation
In particular the iteration code in for ArbitraryPlaneCrossing has been "linearized"
The maximum number of iterations reduced to 20 (for good solution never seen exceeding 7)
log info degraded to log debug
15% speedup of Analytical propagator
Regression:
As usual, when touching arithmetic code regression are observed.
Conversions are, as usual, affected more than others...
Fully consistent with what already observed when changes occur in optimization by the compiler