TLS problem running on CMSSW_7_X_X

[shervin86]

Running ECALELF (that includes GBRLikelihood) we got:

Exception Message:
unable to load /afs/cern.ch/work/l/lbrianza/PHD/ECALELF_EP/CMSSW_7_2_0_pre7/lib/slc6_amd64_gcc481/pluginCalibrationEleNewEnergiesProducer.so because dlopen: cannot load any more
object with static TLS

The EleNewEnergiesProducer plugin calles the GRBLikelihoodEGTools to calculate retrieve energy corrections from regression for electrons.

We are wondering if this error is related to the
-fopenmp option in the BuildFile.xml

Any idea?

[bendavid]

Can you try removing -fopenmp from the compile flags? (This is only needed for the training step and has no effect or benefit for applying the regression)

[shervin86]

I tried and it does not compile

Shervin
Da: Josh Bendavid
Inviato: giovedì 23 ottobre 2014 12:56
A: bendavid/GBRLikelihood
Rispondi a: bendavid/GBRLikelihood
Cc: Shervin Nourbakhsh
Oggetto: Re: [GBRLikelihood] TLS problem running on CMSSW_7_X_X (#7)

Can you try removing -fopenmp from the compile flags? (This is only needed for the training step and has no effect or benefit for applying the regression)

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/7#issuecomment-60222339.

[bendavid]

You get the same error or a different error?

[shervin86]

undefined reference to something.
@lbrianza can reproduce the error.

Shervin
Da: Josh Bendavid
Inviato: giovedì 23 ottobre 2014 13:01
A: bendavid/GBRLikelihood
Rispondi a: bendavid/GBRLikelihood
Cc: Shervin Nourbakhsh
Oggetto: Re: [GBRLikelihood] TLS problem running on CMSSW_7_X_X (#7)

You get the same error or a different error?

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/7#issuecomment-60222833.

[lbrianza]

Hi,
this is the output of the compilation:

Building shared library tmp/slc6_amd64_gcc481/src/HiggsAnalysis/GBRLikelihood/src/HiggsAnalysisGBRLikelihood/libHiggsAnalysisGBRLikelihood.so
tmp/slc6_amd64_gcc481/src/HiggsAnalysis/GBRLikelihood/src/HiggsAnalysisGBRLikelihood/RooHybridBDTAutoPdf.o: In function RooHybridBDTAutoPdf::UpdateTargets(int, int)': RooHybridBDTAutoPdf.cc:(.text+0x2491): undefined reference toomp_get_thread_num'
tmp/slc6_amd64_gcc481/src/HiggsAnalysis/GBRLikelihood/src/HiggsAnalysisGBRLikelihood/RooHybridBDTAutoPdf.o: In function RooHybridBDTAutoPdf::UpdateCurrentNodes(std::vector<HybridGBREvent*, std::allocator<HybridGBREvent*> > const&, HybridGBRTreeD&, int)': RooHybridBDTAutoPdf.cc:(.text+0x2f01): undefined reference toomp_get_thread_num'
tmp/slc6_amd64_gcc481/src/HiggsAnalysis/GBRLikelihood/src/HiggsAnalysisGBRLikelihood/RooHybridBDTAutoPdf.o: In function RooHybridBDTAutoPdf::EvalLossRooFit()': RooHybridBDTAutoPdf.cc:(.text+0x3442): undefined reference toomp_get_thread_num'
tmp/slc6_amd64_gcc481/src/HiggsAnalysis/GBRLikelihood/src/HiggsAnalysisGBRLikelihood/RooHybridBDTAutoPdf.o: In function RooHybridBDTAutoPdf::EvalLoss(double, TVectorT<double> const&, int)': RooHybridBDTAutoPdf.cc:(.text+0x3d01): undefined reference toomp_get_thread_num'
tmp/slc6_amd64_gcc481/src/HiggsAnalysis/GBRLikelihood/src/HiggsAnalysisGBRLikelihood/RooHybridBDTAutoPdf.o: In function RooHybridBDTAutoPdf::RecomputeTargets()': RooHybridBDTAutoPdf.cc:(.text+0x59ba): undefined reference toomp_get_thread_num'
tmp/slc6_amd64_gcc481/src/HiggsAnalysis/GBRLikelihood/src/HiggsAnalysisGBRLikelihood/RooHybridBDTAutoPdf.o: In function RooHybridBDTAutoPdf::RooHybridBDTAutoPdf(char const*, char const*, RooArgList const&, RooAbsReal&, RooRealVar&, std::vector<RooAbsData*, std::allocator<RooAbsData*> > const&, std::vector<RooAbsReal*, std::allocator<RooAbsReal*> > const&)': RooHybridBDTAutoPdf.cc:(.text+0xee62): undefined reference toomp_get_max_threads'

[bendavid]

Ok, this can be fixed with #ifndefs.

On the other hand the regression for 73x will probably avoid the GBRLikelihood dependency (we will do the training in GBRLikelihood but then copy the output into a CMSSW condformat)

Since the existing 5x regression is not very meaningful on 7x reconstruction, maybe one can simply comment out the GBRLikelihood dependency for now?