search

SegmentLinking / TrackLooper

Apache License 2.0
5 stars 14 forks source link

Update library paths #336

Closed ariostas closed 1 year ago

ariostas commented 1 year ago

While trying to build a self-contained image for Github Actions to use, I noticed that the library paths don't match what the chosen version of SCRAM/CMSSW uses. Everything seems to be working fine, but it might be possible to cause unexpected crashes due to version differences.

I only updated the paths to match the current SCRAM/CMSSW version (el8_amd64_gcc11/13_0_0_pre4), but it would be nice to have a more permanent fix. For CUDA/Boost/ROOT we could do something like this.

export CUDA_ROOT="$(readlink -f /cvmfs/cms.cern.ch/$SCRAM_ARCH/cms/cmssw/$CMSSW_VERSION/external/$SCRAM_ARCH/bin/nvcc | rev | cut -d'/' -f 1,2 --complement | rev)"

However, this doesn't work with Alpaka. In general, the selected versions are in xml files in /cvmfs/cms.cern.ch/$SCRAM_ARCH/cms/cmssw/$CMSSW_VERSION/config/toolbox/$SCRAM_ARCH/tools/selected/, but I don't know of an easy way to extract them.

GNiendorf commented 1 year ago

I chose this Alpaka version because it was the most up to date version available within the el8_amd64_gcc11 folder. Changing it to an older version of Alpaka would also cause a mismatch between the caching allocator we use (since it is based on our current Alpaka version). I would be surprised if this PR runs to be honest, because I thought there were some things only introduced in the most recent version of Alpaka that we make use of in our code.

ariostas commented 1 year ago

@GNiendorf Okay, then we definitely shouldn't change the version of Alpaka. I'm also surprised that it seems to run fine considering the mismatch. What do you think about the other changes?

GNiendorf commented 1 year ago

I like your idea of getting rid of hard-coded paths. We can always move to a higher version of CMSSW in our setup scrips that uses the most recent version of Alpaka if we want to go down that road. Manos and Slava probably know what is best here though.

slava77 commented 1 year ago

If the goal is to simplify the setup substantially, one option is to simply use the CMSSW software stack and not decompose the parts.

ariostas commented 1 year ago

I just realized that you can use scram tool info <toolname> to read the info in the associated xml file. So we should be able to use this to get the right paths to libraries.

ariostas commented 1 year ago

This PR was superseded by #348.