NNPDF / pinecards

Runcards needed to generate PineAPPL grids for NNPDF processes
3 stars 1 forks source link

Error during linking with CC #137

Closed Radonirinaunimi closed 2 years ago

Radonirinaunimi commented 2 years ago

For some reasons I couldn't exactly figure out yet, there is an error when linking to the cc during the installation of pineappl. This causes the command ./rr run <folder> <theory> to crash. Here are the logs from running that command. In case it might be useful, here are the logs during the MadGraph run. I can install PineAPPL independently but I was wondering what could the cause of this?

PS: All the dependencies are installed in a CONDA environment and the only modification I had to do, given that i do not have permission write on the user's directory on the cluster, was at the level of the rr script:

if __name__ == "__main__":
    os.environ["HOME"] = "/data/theorie/tanjona/"
    os.environ["PATH"] += os.pathsep + "/data/theorie/tanjona/.local/bin"
    # os.environ["PATH"] += os.pathsep + "/data/theorie/tanjona/anaconda3/envs/madpineappl/bin/"
cschwan commented 2 years ago

This seems to be caused by having two (?) different compilers and possibly incomplete environment variables. Could you try to install/run everything outside your CONDA environment?

Radonirinaunimi commented 2 years ago

@cschwan This issue is still unresolved as I have not had ample time to look into it before. Nevertheless I have started running (since ~2 days ago) in parallel the MT and PTL on my personal desktop computer with 24 threads (each with 12 threads) and they are almost done. Once these are done, I will run the PTNU. But I suspect that this won't be enough at all since we want to generate many templates/pseudo-experiments?

This seems to be caused by having two (?) different compilers and possibly incomplete environment variables. Could you try to install/run everything outside your CONDA environment?

The way of installing it you mentioned above is indeed was I was planning to do at the very first. The reason that refrained me from considering this a first option was that our cluster runs on slightly out of date centos where installing all the non-python dependencies will be a pain in the neck (considering that not even gfortran is pre-installed). This is why I went for CONDA. But I will have the entire evening to try to fix this, so will update you tomorrow.

cschwan commented 2 years ago

@Radonirinaunimi it's not enough, but it's a check of whether we have enough statistics. For each bin the MC uncertainty should be well below the change induced by changing MW.

cschwan commented 2 years ago

For the final runs we'll have to decide which values of MW we'd like to scan, but we can talk about that tomorrow during the PC.

Radonirinaunimi commented 2 years ago

@Radonirinaunimi it's not enough, but it's a check of whether we have enough statistics. For each bin the MC uncertainty should be well below the change induced by changing MW.

Yes! Let's indeed see how the results look like once the runs are over.

For the final runs we'll have to decide which values of MW we'd like to scan, but we can talk about that tomorrow during the PC.

Sounds good!

enocera commented 2 years ago

@Radonirinaunimi Just for the record, here's the link to the Nikhef wiki web page with explanations on how to make use of modern compilers and other pieces of code https://wiki.nikhef.nl/grid/Modern_C_Compiler,_ROOT,_Geant

Radonirinaunimi commented 2 years ago

@Radonirinaunimi Just for the record, here's the link to the Nikhef wiki web page with explanations on how to make use of modern compilers and other pieces of code https://wiki.nikhef.nl/grid/Modern_C_Compiler,_ROOT,_Geant

Thanks a lot @enocera ! This documentation is incredibly useful.