search

LatticeQCD / SIMULATeQCD

SIMULATeQCD is a multi-GPU Lattice QCD framework that makes it easy for physicists to implement lattice QCD formulas while still providing competitive performance.
https://latticeqcd.github.io/SIMULATeQCD/
MIT License
30 stars 12 forks source link

Updated wilson corr #146

Closed RasmusNL closed 10 months ago

RasmusNL commented 10 months ago

The branch contains updates to the wilson line correlator such that hyp smearing can be used and a maximum radius can be specified.

A maximal center gauge fixing has been implemented into the gauge fixing class. The gauge fixing uses a QR algorithm defined in a class called matrix4x4_notSym.h

A quick wilon loop was also created that is now in the experimental folder.

test for the wilson line and maximal center gauge fixing has been made.

All test has been run. The following error was seen on main and on branch. OUT_HBOR_multiple_test:4 total processes failed to start OUT_HBOR_multiple_test:4 total processes failed to start OUT_HBOR_multiple_test:4 total processes failed to start OUT_RndMultipleTest:2 total processes failed to start

though no new errors has appeared, which makes sense as the changes should not affect other applications.

Also the checksum for milc does not exist, and is therefore commented out for now.

Regards Rasmus

dbollweg commented 10 months ago

Looks almost ready to merge. Two things: you commented out the Git Hash in the mpi CommunicationBase constructor. Is that causing issues or can you re-enable that? And: you commented out the checksum check for milc configurations. Can you instead make it optional so that the check is on by default but you can turn it off for your use-case?

dbollweg commented 10 months ago

Should've read your description more closely... Forget about my milc checksum comment.

RasmusNL commented 10 months ago

I dont remember the exact setup, but I think the git hash was failing for some specific modules on Marconi at some point which was the reason for commenting it out. Have uncommented it again as I have not seen the problem with the main branch. Also ran the correlator with git hash on influx and no problems.

I have also added a warning that says: rootLogger.info("WARNING: milc checksum not implemented, use with caution");

at where the checksum comparison is normally performed. When I made the milc reader I made a comparison of plaquettes with the info file. This is however performed after reading in the configuration and can be seen in the application main_wilsonLinesCorrelatorMultiGPUStacked.cpp.