Re: [AMBER-Developers] Just ran make test.cuda.serial.SPFP on master TOT, TLDR 37 test failures

From: Jason Swails <jason.swails.gmail.com>
Date: Tue, 3 Nov 2020 19:08:57 -0500

--
Jason M. Swails 
> On Nov 3, 2020, at 6:29 PM, Scott Le Grand <varelse2005.gmail.com> wrote:
> 
> 230 file comparisons passed
> 37 file comparisons failed (3 of which can be ignored)
> 0 tests experienced errors
> Test log file saved as
> /media/work/slegrand/amber/logs/test_amber_cuda/2020-11-03_15-16-44.log
> Test diffs file saved as
> /media/work/slegrand/amber/logs/test_amber_cuda/2020-11-03_15-16-44.diff
> slegrand.samaritan:/media/work/slegrand/amber/test$ git branch
> * master
>  slegrand-extendedminimization
This has been this way for years. The unit tests used to just run against DPFP because SPFP varied too much between GPU models to get reliable passing tests. 
But doing that masked a bug which caused SPFP to crash. So the CUDA tests are currently required to pass all DPFP tests and pass more SPFP tests than fail and finish with 0 errors. 
The alternative was basically to turn off CI for CUDA or turn off SPFP tests completely. 
Thanks,
Jason
> _______________________________________________
> AMBER-Developers mailing list
> AMBER-Developers.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber-developers
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Tue Nov 03 2020 - 16:30:02 PST
Custom Search