FCHL19 with periodic boundary conditions and OpenMP.

[kvkarandashev]

FCHL19 now works with periodic boundary conditions and is parallelized with respect to atoms.

Minor OpenMP addition in kernels/fgradient_kernels.f90.

[andersx]

Fantastic contribution, thank you so much for this!

• I don't know why that OMP parallel loop was only running in serial, thanks for fixing that
• I am also not sure why the CI is not running properly, I will investigate!
• In the meantime could you add your name to the headers of the files you've changed? I'll also add you as co-author on the qmlcode.org website

Thanks again, this is awesome!

[kvkarandashev]

Thank you for the reply!

• The commented-out draft of OpenMP parallelization included applying a "REDUCTION" clause to arrays whose size scale as natoms and natoms^2, which can drastically increase memory requirements during parallel runs; I suspect that's why it wasn't used. I ended up allocating a separate temporary array with a size scaling as natoms^2 to avoid the problem.
• I've updated the file headers.

[andersx]

Thank you for the update! I'll review this tonight, so should be able to merge this in!

[kvkarandashev]

Great! Last minute update that I decided to make: I added a revised test for PBC implementation of FCHL19, the script "test_fchl_acsf_pbc.py". Before this update the fork contained first draft of the test created by Max Schwilk from benchmarks I used earlier to check that the code works correctly (it's integrated into test_fchl_acsf_force.py script and adds some LiH* files to data); I hope the duplication does not create too much confusion.