Optimize2

[lbluque]

Summary

• Enables openmp based parallelism for computing correlation vectors, cluster interactions, and their changes from single flips.

• Partially addresses Improve monte-carlo speed further by reducing python overheads #327
  Speed-ups can be substantial, up to 4x in the systems I tested. However with this speed-up the bottleneck in most MC will now be in the remaining python code that is run at each step (especially MCUsher.propose_step)

• This required a re-writing of the cython functions into cython extension classes, and implementing C-struct containers to allow accessing the correlation/interaction tensors that are saved as tuples of ndarrays of different dimensions in Python.

• Moved packaging to use pyproject.toml + setup.py

Additional dependencies introduced (if any)

• OpenMP is needed for parallelism (smol should still build in sequential mode if openmp is not available).

TODO

• Building wheels for macos in CI since C complilers do not have openmp as default.
• Add how to build smol from source with openmp in the Developing section of the documentation.
• Include an example notebook setting the different numbers of threads for calculations and plot resulting speed-ups.

Checklist

• Code is in the standard Python style.
  Run pycodestyle and flake8
  on your local machine.
• Docstrings have been added in the Google docstring format.
  Run pydocstyle on your code.
• Tests have been added for any new functionality or bug fixes.
• All existing tests pass.

[lbluque]

I noticed an issue with correctly checking openmp support on Debian Linux, where the check_openmp_support function used to install returns False, but forcing the install by skipping the check successfully builds with openmp support.

I do not think it is critical but will investigate further.

[lbluque]

I noticed an issue with correctly checking openmp support on Debian Linux, where the check_openmp_support function used to install returns False, but forcing the install by skipping the check successfully builds with openmp support.

I do not think it is critical but will investigate further.

The problem comes from using the system gcc and openmp libraries with conda's linker. Since that is not an issue with our code, I think we can proceed to consider merging this.

[qchempku2017]

These are a great amount of excellent work! Thank you so much for doing this to the group!