Description

This PR generalizes the MPCD cell list code to allow triclinic simulation boxes. The cells are aligned to the lattice vectors rather than being a simple cubic lattice. This means that the MPCD cells can be skewed by a tilt factor as well as orthorhombic non-cubic.

Benchmark

SRD solvent with density 10, collision angle 130, and kT = 1 in cube with L = 120.

CPU: 2x Intel Xeon Gold 6248R = 48 cores (Auburn University Easley)

• 4.8.2: 19.5 TPS
• this branch: 19.0 TPS

GPU: 1x A100 (NCSA Delta GPU)

• 4.9.1: 200.2 TPS
• this branch: 200.5 TPS

No change on GPU (A100 is compute hungry), and only a small penalty on CPU. This performance is acceptable to me in order to enable more flexibility.

Motivation and context

This generalization of the code is necessary to enable simulation methods that deform the box, e.g., Lees-Edwards shear flow implemented using the Lagrangian rhomboid method. It also puts less restrictions on the user to make their cells exactly cubic.

Resolves #773

How has this been tested?

Existing C++ tests of the cell list have been generalized to triclinic boxes. There were some cases that no longer worked as originally written because of floating-point rounding. I am also planning a subsequent PR that will change the way MPCD does domain decomposition so that there will be no cells crossing boundaries anymore. Hence, these cases should become irrelevant.

Checklist:

• I have reviewed the Contributor Guidelines.
• I agree with the terms of the HOOMD-blue Contributor Agreement.
• My name is on the list of contributors (sphinx-doc/credits.rst) in the pull request source branch.
• I have summarized these changes in CHANGELOG.rst following the established format.