GitLab

The previous version of this code would compute
the values of d0 and m0 for every iteration. This
was unnecessary as these values are fixed at the
start of the calculation by the radii of the particles
involved. This resulted in unnecessary computation
and memory access that dramatically slowed simulations
using the GBn or GBn2 solvation models.

The new version pre-computes the values based on the
radii that are present in the system. Linear interpolation
is used, which is consistent with Amber. The previous
version of OpenMM used cubic, which gives slightly
different results.

Platform::getOpenMMVersion() includes patch version

Improved method for resizing neighbor list

Fixed edge case with 2D and 3D continuous functions

Fixed bug in PME with triclinic boxes

Fix error check for GBneck parameters

this PR

This doesn't lead to incorrect results, but it does erroneously flag some radii
as being illegal when they aren't (making it impossible to run some suggested
systems).

The parameters need to be compared to the table limits *before* subtracting off
the offset, whereas it's currently being done after.

Fixed error using CMAP with GRO files on Python 3

CPU platform shares neighbor list between forces

Changing box vectors does not change positions

Unified names of platform properties

Optimized CPU implementation of CustomIntegrator

Improved minimization with strict constraint tolerance

Changed mechanism for setting barostat temperature

Periodic boundary conditions for AMOEBA bonded forces

Fixed error in Continuous3DFunction on CUDA and OpenCL

Optimized reference implementation of continuous tabulated functions