GitLab

Variable time step integrators correctly report the size of the last step taken.  Also improved accuracy of integration unit tests.

NonbondedForce sometimes included energy terms that weren't appropriate to the specified force groups

Eliminated having CMake run programs to check for CUDA and OpenCL compatible GPUs.  They mostly just caused problems.

Reimplemented SETTLE velocity constraints with a different algorithm.  The old one produced incorrect results.

Implemented per-DOF uniform random numbers for OpenCL.  Also fixed some bugs in getting and setting per-DOF variables.

Continuing to implement OpenCL CustomIntegrator: added velocity constraints (not yet fully debugged)

Continuing to implement OpenCL CustomIntegrator: added position constraints, and use double precision where possible