GitLab

When computing kinetic energy, make sure the shifted velocities are perpendicular to the constraints

Reduced memory use for random numbers, eliminated unnecessary calculation when generating random numbers

Fixed potential energy discrepancies in large systems between CUDA and Reference platform. The energy calculation now uses the same method as that of reciprocalConvolution so it mirrors the reference implementation. OpenCL is fine.  

When converting to fixed point, multiply by 0x100000000 instead of 0xFFFFFFFF.  This should be (very very slightly) more accurate, since its reciprocal can be exactly represented in floating point.

Began implementing new mixed precision model that does integration in double precision and force evaluation in single precision

Tony's changes to error handling in the CUDA platform so it throws exceptions instead of calling exit()

Fixed bugs that caused constraints to fail when some were implemented with CCMA and some with a different method