GitLab

On Fermi, store forces as 64 bit fixed point and update with atomic operations.  Also various other minor optimizations.

Change suggested by Tony Tye to simplify debugging by including all compilation options in the kernel source

Only use the atomic version of charge spreading on compute level 2.0 or later, since it is slower than the other version on earlier GPUs.

Reduced number of force buffers to one per work group.  Increased work group size for nonbonded interactions.

OpenCL kernels are now stored into a C++ class at build time rather than installing them as text files.

Add updated platform support for the Khronos approved ICD model so AMD and newer Nvidia drivers on Windows/Linux will work.  Haven't tested Apple

Created (not yet fully debugged) OpenCL implementation of VariableVerletIntegrator and VariableLangevinIntegrator.