GitLab

Workaround for Nvidia compiler bug where compute level 1.2 processors try to compile double precision code

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.