GPU Library Update with PPPM...

Š available for testing. It can be downloaded here:

http://users.nccs.gov/~wb8/gpu/download.htm

Follow the instructions on the page to update a current LAMMPS
distribution. Note that these instructions will OVERWRITE default
Makefiles in lib/gpu, so make a copy of any edited Makefiles you need.

The update includes

- gpu acceleration for pppm, lj/expand, and morse
- support for multiple gpu styles in hybrid pair styles

PPPM can be used with or without pair-styles that support gpu
acceleration. If pppm is the only gpu style, newton can be on, otherwise
it should be off. If pppm is used with another gpu pair style, part of the
pppm calculation can be performed simultaneously with CPU non-bonded and
bonded calculations. The gpu-accelerated PPPM styles are:

pppm/gpu/single
pppm/gpu/double

Where single and double are the precision used for mesh calculations. Both
options are available regardless of the precision used for the rest of the
gpu library. FFTs are still performed on the CPU. Someone could write
wrappers to use the CUDA fft library in LAMMPS, but I don't think this
would help much for most parallel simulations.

Please send any issues, comments, suggestions, etc. directly to me until
this is available in the main LAMMPS distribution.

Thanks.

- Mike