CUDA pair_sw bug/limitation

Hi all

I was having trouble when using the CUDA SW potential. Eventually I found out that the atom type was being mapped to random elements when the atom type > 4.

The problem disappeared when the declaration

device constant int map[MANYBODY_NPAIR + 2];

in /lib/cuda/pair_manybody_const.h was changed to

device constant int map[number_of_different_atoms_types + 1];

where number_of_different_atoms_types was hard-coded. MANYBODY_NPAIR is #defined 3 in the same file.


Christian - can you say something about any limitations on atoms types in pair_style sw/cuda ?