I am developing a python script that will import LAMMPS as a module and perform a series of energy minimizations. The employed force field is specified via the kim commands of the OpenKIM interface. I have compiled LAMMPS as library (with the GPU package enabled) and I have been able to run the python script in serial mode successfully.
How can I port the calculations to the GPU package (provided that the force field is supported)? I think that I cannot explicitly specify the name of the pair style due to OpenKIM, i.e. I cannot write down a command of the style “pair_style lj/cut/gpu”.
I am developing a python script that will import LAMMPS as a module and perform a series of energy minimizations. The employed force field is specified via the kim commands of the OpenKIM interface. I have compiled LAMMPS as library (with the GPU package enabled) and I have been able to run the python script in serial mode successfully.
How can I port the calculations to the GPU package (provided that the force field is supported)? I think that I cannot explicitly specify the name of the pair style due to OpenKIM, i.e. I cannot write down a command of the style “pair_style lj/cut/gpu”.
not sure if this is within the conceptual boundaries of the OpenKIM philosophy, as switching to a different implementation (i.e. GPU code version versus the CPU code version) invalidates the reproducibility goal of the OpenKIM project. so the logical (to me) approach would be to submit the corresponding GPU accelerated model to the OpenKIM database.
that said, have you tried using the package command to enable the gpu library acceleration and the suffix command to enable and set transparent appending of a designated suffix?