I’m trying to use QEQ model in lammps,but it wrote in the manual that：“These qeq fixes are not compatible with the GPU and USER-INTEL packages”,so if I want to achieve GPU acceleration in QEQ moudule, how I could modify the code?
Neither the INTEL nor the GPU package are set up to allow acceleration during fixes.
The limitation that the QEQ package currently imposes is - at least for the GPU package - no longer applicable, since the GPU package may be used with the newton on setting now (QEQ is incompatible with newton off which used to be required for the GPU package). But removing the check from the QEQ code will not make the QEQ calculation itself GPU accelerated.
Porting LAMMPS code to GPU acceleration is a non-trivial process and beyond what can be easily explained here. For the most part, you need to learn from the existing source code, its comments and corresponding publications how it is done.
In this specific case, probably posting QEQ fixes to the KOKKOS package may be the more straightforward approach since its approach includes accelerating fixes from the fundamental design (in fact, its design requires it). But even that requires a good understanding of advanced C++ functionality, careful study of existing code, and typically some help and notes from the LAMMPS developers.
If I want to use qeq/dynamic function of qeq module(as I understand it, it should be same as the classic MD), May it be not difficult to achieve GPU acceleration?
I don’t understand what you are asking here that wasn’t asked in your original post.
I have given you an assessment of whether charge equilibration can be used with GPU accelerated force computation and what would be required to port the charge equilibration itself to GPUs. I have nothing to add to that. If you have questions beyond that, you need to ask in a far more specific way with far more details.
OK,Thanks for your reply.
The KOKKOS package (runs on GPUs) already has support for QEq in ReaxFF. It should be fairly straightforward to extend that to general QEq.