I am developing an MPI-parallel program for Computational Fluid Dynamics with Fortran. I do have a few computers at home, and even a little cluster at my work, which host Nvidia card which enable CUDA programming and computing on GPUs.
So, I thought that I could gain some performance if I linked my CFD program with Nvidia’s Algebraic Multigrid library Amgx (AmgX | NVIDIA Developer) I downloaded AmgX, successfully compiled and ran some tests, and I can see it created a libraries called libamgx.a and libamgxsh.so. Would it be possible to link one of those libraries with my Fortran program and use the solver from AmgX?
Has anyone done something like that yet?