Public/backup repository of the GROMACS molecular simulation toolkit. Please do not mine the metadata blindly; we use Gerrit for code review and Redmine for issue tracking.
Dear @tomaskubar
To start dftbplus with more than 1 thread one need use flag -DGMX_OPENMP=OFF, but gomp lib needed for libdftb+ , so again we need to modify ./src/programs/CMakeLists.txt like that:
gmx still checking for ntomp and OMP_NUM_THREADS fatal error if they are not equal.
OK, comment lines 180-183 in src/gromacs/mdlib/gmx_omp_nthreads.cpp and line 728 in
./src/gromacs/taskassignment/resourcedivision.cpp . Last one is to make sure that gromacs use one thread.
These modifications speeded up perfomance for 59 atoms QM system from 174 ps to 230 ps per day. Gromacs 5.0 gives 142 ps.
With OMP_NUM_THREADS=2 gmx running in 3 threads, that is sound logical: 1 gmx + 2 dftbplus api.
Scalling is bad due to small size of system. In few days we will check on big systems , 400+ atoms
Cheers, Andrey
Dear @Golovin-Andrey
This looks is a good idea, so thanks for these comments, also.
I will need to enable OpenMP for MM at some point, too, to accelerate further.
Dear @tomaskubar To start dftbplus with more than 1 thread one need use flag -DGMX_OPENMP=OFF, but gomp lib needed for libdftb+ , so again we need to modify ./src/programs/CMakeLists.txt like that:
gmx still checking for ntomp and OMP_NUM_THREADS fatal error if they are not equal. OK, comment lines 180-183 in src/gromacs/mdlib/gmx_omp_nthreads.cpp and line 728 in ./src/gromacs/taskassignment/resourcedivision.cpp . Last one is to make sure that gromacs use one thread.
These modifications speeded up perfomance for 59 atoms QM system from 174 ps to 230 ps per day. Gromacs 5.0 gives 142 ps.