GROMACS

HPCCM recipes for generating FFTW/GROMACS contianer specification file.

The recipe contains two sub-commands, one for FFTW and the other for GROMACS :

$ ./generate_specifications_file.py -h
./generate_specifications_file.py [-h] {fftw,gmx} ...

Generating Container Specification File for FFTW

$ ./generate_specifications_file.py fftw -h/--help
./generate_specifications_file.py fftw [-h] [--format {docker,singularity}] [--gcc {5,6,7,8,9}] [--double]
                                          (--ubuntu {16.04,18.04,19.10,20.4} | --centos {5,6,7,8}) [--fftw {3.3.7,3.3.8}] --simd
                                          {sse2,avx,avx2,avx512} [{sse2,avx,avx2,avx512} ...]

Sample command to Generate Container Specification File for Docker :

./generate_specifications_file.py fftw --format docker --ubuntu 18.04  --fftw 3.3.7 --gcc 8 --simd avx avx2 sse2 avx512 > Dockerfile

Generating Container Specification File for GROMACS

Note that the options listed here for gcc, ubuntu, centos, gromacs, fftw, cuda, cmake, openmpi, impi are guides to what can be used, and these options are not set to specific choices. engine and simd choices are constrained though, and the source code will need to be modified if you which to extend these.

$ ./generate_specifications_file.py gmx -h/--help

./generate_specifications_file.py gmx [-h] [--format {docker,singularity}] [--gcc {5,6,7,8,9}] [--double]
                                       (--ubuntu {16.04,18.04,19.10,20.4} | --centos {5,6,7,8}) [--gromacs {2019.2,2020.1,2020.2,2020.3}]
                                       [--fftw {3.3.7,3.3.8} | --fftw-container FFTW_CONTAINER] [--cuda {9.1,10.0,10.1}] [--regtest]
                                       [--cmake {3.14.7,3.15.7,3.16.6,3.17.1}] [--openmpi {3.0.0,4.0.0} | --impi {2018.3-051,2019.6-088}]
                                       [--engines simd=avx_512f|avx2|avx|sse2:rdtscp=on|off [simd=avx_512f|avx2|avx|sse2:rdtscp=on|off ...]]

Sample command to Generate Container Specification File for Docker provided with fftw version :

./generate_specifications_file.py gmx --format docker --gromacs 2020.1 --ubuntu 18.04 --gcc 9 --cmake 3.17.1 --engines simd=sse2:rdtscp=off simd=sse2:rdtscp=on  --openmpi 3.0.0 --regtest --fftw-container gromacs/fftw > Dockerfile

Sample command to Generate Container Specification File for Docker provided with fftw container :

./generate_specifications_file.py gmx --format docker --gromacs 2020.1 --ubuntu 18.04 --gcc 9 --cmake 3.17.1 --engines simd=sse2:rdtscp=off simd=sse2:rdtscp=on  --openmpi 3.0.0 --regtest --fftw 3.3.7> Dockerfile

Choosing SIMD and RDTSCP instruction for GROMACS build using the option --engines :

Value format:

 simd=avx_512f|avx2|avx|sse2:rdtscp=on|off

Example (Warning: There should be no space in --engines option value)

 simd=avx2:rdtscp=on

It is possible to choose multiple value for --engines to have multiple GROMACS installation within the same container as follows:

--engines simd=sse2:rdtscp=on simd=avx2:rdtscp=on

Generating Docker Image

docker build -t <image_name> .

The above commands assumes that it is run from the directory gromacs-hpccm-recipes-mult-stages and the Dockerfile lives in this directory. The reason is that, directory gromacs-hpccm-recipes-mult-stages contains some utility scripts what will be copied to the final image. In the essence, one have to use this directory as build context.

Running Image

The Available GROMACS wrapper binaries will be the followings based on mpi enabled or disabled (enabling double precision not tested yet):

• gmx
• gmx_mpi

With Singularity

Build Singularity image from Docker:

singularity build <name of image to be built> docker://<docker_image>

Example of running mdrun with gmx_mpi wrapper:

mpirun -np <no of processes> singularity exec -B <host directory to bind> <singularity image> gmx_mpi mdrun -s <.tpr file> -deffnm <ouput_file_name>

To enable NVIDIA GPU support, please add --nv flag to singularity exec command as follows:

mpirun -np <no of processes> singularity exec --nv -B <host directory to bind> <singularity image> gmx_mpi mdrun -s <.tpr file> -deffnm <ouput_file_name>

Before running the above command, you have to make sure that you have added appropriate module for gcc, openmpi and cuda.

Without Singularity

Bind the directory that you want Docker to get access to. Below is an example of running mdrun module using gmx wrapper:

mkdir $HOME/data
docker run -v $HOME/data:/data -w /data -it <image_name> gmx mdrun -s <.tpr file> -deffnm <ouput_file_name>

Dependencies

• python3
• libgomp1
• openmpi
• fftw