Installation

Download

The source code is hosted on github.

Prerequisites

To compile (and run) GPUMD one requires an Nvidia GPU card with compute capability no less than 3.5 and CUDA toolkit 9.0 or newer. On Linux systems, one also needs a C++ compiler supporting at least the C++11 standard. On Windows systems, one also needs the cl.exe compiler from Microsoft Visual Studio and a 64-bit version of make.exe.

Compilation

In the src directory run make, which generates two executables, nep and gpumd. Please check the comments in the beginning of the makefile for some compiling options.

Examples

You can find several examples for how to use both the gpumd and nep executables in the examples directory of the GPUMD repository.

NetCDF Setup Instructions

To use NetCDF (see dump_netcdf keyword) with GPUMD, a few extra steps must be taken before building GPUMD. First, you must download and install the correct version of NetCDF. Currently, GPUMD is coded to work with netCDF-C 4.6.3 and it is recommended that this version is used (not newer versions).

The setup instructions are below:

Download netCDF-C 4.6.3
Configure and build NetCDF. It is best to follow the instructions included with the software but, for the configuration, please use the following flags seen in our example line
```
./configure --prefix=<path> --disable-netcdf-4 --disable-dap
```
Here, the --prefix determines the output directory of the build. Then make and install NetCDF:
```
make -j && make install
```
Enable the NetCDF functionality. To do this, one must enable the USE_NETCDF flag. In the makefile, this will look as follows:
```
CFLAGS = -std=c++14 -O3 $(CUDA_ARCH) -DUSE_NETCDF
```
In addition to that line the makefile must also be updated to the following:
```
INC = -I<path>/include -I./
LDFLAGS = -L<path>/lib
LIBS = -lcublas -lcusolver -l:libnetcdf.a
```
where <path> should be replaced with the installation path for NetCDF (defined in --prefix of the ./configure command).
Follow the remaining GPUMD installation instructions

Following these steps will enable the dump_netcdf keyword.

PLUMED Setup Instructions

To use PLUMED (see plumed keyword) with GPUMD, a few extra steps must be taken before building GPUMD. First, you must download and install PLUMED.

The setup instructions are below:

Download the latest version of PLUMED, e.g. the plumed-src-2.8.2.tgz tarball.
Configure and build PLUMED. It is best to follow the instructions, but for a quick installation, you may use the following setup:
```
./configure --prefix=<path> --disable-mpi --enable-openmp --enable-modules=all
```
Here, the --prefix determines the output directory of the build. Then make and install PLUMED:
```
make -j6 && make install
```
Then update your environment variables (e.g., add the following lines to your bashrc file):
```
export PLUMED_KERNEL=<path>/lib/libplumedKernel.so
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:<path>/lib
export PATH=$PATH:<path>/bin
```
where <path> should be replaced with the installation path for PLUMED (defined in --prefix of the ./configure command). Finally, reopen your shell to apply changes.
Enable the PLUMED functionality. To do this, one must enable the USE_PLUMED flag. In the makefile, this will look as follows:
```
CFLAGS = -std=c++14 -O3 $(CUDA_ARCH) -DUSE_PLUMED
```
In addition to that line the makefile must also be updated to the following:
```
INC = -I<path>/include -I./
LDFLAGS = -L<path>/lib -lplumed -lplumedKernel
```
where <path> should be replaced with the installation path for PLUMED (defined in --prefix of the ./configure command).
Follow the remaining GPUMD installation instructions

Following these steps will enable the plumed keyword.