Preparing the environment¶
The following actions are essential for the build of SIESTA to succeed. They have to be performed at least once. Then, depending on the installation method used, updating the available toolchains and libraries will be either automatic or manual.
On Linux¶
Debian-based distributions with GCC¶
On distributions like Debian, Mint or Ubuntu, the minimum requirements can be installed system-wide with the apt package manager. To enable the default GCC-based toolchain, it is as simple as typing:
sudo apt install build-essential gfortran
The build-essential package makes sure that the C and C++ compilers from GCC, as well as GNU Make, are always installed.
If you want to benefit from enhanced performance, you can also install a MPI distribution. On Debian and its derivatives, the most widely used implementation is OpenMPI:
sudo apt install libopenmpi-dev openmpi-bin
We will assume in the following that you have installed OpenMPI. If this is not the case, simply remove the -mpi- packages from the install arguments of apt.
Although SIESTA 4.x comes with an embedded linear algebra implementation, it is highly recommended to use a native one, e.g. a recent version from Netlib. With apt, this reads:
sudo apt install libblas-dev liblapack-dev libscalapack-mpi-dev
For platform-independent I/O with HDF5 and NetCDF, the following packages are necessary:
sudo apt install libhdf5-dev libhdf5-mpi-dev libnetcdf-dev libnetcdff-dev netcdf-bin
Please note that HDF5 is a very complex set of libraries, in particular when it comes to configuring and using the correct ones. You can expect a few teething problems if you are not yet familiar with it.
If you intend to download the source code of SIESTA from internet, we recommend you to install wget or curl. Both work very well, choosing one over the other is mostly a question of personal preference. If you are not sure, you can install both:
sudo apt install curl wget
In some cases, you may have to apply patches to some source file before compiling SIESTA. For this, you need difference-related utilities:
sudo apt install diffutils patch
Although this is not something you will use daily, these tools are still quite handy.
RPM-based distributions with GCC¶
On distributions like RedHat, CentOS or Fedora, the minimum requirements can be installed system-wide with the yum package manager.
On CentOS, the first step is to install the epel-release repository:
sudo yum install epel-release
It will be useful to get up-to-date libraries. After that, you can install the Development Tools set of packages and gfortran:
sudo yum groupinstall "Development Tools"
sudo yum install gcc-gfortran
In order to benefit from a MPI distribution like OpenMPI, you have to install the corresponding packages:
sudo yum install openmpi-devel
Linear algebra libraries can then be installed with:
sudo yum install blas-devel lapack-devel blacs-openmpi-devel scalapack-openmpi-devel
For platform-independent I/O with HDF5 and NetCDF, you have to install the following packages:
sudo yum install hdf5-devel hdf5-openmpi-devel netcdf-devel netcdf-fortran-devel
Please note that HDF5 is a very complex set of libraries, in particular when it comes to configuring and using the correct ones. You can expect a few teething problems if you are not yet familiar with it.
If you intend to download the source code of SIESTA from internet, we recommend you to install wget or curl. Both work very well, choosing one over the other is mostly a question of personal preference. If you are not sure, you can install both:
sudo yum install curl wget
The RedHat Development Tools come with utilities to patch source files when necessary. There is thus no need to install any additional package.
Common issues¶
What happens if you cannot become root on the computer where you would like to use SIESTA? When possible, you can point your system administrator to these instructions. Otherwise, if you have enough disk space available, you can try the approach proposed by EasyBuild, as explained in Building SIESTA with EasyBuild.
On macOS¶
Building and using Fortran programs on macOS is not the easiest of things, since it departs strongly from the usage policy of Apple. However, if you are ready for a few bumps on the road, in particular when updating the system, it is possible to develop scientific software on a macOS-based computer.
Please note that the recommended solution to set the build environment varies with time, depending on the issues encountered on the current releases of the operating systems. In 2021, it seems that Homebrew is the most satisfactory solution. In any case, there are steps that you have to perform beforehand, no matter how you install the rest of the packages.
Prerequisites¶
Before doing anything, you have to check that you have enough free disk space on your computer. You will need 30 to 40 Gb free for everything to work. If you use a Macbook Air, this can be problematic, since you might have to extend the available storage space, which has some limitations (XXX put link here).
Before developing anything, you have to install and activate XCode. Please be aware that it requires a lot of disk space to be installed and work properly. There are several ways to do it. If you follow the instructions from FreeCodeCamp, you’ll be ready to go.
Installing packages with Homebrew¶
XXX Help wanted!
On Windows¶
Coming soon …