abinit channel

This repository contains conda recipes to build Abinit-related packages. The pre-compiled libraries and executables for Linux and MacOSx are available on the abinit channel. The goal is to facilitate the installation of abinit to end-users who want to run the code on their personal computers without having to pass through the compilation process. This channel also provides conda packages for AbiPy to install a powerful Python + Fortran ecosystem for materials science research.

For a quick howto with the five commands required to install Abinit on your machine (Linux or MacOSx), jump immediately to the next section. A more detailed discussion about the installation with conda, and the use of conda environments is given in this section. For further information on the conda package manager, please consult the official conda documentation.

Note that these pre-compiled executables are useful if you want to try Abinit on your machine but they are not supposed to be used for high-performance calculations.

For examples of configuration files to configure/compile Abinit on clusters, please visit the abiconfig repository. If you need a real package manager able to support multiple versions and configurations of software, consider the following projects:

• spack
• easybuild

Both projects are designed for large supercomputing centers and they already provide configuration files to build Abinit.

How to install Abinit in five steps

If you are a Linux user, download and install miniconda on your local machine with:

wget https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh

while for MacOSx use:

curl -o https://repo.continuum.io/miniconda/Miniconda3-latest-MacOSX-x86_64.sh
bash Miniconda3-latest-MacOSX-x86_64.sh

Answer yes to the question:

Do you wish the installer to prepend the Miniconda3 install location
to PATH in your /home/gmatteo/.bashrc ? [yes|no]
[no] >>> yes

Source your .bashrc file to activate the changes done by miniconda to your $PATH:

source ~/.bashrc

Add conda-forge to the conda channels:

conda config --add channels conda-forge

Install the parallel version of abinit from the abinit channel with:

conda install abinit --channel abinit
abinit -v

The troubleshooting section discusses how to solve typical problems.

Getting started

Download anaconda for your operating system from https://www.continuum.io/downloads. anaconda is a distribution with the most popular Python packages for data science and includes conda the cross-platform, language-agnostic package-manager required to install Abinit. If you don't need the entire anaconda distribution, start with miniconda which contains only conda and Python.

By default, the installer adds the following line to your .bash_profile:

export PATH="/Users/gmatteo/anaconda3/bin:$PATH"

so we have to source ~/.bash_profile before continuing in order to have the conda executable in our $PATH.

Create a new conda environment (let's call it abienv) with:

conda create -n abienv

and activate it with:

source activate abienv

It's always a good idea to install Abinit and its dependencies in a separate environment because installing one program at a time can lead to dependency conflicts. For more information about conda environment, consult the official documentation.

The most important (pre-compiled) libraries will be obtained by conda-forge so we have to add conda-forge to the list of default channels with:

conda config --add channels conda-forge

Now we can install abinit from the abinit channel with:

conda install abinit --channel abinit

This command downloads and installs the last version of Abinit from the abinit channel. The Abinit executables are placed inside the anaconda directory associated to the abienv environment:

which abinit
/Users/gmatteo/anaconda3/envs/abienv/bin/abinit

Linux users can use the shell command:

ldd `which abinit`

to get the list of dynamic libraries linked to the application whereas macOSx can use:

otool -L `which abinit`

/Users/gmatteo/anaconda3/envs/abinit/bin/abinit:
        /usr/lib/libc++.1.dylib (compatibility version 1.0.0, current version 307.4.0)
        @rpath/libnetcdff.6.dylib (compatibility version 6.0.0, current version 6.1.1)
        @rpath/libnetcdf.11.dylib (compatibility version 11.0.0, current version 11.4.0)
        @rpath/libhdf5_hl.10.dylib (compatibility version 12.0.0, current version 12.0.0)
        @rpath/libhdf5.10.dylib (compatibility version 13.0.0, current version 13.0.0)
        @rpath/libfftw3f.3.dylib (compatibility version 9.0.0, current version 9.6.0)
        @rpath/libfftw3.3.dylib (compatibility version 9.0.0, current version 9.6.0)
        @rpath/libopenblasp-r0.2.19.dylib (compatibility version 0.0.0, current version 0.0.0)
        @rpath/libmpifort.12.dylib (compatibility version 14.0.0, current version 14.0.0)
        @rpath/libmpi.12.dylib (compatibility version 14.0.0, current version 14.0.0)
        @rpath/libpmpi.12.dylib (compatibility version 14.0.0, current version 14.0.0)
        @rpath/./libgfortran.3.dylib (compatibility version 4.0.0, current version 4.0.0)
        /usr/lib/libSystem.B.dylib (compatibility version 1.0.0, current version 1238.0.0)
        @rpath/./libquadmath.0.dylib (compatibility version 1.0.0, current version 1.0.0)

The output of otool indicates that this executable is linked against openblas, MPI (mpich), FFTW and netcdf4+hdf5. The BLAS library is threaded and the number of threads used at runtime is defined by the environment variable OMP_NUM_THREADS. We strongly suggest to add:

export OMP_NUM_THREADS=1

to your .bash_profile because by default the OpenMP library uses all the available cores if this variable is not defined.

Alternatively, one can install the sequential version (minimal dependencies, no parallelism) with:

conda install abinit_seq --channel abinit

To get the list of Abinit versions available in the abinit channel:

conda search abinit --channel abinit

    Fetching package metadata ...........
    abinit                       8.0.8                         0  abinit
                              *  8.2.2                         0  abinit
    abinit_seq                *  8.2.2                         0  abinit

To install a particular version of Abinit use:

$ conda install abinit=8.2.0 --channel abinit

Packages

abinit:

Version with MPI support, libxc, fftw3, openblas and netcdf4 + hdf5

abinit_seq:

Sequential version with internal fallbacks for libxc and netcdf3 (statically linked).

oncvpsp:

Generator for norm-conserving pseudopotentials compiled with libxc support

atompaw:

Generator for PAW datasets compiled with libxc support

Troubleshooting

• All the conda applications should use libraries installed inside the conda environment. As a consequence, the use of the $LD_LIBRARY_PATH (linux) or $DYLD_LIBRARY_PATH (MacOsx) environment variables is strongly discouraged as it can lead to runtime errors and malfunctioning. So it is recommended to unset them if they are set, unless you know what you are doing (use conda info -a to get info on your environment).

• The parallel version of Abinit should be launched with the mpirun executable provided by conda so make sure that the bin directory of anaconda comes before the other directories and use:

  $ which mpirun
  ~/anaconda3/envs/abienv/bin/mpirun
  

• If the parallel version of Abinit aborts with the following error:

    Fatal error in MPI_Init: Other MPI error, error stack:
    MPIR_Init_thread(474)..............:
    MPID_Init(190).....................: channel initialization failed
    MPIDI_CH3_Init(89).................:
    MPID_nem_init(320).................:
    MPID_nem_tcp_init(173).............:
    MPID_nem_tcp_get_business_card(420):
    MPID_nem_tcp_init(379).............: gethostbyname failed, gmac2 (errno 1)

open /etc/hosts with e.g. sudo vi /etc/hosts and add a new entry mapping the 127.0.0.1 ip address to the name of your machine e.g.:

    127.0.0.1       localhost
    127.0.0.1       gmac2      # Add this line. Replace gmac2 with the name of your machine.

• Most of the dependencies including libgcc and libgfortran are automatically installed in your conda environment when you issue conda install APPNAME -c abinit. In principle, these libraries should be compatible with the abinit executables but incompatibilities may appear when the conda developers decided to upgrade the gcc version or if other tricky dependencies such as the MPI library are upgraded upstream. In this case, contact us and we will try to provide new pre-compiled versions compatible with the new anaconda software stack. Alternatively, you may try the sequential version abinit_seq in which the number of external dependencies and therefore the probability of linkage problems is significantly reduced.

• All the dependencies are provided by conda with the exception of the C standard library (libc on linux, libSystem.B.dylib on MacOsx). If the C library provided by your OS is too old, you will get error messages such as:

  abinit
  abinit: /lib64/libc.so.6: version `GLIBC_2.14' not found (required by abinit)
  

  when you try to execute the application from the terminal. ldd indeed shows that our system uses libc 2.12

  ldd --version
  ldd (GNU libc) 2.12
  

  This means that our executable requires a C library that is not compatible with the one available on your system (this usually happens when your library is too old). At the time of writing, we build executables and libraries for linux with GNU libc 2.12 while MacOsx applications are built with MacOS 10.11.2. This should cover the most common cases but it your OS is too old you will have to compile from source.

  Note that supporting all the possible libc versions is not easy since we should set up a machines with the same C-library as the one used on your system, find a version of conda that works with this configuration and finally try to recompile the application and the corresponding libraries)

For more info, please consult the official conda documentation