Installation

Attention

Due to classy not being available on conda-forge, we do not support installation via conda from v4 (it was supported for v3.x.x). Please use pip or uv installation as described below.

Non-Python Dependencies

21cmFAST relies on some C libraries which must be installed on your system prior to installation.

The first thing you will need is a C compiler. We support gcc (with -fopenmp support) and clang (with -fopenmp support via libomp).

The C libraries required are:

  • gsl

  • fftw (compiled with floating-point enabled, and --enable-shared)

As it turns out, though these are fairly common libraries, getting them installed in a way that 21cmFAST understands on various operating systems can be slightly non-trivial.

See the FAQ for more detailed questions on installation. If you are on MacOS and are having trouble with installation (or would like to share a successful installation strategy!) please see this discussion. Note also that we have a CI workflow (test_build_strategies.yaml) that tests various installation strategies on various OSes, so if you are having trouble with installation, please check the CI results to see if your OS and installation strategy is covered by the tests (and if not, please let us know!). Following the steps in that YAML file should lead to a successful installation, and if not, the CI results may give some clues as to what is going wrong.

One of the main things to note is that all the compiled code needs to be compiled with the same toolchain, and this toolchain also needs to be the same as the one used at import time. Very often this works out of the box (according to the below instructions), but if you have multiple compilers installed, or if you are on a HPC environment where the compilers are not in the standard location, then you may need to specify the compiler and library locations explicitly (see below). This includes the CXX compiler used to compile the C++ code in classy, which is a dependency.

If you are using conda, the easiest way to get the dependencies is to install them via conda itself. You can do this with:

conda install -c conda-forge gsl fftw compilers pkg-config

Most linux distros come with packages for the requirements, and also gcc by default. As long as these packages install into the standard location, a standard installation of 21cmFAST will be automatically possible (see below).

E.g. on Arch-based distros:

sudo pacman -S fftw gsl

On Debian-based distros:

sudo apt-get install libfftw3-dev libgsl-dev libomp-dev pkg-config

On Ubuntu (but not, apparently, Arch-based distros) if using clang instead of gcc, you will also need to point the linker to the specific version of libomp installed by apt-get. You can do this (for example, with clang v15) with:

export OMP_LIB="/usr/lib/llvm-15/lib"

The easiest way to get the dependencies (other than conda) is via homebrew. You can install them with:

brew install gsl fftw libomp

We have found that using brew tends to install compilers etc into versioned subdirectories, so if you want to use this installation strategy, you will likely need to specify the compiler explicitly. For example, in our CI tests, we have to specify (e.g.) CC=gcc-15 and CXX=g++-15 for the installation to work. See below for how to specify the compiler at installation time. On modern MacOS versions, gcc actually points to clang, which is also fine. One thing to watch out for is specifically using clang as the compiler, and having gcc not be a simple shim to clang. In this case, the classy dependency will fail, because it (incorrectly) points to gcc instead of using the CC variable.

These libraries will often be available on a HPC environment by using module load gsl fftw3 (or similar) commands.

Note also that while fftw may be available to load, it may not have the correct compilation options (i.e. float-enabled and multiprocessing-enabled). In that case, you may need to install your own local copy of fftw (and possibly gsl too) from source.

Setting correct env variables for installation

It is not guaranteed that the libraries you just installed will be able to be found by the compiler when installing 21cmFAST. If they are installed to a place not on the LD_LIBRARY/INCLUDE paths, then you must use compilation options to specify where they are. The following options will be set as options at the front of the installation command (see below).

To specify the C compiler, use CC (e.g. CC=gcc or CC=clang). Usually this will not be necessary since the default compiler should work. However, if you’d like to use clang on linux, for example, you may need to set CC explicitly.

To add a library search path, use <LIBRARY>_LIB where <LIBRARY> is either GSL or FFTW. For example, if gsl is installed in /opt/local/lib, then you would add GSL_LIB=/opt/local/lib to the installation command.

If you don’t know where the library is installed, you can try running:

locate libgsl.so
locate libfftw3.so

to find the libs, or:

locate gsl.h
locate fftw3.h

to find the include files.

Note

On MacOS (at least, 10.14-10.15), you may need to also add the following installation flag (see here):

CFLAGS="-isysroot /Library/Developer/CommandLineTools/SDKs/MacOSX<input version>.sdk"

This does not appear to be necessary for MacOS 14+, at least, not if installing the dependencies with homebrew or conda.

Other Compile Options

Similar to the CC, *_LIB and *_INC options above, there are some other options that can be set at compile time.

Debug Mode

To compile in debug mode (which enables tools like valgrind and gdb to give useful output), set the DEBUG environment variable to True (or 1).

Logging in C-Code

By default, the C-code will only print to stderr when it encounters warnings or critical errors. However, there exist several levels of logging output that can be switched on, but only at compilation time. To enable these, use the following compilation option:

LOG_LEVEL=<log_level>

The <log_level> can be any non-negative integer, or one of the following (case-insensitive) identifiers:

NONE, ERROR, WARNING, INFO, DEBUG, SUPER_DEBUG, ULTRA_DEBUG

If an integer is passed, it corresponds to the above levels in order (starting from zero). Be careful if the level is set to 0 (or NONE), as useful error and warning messages will not be printed. By default, the log level is 2 (or WARNING), unless the DEBUG=1 environment variable is set, in which case the default is 4 (or DEBUG). Using very high levels (eg. ULTRA_DEBUG) can print out a lot of information and make the run time much longer, but may be useful in some specific cases.

Installation Instructions

After following the above instructions, you are ready to install 21cmFAST. How you install 21cmFAST depends on whether you are a user or a developer, and which tool you’d like to use to install it.

First, make an isolated virtual environment and activate it (note the python version should be the latest compatible version):

conda create -n 21cmfast_env python=3.13 fftw gsl compilers pkg-config numpy scipy click pyyaml cffi astropy h5py matplotlib attrs
conda activate 21cmfast_env

uv venv --python 3.13
source activate .venv/bin/activate

uv init --name project-name --python=3.13  # see other uv options
cd project-name

python3 -m venv 21cmfast_env
source 21cmfast_env/bin/activate

Then install the latest stable release from PyPI with (see above for compile options):

[COMPILE OPTIONS] pip install 21cmFAST

[COMPILE OPTIONS] uv pip install 21cmfast

[COMPILE OPTIONS] uv add 21cmfast