Installation instructions

Windows/Linux, using conda for Python packages

Requirements: conda Python environment, with 64 bit Python 3.9-3.13.

We provide packages for the ASTRA Toolbox in the astra-toolbox channel for the conda package manager. We depend on CUDA packages available from the nvidia channel. To install ASTRA into the current conda environment, run:

conda install -c astra-toolbox -c nvidia astra-toolbox

We also provide development packages between releases occasionally:

conda install -c astra-toolbox/label/dev -c nvidia astra-toolbox

Linux, using pip for Python packages

Requirements: Python environment with 64 bit Python 3.9-3.13.

pip install astra-toolbox

Note that, unlike conda packages, we only provide packages built for Linux platform, and only with a single reasonably recent version of CUDA toolkit. These packages depend on PyPI CUDA distribution provided by NVIDIA.

Windows, binary

Download and unpack the .zip archive for the desired version from the Downloads. Add the mex\ and tools\ subdirectories to your MATLAB path, or install the Python wheel using pip. We require the Microsoft Visual Studio 2017 redistributable package. If this is not already installed on your system, it is included as vc_redist.x64.exe in the ASTRA zip file.

Linux, from source

Requirements: automake, libtool, g++ (7 or higher), CUDA (11.0 or higher)

Build dependencies can be obtained via the OS package manager or via conda. For example, a conda environment with a full set of dependencies can be created with:

conda create -n astra-build automake libtool gxx_linux-64 cuda-minimal-build libcufft-dev python cython scipy -c conda-forge

You can then do conda activate astra-build and use --with-cuda=$CONDA_PREFIX in the build configuration.

For MATLAB

Additional requirements: MATLAB (R2012a or higher)

cd build/linux
./autogen.sh   # when building a git version
./configure --with-cuda=/usr/local/cuda \
            --with-matlab=/usr/local/MATLAB/R2012a \
            --prefix=$HOME/astra \
            --with-install-type=module
make
make install

Add $HOME/astra/matlab and its subdirectories (tools, mex) to your MATLAB path.

If you want to build the Octave interface instead of the MATLAB interface, specify --enable-octave instead of --with-matlab=.... The Octave files will be installed into $HOME/astra/octave . On some Linux distributions building the Astra Octave interface will require the Octave development package to be installed (e.g., liboctave-dev on Ubuntu).

NB: Each MATLAB version only supports a specific range of g++ versions. Despite this, if you have a newer g++ and if you get errors related to missing GLIBCXX_3.4.xx symbols, it is often possible to work around this requirement by deleting the version of libstdc++ supplied by MATLAB in MATLAB_PATH/bin/glnx86 or MATLAB_PATH/bin/glnxa64 (at your own risk), or setting LD_PRELOAD=/usr/lib64/libstdc++.so.6 (or similar) when starting MATLAB.

For Python

Additional requirements: Python (3.x), setuptools, Cython, scipy

cd build/linux
./autogen.sh   # when building a git version
./configure --with-cuda=/usr/local/cuda \
            --with-python \
            --with-install-type=module
make
make install

This will install Astra into your current Python environment.

As a C++ library

cd build/linux
./autogen.sh   # when building a git version
./configure --with-cuda=/usr/local/cuda
make
make install-dev

This will install the Astra library and C++ headers.

macOS, from source

Use the Homebrew package manager to install libtool, autoconf, automake.

cd build/linux
./autogen.sh
CPPFLAGS="-I/usr/local/include" NVCCFLAGS="-I/usr/local/include" ./configure \
  --with-cuda=/usr/local/cuda \
  --with-matlab=/Applications/MATLAB_R2016b.app \
  --prefix=$HOME/astra \
  --with-install-type=module
make
make install

Windows, from source using Visual Studio 2017

Requirements: Visual Studio 2017 (full or community), CUDA (11.0 or higher), MATLAB (R2012a or higher) and/or Python 3.x + setuptools + Cython + scipy.

Using the Visual Studio IDE:

Set the environment variable MATLAB_ROOT to your MATLAB install location.
Open build\msvc\astra_vc14.sln in Visual Studio.
Select the appropriate solution configuration (typically Release_CUDA|x64).
Build the solution.
Install by copying AstraCuda64.dll and all .mexw64 files from build\msvc\bin\x64\Release_CUDA and the entire matlab\tools directory to a directory to be added to your MATLAB path.

Using .bat scripts in `build\msvc`:

Edit build_env.bat and set up the correct library versions and paths.
- For MATLAB: Run build_matlab.bat. The .dll and .mexw64 files will be in build\msvc\bin\x64\Release_Cuda.
- For Python: Run build_python3.bat. This will produce a Wheel file in python\dist directory, which can be installed using pip.

Building conda packages

Linux

Requirements: podman and buildah.

Change to astra-toolbox/build/conda directory
Build container images by running the containers/setup*.sh scripts
Run ./release.sh

Windows

Requirements: conda-build, git, Visual Studio 2017 with Build Tools and Native Desktop workload, Windows SDK version 10.0.22621.2, CUDA toolkit of desired version(s).

Activate conda: C:\tools\miniconda3\condabin\activate.bat
Activate VS Build Tools:

"C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\VC\Auxiliary\Build\vcvarsall.bat" amd64 10.0.22621.0 -vcvars_ver=14.1

Change to astra-toolbox\build\conda directory
Build libastra packages, skipping the testing phase:

conda build -m libastra\win64_build_config.yaml -c nvidia --no-test libastra

Build and test astra-toolbox packages:

conda build -m astra-toolbox\win64_build_config.yaml -c nvidia --no-test astra-toolbox

Test the previously built libastra packages:

conda build -c nvidia --test C:\tools\miniconda3\conda-bld\win-64\libastra*.tar.bz2

Local installation

The built packages can be installed locally using conda install astra-toolbox -c nvidia -c local.

Testing your installation

To perform a (very) basic test of your ASTRA installation in Python, you can run the following command:

import astra
astra.test()

astra_test