Installation

Note

About version 2.0.0 and earlier releases

libparallelproj version 2.0.0 introduced a significant restructuring of the project. Versions earlier than 2.0.0 came from a different repository called parallelproj, which provided a mixed package containing both:

  • A minimal C and Python API for core 3D projection functions

  • Higher-level Python tools for scanner geometry, sinogram handling, and image reconstruction

Starting with version 2.0.0, libparallelproj focuses exclusively on the core projection functions and their APIs. The parallelproj-core package provides the minimal Python API to the core projectors using the stable Python ABI (python-abi3), ensuring compatibility across Python versions, without the need for recompilation.

Future versions of the existing parallelproj python package (v>= 2.0) will build upon parallelproj-core as the standard API to the projectors, providing higher-level tools for scanner geometry, sinogram handling, and image reconstruction.

The project consolidation provides a cleaner separation of concerns and improved maintainability. Version 2.0.0 and later are the recommended versions for all new projects and updates.

Install pre-compiled package from conda-forge

We strongly recommend setting up a new conda environment for using libparallelproj. This ensures that the correct dependencies are installed and prevents conflicts with other packages in existing environments (especially when using CUDA).

C-API only

mamba create -n my_new_env -c conda-forge "libparallelproj>=2"

conda create -n my_new_env -c conda-forge "libparallelproj>=2"

Note

On systems with a CUDA device, the conda-forge solver should select a CUDA build of libparallelproj automatically. CUDA 12 and 13 builds are available, and the solver should select the correct build. In case you want to force a specific build (e.g. CUDA 12 vs CUDA 13 vs CPU), you can use

mamba create -n my_new_env -c conda-forge cuda-version=12.9 "libparallelproj>=2"

mamba create -n my_new_env -c conda-forge cuda-version=13 "libparallelproj>=2"

mamba create -n my_new_env -c conda-forge "libparallelproj>=2=cpu*"

Python API (includes C-API)

Install the conda-forge package parallelproj-core:

mamba create -n my_new_env -c conda-forge parallelproj-core

conda create -n my_new_env -c conda-forge parallelproj-core

Note

parallelproj-core depends on libparallelproj. In case the solver wants to install the wrong build (e.g. CUDA vs CPU), use the note above to force the correct build of libparallelproj by adding the it as a dependency in the command above, e.g.:

mamba create -n my_new_env -c conda-forge cuda-version=12.9 parallelproj-core

mamba create -n my_new_env -c conda-forge cuda-version=13 parallelproj-core

mamba create -n my_new_env -c conda-forge "libparallelproj=*=cpu*" parallelproj-core

Compile from source

For source builds, follow the instructions in the GitHub project README