A software toolkit for generating CPU and GPU code to compute chemical kinetics, thermodynamics, and transport properties from a chemical reaction mechanism.

• Python v3.8 and later

Download or clone the repository.

Using the downloaded source code, kinetix can be installed as a Python module:

python3 install . 

python3 -m kinetix --mechanism kinetix/mechanisms/gri30.yaml --output kinetix/out/mechanisms/gri

python3 kinetix/__main__.py --mechanism kinetix/mechanisms/gri30.yaml --output kinetix/out/mechanisms/gri

Benchmark kernels are provided to evaluate the performance of the routines generated by kinetix for different computing platforms. These kernels use the OCCA library to support various hardware targets such as CPUs (SERIAL backend) and GPUs (CUDA or HIP backend).

• Linux, Mac OS X (Microsoft Windows is not supported)
• C++17/C99 compatible compilers + GNU/Intel Fortran
• MPI-3.1 or later
• CMake version 3.21 or later

• CUDA 9 or later
• HIP 4.2 or later
• SYCL 2020 or later
• Cantera 2.5 or later

To install the benchmark kernels, use the provided setup script:

cd benchmark
./install_bk.sh

For custom installation options, such as changing the installation directory or changing the amrex backend, run:

source ./install_bk.sh --help

export OCCA_CXX=mpic++
mpirun -np 1 ${KINETIX_PATH}/bin/kinetix_bk --backend SERIAL --n-states 10 --mode 1 --yaml-file kinetix/mechanisms/gri30.yaml

mpirun -np 1 ${KINETIX_PATH}/bin/kinetix_bk --backend CUDA --n-states 10 --mode 1 --yaml-file kinetix/mechanisms/gri30.yaml --unroll-loops

export OCCA_CXX=mpic++
mpirun -np 1 ${KINETIX_PATH}/bin/kinetix_bk --backend SERIAL --n-states 1000000 --mode 2  --yaml-file kinetix/mechanisms/gri30.yaml

mpirun -np 1 ${KINETIX_PATH}/bin/kinetix_bk --backend CUDA --n-states 1000000 --mode 2  --yaml-file kinetix/mechanisms/gri30.yaml --unroll-loops

Check the examples/ directory to see how the generated routines by KinetiX (or PelePhysics or Cantera) can be integrated in a simple MPI program without OCCA dependency (CPU only).

To cite KinetiX, please use this Computer Physics Communications article:

@article{Danciu2025,
  title = {KinetiX: A performance portable code generator for chemical kinetics and transport properties},
  volume = {310},
  ISSN = {0010-4655},
  url = {http://dx.doi.org/10.1016/j.cpc.2025.109504},
  DOI = {10.1016/j.cpc.2025.109504},
  journal = {Computer Physics Communications},
  publisher = {Elsevier BV},
  author = {Danciu,  Bogdan A. and Frouzakis,  Christos E.},
  year = {2025},
  month = may,
  pages = {109504}
}

KinetiX is licensed under the BSD-2 Clause License - see the LICENSE file for details.