PyTNL

PyTNL provides Python bindings for the Template Numerical Library (TNL) — a modern C++ library for building efficient numerical solvers and HPC algorithms.

TNL targets multicore CPUs, GPUs (CUDA and ROCm/HIP) and distributed-memory systems (MPI) behind a unified programming model. PyTNL exposes selected TNL building blocks to Python, enabling Python-driven workflows while keeping performance-critical kernels in compiled backends.

Scope

PyTNL is intended for:

• interoperability with TNL components where Python bindings are available,
• prototyping and orchestration of numerical solvers or HPC workflows,
• combining Python ergonomics with native code performance characteristics.

The exact set of exported classes and functions may evolve over time.

Highlights

TNL (and therefore PyTNL where bindings exist) provides building blocks such as:

• CPU/GPU-aware data structures and memory utilities
• parallel primitives (iteration patterns, reductions, scans)
• linear algebra foundations (templated vectors, dense and sparse matrices, sparse data formats (segments))
• iterative linear solvers and preconditioners (CG, BiCGStab, GMRES, Jacobi, etc.)
• ODE solvers (Euler, Fehlberg, Kutta, Heun, Runge-Kutta-Merson, Matlab, etc.)
• structured grid and unstructured mesh data structures (including distributed) and utilities, including import from FPMA, Netgen, VTK, VTU and PVTU formats

Documentation

PyTNL documentation (including installation instructions) is hosted on GitLab Pages: https://tnl-project.gitlab.io/pytnl/

License

PyTNL is provided under the terms of the MIT License.