muGrid 0.111.0

MPI-parallel regular grids

µGrid

µGrid is a C++ library for discrete representation of fields on structured grids, with Python bindings. It provides efficient data structures and algorithms for solving partial differential equations on regular grids, with support for MPI parallelization and GPU acceleration.

Features

• Field collections: Manage scalar, vector, and tensor fields on structured grids with flexible memory layouts
• Domain decomposition: Cartesian decomposition with ghost cell communication for stencil operations
• Linear operators: Discrete differential operators including Laplacian and FEM gradient operators for spectral methods
• FFT engine: Built-in Fast Fourier Transform with MPI-parallel support (auto-selected slab or pencil decomposition) and native cuFFT/rocFFT N-D transforms on the GPU
• GPU support: Optional CUDA and HIP backends for GPU-accelerated computation
• NetCDF I/O: Serial and parallel file I/O for checkpointing and analysis

µGrid is written in C++20 and has language bindings for Python via pybind11.

This README contains only a small quick start guide. Please refer to the full documentation for more help.

Quick start

To install µGrid, run

pip install muGrid

Note that on most platforms this will install a binary wheel, that was compiled with a minimal configuration. To compile for your specific platform use

pip install -v --no-binary muGrid muGrid

which will compile the code. µGrid will autodetect MPI. GPU support (CUDA/ROCm) is off by default and must be enabled explicitly at build time (see the documentation); it is not autodetected. For I/O, it will try to use Unidata NetCDF for serial builds and PnetCDF for MPI-parallel builds. Monitor output to see which of these options were automatically detected.

Funding

This development has received funding from the Swiss National Science Foundation, the European Research Council, and the Deutsche Foschungsgemeinschaft.