Home

PyQUDA

Python wrapper for QUDA written in Cython. Current version: 0.10.48

This project aims to benefit from the optimized linear algebra library CuPy in Python based on CUDA. CuPy and QUDA will allow us to perform most lattice QCD research operations with high performance. PyTorch and DPNP (for Intel GPU) are alternative options.

This project is based on the latest QUDA develop branch. PyQUDA should be compatible with any commit of QUDA after PR #1489, but leave some features disabled.

Package Structure

PyQUDA is organized into four packages:

Package	PyPI Name	Description
pyquda_core	pyquda	Core QUDA wrapper: Cython bindings, lattice field classes, Dirac operators, HMC framework, gauge operations
pyquda_utils	pyquda-utils	High-level utilities: fermion inversion helpers, source generation, phase, FFT, gamma algebra, HMC parameters, I/O wrappers
pyquda_io	(included in pyquda-utils)	Low-level file I/O: Chroma, MILC, KYU, XQCD, ILDG, NERSC, OpenQCD, NPY format readers/writers
pyquda_plugins	(included in pyquda-utils)	Plugin system: auto-generate Cython wrappers from C header files

Requirements

• Python >= 3.8
• GPU backend: CuPy >= 12, PyTorch >= 2, or DPNP (experimental)
• mpi4py for multi-GPU support
• QUDA compiled as a shared library

Features

• Multi-GPU supported (MPI)
• Multiple GPU backends: CuPy, PyTorch, NumPy (CPU-only), DPNP
• Quark propagator: Wilson, Clover (with multigrid), HISQ, Staggered, Laplace
• HMC: pure gauge, 1/2-flavor Clover, N-flavor HISQ
• Gauge smearing: APE, stout, HYP (3D/4D)
• Fermion smearing: Gaussian (Wuppertal)
• Gradient flow: Wilson/Symanzik (with adjoint flow for fermions)
• Gauge fixing: over-relaxation, FFT
• Rich file I/O: Chroma QIO, MILC, KYU, XQCD, ILDG, NERSC, OpenQCD, NPY, HDF5
• Plugin system for wrapping external C libraries