PyNX: Python tools for Nano-structures Xtallography

Introduction

PyNX stands for Python tools for Nano-structures Crystallography. It is a python library with the following main modules:

pynx.gpu: X-ray scattering computing using graphical processing units, allowing up to 2.5x10^11 reflections/atoms/seconds (single nVidia Titan X)
pynx.gid: module using pynx.gpu for Grazing Incidence Diffraction calculations, using the Distorted Wave Born Approximation
pynx.wavefield: X-ray wavefield propagation in the near, far field, or continuous (examples available at the end of wavefield.py)
pynx.ptycho : simulation and analysis of experiments using the ptychography technique, using either CPU or GPU using OpenCL. Examples are available in the pynx/Examples directory.

Download

PyNX is available from:

http://ftp.esrf.fr/pub/scisoft/PyNX/
All modules except pynx.ptycho (see http://ftp.esrf.fr/pub/scisoft/PyNX/README.txt) are also available from:
- PyPI (pip install pynx)
- http://pynx.sf.net

Pynx.gpu and pynx.gid: X-ray scattering GPU computing

Description

This module aims to help computing scattering (X-ray or neutrons) for atomic structures, especially if they are distorted or disordered.

The library uses GPU computing (although parallel CPU computing is also available as a fall-back), with the following platforms:

nVidia's CUDA toolkit and the pyCUDA library
OpenCL language, along with pyOpenCL library.

Using GPU computing, PyNX provides fast parallel computation of scattering from large assemblies of atoms (>>1000 atoms) and 1D, 2D or 3D coordinates (>>1000) in reciprocal space.

Typical computing speeds on GPUs more than 10^11 reflections.atoms/s on nVidia cards (3.5x10^11 on 2xTitan, 2x10^11 on a GTX 690, 5x10^10 on a GTX295, 2.5x10^10 on a GTX465), more than 2 orders of magnitude faster than on a CPU.

Note that the main interest of pynx.gpu is the ability to compute scattering from any assembly of atoms (not regularly-spaced) to any set of points in reciprocal space. While a FFT will always be faster, it is much more restrictive since the FFT imposes a strict relation between the sampling in real (atomic positions) and reciprocal (hkl coordinates) space.

News

2016/03: more changes to the OpenCL code, which should now avoid any issue with initialization
2012/06/26: the OpenCL code has been improved, and should provide better performance and stability (avoid 'out of resources' issues)
2011/10/01: the PyNX code has been updated to work both with CUDA and OpenCL

Citation & Bibliography

If you use PyNX for scientific work, please consider including a citation:

The PyNX article has been published: J. Appl. Cryst. 44(2011), 635-640. A preprint version is also available on ArXiv:1010.2641
Give a link to the project: http://pynx.sourceforge.net

License

The PyNX library is distributed with a CeCILL-B license (an open-source license similar to the FreeBSD one).

Pynx.wavefield: Wavefield propagation

This module allows to propagate 2D wavefield using either:

near field propagation
far field propagation
continuous propagation using the fractional Fourier Transform approach

All calculations can either be done using CPU or GPU (OpenCL)

Note that this module is in heavy development (2016/05)

Pynx.ptycho: Ptychography simulation and analysis

This modules allows the simulation and analysis of ptychography experiments, with the following features:

2D ptychography using the difference map (Thibault et al.) approach
Works with any type of illumination (sharp or not)
Object and/or Probe reconstruction
Refinement of scanning positions
GPU implementation using OpenCL is also available (in heavy development as of mid-2016):
- example speed on single Titan X as of 2016/06: 0.25s/cycle for 256 frames of 1024x1024 pixels
- GPU implementation allows using modes for probe and object