SSC-CDI: Coherent Diffractive Imaging package

Authors

• Eduardo X. Miqueles, LNLS/CNPEM
• Yuri R. Tonin
• Giovanni Baraldi

Contributors

• Alan Zanoni Peixinho, LNLS/CNPEM
• Leonardo M. Corrêa, LNLS/CNPEM
• Lucas Antonio Pelike, LNLS/CNPEM
• Paola Ferraz, LNLS/CNPEM

Past contributors

• Yuri R. Tonin
• Giovanni Baraldi
• Mauro Luiz Brandao-Junior, LNLS/CNPEM
• Camila F. A. Lages
• Julia C. Carvalho

Acknowledgements

We would like to acknowledge the Brazilian Ministry of Science, Technology, and Innovation MCTI for supporting this work through the Brazilian Center for Research in Energy and Materials (CNPEM).

Contact

Sirius Scientific Computing Team: [email protected]

Documentation

The package documentation can be found on the GCC website https://gcc.lnls.br/ssc/ssc-cdi/index.html inside the CNPEM network. Also, the HTML documentation can be found in the source directory ./docs/build/index.html and can be opened with your preferred brownser.

Citation

If you use this package in your research, please cite the following publication:

@Article{jimaging10110286,
  AUTHOR = {Tonin, Yuri Rossi and Peixinho, Alan Zanoni and Brandao-Junior, Mauro Luiz and Ferraz, Paola and Miqueles, Eduardo Xavier},
  TITLE = {ssc-cdi: A Memory-Efficient, Multi-GPU Package for Ptychography with Extreme Data},
  JOURNAL = {Journal of Imaging},
  VOLUME = {10},
  YEAR = {2024},
  NUMBER = {11},
  ARTICLE-NUMBER = {286},
  URL = {https://www.mdpi.com/2313-433X/10/11/286},
  PubMedID = {39590749},
  ISSN = {2313-433X}
}

Install

This package uses C, C++, CUDA and Python3. See bellow for full requirements.

The library sscCdi can be installed with form the source code or by pip/git if inside the CNPEM network.

GITHUB

One can clone our public github repository and install the latest version by:

git clone https://github.com/cnpem/ssc-cdi.git 
cd ssc-cdi
make clean && make

For a specific version, one can use:

    git clone https://github.com/cnpem/ssc-cdi.git --branch v<version> --single-branch
    cd ssc-cdi 
    make clean && make

The <version> is the version of the sscCdi to be installed. Example, to install version 0.14.2

    git clone https://github.com/cnpem/ssc-cdi.git --branch v0.14.2 --single-branch
    cd ssc-cdi 
    make clean && make

Source code from Zenodo

The source code can be downloaded from zenodo website under the DOI:10.5281/zenodo.13693177. On the left panel, one can find the available versions. Select the version want and download the ssc-cdi.tar.gz with the source files, one can decompress by

    tar -xvf ssc-cdi.tar.gz

To compile the source files, enter the following command inside the folder

    make clean && make

PIP

---

Warning: This installation option is available only inside the CNPEM network.

---

One can install the latest version of sscCdi directly from the pip server

    pip install sscCdi==<version> --index-url https://gitlab.cnpem.br/api/v4/projects/1978/packages/pypi/simple

Where <version> is the version number of the sscCdi

    pip install sscCdi==0.14.2 --index-url https://gitlab.cnpem.br/api/v4/projects/1978/packages/pypi/simple

GITLAB

---

Warning: For this installation option is available only inside the CNPEM network.

---

One can clone our gitlab repository and install the latest version by:

git clone https://gitlab.cnpem.br/GCC/ssc-cdi.git 
cd ssc-cdi
make clean && make

For a specific version, one can use:

    git clone https://gitlab.cnpem.br/GCC/ssc-cdi.git --branch v<version> --single-branch
    cd ssc-cdi 
    make clean && make

The <version> is the version of the sscCdi to be installed. Example, to install version 0.14.2

    git clone https://gitlab.cnpem.br/GCC/ssc-cdi.git --branch v0.14.2 --single-branch
    cd ssc-cdi 
    make clean && make

Memory

Be careful using GPU functions due to memory allocation.

Requirements

Before installation, you will need the following packages installed:

• CUDA >= 10.0.0
• C
• C++
• Python >= 3.8.0
• PIP
• libcurl4-openssl-dev
• scikit-build>=0.17.0
• setuptools>=60.0.0
• ninja==1.11.1.1
• wheel==0.45.0
• CMAKE>=3.18

This package supports nvidia GPUs with capabilities 7.0 or superior and a compiler with support to c++17.

The following modules are used:

• CUBLAS
• CUFFT
• PTHREADS

The following Python3 modules are used:

• numpy<2.0
• scikit-image
• scipy
• matplotlib
• SharedArray
• h5py
• cupy
• ipywidgets
• tqdm

Uninstall

To uninstall sscCdi use the command, independent of the instalation method,

    pip uninstall sscCdi