Python C Extension


In this project, we show how to develop a C extension for Python. The C library will be interfaced with Python using ctypes. Additionally, the C library will be accelerated via parallelization using OpenMP. The entire code is finally containerized using Docker for easy deployment.

The following tools will be used in this project:

  • Python
  • C
  • OpenMP
  • Docker


Find the source code in the repository.

Learning Outcome

At the end of this project, we should be able to:

  • Call C/C++ compiled code from Python using the ctypes and numpy.ctypeslib modules
  • Parallelize C code with OpenMP
  • Dockerize Python, C, and OpenMP code

Project Structure

The project structure is as follows:

python-c-extension                   # Repository root
├── example                      
│   └──              # Main Python code
├── lib                       
│   ├── cpp                          # C folder
│   │   ├── build                    # C build system
│   │   ├── lib                     
│   │   │   └── feature               
│   │   |       ├── CMakeLists.txt 
│   │   |       ├── feature.c        # C library
│   │   |       └── feature.h        # C header file
│   │   ├── out 
│   │   |   └──        # C library shared object
│   │   └── CMakeLists                
│   └──                   # Python library
├── .dockerignore
├── .env                             # Config file 
├── .gitignore
├── docker-compose.yml               # Docker deployment
├── Dockerfile                       # To create Docker container
└── requirements.txt                 # Python dependencies


  1. Download the code using git clone
  2. Install Docker and Docker Compose in your computer
  3. Run the code using Docker
     $ cd /path/to/repository/root
     $ docker build -t pycextension .
     $ docker-compose up
  4. If you intend to run the project locally without Docker
    • Change the ROOT variable in the .env file to point to the repository’s root.
    • Build the C code using:
        $ cd /path/to/pythoncextension/lib/cpp/
        $ cmake -E make_directory build 
        $ cmake -E chdir ./build cmake -DCMAKE_BUILD_TYPE=Release ..
        $ cmake --build ./build
    • Run the Python code:
        $ python /path/to/python-c-extension/example/   

System Design

Certain key aspects of the system is further explored in the following sections.

Python code

  1. The main Python code lies in /example/ file.
  2. A dummy Numpy matrix is generated and passed into a Python wrapper function, namely, lib.feature.synthesize_matrix().
  3. The Python wrapper function calls an underlying C shared object to perform some random computation and returns a modified matrix.
    • Note that the data type of the output matrix is float whereas the data type of the input matrix was int.
  4. In the /lib/ file,
    • The shared library is opened using CDLL function.
    • The argument and return types of the C function are specified.
  5. Some translations between Python and C types are as follows:
    Python C/C++ type Comments
    ctypes.c_int int
    ctypes.c_double double
    numpy.ctypeslib.ndpointer(dtype=np.uint8, ndim=1) uchar[ ] Pass a 1 dimensional Numpy array of type numpy.uint8
    numpy.ctypeslib.ndpointer(dtype=np.float32, ndim=2) float[ ][ ] Pass a 2 dimensional Numpy array of type numpy.float32

C library

  1. The C libraries reside in /lib/cpp/lib/ folder.
  2. The code in synthhesize_matrix() function is parallelized using OpenMP. OpenMp #pragma commands are used to set the number of threads, collapse nested for loops, and to implement a worksharing for loop.
  3. The C code is compiled with the help of CMake as defined in CMakeLists.txt file.
  4. In the /lib/cpp/CMakeLists.txt file, we set the output directory of the shared object .so to be /lib/cpp/out/ as follows.

    This simplifies the process of finding and referring to the shared object in our Python code.

Python depedencies

  1. Python dependencies are managed in /requirements.txt file.
  2. Please refer to the programming guide for explanation on how to manage, generate, and install, python dependencies


  1. All necessary Python, pip, CMake, and OpenMP libraries are installed by the /Dockerfile.
  2. The Dockerfile also automatically rebuilds the C library shared object each time the Docker image is built via docker-compose up command.


Leave a comment