The goal of the project is to play around and create a simplistic boilerplate code for Python-C++ and Python-CUDA integrations.
- pip install .
In [1]: import cpp_bindings
In [2]: cpp_bindings.get_result()
Out[2]: 42
In [3]: import python_code
In [4]: python_code.get_result2()
Out[4]: 84
Todo:
- builable C++ code
- Python wrapper for the C++ code
- setup.py for Python package
- simple and runable CUDA kernel
- Python wrapper and setup.py work with CUDA
- find a way to return GPU tensor from C++ directly to Python/pytorch (without copy to host)
Finally:
- simple API
- Dockerfile and docker-compose