C++/CUDA Extensions in PyTorch

An example of writing a C++ extension for PyTorch. See here for the accompanying tutorial.

There are a few "sights" you can metaphorically visit in this repository:

• Inspect the C++ and CUDA extensions in the cpp/ and cuda/ folders,

• Build C++ and/or CUDA extensions by going into the cpp/ or cuda/ folder and executing python setup.py install,

• JIT-compile C++ and/or CUDA extensions by going into the cpp/ or cuda/ folder and calling python jit.py, which will JIT-compile the extension and load it,

• Benchmark Python vs. C++ vs. CUDA by running python benchmark.py {py, cpp, cuda} [--cuda],

• Run gradient checks on the code by running python grad_check.py {py, cpp, cuda} [--cuda].

• Run output checks on the code by running python check.py {forward, backward} [--cuda].

Authors

Peter Goldsborough