CUDA-Torch Starter

This repository is a boilerplate for writing C++ / CUDA extension for PyTorch / libtorch.

1. Download the LibTorch and GoogleTest

1. LibTorch: https://download.pytorch.org/libtorch/cu128/libtorch-cxx11-abi-shared-with-deps-2.7.0%2Bcu128.zip

2. GoogleTest: https:////github.com/google/googletest/archive/03597a01ee50ed33e9dfd640b249b4be3799d395.zip

Place them / symbolic link them to CUDAExtension/lib/libtorch and CUDAExtension/lib/googletest respectively.

2. Use Dev Container for Correct PyTorch and CUDA ToolChain

1. Run docker compose build to build the docker image for development environment on your local machine.

2. Use VSCode devcontainer feature to launch workspace and work.

3. Adding Custom CPP Extension / CUDA Kernel

See ./CUDAExtension/README.md.

4. Running Python Scripts

1. After building the CUDAExtension package, run python -m PythonSrc.<your_script_name> and use import CUDAExtension in script to access the built extension.