Metadata-Version: 2.2
Name: graphgp-cuda
Version: 0.0.3
Summary: Custom CUDA extension for GraphGP package.
Author-Email: Benjamin Dodge <bendodge@stanford.edu>, Philipp Frank <phfrank@stanford.edu>
Project-URL: Source, https://github.com/stanford-ism/graphgp-cuda
Requires-Python: >=3.11
Requires-Dist: jax[cuda12]
Provides-Extra: dev
Requires-Dist: jax[cuda12]==0.9.0; extra == "dev"
Requires-Dist: pytest>=8.3.5; extra == "dev"
Requires-Dist: ruff>=0.11.11; extra == "dev"
Description-Content-Type: text/markdown

# graphgp-cuda

This is the custom CUDA extension for [GraphGP](https://github.com/stanford-ism/graphgp). It is not intended to be used as a standalone package.

The graph construction uses two GPU-friendly tree algorithms [[2](https://arxiv.org/abs/2211.00120), [3](https://arxiv.org/abs/2210.12859)], which were originally implemented in the [cudaKDTree](https://github.com/ingowald/cudaKDTree) library.

For in-place sorts, we use a modified version of the Apache 2.0-licensed [cubit](https://github.com/ingowald/cudaBitonic) bitonic sorting code, which can be found in "graphgp_cuda/sort.h".