Build WholeGraph from source#
WholeGraph is built from the
cuGraph-GNN repository. The default
build installs both libwholegraph and pylibwholegraph.
Prerequisites#
Use a Linux system and CUDA version supported by RAPIDS 26.10. The generated development environment in the repository is the authoritative dependency list.
git clone https://github.com/rapidsai/cugraph-gnn.git
cd cugraph-gnn
conda env create \
--name cugraph_gnn_dev \
--file conda/environments/all_cuda-133_arch-x86_64.yaml
conda activate cugraph_gnn_dev
Choose the generated environment file that matches the desired supported CUDA and architecture combination when more than one is present.
Build the native and Python libraries#
./build.sh clean
./build.sh libwholegraph pylibwholegraph
The build script installs into $PREFIX when it is set, then
$CONDA_PREFIX, and otherwise the system prefix. Useful options include:
-nto build without installing;-vfor verbose output;-gfor a debug build;--pydevelopfor an editable Python install;--allgpuarchto build all supported GPU architectures.
Run ./build.sh --help for the current complete option list.
Run tests#
Build the C++ test targets with:
./build.sh tests
Run the Python tests from the repository root:
python -m pytest python/pylibwholegraph
See the cuGraph-GNN contributing guide for developer workflow and style requirements.