This repository is forked from NVIDIA cuGraph at v26.08.00.
This fork keeps the upstream cuGraph codebase as its base and carries local changes needed by
algeon-datafusion (check diff).
The fork is intentionally limited to changes required by algeon-datafusion:
- Predicate-aware single-GPU breadth-first search across the C++, C, and Python APIs, including vertex and edge filters, target discovery, and device-resident edge-ID lowering.
- Host-side memory preflights for BFS and personalized PageRank side inputs and single-GPU graph construction, allowing downstream admission decisions before GPU allocation.
- Native integration hardening through typed allocation and invalid-vertex errors, explicit symmetric-graph rejection for strongly connected components, and reliable build-helper invocation.