[WIP] feat(sedona-spatial-join-gpu): Implement the GPU-based spatial join #439
Conversation
|
@pwrliang @paleolimbot the 186 commits here look crazy and all authors of these commits will become authors of this specific PR once merged. Is there a way to fix this? |
|
You might be able to rebase and squash commits |
|
Pretty sure when a committer merges this PR they can edit the commit message and log and clean it up and delete all the extra co authors. But still this creates extra work and 186 commits is not the usual workflow. |
|
@pwrliang the pre-commit check is failing here https://github.com/apache/sedona-db/actions/runs/20157117378/job/57861882061?pr=439 pre-commit is really a series of basic and static checks. So have a quick read of the next two links with some docs on pre-commit: https://github.com/apache/sedona-db/blob/main/docs/contributors-guide.md#linting https://github.com/apache/sedona/blob/master/docs/setup/compile.md#pre-commit Let me know if you need more help getting pre-commit working |
|
Close this PR as there's an existing one #465 |
3 participants
Pull Request Description
This pull request implements GPU-based spatial joins by extending
c/sedona-libgpuspatial. Specifically, spatial join queries involving Point-in-Polygon (PIP) operations can now be offloaded to the GPU's Ray-Tracing (RT) cores to significantly boost performance.Performance Benchmarks
The following results compare performance on a g5.2xlarge instance (NVIDIA A10) against [sedona-db on an m7i.2xlarge instance](https://sedona.apache.org/spatialbench/single-node-benchmarks/#__tabbed_2_1).
SF=1
SF=10
Future Roadmap
ST_DWithin.