facebookincubator / velox

A composable and fully extensible C++ execution engine library for data management systems.
https://velox-lib.io/
Apache License 2.0
3.53k stars 1.16k forks source link

fix: Add support for sm5x CUDA archs to breeze #11593

Closed dreveman closed 3 days ago

dreveman commented 5 days ago

Code was limited to sm60 and above previously. This makes it compile for sm50 too, which is the default for nvcc.

Also adds a ubsan build type and fixes some issues that caused building without tracing enabled to fail.

netlify[bot] commented 5 days ago

Deploy Preview for meta-velox canceled.

Name Link
Latest commit 829fded0ef076a6fbe9f88f7b49f758cf3a913cd
Latest deploy log https://app.netlify.com/sites/meta-velox/deploys/673ce3b3c6d6ab0008d7fec8
dreveman commented 5 days ago

@Yuhta some build fixes

dreveman commented 4 days ago

@Yuhta Please review this when you have a chance, thanks!

dreveman commented 4 days ago

Not urgent, but would be nice if you can run some of the tests in https://github.com/facebookincubator/velox/blob/main/.github/workflows/linux-build-base.yml

@Yuhta Yes, happy to add that. Should we limit it to OpenMP or can we also run CUDA tests? And how about also running some tests on MacOS?

Yuhta commented 4 days ago

@dreveman If we can run GPU tests we can also enable CUDA tests, otherwise let's just enable OpenMP for now. Let's just run it for Ubuntu for now, and if we see a lot compatibility issue with MacOS we can add check for that. @assignUser Can we run GPU tests on GitHub action now?

assignUser commented 4 days ago

Can we run GPU tests on GitHub action now?

No that stalled out a while ago #9527

Yuhta commented 4 days ago

@dreveman If your team has bandwidth to help setting up GPU on GitHub action it would be nice. We are only running some very minimal sanity check for breeze/CUDA inside Wave.

dreveman commented 4 days ago

@dreveman If your team has bandwidth to help setting up GPU on GitHub action it would be nice. We are only running some very minimal sanity check for breeze/CUDA inside Wave.

Yes, I'll see what we can do here. But let's start with running OpenMP tests and building for CUDA (and maybe also macos) as a start.

facebook-github-bot commented 4 days ago

@kagamiori has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

facebook-github-bot commented 3 days ago

@kagamiori merged this pull request in facebookincubator/velox@9b3fbcd6d7ded899faf9b1c460fdbbf2b21927b7.