Skip to content

Conversation

@rcoreilly
Copy link
Member

@rcoreilly rcoreilly commented Aug 10, 2025

This updates to wgpu-native v25.0.2.1 (current latest), building on the great work that @obiwac did to update the code in his earlier PR #4. It is now fully tested and working on all the cogentcore.org uses of webgpu (fairly extensive, including graphics and compute), but the libraries (which are very large collectively) are not committed:

  • Use getrelease.goal for current mechanism to get the released wgpu-native libraries downloaded and installed. The libraries will be committed to this repo when v26 comes out, but for now, this is required. This requires:
  • gh tool: https://github.com/cli/cli with authentication configured.
  • goal: go install cogentcore.org/lab/goal/cmd/goal@main

obiwac and others added 30 commits May 2, 2025 14:28
…xed that but now the callback handle is nil.
@rcoreilly rcoreilly marked this pull request as draft August 14, 2025 21:10
@mreinstein
Copy link

looks like upstream is up to v27 now

@rcoreilly
Copy link
Member Author

great! I'm seeing a lot of churn on wgpu upstream releases in last few days but I'm not sure when the dust would settle.. I'll give it a try with the current and see how it goes.

@rcoreilly
Copy link
Member Author

@mreinstein it is all working with the latest v27 wgpu_native -- only one minor change, so it seems that they've stabilized the api finally. Unfortunately it still has roughly 20% worse compute performance on linux nvidia, which is my main use-case.

The main issue is that because we ship the static libs with the library, we don't want to update too frequently, as this will bloat the overall repo size considerably. I was not able to test on other platforms yet so I'll hold off until I can do that (unless you can do so). And we'll see if there are any further updates based on all the upstream updates.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants