Skip to content

Check parallel WPT tests on macOS - #2

Closed
mklimenko-nv wants to merge 34 commits into
mainfrom
macos_wpt_parallel
Closed

mklimenko-nv wants to merge 34 commits into
mainfrom
macos_wpt_parallel

Conversation

@mklimenko-nv

Copy link
Copy Markdown
Owner

No description provided.

@github-actions

Copy link
Copy Markdown

Documentation Build ✅

Documentation builds successfully! 🎉

Preview will be available after merge.

@github-actions

Copy link
Copy Markdown

Documentation Build ✅

Documentation builds successfully! 🎉

Preview will be available after merge.

@mklimenko-nv

Copy link
Copy Markdown
Owner Author

Despite having 3 cores in GitHub-hosted runner, there's no speedup in WPT tests vs the baseline. Closing

@mklimenko-nv
mklimenko-nv deleted the macos_wpt_parallel branch July 14, 2026 16:40
mklimenko-nv pushed a commit that referenced this pull request Sep 16, 2026
## Summary

- lower `resample2d` for every valid pair/order of WebNN axes by moving
the selected axes to MIL's spatial dimensions, upsampling, and restoring
the original layout
- remove the final two CoreML `resample2d` expected failures (`axes: [0,
1]` and `[1, 0]`)
- add regression coverage proving WebNN `padding` is emitted as CoreML
MIL's required `pad` input, rather than the ONNX-only `pads` spelling

Fixes #2.
Fixes rustnn#54.

## Validation

- M4, Rust 1.97.0: `make fmt-check`, `make lint`, and strict Clippy with
`coreml-runtime,dynamic-inputs` pass
- Rust library tests: 354 default; 367 with
`coreml-runtime,dynamic-inputs`; all pass
- CoreML resample WPT: 13 passed, 0 skipped, 0 failed
- full CoreML WPT at `04cd4a830e26add75d385f28d95a080600b5d70f`: 2,460
passed, 41 expected skips, 20 existing negative-index scatter failures.
A clean main control reproduces the same 20 failures; the only
case-status changes are the two resample improvements
- backend operator-support documentation drift check passed
- earlier physical iPhone XS Max validation: RustNN converted the WebNN
graph to `.mlmodel` on the phone; `MLModel.compileModel(at:)` compiled
it to `.mlmodelc` on that same phone; CPU-only and normal CoreML
predictions both matched all 24 expected nearest-neighbor values exactly

---------

Co-authored-by: Markus Tavenrath <mtavenrath@users.noreply.github.com>
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.

1 participant