[wip][Wasm] Emit R2R branch hints from profile data

[lewing]

Summary

Emit the standardized WebAssembly metadata.code.branch_hint custom section from CoreCLR ReadyToRun compilation when static PGO identifies strongly biased conditional branches.

This is an experimental implementation for #133794. Engines that do not consume branch hints safely ignore the custom section.

Design

• Select hints in RyuJIT after final Wasm CFG layout and condition reversal.
• Require static, sufficiently trustworthy PGO and an 80/20 likelihood threshold in production.
• Attach the final hint to the emitted br_if instruction descriptor.
• Report function-relative instruction offsets through the existing JIT metadata channel.
• Retain hints as ReadyToRun per-method side data and resolve final function indices in the Wasm object writer.
• Emit one sorted metadata.code.branch_hint custom section immediately before the code section.
• Adjust hint offsets when Webcil relocation resolution shrinks variable-length LEB encodings.
• Reset reported metadata on internal JIT retries so discarded compilation attempts cannot leak stale hints.
• Keep relocatable Wasm output explicitly unsupported because its function indices and code offsets are not final.

A JitWasmBranchHintStress switch permits deterministic tests using synthesized profile weights without changing production gating.

Validation

• ./build.sh clr
• PATH=/opt/homebrew/bin:$PATH ./build.sh -os browser -c Debug -subset clr+libs
• python3 src/coreclr/scripts/jitformat.py -r <runtime> -o osx -a arm64 -b Checked --verbose --fix --projects dll
• Full browser-wasm ILCompiler.ReadyToRun.Tests: 75 passed, 37 skipped, 0 failed
• Independent Wasm parsing confirmed the custom section is present before the code section.
• Tests verify function and branch ordering, metadata shape, final function indices, funclet handling, relocation-adjusted offsets, and that each hint points to an actual if or br_if opcode.
• Tests also verify the section is absent without qualifying profile data or the test stress switch.

Size impact

For the test module, identical synthesized profile input produced:

Configuration	Size
Profile data, branch hints disabled	6,528 bytes
Profile data, branch hints enabled	6,560 bytes

The custom section and hints add 32 bytes in this sample.

Follow-up measurements

This draft establishes correct metadata generation. Before considering it ready to merge, it needs end-to-end measurements on Wasm engines that consume branch hints, including representative application startup/steady-state results, engine compilation time, generated native-code size, and broader R2R image-size impact.

Note

This pull request description was drafted with GitHub Copilot.

[azure-pipelines]

Azure Pipelines:
Successfully started running 6 pipeline(s).
10 pipeline(s) were filtered out due to trigger conditions.
There may be pipelines that require an authorized user to comment /azp run to run.

[dotnet-policy-service]

Tagging subscribers to this area: @dotnet/crossgen-contrib
See info in area-owners.md if you want to be subscribed.

[dotnet-policy-service]

Tagging subscribers to 'arch-wasm': @lewing, @pavelsavara
See info in area-owners.md if you want to be subscribed.

[lewing]

I'm exploring tooling changes against this prototype.

[AndyAyersMS]

We don't want to call report here, it will get called automatically when the jit is done with the method. Instead, increment the metric.

[AndyAyersMS]

Omit this (see below).

If you want to know what % of branches got hinted it will take more work.

[AndyAyersMS]

I would just call fgHaveProfileWeights here and not worry about the source or quantity.

[AndyAyersMS]

Instead of forcing on hints then you could force them off here.