[Wasm RyuJit] Add Wasm SIMD NYIs, fix PushArgs assert, fix Async/Try/SCC issue. - #129494
#129494[Wasm RyuJit] Add Wasm SIMD NYIs, fix PushArgs assert, fix Async/Try/SCC issue.#129494AndyAyersMS merged 2 commits intodotnet:maindotnet/runtime:mainfrom AndyAyersMS:wasm-scc-isolated-try-entryAndyAyersMS/runtime:wasm-scc-isolated-try-entryCopy head branch name to clipboard
Conversation
AndyAyersMS
commented
Jun 16, 2026
- Report SIMD16 local loads as Wasm SIMD NYIs so R2R compilation can skip them when configured.
- Avoid assuming retbuf call arguments are PUTARG nodes on targets without a fixed register set (Wasm).
- Add temporary bidirectional edges for isolated async try side entries so SCC rewriting can see them. This handles cases where the in-try resumption target is not reachable from the try entry and throws.
…SCC issue. - Report unsupported SIMD16 local loads as Wasm SIMD NYIs so R2R compilation can skip them when configured. - Avoid assuming retbuf call arguments are PUTARG nodes on targets without a fixed register set (Wasm). - Add temporary bidirectional edges for isolated async try side entries so SCC rewriting can see them. This handles cases where the in-try resumption target is not reachable from the try entry and throws. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
|
@adamperlin PTAL Found these while working on async support for Wasm. I was seeing some SIMD16 leaking into the IR unexpectedly, so added a few more safeguards. For the Async/Try/SCC change, we had a case where an async resumption target T was in a try and immediately threw, so adding a fake edge from T to the try entry did not create a loop. So we now also add a fake edge from the try entry to T. SCC breakup then ensures that the real flow edge (from method entry to the resumption target) routes through the try header. |
There was a problem hiding this comment.
Pull request overview
This PR updates the CoreCLR JIT (with a focus on the Wasm backend) to (1) treat certain SIMD16 load patterns as Wasm-SIMD NYIs (so R2R compilation can mark methods unsupported when configured), (2) relax an argument-shape assertion for retbuf handling on targets without a fixed register set (Wasm), and (3) adjust try-region temporary edge construction to help SCC-based EH rewriting handle isolated async side entries.
Changes:
- Prevent 16-byte struct layouts from being treated as SIMD16 register types on Wasm (
ClassLayout::GetRegisterType). - Add
NYI_WASM_SIMD(...)guards for several SIMD16 load shapes in Wasm codegen (lcl var/fld, indir, load-local-into-reg). - Add temporary header→dest edges for certain multiple-entry try regions when the side-entry destination isn’t reachable within the try (to form SCCs), and relax the retbuf PUTARG assertion for !fixed-reg-set targets.
Reviewed changes
Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.
| File | Description |
|---|---|
| src/coreclr/jit/layout.h | Avoids reporting 16-byte layouts as TYP_SIMD16 register types on Wasm. |
| src/coreclr/jit/flowgraph.cpp | Adds additional temporary edges to help SCC discovery for multiple-entry try regions; includes a comment typo fix. |
| src/coreclr/jit/codegenwasm.cpp | Adds Wasm SIMD NYI reporting for several SIMD16 load paths (with a noted gap for reg-candidate locals). |
| src/coreclr/jit/codegencommon.cpp | Avoids assuming the retbuf arg is always a PUTARG node when HAS_FIXED_REGISTER_SET==0 (Wasm). |
adamperlin
left a comment
There was a problem hiding this comment.
This looks good to me from what I understand, I just had one suggestion for a comment. Thanks for the NYI SIMD fixes.
|
Tagging subscribers to 'arch-wasm': @lewing, @pavelsavara |
|
/ba-g persistent build timeouts |
…SCC issue. (#129494) - Report SIMD16 local loads as Wasm SIMD NYIs so R2R compilation can skip them when configured. - Avoid assuming retbuf call arguments are PUTARG nodes on targets without a fixed register set (Wasm). - Add temporary bidirectional edges for isolated async try side entries so SCC rewriting can see them. This handles cases where the in-try resumption target is not reachable from the try entry and throws. --------- Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>