[Wasm RyuJit] Debug SPC Wasm Validation Fixes

[AndyAyersMS]

Fixes for validation issues 2, 3, 4, 5, and 7 for Debug SPC.

See #125756 (comment)

2. extend shift amount to i64 if necessary
3. fix type used to pass 8 byte structs in registers
4. Wasm relops always produce i32
5. add generic context and async continuation args for Wasm method sig

Also add another bail-out option JitR2RUnsupportedRange that causes the JIT to fail R2R compilations for methods with the indicated JIT hashes. This is useful for selectively suppressing JIT codegen that has validation failures so we can find the full set without having fixes (since the validator will stop at the first error).

[AndyAyersMS]

FYI @dotnet/jit-contrib

With these, #125812, workarounds for the asserts in #125756 and JitWasmNyiToR2RUnsupported=1, a debug SPC will validate. It contains about 14K method bodies.

[Copilot]

Pull request overview

This PR targets WebAssembly RyuJit “Debug SPC” validation failures by adjusting Wasm calling signatures and fixing/guarding several codegen and lowering cases that produce invalid Wasm.

Changes:

• Update Wasm method signature lowering to include generic-context and async-continuation hidden arguments.
• Fix/adjust Wasm codegen for relop result typing, shift amount typing, and selected intrinsics with mixed operand/result widths.
• Add a Wasm-only JIT config (JitR2RUnsupportedRange) to force R2R-unsupported bailouts for selected method hashes.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
src/coreclr/tools/Common/JitInterface/WasmLowering.cs	Extends Wasm signature construction to include async continuation + generic context hidden args.
src/coreclr/jit/morph.cpp	Tweaks struct arg field list creation for Wasm to use the segment register type.
src/coreclr/jit/jitconfigvalues.h	Adds JitR2RUnsupportedRange config string under TARGET_WASM.
src/coreclr/jit/codegenwasm.cpp	Fixes several Wasm instruction typing issues; adds CopyBlk NYI guard; updates intrinsic handling for mixed types.
src/coreclr/jit/codegencommon.cpp	Adds debug-only Wasm hook to fail R2R codegen for method-hash ranges.

[SingleAccretion]

This looks like something that needs to be handled in createSlotAccess, as another instance of LOWER_DECOMPOSE_LONGS == 0 problem. As-is we'll lose the GC types here (I don't have a clean change in mind though, it looks tricky for the general multi-reg case, and like we should really be passing regType down and making decisions based on that).

[AndyAyersMS]

I meant to bypass this for GC structs; I think the createSlotAccess logic works there, at least for now.

[jakobbotsch]

I would suggest adding an overload of ABIRegisterSegment::GetRegisterType that takes ClassLayout and teaching it to return GC types appropriately, then switch the code here to always use the regType returned by that. I think multiple places can benefit from that.

[AndyAyersMS]

GetRegisterType(ClassLayout*, unsigned offset)?

[AndyAyersMS]

Ah I guess we can assume offset is part of the seg. Seems like we already have that overload.

[jakobbotsch]

I don't think you need an offset version here, you can use the layout from the signature. Looks like the overload actually already exists. So probably

var_types regType = arg->GetSignatureLayout() != null ? seg.GetRegisterType(arg->GetSignatureLayout()) : seg.GetRegisterType()

would work.

[jakobbotsch]

(actually layout can be used in place of arg->GetSignatureLayout())

[AndyAyersMS]

Yes. How do you feel about allowing GetRegisterType(nullptr)?

[jakobbotsch]

That sounds good to me.

[AndyAyersMS]

Just changed this one spot. Verified no x64 diffs locally.

[dotnet-policy-service]

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

[jakobbotsch]

Morph changes look good to me. I think the arm64 diffs are fine (we stop doing some "out-of-bounds" loads resulting in fewer ldp's, I assume).