[Wasm RyuJIT] Codegen for reverse pinvoke/unmanagedcallersonly

[kg]

Sufficient to compile

    [MethodImpl(MethodImplOptions.NoInlining)]
    static void * getUco () {
        return (delegate* unmanaged[Cdecl]<int>)&uco;
    }

    [UnmanagedCallersOnly(CallConvs = new[] { typeof(CallConvCdecl) })]
    static int uco () {
        return 42;
    }

[dotnet-policy-service]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

[Copilot]

Pull request overview

This PR updates the WASM RyuJIT prolog to support methods where the WASM shadow stack pointer is not passed as an explicit argument (reverse P/Invoke / UnmanagedCallersOnly entrypoints), allowing the sample in the PR description to compile.

Changes:

• Introduces a WASM global index constant intended to represent the stack pointer global.
• Updates CodeGen::genAllocLclFrame to initialize the SP local from a global when the SP “arg” is not actually a parameter (RPI case).
• Keeps the existing SP-as-first-arg path for normal managed calling convention.

[Copilot]

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 1 comment.

[kg]

PTAL @dotnet/jit-contrib

[AndyAyersMS]

I think we may run into some trouble when lvaWasmSpArg does not represent an arg.

In particular compMapILargNum and compMap2ILvarNum are not prepared to handle this case.

[kg]

Will test with a more complex method and see if I can get things to fail.

[kg]

I think we may run into some trouble when lvaWasmSpArg does not represent an arg.

In particular compMapILargNum and compMap2ILvarNum are not prepared to handle this case.

I couldn't come up with a way to get anything to fail so I tried these speculative fixes, do you think more is necessary?
e5e615f

[AndyAyersMS]

I think that's what was needed for those two. Not sure if we do this sort of thing anywhere else.

[kg]

I think that's what was needed for those two. Not sure if we do this sort of thing anywhere else.

I didn't see any other references to it that looked suspect, so I think we're good.