[Wasm RyuJIT] Insert r2r ind cell at the end of the args list instead of after 'this'

[kg]

Without this change, the generated arglist for a ftn call looks like (sp) (indcell) (arg0) (arg1) ... when what we want is (sp) (arg0) (arg1) (indcell).

With this it's possible to evaluate a recursive fibonacci as long as you prime the heap with the right ftn ptr, i.e.

    // note: 3rd method in the assembly, so i use the ftn ptr '2' below
    [MethodImpl(MethodImplOptions.NoInlining)]
    public static int fib (int n) {
        return (n < 2) 
            ? n 
            : fib(n - 1) + fib(n - 2);
    }

> dotnet run wasm-ryujit-runner.cs -- --auto-build --config Debug --checkout Z:\runtime --assembly Z:\wasm-test\bin\Debug\net10.0\wasm-test.dll --js-expr "HEAPU32[0] = 2, exports.wasm_test_Program__fib(stackPointer.value, 31, 0)

...

running 'HEAPU32[0] = 2, exports.wasm_test_Program__fib(stackPointer.value, 31, 0)...
result was 1346269

For this test I have to write the ftn ptr 2 at offset 0 in the heap because morph generates indcells that turn into i32.const 0 for some reason I've yet to identify. We can fix that later.

[kg]

The LIR for a call in fib looks like this with my fix:

Generating: N007 (  1,  1) [000023] -----+-----                   t23 =    PHYSREG   int    $0
IN0010:             local.get 0
Generating: N008 (  1,  1) [000010] -----+-----                   t10 =    LCL_VAR   int    V01 arg0         u:1 $1 (last use) $81
IN0011:             local.get 1
Generating: N009 (  1,  2) [000011] -----+-----                   t11 =    CNS_INT   int    -2 $44
IN0012:             i32.const -2
                                                                        /--*  t10    int    
                                                                        +--*  t11    int    
Generating: N010 (  3,  4) [000012] -----+-----                   t12 = *  ADD       int    $102
Debug: Closing V01 debug range.
							Live vars after [000010]: {V00 V01} -{V01} => {V00}
IN0013:             i32.add
Generating: N011 (  1, 11) [000022] H----+-----                   t22 =    CNS_INT(h) int    0x420078 ftn $140
IN0014:             i32.const 0
Generating: N001 (  1, 11) [000028] H----------                   t28 =    CNS_INT(h) int    0x420078 ftn
IN0015:             i32.const 0
                                                                        /--*  t28    int    
Generating: N002 (  4, 13) [000029] n---G------                   t29 = *  IND       int   
IN0016:             i32.load 0 0
                                                                        /--*  t23    int    wasm sp $0
                                                                        +--*  t12    int    arg1 $1
                                                                        +--*  t22    int    r2r cell $2
                                                                        +--*  t29    int    control expr
Generating: N012 ( 19, 21) [000013] --CXG+-----                   t13 = *  CALL r2r_ind int    Program:fib(int):int $83
IN0017:             call_indirect 4611686018431713408 0      ;; Program:fib(int):int

[kg]

cc @AndyAyersMS @SingleAccretion Not sure this is the right way to go about the fix.

[dotnet-policy-service]

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

[Copilot]

Pull request overview

Adjusts CoreCLR JIT call-argument construction for WASM ReadyToRun indirect calls so the R2R indirection cell argument is appended at the end of the argument list (matching the expected LIR evaluation/stack order for the WASM backend).

Changes:

• For TARGET_WASM, append the WellKnownArg::R2RIndirectionCell argument via PushBack instead of inserting it after this/first arg.
• Keep existing behavior for non-WASM targets (still uses InsertAfterThisOrFirst).
• Clarify the FEATURE_READYTORUN #endif comment in this area.

---

You can also share your feedback on Copilot code review. Take the survey.

[SingleAccretion]

Approach LGTM.

[kg]

The indcell always being 0 is likely due to this:

runtime/src/coreclr/jit/codegenwasm.cpp

Lines 1405 to 1409 in 93ec72d

	if (icon->ImmedValNeedsReloc(m_compiler))
	{
	// WASM-TODO: Generate reloc for this handle
	ins = INS_I_const;
	bits = 0;

[AndyAyersMS]

Let's take this and work on the broader change separately.

[kg]

/ba-g wasm only change, red lanes are known failures with tracking issues