Generic math does not work in WASM AOT

[kg]

Description

Any code using Generic Math will produce runtime errors in an AOT'd wasm application. It works in the interpreter.

Reproduction Steps

    private static T LerpGeneric<T> (T a, T b, T t) 
        where T : IAdditionOperators<T, T, T>, 
            ISubtractionOperators<T, T, T>, 
            IMultiplyOperators<T, T, T>
    {
        var size = b - a;
        var offset = size * t;
        return a + offset;
    }
    
    [BenchmarkCategory(Categories.Runtime)]
    [Benchmark]
    public double Lerp_Generic ()
    {
        double result = 0;
        for (int i = 0; i < IterationCount; i++) {
            double accumulator = 0;
            for (int j = 0; j < ArrayLength; j++) {
                double t = j / dArrayLength;
                accumulator += LerpGeneric(A[j], B[LengthMinusOne - j], t);
            }
            result += accumulator;
        }
        return result;
    }

Expected behavior

Working generic math (even if it falls back to the interpreter)

Actual behavior

console.error: RuntimeError: null function or function signature mismatch
console.error: RuntimeError: null function or function signature mismatch
    at MicroBenchmarks_GenericMathTest_Lerp_Generic (wasm://wasm/2627c6ae:wasm-function[290160]:0x4b57467)
    at b8f19783_ecdd_483a_a5be_f2cfb1051f26_wrapper_delegate_invoke__Module_invoke_double (wasm://wasm/2627c6ae:wasm-function[138969]:0x20df192)
    at b8f19783_ecdd_483a_a5be_f2cfb1051f26_BenchmarkDotNet_Autogenerated_Runnable_3_WorkloadActionNoUnroll_long (wasm://wasm/2627c6ae:wasm-function[139003]:0x20e20d5)
    at b8f19783_ecdd_483a_a5be_f2cfb1051f26_wrapper_delegate_invoke_System_Action_1_long_invoke_void_T_long (wasm://wasm/2627c6ae:wasm-function[138974]:0x20dfe10)
    at BenchmarkDotNet_BenchmarkDotNet_Engines_Engine_RunIteration_BenchmarkDotNet_Engines_IterationData (wasm://wasm/2627c6ae:wasm-function[314273]:0x5258a20)
    at BenchmarkDotNet_BenchmarkDotNet_Engines_EngineFactory_Jit_BenchmarkDotNet_Engines_Engine_int_int_int (wasm://wasm/2627c6ae:wasm-function[14040]:0x4edbcb)
    at BenchmarkDotNet_BenchmarkDotNet_Engines_EngineFactory_CreateReadyToRun_BenchmarkDotNet_Engines_EngineParameters (wasm://wasm/2627c6ae:wasm-function[314292]:0x5259ef8)
    at BenchmarkDotNet_Diagnostics_Windows_aot_wrapper_gsharedvt_out_sig_obj_this_obj (wasm://wasm/2627c6ae:wasm-function[922]:0x18c201)
    at jit_call_cb (wasm://wasm/2627c6ae:wasm-function[36553]:0xbe0d89)
    at invoke_vi (dotnet.js:1:234708)
    ```

### Regression?

_No response_

### Known Workarounds

_No response_

### Configuration

latest ```main```

### Other information

_No response_

[ghost]

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

Issue Details

---

Description

Any code using Generic Math will produce runtime errors in an AOT'd wasm application. It works in the interpreter.

Reproduction Steps

    private static T LerpGeneric<T> (T a, T b, T t) 
        where T : IAdditionOperators<T, T, T>, 
            ISubtractionOperators<T, T, T>, 
            IMultiplyOperators<T, T, T>
    {
        var size = b - a;
        var offset = size * t;
        return a + offset;
    }
    
    [BenchmarkCategory(Categories.Runtime)]
    [Benchmark]
    public double Lerp_Generic ()
    {
        double result = 0;
        for (int i = 0; i < IterationCount; i++) {
            double accumulator = 0;
            for (int j = 0; j < ArrayLength; j++) {
                double t = j / dArrayLength;
                accumulator += LerpGeneric(A[j], B[LengthMinusOne - j], t);
            }
            result += accumulator;
        }
        return result;
    }

Expected behavior

Working generic math (even if it falls back to the interpreter)

Actual behavior

console.error: RuntimeError: null function or function signature mismatch
console.error: RuntimeError: null function or function signature mismatch
    at MicroBenchmarks_GenericMathTest_Lerp_Generic (wasm://wasm/2627c6ae:wasm-function[290160]:0x4b57467)
    at b8f19783_ecdd_483a_a5be_f2cfb1051f26_wrapper_delegate_invoke__Module_invoke_double (wasm://wasm/2627c6ae:wasm-function[138969]:0x20df192)
    at b8f19783_ecdd_483a_a5be_f2cfb1051f26_BenchmarkDotNet_Autogenerated_Runnable_3_WorkloadActionNoUnroll_long (wasm://wasm/2627c6ae:wasm-function[139003]:0x20e20d5)
    at b8f19783_ecdd_483a_a5be_f2cfb1051f26_wrapper_delegate_invoke_System_Action_1_long_invoke_void_T_long (wasm://wasm/2627c6ae:wasm-function[138974]:0x20dfe10)
    at BenchmarkDotNet_BenchmarkDotNet_Engines_Engine_RunIteration_BenchmarkDotNet_Engines_IterationData (wasm://wasm/2627c6ae:wasm-function[314273]:0x5258a20)
    at BenchmarkDotNet_BenchmarkDotNet_Engines_EngineFactory_Jit_BenchmarkDotNet_Engines_Engine_int_int_int (wasm://wasm/2627c6ae:wasm-function[14040]:0x4edbcb)
    at BenchmarkDotNet_BenchmarkDotNet_Engines_EngineFactory_CreateReadyToRun_BenchmarkDotNet_Engines_EngineParameters (wasm://wasm/2627c6ae:wasm-function[314292]:0x5259ef8)
    at BenchmarkDotNet_Diagnostics_Windows_aot_wrapper_gsharedvt_out_sig_obj_this_obj (wasm://wasm/2627c6ae:wasm-function[922]:0x18c201)
    at jit_call_cb (wasm://wasm/2627c6ae:wasm-function[36553]:0xbe0d89)
    at invoke_vi (dotnet.js:1:234708)
    ```

### Regression?

_No response_

### Known Workarounds

_No response_

### Configuration

latest ```main```

### Other information

_No response_

<table>
  <tr>
    <th align="left">Author:</th>
    <td>kg</td>
  </tr>
  <tr>
    <th align="left">Assignees:</th>
    <td>-</td>
  </tr>
  <tr>
    <th align="left">Labels:</th>
    <td>

`arch-wasm`, `area-Codegen-AOT-mono`

</td>
  </tr>
  <tr>
    <th align="left">Milestone:</th>
    <td>-</td>
  </tr>
</table>
</details>

[kg]

The generated llvm (I think) and wasm both attempt to do an indirect call to try and invoke the generic method:

NOEX_BB25:                                        ; preds = %NOEX_BB21
  %t28 = shl i32 %t27, 3
  %t29 = add i32 %t24, %t28
  %t61 = add i32 %t29, 16
  %25 = inttoptr i32 %t61 to double*
  %t30 = load double, double* %25, align 8
  %26 = load i32*, i32** @aotconst_method_rgctx_346, align 4
  %27 = load i32*, i32** getelementptr inbounds ([16 x i32*], [16 x i32*]* @dummy_got, i32 0, i32 0), align 4
  %28 = load double (double, double, double, i32*)*, double (double, double, double, i32*)** @aotconst_method_16324, align 4
  %29 = bitcast i32* %27 to double (double, double, double, i32*)*
  %30 = notail call double %28(double %t22, double %t30, double %t15, i32* %26), !managed_name !153
  %t38 = fadd double %6, %30
  %t63 = add i32 %5, 1
  br label %BB6

418899EE0D            i32_const 0x01BB8C88 29068424
280200                i32_load [4] +0
41D8FFF50D            i32_const 0x01BD7FD8 29196248
280200                i32_load [4] +0
11A50100              call_indirect tables[0] d(dddi)
A0                    f64_add 
2107                  set_local f64 local5
0C00                  br 0

But the AOT compiler skipped it:

Mono Ahead of Time compiler - compiling assembly /home/kate/Projects/performance/artifacts/bin/MicroBenchmarks/Release/net6.0/08c83c26-9e6f-4b98-8840-11ebcb3afa6f/bin/net6.0/browser-wasm/AppBundle/managed/MicroBenchmarks.dll

Skip (gshared failure): T_REF GenericMathTest:LerpGeneric<T_REF> (T_REF,T_REF,T_REF) ((null))

Skip (gshared failure): T_DOUBLE GenericMathTest:LerpGeneric<T_DOUBLE> (T_DOUBLE,T_DOUBLE,T_DOUBLE) ((null))

It's not clear to me why we're doing this and exploding at runtime instead of jumping into the interpreter to run it.

[kg]

From dumping verbose AOT logs, it looks like this is an expected FIXME failure:
https://github.com/dotnet/runtime/blob/main/src/mono/mono/mini/method-to-ir.c#L7210-L7213
Because it's constrained and static:

.method private hidebysig static 
		!!T LerpGeneric<(class [System.Runtime]System.IAdditionOperators`3<!!T, !!T, !!T>, class [System.Runtime]System.ISubtractionOperators`3<!!T, !!T, !!T>, class [System.Runtime]System.IMultiplyOperators`3<!!T, !!T, !!T>) T> (
			!!T a,
			!!T b,
			!!T t
		) cil managed 
	{
		// Header Size: 12 bytes
		// Code Size: 42 (0x2A) bytes
		// LocalVarSig Token: 0x11000059 RID: 89
		.maxstack 2
		.locals init (
			[0] !!T,
			[1] !!T
		)

		/* 0x00008AC8 03           */ IL_0000: ldarg.1
		/* 0x00008AC9 02           */ IL_0001: ldarg.0
		/* 0x00008ACA FE164E00001B */ IL_0002: constrained. !!T
		/* 0x00008AD0 28DD00000A   */ IL_0008: call      !2 class [System.Runtime]System.ISubtractionOperators`3<!!T, !!T, !!T>::op_Subtraction(!0, !1)
		/* 0x00008AD5 0A           */ IL_000D: stloc.0
		/* 0x00008AD6 06           */ IL_000E: ldloc.0
		/* 0x00008AD7 04           */ IL_000F: ldarg.2
		/* 0x00008AD8 FE164E00001B */ IL_0010: constrained. !!T
		/* 0x00008ADE 28DE00000A   */ IL_0016: call      !2 class [System.Runtime]System.IMultiplyOperators`3<!!T, !!T, !!T>::op_Multiply(!0, !1)
		/* 0x00008AE3 0B           */ IL_001B: stloc.1
		/* 0x00008AE4 02           */ IL_001C: ldarg.0
		/* 0x00008AE5 07           */ IL_001D: ldloc.1
		/* 0x00008AE6 FE164E00001B */ IL_001E: constrained. !!T
		/* 0x00008AEC 28DF00000A   */ IL_0024: call      !2 class [System.Runtime]System.IAdditionOperators`3<!!T, !!T, !!T>::op_Addition(!0, !1)
		/* 0x00008AF1 2A           */ IL_0029: ret
	} // end of method GenericMathTest::LerpGeneric