I would prefer to have written this on top of our tail-call logic (even though it would be slower), so as to avoid adding more complexity to the interpreter execution logic for a scenario which is vanishingly rare.

There is still something not right, based on debugging some other test that accidentally also uses CEE_JMP. I am setting no-merge label until I sort it out.

Problem fixed

This makes around 20 more coreclr tests pass with the interpreter.

@janvorli, is there a tracking issue of remaining instructions? The list at #112158 seems like it was completed but not updated.

Can the JMP target be a generic method? If so, do we need to do something we aren't doing here? I don't see any logic that would handle it, but maybe it would just automatically work?

@kg In principle, yes. And this implementation wouldn't work for that. I would be ok with making jmp to a generic method or static method on a generic type cause a BADCODE, since I don't think we have evidence that anyone has ever used jmp with a generics. Looking at the implementation of JMP in the jit, I'm not confident it handles that situation correctly either. This is another reason, I would prefer an implementation that used the general purpose tail-call logic we've built, as that is something we have real testing for.

Looking at the implementation of JMP in the jit, I'm not confident it handles that situation correctly either

I think for cross-genericness jumps using generic context the JIT will badcode too here:

It might work if both the caller and callee use a generic context parameter.

As an aside, I would echo what @davidwrighton suggests to represent this via tailcalls, if possible. In the JIT we have a similar explicit representation for CEE_JMP and it has caused a number of issues over the years. One of the major ones is that the JIT's internal representation of these jumps does not list the arguments being passed as operands explicitly, so it is not immediately obvious to transformations that there is a use of some values when it sees the internal jump operation.

Of course I don't know if the interpreter would be inclined to ever run into similar issues -- they have been mostly in optimized code.

@janvorli, is there a tracking issue of remaining instructions? The list at #112158 seems like it was completed but not updated.

@am11 the CEE_JMP is the last one that was not implemented yet. We've forgotten to update #112158 w.r.t. to the other implemented ones. I've just updated it.

This makes around 20 more coreclr tests pass with the interpreter.

It is actually 55, not 20. The 20 was with an issue I had in the initial state of the PR and fixed later.

Based on the comments from @davidwrighton and @jakobbotsch, I am working on moving the implementation to what @davidwrighton suggested.

Looking at the implementation of JMP in the jit, I'm not confident it handles that situation correctly either

Would it make sense to add a test here that ensures both the JIT and Interpreter behave the same?

I've pushed another commit that changes the CEE_JMP handling based on @davidwrighton and @jakobbotsch comments. It is now implemented purely at the compilation side. It loads the current method's args and then emulates the CEE_JMP using a tail call.

Looking at the implementation of JMP in the jit, I'm not confident it handles that situation correctly either

Would it make sense to add a test here that ensures both the JIT and Interpreter behave the same?

⚙️ Test 1 — Jump from one regular method to another

```msil
.assembly JmpTest1 {}
.assembly extern mscorlib {}

.class public auto ansi beforefieldinit Program extends [mscorlib]System.Object
{
    .method public static void Target() cil managed
    {
        .maxstack 8
        ldstr "Target called"
        call void [mscorlib]System.Console::WriteLine(string)
        ret
    }

    .method public static void Source() cil managed
    {
        jmp void Program::Target()
    }

    .method public static void Main() cil managed
    {
        call void Program::Source()
        ret
    }
}
```

🟢 Expected behavior:
```
Target called
```

⚙️ Test 2 — Jump from generic method to generic method

```msil
.assembly JmpTest2 {}
.assembly extern mscorlib {}

.class public auto ansi beforefieldinit Program extends [mscorlib]System.Object
{
    .method public static void Target<T>() cil managed
    {
        .maxstack 8
        ldstr "Target<T> called"
        call void [mscorlib]System.Console::WriteLine(string)
        ret
    }

    .method public static void Source<T>() cil managed
    {
        jmp void Program::Target<!T>()
    }

    .method public static void Main() cil managed
    {
        call void Program::Source<int32>()
        ret
    }
}
```

🟢 Expected:
```
Target<T> called
```

⚙️ Test 3 — Jump from generic method → regular method
```msil
.assembly JmpTest3 {}
.assembly extern mscorlib {}

.class public auto ansi beforefieldinit Program extends [mscorlib]System.Object
{
    .method public static void Target() cil managed
    {
        .maxstack 8
        ldstr "Regular Target called"
        call void [mscorlib]System.Console::WriteLine(string)
        ret
    }

    .method public static void Source<T>() cil managed
    {
        jmp void Program::Target()
    }

    .method public static void Main() cil managed
    {
        call void Program::Source<int32>()
        ret
    }
}
```

🟢 Expected:

```
Regular Target called
```

⚙️ Test 4 — Jump from regular method → generic method
```msil
.assembly JmpTest4 {}
.assembly extern mscorlib {}

.class public auto ansi beforefieldinit Program extends [mscorlib]System.Object
{
    .method public static void Target<T>() cil managed
    {
        .maxstack 8
        ldstr "Generic Target<T> called"
        call void [mscorlib]System.Console::WriteLine(string)
        ret
    }

    .method public static void Source() cil managed
    {
        jmp void Program::Target<int32>()
    }

    .method public static void Main() cil managed
    {
        call void Program::Source()
        ret
    }
}
```

🟢 Expected:
```
Generic Target<T> called
```

AI generated these four test cases.

Thank you @am11! I am going to try these tests and add them to this PR

@MichalPetryka I've added the tests based on the code @am11 shared above and made the interpreter behave like the JIT.

/ba-g The windows arm64 testing is down