Generic method constrained by struct & interface can not see readonly modifier

[bollhals]

Description

If you have a generic method with a struct & interface constrain and that interface contains a method, it is not able to see that the method implemented in the struct is readonly, so it ends up creating defensive copies even though it shouldn't need to.

Configuration

Any version, even on main

Regression?

No.

Data

Example on SharpLab:

using System;
using SharpLab.Runtime;
using System.Runtime.CompilerServices;

public class C {
    [JitGeneric(typeof(S))]
    public void MIn<T>(in T s)
        where T : struct, IInterface
    {
            s.Test();
    }
    
    [JitGeneric(typeof(S))]
    public void MRef<T>(ref T s)
        where T : struct, IInterface
    {
            s.Test();
    }
}

public interface IInterface {
    void Test();
}

public readonly struct S : IInterface
{
    public readonly long P1;
    public readonly long P2;
    public readonly long P3;
    public readonly long P4;
    
    [MethodImpl(MethodImplOptions.NoInlining)]
    public readonly void Test()
    {
    }
}

C.MIn[[S, _]](S ByRef)
    L0000: sub esp, 0x20
    L0003: vzeroupper
    L0006: vmovdqu xmm0, [edx]
    L000a: vmovdqu [esp], xmm0
    L000f: vmovdqu xmm0, [edx+0x10]
    L0014: vmovdqu [esp+0x10], xmm0
    L001a: lea ecx, [esp]
    L001d: call dword ptr [0xf51ccd0]
    L0023: add esp, 0x20
    L0026: ret

C.MRef[[S, _]](S ByRef)
    L0000: mov ecx, edx
    L0002: call dword ptr [0xf51ccd0]
    L0008: ret

[dotnet-issue-labeler]

I couldn't figure out the best area label to add to this issue. If you have write-permissions please help me learn by adding exactly one area label.

[ghost]

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

Issue Details

---

Description

If you have a generic method with a struct & interface constrain and that interface contains a method, it is not able to see that the method implemented in the struct is readonly, so it ends up creating defensive copies even though it shouldn't need to.

Configuration

Any version, even on main

Regression?

No.

Data

Example on SharpLab:

using System;
using SharpLab.Runtime;
using System.Runtime.CompilerServices;

public class C {
    [JitGeneric(typeof(S))]
    public void MIn<T>(in T s)
        where T : struct, IInterface
    {
            s.Test();
    }
    
    [JitGeneric(typeof(S))]
    public void MRef<T>(ref T s)
        where T : struct, IInterface
    {
            s.Test();
    }
}

public interface IInterface {
    void Test();
}

public readonly struct S : IInterface
{
    public readonly long P1;
    public readonly long P2;
    public readonly long P3;
    public readonly long P4;
    
    [MethodImpl(MethodImplOptions.NoInlining)]
    public readonly void Test()
    {
    }
}

C.MIn[[S, _]](S ByRef)
    L0000: sub esp, 0x20
    L0003: vzeroupper
    L0006: vmovdqu xmm0, [edx]
    L000a: vmovdqu [esp], xmm0
    L000f: vmovdqu xmm0, [edx+0x10]
    L0014: vmovdqu [esp+0x10], xmm0
    L001a: lea ecx, [esp]
    L001d: call dword ptr [0xf51ccd0]
    L0023: add esp, 0x20
    L0026: ret

C.MRef[[S, _]](S ByRef)
    L0000: mov ecx, edx
    L0002: call dword ptr [0xf51ccd0]
    L0008: ret

Author:	bollhals
Assignees:	-
Labels:	tenet-performance, area-CodeGen-coreclr, untriaged
Milestone:	-

[tfenise]

This is rather a problem with C#. The IL already includes codes to create a defensive copy, as the C# compiler does not know whether T is readonly, so it generates codes according to the worst case. struct being readonly is only a concept of C#, not the runtime. You could argue that maybe the C# feature should have better support from the runtime, though.

[tfenise]

See also dotnet/csharplang#3055.

[bollhals]

Ah, there was that issue that I once saw ... I know I read it somewhere, but I couldn't find it anymore. Thanks for linking it.

[jakobbotsch]

I don't think there is much the JIT can do here. From the IL it sees the proposed optimization would be illegal -- it would change output of a program such as the following:

public class C
{
    public void MIn<T>(in T s)
        where T : struct, IInterface
    {
        s.Test();
    }

    public void MRef<T>(ref T s)
        where T : struct, IInterface
    {
        s.Test();
    }

    public unsafe static void Main()
    {
        S s = new S();
        S.F = &s;
        new C().MIn(in s);
    }
}

public interface IInterface
{
    void Test();
}

public unsafe readonly struct S : IInterface
{
    public readonly long P1;
    public readonly long P2;
    public readonly long P3;
    public readonly long P4;

    [MethodImpl(MethodImplOptions.NoInlining)]
    public readonly void Test()
    {
        fixed (S* pThis = &this)
            Console.WriteLine(pThis == F);
    }

    public static S* F;
}

[JulieLeeMSFT]

Closing the issue. @bollhals, please reopen if you still think that JIT has to change something.

I don't think there is much the JIT can do here. From the IL it sees the proposed optimization would be illegal -- it would change output of a program such as the following: