Allow static types to implement interfaces and appear as type arguments

[dmchurch]

Now that interfaces can specify abstract static members, it's no longer meaningless to have a static class implement an interface or be used as a type argument. For example, the static classes Vector128, Vector256, et al in the System.Runtime.Intrinsics namespace all have the same generic method names, with parameters of VectorNNN<T> corresponding to the type. Since they all have the same API:

public static class Vector128
{
    public static Vector128<T> Abs<T>(Vector128<T> vector) where T : struct { /* ... */ }
    public static Vector128<T> Add<T>(Vector128<T> left, Vector128<T> right) where T : struct { /* ... */ }
    public static Vector128<T> AndNot<T>(Vector128<T> left, Vector128<T> right) where T : struct { /* ... */ }
    // ...
}
public static class Vector256
{
    public static Vector256<T> Abs<T>(Vector256<T> vector) where T : struct { /* ... */ }
    public static Vector256<T> Add<T>(Vector256<T> left, Vector256<T> right) where T : struct { /* ... */ }
    public static Vector256<T> AndNot<T>(Vector256<T> left, Vector256<T> right) where T : struct { /* ... */ }
    // ...
}
public static class Vector512
{
    public static Vector512<T> Abs<T>(Vector512<T> vector) where T : struct { /* ... */ }
    public static Vector512<T> Add<T>(Vector512<T> left, Vector512<T> right) where T : struct { /* ... */ }
    public static Vector512<T> AndNot<T>(Vector512<T> left, Vector512<T> right) where T : struct { /* ... */ }
    // ...
}
public readonly struct Vector128<T> : IEquatable<Vector128<T>> where T : struct { /* ... */ }
public readonly struct Vector256<T> : IEquatable<Vector256<T>> where T : struct { /* ... */ }
public readonly struct Vector512<T> : IEquatable<Vector512<T>> where T : struct { /* ... */ }

it might make sense for them all to implement the same CRTP interface:

public interface IVectorOperations<TOperations>
    where TOperations: IVectorOperations<TOperations>
{
    abstract static TVector Abs<TVector, TElement>(TVector vector)
        where TVector : IVectorIntrinsic<TVector, TOperations, TElement>
        where TElement : struct;
    abstract static TVector Add<T>(TVector left, TVector right)
        where TVector : IVectorIntrinsic<TVector, TOperations, TElement>
        where TElement : struct;
    abstract static TVector AndNot<T>(TVector left, TVector right)
        where TVector : IVectorIntrinsic<TVector, TOperations, TElement>
        where TElement : struct;
}

public interface IVectorIntrinsic<TVector, TOperations, TElement> : IEquatable<TVector>
    where TVector: IVectorIntrinsic<TVector, TOperations, TElement>
    where TOperations: IVectorOperations<TOperations>
    where T : struct
{
}

public static class Vector128 : IVectorOperations<Vector128> { /* ... */ }
public static class Vector256 : IVectorOperations<Vector256> { /* ... */ }
public static class Vector512 : IVectorOperations<Vector512> { /* ... */ }
public readonly struct Vector128<T> : IVectorIntrinsic<Vector128<T>, Vector128, T> where T : struct { /* ... */ }
public readonly struct Vector256<T> : IVectorIntrinsic<Vector256<T>, Vector256, T> where T : struct { /* ... */ }
public readonly struct Vector512<T> : IVectorIntrinsic<Vector512<T>, Vector512, T> where T : struct { /* ... */ }

Unfortunately, C# won't let you do this, because static classes can't implement interfaces, and static classes can't appear as type arguments. (This is for good reasons, of course: a static type, unlike any other type, doesn't have a default, so if you had one as a type argument, any generic parameter would run the risk of creating nonsense code.)

To address this, I'd like to propose static as a generic constraint, alongside class, struct, and unmanaged, with the obvious meaning. That would allow the where TOperations clauses to read:

where TOperations : static, IVectorOperations<TOperations>

and then you could write generic methods that would operate on SIMD intrinsics of any size 🙂

[dmchurch]

cc @tannergooding, as this relates to what we'd been discussing earlier

[huoyaoyuan]

Previous discussions: #4840 #5517

[dmchurch]

Aha, thank you! I'm not sure how I missed those on my search.