Avx2.BroadcastScalarToVector128(T*) overloads are missing

[tannergooding]

Avx2 is currently only exposing the Avx2.BroadcastScalarToVector128(Vector128<T>) overloads.

It should also expose the Avx2.BroadcastScalarToVector128(T*) overloads (exploding T, due to the limitations on using T*) to have parity with the overloads available for float/double in Avx and to ensure that the VPBROADCASTB xmm, m8 encodings are properly exposed.

[tannergooding]

FYI. @eerhardt, @CarolEidt, @fiigii

[tannergooding]

Found this API missing when porting the DirectX math library: https://github.com/Microsoft/DirectXMath

[fiigii]

@tannergooding Can we implement these overloads by containment optimization?

[tannergooding]

@fiigii, not easily.

Users would have to convert a ref T into a ref Vector128<T> and understand that it is safe to do that since this operation only reads T bytes.

I think it is more understandable, and easier to handle, if we just add T* overloads.

[fiigii]

Users would have to convert a ref T into a ref Vector128

Agree.

if we just add T* overloads.

Don't you mean the overloads of pointers to all integer types (e.g., byte*, short*, etc) or generic ref? As I know, C# does not have generic pointers.

[tannergooding]

Don't you mean the overloads of pointers to all integer types

Right, hence the exploding T, due to the limitations on using T* comment in the original post.

As I know, C# does not have generic pointers.

It does, as of C# 7.3 with the unmanaged constraint. But there are issues with doing that (such as requiring the compiler to understand the new modreq attribute) that make it less desirable to use in the framework.