[SM6.10][LinAlg] Impl MatrixRef, CreateMatrix builtins

[V-FEXrt]

Fixes #7846

Implments:

• the HLSL type __builtin_la_MatrixRef
• the HLSL intrinsic __builtin_la_CreateMatrix()
• the DXIL type %dx.types.MatrixRef
• the DXIL op @dx.op.createMatrix(i32 312)

Note: The DXIL op codes will change in response to microsoft/hlsl-specs#698

[V-FEXrt]

Pretty sure I also want a .ll test but I don't know the magic incantation to generate the IR upto but right before the dxilgen pass

[damyanp]

I've read through this and didn't spot any obvious typos.

Do we need to apply the same "experimental DXIL" stuff to this that we're talking about for other SM 6.10 features?

[damyanp]

Does it matter that these values have changed?

I've no reason to think it does, just checking.

[V-FEXrt]

I'm assuming no but maybe someone else can weigh in. The ifdef basically forces them to change though since the last number is only incremented for SPIRV. Beyond that this seems to be an internal enum so changing it shouldn't be observable to end users.

[tex3d]

These are mostly internal details, but the built-in intrinsic extension mechanism relies on this header (and this enum) as well. That's currently only used by Xbox backend, and wouldn't be impacted by these changes unless updated to depend on them, and even then, using the latest header would keep it in sync.

Ideally, I think we should get rid of the ifdef here (so LICOMPTYPE_VK_BUFFER_POINTER is always part of the enumeration), and update the corresponding table g_LegalIntrinsicCompTypes in SemaHLSL.cpp. But that should be a separate change.

[V-FEXrt]

Do we need to apply the same "experimental DXIL" stuff to this that we're talking about for other SM 6.10 features?

Yep