[llvm] [AMDGPU][wmma] - Add tied wmma intrinsic (PR #69903)

[Piotr Sobczak via llvm-commits]

piotrAMD wrote:

> If this is strictly more expressive than what we already have for the same instructions, can we just auto-upgrade the existing intrinsic to the version with the tied operand?

This is more expressive, but can produce worse code in the cases where you do not care about the other half.

https://github.com/llvm/llvm-project/pull/69903