[llvm] [AMDGPU] Unpack V_PK_MOV_B32 for MFMA co-issue (PR #211937)

[Jay Foad via llvm-commits]

jayfoad wrote:

> The patch also avoids unpacking when writing the low destination would clobber the source needed by the high move.

In that case wouldn't it be trivial to do the high move first?

https://github.com/llvm/llvm-project/pull/211937