[llvm] [AMDGPU] Form V_MAD_U64_U32 from mul24/mulhi24 (PR #72393)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> You might both be right (I'm not super familiar with CGP) but I don't really understand why. On a "full rate doubles" machine (i.e. SIDPFullSpeedModel or SIDPGFX940FullSpeedModel), 24-bit multiplies are no faster than 32-bit, so why would CGP need to do anything at all?

I believe they also use less power. Plus better to not depend on the specific speed model, the more common looking ISA the better 

https://github.com/llvm/llvm-project/pull/72393