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

[Matt Arsenault via llvm-commits]

arsenm wrote:

> For background: my understanding is that the only reason CGP needs to help select 24-bit multiplies is that you can't do it during instruction selection because there is no way for a target to simplify a generic opcode like ISD::MUL based on demanded 

I think I originally moved this because computeKnownBits in the IR is significantly better than on the DAG. In particular because of the cross block copy issue

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