[llvm] [CGP]: Optimize mul.overflow. (PR #148343)

[Nikita Popov via llvm-commits]

https://github.com/nikic commented:

I question the value of this optimization, especially as implemented right now. This *might* (and I'm really not convinced) make more sense if you *only* specialize for the case of both high halves being zero without covering mixed cases, which blows up code size a lot.

https://github.com/llvm/llvm-project/pull/148343