[llvm] [X86] Fold vpmadd52h/l for pattern X * 0 + Y --> Y (PR #156086)

[Simon Pilgrim via llvm-commits]

RKSimon wrote:

> Folds like `X * 1 + Y -> X + Y`, `C1 * C2 + Y -> C3 + Y` may help too.

We probably need to start computing KnownBits for the intermediate mulu52/104 - a lot of these will be trivial after that.

https://github.com/llvm/llvm-project/pull/156086