[llvm] [VectorCombine] Fold binary op of reductions. (PR #121567)

[Mikhail Gudim via llvm-commits]

mgudim wrote:

> Is it possible to make the cost model detect and make use getExtendedReductionCost / getMulAccReductionCost? Otherwise it might not correctly detect that the extending reductions are cheap when they have a `vecreduce.add(s/zext())` or `vecreduce.add(mul(zext(), zext()))` pattern.

I'll do that when I come back (in about a week).

https://github.com/llvm/llvm-project/pull/121567