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

[Mikhail Gudim via llvm-commits]

mgudim wrote:

@davemgreen

I looked  at `LoopVectorizer` and there are at least three IR patterns to which the `getMulAccReductionCost` applies:
`reduce.add(ext(mul(ext(A), ext(B)))`,
`reduce.add(mul(ext(A), ext(B)))`, 
`reduce.add(mul())`.

In all three cases the code looks quite similar. Also, it seems like I would have to repeat most of it in my patch too. Should we first come up with some API that simplifies `LoopVectorizer` and can be reused in my patch?

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