[llvm] [VPlan] Add NegatedMulAccReduction expression-recipe variant (PR #199413)

[Florian Hahn via llvm-commits]

https://github.com/fhahn commented:



>> What makes you interested in inloop reductions, out of interest?
>    Nothing special in particular.
>
> @fhahn How should we proceed on this?

If we can now compute costs more accurately, it would be good to add tests covering the cost code path (like the one I shared, but maybe there are other interesting cases). With that it should be good to go, assuming the new decisions look good to @davemgreen 

https://github.com/llvm/llvm-project/pull/199413