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

[David Green via llvm-commits]

davemgreen wrote:

Sorry I didn't see this. I think sub reductions, providing they are not extending, will look a little different and keep vectorizing. We can change the cost model if needed, but after looking at it I'm not too worried.

What makes you interested in inloop reductions, out of interest?

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