[llvm] [LV] Fix MVE regression from #132190 (PR #141736)

[David Green via llvm-commits]

https://github.com/davemgreen commented:

I think in the long run it makes sense to treat this separately to MaximizeVectorBandwidth, perhaps through a another target control. And in the longer-run adding them as part of the cost-model. In the mean time this fixes a number of performance regressions reported, some of which were quite large for MVE. I'd recommend getting this in to be back to where we were, performance-wise.

https://github.com/llvm/llvm-project/pull/141736