[llvm] [WebAssembly] Add getMinVectorRegisterBitWidth overriding (PR #189465)

[Sam Parker via llvm-commits]

sparker-arm wrote:

> with no regression whatsover in other tests 

And, it seems, no regressions in the benchmarks. In the current pool of benchmarks, libyuv is the one that is most likely to show the affects of vectorization changes. I don't know in general how much the SLP vectorizer triggers on them, since most of them probably comprise of relatively small loops.

When I'm back from holiday, I will prioritize getting more benchmarks upstream. 

> I'm not sure if its better to measure codesize

Sure, that's an important metric too.

https://github.com/llvm/llvm-project/pull/189465