[llvm] [WebAssembly] Support partial-reduce accumulator (PR #158060)

[Derek Schuff via llvm-commits]

https://github.com/dschuff approved this pull request.

Code looks good.
One question on the context and testing; the TTI change causes the loop vectorizer to vectorize where it wouldn't before, and I assume that the default (CHECK) cases cover that (i.e. without this change we wouldn't get the intrinsics in the IR). what's the purpose of the MAX-BANDWIDTH case? how would real code end up covering this case?

https://github.com/llvm/llvm-project/pull/158060