[llvm] [CodeGen] Expansion of scalable vector reductions (PR #129214)

[David Green via llvm-commits]

davemgreen wrote:

Hi - If you want a way forward with this patch then I think that:
 - If you want to do this for specific vscale_range's (and it doesn't require branching) then that could be done in DAG. (Although I do think we should be considering canonicalizing to fixed length vectors if the vector_range is a single value).
 - If you want to do this for all scalable vectors in a loop, then adding a new pass to do that would make sense to me. It is hopefully the same logic you have here. (i.e. We could have a pass that lowers, pre-isel, the instructions that the backend cannot otherwise handle. It could then be extended in the future to handle other operations that have no way of lowering successfully).

https://github.com/llvm/llvm-project/pull/129214