[llvm] [LoopVectorize] Refine runtime memory check costs when there is an outer loop (PR #76034)

[David Sherwood via llvm-commits]

david-arm wrote:

> > To be honest either approach is unsatisfactory, but I feel option 2) gives the reviewer a more fighting chance of seeing cause and effect!
> 
> IMO including a separate commit for precommitting tests in a feature PR isn't ideal, as the default view in GH shows the diff for all changes compared to the main branch and when merging the PR all commits get squashed into one.
> 
> It's now possible to get something close to Phabricator's stacked reviews discussed here https://discourse.llvm.org/t/update-on-github-pull-requests/71540/146.

That's true, but to be honest the idea of creating loads of user branches on main fills me with dread. :) I've not tried it myself so I don't know how much extra overhead is involved. If you have examples where you have done this successfully/efficiently I'd be genuinely curious to see how it's done?

https://github.com/llvm/llvm-project/pull/76034