[llvm] [LoopVectorize] LLVM fails to vectorise loops with multi-bool varables (PR #89226)

[David Sherwood via llvm-commits]

================
@@ -0,0 +1,794 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
----------------
david-arm wrote:

The only problem is that you'll probably still end up with lots of comments or requests to change the tests once reviewers see the actual code changes and, for example, realise that a test case is missing for a particular change. Then you still end up changing the tests in the second patch anyway and at that point it's too late to change the first test-only patch because it has already landed.

I realise the official solution to this is to create branches in the main llvm repo in order to do stacked pull requests. That technically is a solution that probably works well for some cases, but given this is a very frequent thing to do it's also more costly and it seems that people aren't deleting their branches afterwards. `git branch -r -l | grep users | wc -l` returns a count of 638.

Not trying to say which way is right or wrong, or what works best for the individual developer, but just pointing out there are pros and cons to each approach.

https://github.com/llvm/llvm-project/pull/89226