[llvm] [AArch64] Add MATCH loops to LoopIdiomVectorizePass (PR #101976)

[Ricardo Jesus via llvm-commits]

rj-jesus wrote:

> The patch is looking mostly good. I've still got a bit more to review, but at the moment there is a serious bug with this so can you hold off landing this until it's fixed?

Of course, I will do, thank you very much for the heads up!

> The inner loop can exit early before you've loaded the rest of the needle so it needs memory checks before entering the vector loop to ensure the needle doesn't cross a page boundary.

> In fact, I think we need page boundary checks for the pointer in the outer loop as well. Suppose you find a match in your needle after only two iterations in your scalar outer loop? Your third iteration could cross a page boundary and seg fault. The scalar version of the loop will be fine because we exited before the fault, but the vector version will crash.

I think I'm missing something. We should never be reading beyond the end of the search or needle arrays (if we do, that's definitely a bug), so I don't think we would be prone to segfaulting. The idea was to use `llvm.get.active.lane.mask` to create a valid predicate for the loads, shouldn't that work?

https://github.com/llvm/llvm-project/pull/101976