[llvm] [LAA] Hoist setting condition for RT-checks (PR #128045)

[Igor Kirillov via llvm-commits]

igogo-x86 wrote:

I looked at the `retry_after_dep_check_with_unknown_offset` test, and I’m a bit unclear about the current logic behind when `shouldRetryWithRuntimeCheck()` returns true.

>From what I can tell, we can have multiple `DepType::Unknown` dependences that, by themselves, do not trigger the retry with runtime memory checks, and therefore block vectorisation. But once a single `FoundNonConstantDistanceDependence` is found, the analysis seems to switch to the **"LAA: Retrying with memory checks"** path and allows all these dependences to be handled via runtime checks.

Is this the intended behaviour? If so, is there a rationale for treating `DepType::Unknown` differently depending on the presence of a non-constant distance dependence?

Also, as noted above, the patch does not appear to be NFC in practice, but that might be acceptable. Making it truly NFC would require hoisting quite a bit of logic from `MemoryDepChecker::isDependent` into `getDependenceDistanceStrideAndSize`, which doesn’t seem sensible.

In any case, I’d appreciate some insight into the retry logic.

https://github.com/llvm/llvm-project/pull/128045