[llvm] [DA] Rewrite the formula in the Strong SIV test (PR #179665)

[Ehsan Amiri via llvm-commits]

amehsan wrote:

> FWIW, leaving my thoughts here.

Thanks for your comments. Two points:

1. I will try to clarify my reasoning based on the "future changes". If others find it not convincing, I am OK to take it back. Think about this situation: We need to allow per-use nsw flags 5 years down the road to catch some loop opt opportunity. We would need to rewrite the formula back to the existing form and also realize that this two forms are not equivalent. (Also consider that this situation likely exist for multiple other tests). _**While I agree that sacrificing precision for correctness is a good strategy I think when we have a provably correct solution that is also more precise, we should choose that.**_ Again, I am open to ignore this argument if others find it non-convincing.


2- The rewritten formula is not "Strong SIV" anymore as discussed in the literature. It is a different test close to one of the two parts of in Strong SIV.






https://github.com/llvm/llvm-project/pull/179665