[llvm] [LoopIdiom] Perform loop versioning to use memcpy (PR #125043)

[Ryotaro Kasuga via llvm-commits]

kasuga-fj wrote:

Thanks for your advice.

I think you're absolutely right. This patch would make the IR output complex, which is not appropriate for simplifying the function. Also, I have not fully considered the concerns you raise. I will close this PR because the approach doesn't look so good.

Let me ask you a question. If we'd like to address the original issue, what should we do? Should we add a new pass that does versioning and place it in the after the inline module optimization pipeline?

https://github.com/llvm/llvm-project/pull/125043