[llvm] Revert "[SimplifyCFG] Deduce paths unreachable if they cause div/rem UB" (PR #137741)

[Kees Cook via llvm-commits]

kees wrote:

> The kernel has a lot of highly dubious, incredibly fragile analysis tooling, which keeps producing false positives every time optimization behavior shifts in the most minuscule ways. I am very much not willing to revert correct changes to satisfy it.

If there was a "-fno-undefined-behavior" or a "-Wundefined-behavior", Linux wouldn't need to have so many careful analysis tools. There keep being surprise elisions, and that's the real problem. :(


https://github.com/llvm/llvm-project/pull/137741