[llvm] [X86] Try Folding icmp of v8i32 -> fcmp of v8f32 on AVX (PR #82290)

[via llvm-commits]

goldsteinn wrote:

> We should probably disable it regardless of the compiler settings because it'd be weird for FP flags to invalidate what looks like integer comparisons.... In my comment where I brought this up I posted a way to compare i32 for equality/inequality regardless of range, but it fails to generalize. I think your converting-to-float method is totally fine though.

What do you mean by invalidate? The issue is simplifying a float signbit comparison to just return the float. DAZ can make the simplification incorrect.

https://github.com/llvm/llvm-project/pull/82290