[libc-commits] [libc] [libc] Work around incorrect fmin/fmax results for +/-x (PR #83158)

[Matt Arsenault via libc-commits]

arsenm wrote:

> https://godbolt.org/z/oP8zrYh6T

This change should be reverted

1. The library shouldn't be working around instruction behavior 
2. C does not require this
3. The instruction (and expand-with-quieting implementation) does have the correct signed behavior, I don't see how your godbolt example demonstrates it does not
4.  We would be better off with a separate set of intrinsics which do have the 2019 behavior

https://github.com/llvm/llvm-project/pull/83158