[llvm-branch-commits] [llvm] Prevent optimizations of math intrinsics in strictfp functions (PR #201433)

[Serge Pavlov via llvm-branch-commits]

================

----------------
spavloff wrote:

Hopefully, this is a temporary solution. When new properties appear (like "signaling_nans"), some optimization would be allowed even in strictfp function. In this case checking conditions for each case has an advantage - it marks the place where conditions in strictfp function should be checked. But is some cases splitting conditions would give a cumbersome code, `foldPowiReassoc` is such.

https://github.com/llvm/llvm-project/pull/201433