[llvm] [KnownFPClass] Add output/input denormal helpers (PR #225575)

[Eric Ross via llvm-commits]

ZERICO2005 wrote:

Also, `applyInputDenormalMode` and `applyOutputDenormalMode` are different from `propagateDenormal`:
- `applyInputDenormalMode` input flushing only
- `applyOutputDenormalMode` output flushing only
- `propagateDenormal` handles both output/input simultaneously, but assumes `DenormalMode::Dynamic` if the output/input mode do not match.

The only in-tree use of `propagateDenormal` is `propagateCanonicalizingSrc`, and the only in-tree use of `propagateCanonicalizingSrc` is in `KnownFPClass::ldexp` for the case of `ldexp(x, 0)`. I think it may be best to remove/replace `propagateDenormal`/`propagateCanonicalizingSrc` in a follow up PR.

https://github.com/llvm/llvm-project/pull/225575