| /* SPDX-License-Identifier: MIT OR Apache-2.0 */ |
| //! IEEE 754-2011 `maxNum`. This has been superseded by IEEE 754-2019 `maximumNumber`. |
| //! |
| //! Per the spec, returns the canonicalized result of: |
| //! - `x` if `x > y` |
| //! - `y` if `y > x` |
| //! - The other number if one is NaN |
| //! - Otherwise, either `x` or `y`, canonicalized |
| //! - -0.0 and +0.0 may be disregarded (unlike newer operations) |
| //! |
| //! Excluded from our implementation is sNaN handling. |
| //! |
| //! More on the differences: [link]. |
| //! |
| //! [link]: https://grouper.ieee.org/groups/msc/ANSI_IEEE-Std-754-2019/background/minNum_maxNum_Removal_Demotion_v3.pdf |
| |
| use crate::support::Float; |
| |
| #[inline] |
| pub fn fmax<F: Float>(x: F, y: F) -> F { |
| let res = if x.is_nan() || x < y { y } else { x }; |
| res.canonicalize() |
| } |
