Auto merge of #126608 - tgross35:f16-f128-library, r=Mark-Simulacrum
Add more constants, functions, and tests for `f16` and `f128` This adds everything that was in some way blocked on const eval, since https://github.com/rust-lang/rust/pull/126429 landed. There is a lot of `cfg(bootstrap)` since that is a fairly recent change. `f128` tests are disabled on everything except x86_64 and Linux aarch64, which are two platforms I know have "good" support for these types - meaning basic math symbols are available and LLVM doesn't hit selection crashes. `f16` tests are enabled on almost everything except for known LLVM crashes. Doctests are only enabled on x86_64. Tracking issue: https://github.com/rust-lang/rust/issues/116909
This commit is contained in:
bors
commit
4bdf8d2d58
12 changed files with 2638 additions and 148 deletions
|
|
@ -11,6 +11,7 @@
|
|||
|
||||
#![unstable(feature = "f128", issue = "116909")]
|
||||
|
||||
use crate::convert::FloatToInt;
|
||||
use crate::mem;
|
||||
|
||||
/// Basic mathematical constants.
|
||||
|
|
@ -220,21 +221,145 @@ impl f128 {
|
|||
#[unstable(feature = "f128", issue = "116909")]
|
||||
pub const MAX_10_EXP: i32 = 4_932;
|
||||
|
||||
/// Not a Number (NaN).
|
||||
///
|
||||
/// Note that IEEE 754 doesn't define just a single NaN value;
|
||||
/// a plethora of bit patterns are considered to be NaN.
|
||||
/// Furthermore, the standard makes a difference
|
||||
/// between a "signaling" and a "quiet" NaN,
|
||||
/// and allows inspecting its "payload" (the unspecified bits in the bit pattern).
|
||||
/// This constant isn't guaranteed to equal to any specific NaN bitpattern,
|
||||
/// and the stability of its representation over Rust versions
|
||||
/// and target platforms isn't guaranteed.
|
||||
#[cfg(not(bootstrap))]
|
||||
#[allow(clippy::eq_op)]
|
||||
#[rustc_diagnostic_item = "f128_nan"]
|
||||
#[unstable(feature = "f128", issue = "116909")]
|
||||
pub const NAN: f128 = 0.0_f128 / 0.0_f128;
|
||||
|
||||
/// Infinity (∞).
|
||||
#[cfg(not(bootstrap))]
|
||||
#[unstable(feature = "f128", issue = "116909")]
|
||||
pub const INFINITY: f128 = 1.0_f128 / 0.0_f128;
|
||||
|
||||
/// Negative infinity (−∞).
|
||||
#[cfg(not(bootstrap))]
|
||||
#[unstable(feature = "f128", issue = "116909")]
|
||||
pub const NEG_INFINITY: f128 = -1.0_f128 / 0.0_f128;
|
||||
|
||||
/// Sign bit
|
||||
#[cfg(not(bootstrap))]
|
||||
pub(crate) const SIGN_MASK: u128 = 0x8000_0000_0000_0000_0000_0000_0000_0000;
|
||||
|
||||
/// Minimum representable positive value (min subnormal)
|
||||
#[cfg(not(bootstrap))]
|
||||
const TINY_BITS: u128 = 0x1;
|
||||
|
||||
/// Minimum representable negative value (min negative subnormal)
|
||||
#[cfg(not(bootstrap))]
|
||||
const NEG_TINY_BITS: u128 = Self::TINY_BITS | Self::SIGN_MASK;
|
||||
|
||||
/// Returns `true` if this value is NaN.
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f128)]
|
||||
/// # // FIXME(f16_f128): remove when `unordtf2` is available
|
||||
/// # #[cfg(all(target_arch = "x86_64", target_os = "linux"))] {
|
||||
///
|
||||
/// let nan = f128::NAN;
|
||||
/// let f = 7.0_f128;
|
||||
///
|
||||
/// assert!(nan.is_nan());
|
||||
/// assert!(!f.is_nan());
|
||||
/// # }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[must_use]
|
||||
#[cfg(not(bootstrap))]
|
||||
#[unstable(feature = "f128", issue = "116909")]
|
||||
#[allow(clippy::eq_op)] // > if you intended to check if the operand is NaN, use `.is_nan()` instead :)
|
||||
pub const fn is_nan(self) -> bool {
|
||||
self != self
|
||||
}
|
||||
|
||||
// FIXME(#50145): `abs` is publicly unavailable in core due to
|
||||
// concerns about portability, so this implementation is for
|
||||
// private use internally.
|
||||
#[inline]
|
||||
#[cfg(not(bootstrap))]
|
||||
#[rustc_const_unstable(feature = "const_float_classify", issue = "72505")]
|
||||
pub(crate) const fn abs_private(self) -> f128 {
|
||||
// SAFETY: This transmutation is fine. Probably. For the reasons std is using it.
|
||||
unsafe {
|
||||
mem::transmute::<u128, f128>(mem::transmute::<f128, u128>(self) & !Self::SIGN_MASK)
|
||||
}
|
||||
}
|
||||
|
||||
/// Returns `true` if this value is positive infinity or negative infinity, and
|
||||
/// `false` otherwise.
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f128)]
|
||||
/// # // FIXME(f16_f128): remove when `eqtf2` is available
|
||||
/// # #[cfg(all(target_arch = "x86_64", target_os = "linux"))] {
|
||||
///
|
||||
/// let f = 7.0f128;
|
||||
/// let inf = f128::INFINITY;
|
||||
/// let neg_inf = f128::NEG_INFINITY;
|
||||
/// let nan = f128::NAN;
|
||||
///
|
||||
/// assert!(!f.is_infinite());
|
||||
/// assert!(!nan.is_infinite());
|
||||
///
|
||||
/// assert!(inf.is_infinite());
|
||||
/// assert!(neg_inf.is_infinite());
|
||||
/// # }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[must_use]
|
||||
#[cfg(not(bootstrap))]
|
||||
#[unstable(feature = "f128", issue = "116909")]
|
||||
#[rustc_const_unstable(feature = "const_float_classify", issue = "72505")]
|
||||
pub const fn is_infinite(self) -> bool {
|
||||
(self == f128::INFINITY) | (self == f128::NEG_INFINITY)
|
||||
}
|
||||
|
||||
/// Returns `true` if this number is neither infinite nor NaN.
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f128)]
|
||||
/// # // FIXME(f16_f128): remove when `lttf2` is available
|
||||
/// # #[cfg(all(target_arch = "x86_64", target_os = "linux"))] {
|
||||
///
|
||||
/// let f = 7.0f128;
|
||||
/// let inf: f128 = f128::INFINITY;
|
||||
/// let neg_inf: f128 = f128::NEG_INFINITY;
|
||||
/// let nan: f128 = f128::NAN;
|
||||
///
|
||||
/// assert!(f.is_finite());
|
||||
///
|
||||
/// assert!(!nan.is_finite());
|
||||
/// assert!(!inf.is_finite());
|
||||
/// assert!(!neg_inf.is_finite());
|
||||
/// # }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[must_use]
|
||||
#[cfg(not(bootstrap))]
|
||||
#[unstable(feature = "f128", issue = "116909")]
|
||||
#[rustc_const_unstable(feature = "const_float_classify", issue = "72505")]
|
||||
pub const fn is_finite(self) -> bool {
|
||||
// There's no need to handle NaN separately: if self is NaN,
|
||||
// the comparison is not true, exactly as desired.
|
||||
self.abs_private() < Self::INFINITY
|
||||
}
|
||||
|
||||
/// Returns `true` if `self` has a positive sign, including `+0.0`, NaNs with
|
||||
/// positive sign bit and positive infinity. Note that IEEE 754 doesn't assign any
|
||||
/// meaning to the sign bit in case of a NaN, and as Rust doesn't guarantee that
|
||||
/// the bit pattern of NaNs are conserved over arithmetic operations, the result of
|
||||
/// `is_sign_positive` on a NaN might produce an unexpected result in some cases.
|
||||
/// See [explanation of NaN as a special value](f32) for more info.
|
||||
/// See [explanation of NaN as a special value](f128) for more info.
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f128)]
|
||||
|
|
@ -257,7 +382,7 @@ impl f128 {
|
|||
/// meaning to the sign bit in case of a NaN, and as Rust doesn't guarantee that
|
||||
/// the bit pattern of NaNs are conserved over arithmetic operations, the result of
|
||||
/// `is_sign_negative` on a NaN might produce an unexpected result in some cases.
|
||||
/// See [explanation of NaN as a special value](f32) for more info.
|
||||
/// See [explanation of NaN as a special value](f128) for more info.
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f128)]
|
||||
|
|
@ -278,6 +403,222 @@ impl f128 {
|
|||
(self.to_bits() & (1 << 127)) != 0
|
||||
}
|
||||
|
||||
/// Returns the least number greater than `self`.
|
||||
///
|
||||
/// Let `TINY` be the smallest representable positive `f128`. Then,
|
||||
/// - if `self.is_nan()`, this returns `self`;
|
||||
/// - if `self` is [`NEG_INFINITY`], this returns [`MIN`];
|
||||
/// - if `self` is `-TINY`, this returns -0.0;
|
||||
/// - if `self` is -0.0 or +0.0, this returns `TINY`;
|
||||
/// - if `self` is [`MAX`] or [`INFINITY`], this returns [`INFINITY`];
|
||||
/// - otherwise the unique least value greater than `self` is returned.
|
||||
///
|
||||
/// The identity `x.next_up() == -(-x).next_down()` holds for all non-NaN `x`. When `x`
|
||||
/// is finite `x == x.next_up().next_down()` also holds.
|
||||
///
|
||||
/// ```rust
|
||||
/// #![feature(f128)]
|
||||
/// #![feature(float_next_up_down)]
|
||||
/// # // FIXME(f16_f128): remove when `eqtf2` is available
|
||||
/// # #[cfg(all(target_arch = "x86_64", target_os = "linux"))] {
|
||||
///
|
||||
/// // f128::EPSILON is the difference between 1.0 and the next number up.
|
||||
/// assert_eq!(1.0f128.next_up(), 1.0 + f128::EPSILON);
|
||||
/// // But not for most numbers.
|
||||
/// assert!(0.1f128.next_up() < 0.1 + f128::EPSILON);
|
||||
/// assert_eq!(4611686018427387904f128.next_up(), 4611686018427387904.000000000000001);
|
||||
/// # }
|
||||
/// ```
|
||||
///
|
||||
/// [`NEG_INFINITY`]: Self::NEG_INFINITY
|
||||
/// [`INFINITY`]: Self::INFINITY
|
||||
/// [`MIN`]: Self::MIN
|
||||
/// [`MAX`]: Self::MAX
|
||||
#[inline]
|
||||
#[cfg(not(bootstrap))]
|
||||
#[unstable(feature = "f128", issue = "116909")]
|
||||
// #[unstable(feature = "float_next_up_down", issue = "91399")]
|
||||
pub fn next_up(self) -> Self {
|
||||
// Some targets violate Rust's assumption of IEEE semantics, e.g. by flushing
|
||||
// denormals to zero. This is in general unsound and unsupported, but here
|
||||
// we do our best to still produce the correct result on such targets.
|
||||
let bits = self.to_bits();
|
||||
if self.is_nan() || bits == Self::INFINITY.to_bits() {
|
||||
return self;
|
||||
}
|
||||
|
||||
let abs = bits & !Self::SIGN_MASK;
|
||||
let next_bits = if abs == 0 {
|
||||
Self::TINY_BITS
|
||||
} else if bits == abs {
|
||||
bits + 1
|
||||
} else {
|
||||
bits - 1
|
||||
};
|
||||
Self::from_bits(next_bits)
|
||||
}
|
||||
|
||||
/// Returns the greatest number less than `self`.
|
||||
///
|
||||
/// Let `TINY` be the smallest representable positive `f128`. Then,
|
||||
/// - if `self.is_nan()`, this returns `self`;
|
||||
/// - if `self` is [`INFINITY`], this returns [`MAX`];
|
||||
/// - if `self` is `TINY`, this returns 0.0;
|
||||
/// - if `self` is -0.0 or +0.0, this returns `-TINY`;
|
||||
/// - if `self` is [`MIN`] or [`NEG_INFINITY`], this returns [`NEG_INFINITY`];
|
||||
/// - otherwise the unique greatest value less than `self` is returned.
|
||||
///
|
||||
/// The identity `x.next_down() == -(-x).next_up()` holds for all non-NaN `x`. When `x`
|
||||
/// is finite `x == x.next_down().next_up()` also holds.
|
||||
///
|
||||
/// ```rust
|
||||
/// #![feature(f128)]
|
||||
/// #![feature(float_next_up_down)]
|
||||
/// # // FIXME(f16_f128): remove when `eqtf2` is available
|
||||
/// # #[cfg(all(target_arch = "x86_64", target_os = "linux"))] {
|
||||
///
|
||||
/// let x = 1.0f128;
|
||||
/// // Clamp value into range [0, 1).
|
||||
/// let clamped = x.clamp(0.0, 1.0f128.next_down());
|
||||
/// assert!(clamped < 1.0);
|
||||
/// assert_eq!(clamped.next_up(), 1.0);
|
||||
/// # }
|
||||
/// ```
|
||||
///
|
||||
/// [`NEG_INFINITY`]: Self::NEG_INFINITY
|
||||
/// [`INFINITY`]: Self::INFINITY
|
||||
/// [`MIN`]: Self::MIN
|
||||
/// [`MAX`]: Self::MAX
|
||||
#[inline]
|
||||
#[cfg(not(bootstrap))]
|
||||
#[unstable(feature = "f128", issue = "116909")]
|
||||
// #[unstable(feature = "float_next_up_down", issue = "91399")]
|
||||
pub fn next_down(self) -> Self {
|
||||
// Some targets violate Rust's assumption of IEEE semantics, e.g. by flushing
|
||||
// denormals to zero. This is in general unsound and unsupported, but here
|
||||
// we do our best to still produce the correct result on such targets.
|
||||
let bits = self.to_bits();
|
||||
if self.is_nan() || bits == Self::NEG_INFINITY.to_bits() {
|
||||
return self;
|
||||
}
|
||||
|
||||
let abs = bits & !Self::SIGN_MASK;
|
||||
let next_bits = if abs == 0 {
|
||||
Self::NEG_TINY_BITS
|
||||
} else if bits == abs {
|
||||
bits - 1
|
||||
} else {
|
||||
bits + 1
|
||||
};
|
||||
Self::from_bits(next_bits)
|
||||
}
|
||||
|
||||
/// Takes the reciprocal (inverse) of a number, `1/x`.
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f128)]
|
||||
/// # // FIXME(f16_f128): remove when `eqtf2` is available
|
||||
/// # #[cfg(all(target_arch = "x86_64", target_os = "linux"))] {
|
||||
///
|
||||
/// let x = 2.0_f128;
|
||||
/// let abs_difference = (x.recip() - (1.0 / x)).abs();
|
||||
///
|
||||
/// assert!(abs_difference <= f128::EPSILON);
|
||||
/// # }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[cfg(not(bootstrap))]
|
||||
#[unstable(feature = "f128", issue = "116909")]
|
||||
#[must_use = "this returns the result of the operation, without modifying the original"]
|
||||
pub fn recip(self) -> Self {
|
||||
1.0 / self
|
||||
}
|
||||
|
||||
/// Converts radians to degrees.
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f128)]
|
||||
/// # // FIXME(f16_f128): remove when `eqtf2` is available
|
||||
/// # #[cfg(all(target_arch = "x86_64", target_os = "linux"))] {
|
||||
///
|
||||
/// let angle = std::f128::consts::PI;
|
||||
///
|
||||
/// let abs_difference = (angle.to_degrees() - 180.0).abs();
|
||||
/// assert!(abs_difference <= f128::EPSILON);
|
||||
/// # }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[cfg(not(bootstrap))]
|
||||
#[unstable(feature = "f128", issue = "116909")]
|
||||
#[must_use = "this returns the result of the operation, without modifying the original"]
|
||||
pub fn to_degrees(self) -> Self {
|
||||
// Use a literal for better precision.
|
||||
const PIS_IN_180: f128 = 57.2957795130823208767981548141051703324054724665643215491602_f128;
|
||||
self * PIS_IN_180
|
||||
}
|
||||
|
||||
/// Converts degrees to radians.
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f128)]
|
||||
/// # // FIXME(f16_f128): remove when `eqtf2` is available
|
||||
/// # #[cfg(all(target_arch = "x86_64", target_os = "linux"))] {
|
||||
///
|
||||
/// let angle = 180.0f128;
|
||||
///
|
||||
/// let abs_difference = (angle.to_radians() - std::f128::consts::PI).abs();
|
||||
///
|
||||
/// assert!(abs_difference <= 1e-30);
|
||||
/// # }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[cfg(not(bootstrap))]
|
||||
#[unstable(feature = "f128", issue = "116909")]
|
||||
#[must_use = "this returns the result of the operation, without modifying the original"]
|
||||
pub fn to_radians(self) -> f128 {
|
||||
// Use a literal for better precision.
|
||||
const RADS_PER_DEG: f128 =
|
||||
0.0174532925199432957692369076848861271344287188854172545609719_f128;
|
||||
self * RADS_PER_DEG
|
||||
}
|
||||
|
||||
/// Rounds toward zero and converts to any primitive integer type,
|
||||
/// assuming that the value is finite and fits in that type.
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f128)]
|
||||
/// # // FIXME(f16_f128): remove when `float*itf` is available
|
||||
/// # #[cfg(all(target_arch = "x86_64", target_os = "linux"))] {
|
||||
///
|
||||
/// let value = 4.6_f128;
|
||||
/// let rounded = unsafe { value.to_int_unchecked::<u16>() };
|
||||
/// assert_eq!(rounded, 4);
|
||||
///
|
||||
/// let value = -128.9_f128;
|
||||
/// let rounded = unsafe { value.to_int_unchecked::<i8>() };
|
||||
/// assert_eq!(rounded, i8::MIN);
|
||||
/// # }
|
||||
/// ```
|
||||
///
|
||||
/// # Safety
|
||||
///
|
||||
/// The value must:
|
||||
///
|
||||
/// * Not be `NaN`
|
||||
/// * Not be infinite
|
||||
/// * Be representable in the return type `Int`, after truncating off its fractional part
|
||||
#[inline]
|
||||
#[unstable(feature = "f128", issue = "116909")]
|
||||
#[must_use = "this returns the result of the operation, without modifying the original"]
|
||||
pub unsafe fn to_int_unchecked<Int>(self) -> Int
|
||||
where
|
||||
Self: FloatToInt<Int>,
|
||||
{
|
||||
// SAFETY: the caller must uphold the safety contract for
|
||||
// `FloatToInt::to_int_unchecked`.
|
||||
unsafe { FloatToInt::<Int>::to_int_unchecked(self) }
|
||||
}
|
||||
|
||||
/// Raw transmutation to `u128`.
|
||||
///
|
||||
/// This is currently identical to `transmute::<f128, u128>(self)` on all platforms.
|
||||
|
|
@ -287,6 +628,14 @@ impl f128 {
|
|||
///
|
||||
/// Note that this function is distinct from `as` casting, which attempts to
|
||||
/// preserve the *numeric* value, and not the bitwise value.
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f128)]
|
||||
///
|
||||
/// # // FIXME(f16_f128): enable this once const casting works
|
||||
/// # // assert_ne!((1f128).to_bits(), 1f128 as u128); // to_bits() is not casting!
|
||||
/// assert_eq!((12.5f128).to_bits(), 0x40029000000000000000000000000000);
|
||||
/// ```
|
||||
#[inline]
|
||||
#[unstable(feature = "f128", issue = "116909")]
|
||||
#[must_use = "this returns the result of the operation, without modifying the original"]
|
||||
|
|
@ -326,6 +675,16 @@ impl f128 {
|
|||
///
|
||||
/// Note that this function is distinct from `as` casting, which attempts to
|
||||
/// preserve the *numeric* value, and not the bitwise value.
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f128)]
|
||||
/// # // FIXME(f16_f128): remove when `eqtf2` is available
|
||||
/// # #[cfg(all(target_arch = "x86_64", target_os = "linux"))] {
|
||||
///
|
||||
/// let v = f128::from_bits(0x40029000000000000000000000000000);
|
||||
/// assert_eq!(v, 12.5);
|
||||
/// # }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[must_use]
|
||||
#[unstable(feature = "f128", issue = "116909")]
|
||||
|
|
@ -335,4 +694,315 @@ impl f128 {
|
|||
// Stability concerns.
|
||||
unsafe { mem::transmute(v) }
|
||||
}
|
||||
|
||||
/// Return the memory representation of this floating point number as a byte array in
|
||||
/// big-endian (network) byte order.
|
||||
///
|
||||
/// See [`from_bits`](Self::from_bits) for some discussion of the
|
||||
/// portability of this operation (there are almost no issues).
|
||||
///
|
||||
/// # Examples
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f128)]
|
||||
///
|
||||
/// let bytes = 12.5f128.to_be_bytes();
|
||||
/// assert_eq!(
|
||||
/// bytes,
|
||||
/// [0x40, 0x02, 0x90, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
/// 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00]
|
||||
/// );
|
||||
/// ```
|
||||
#[inline]
|
||||
#[unstable(feature = "f128", issue = "116909")]
|
||||
#[must_use = "this returns the result of the operation, without modifying the original"]
|
||||
pub fn to_be_bytes(self) -> [u8; 16] {
|
||||
self.to_bits().to_be_bytes()
|
||||
}
|
||||
|
||||
/// Return the memory representation of this floating point number as a byte array in
|
||||
/// little-endian byte order.
|
||||
///
|
||||
/// See [`from_bits`](Self::from_bits) for some discussion of the
|
||||
/// portability of this operation (there are almost no issues).
|
||||
///
|
||||
/// # Examples
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f128)]
|
||||
///
|
||||
/// let bytes = 12.5f128.to_le_bytes();
|
||||
/// assert_eq!(
|
||||
/// bytes,
|
||||
/// [0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
/// 0x00, 0x00, 0x00, 0x00, 0x00, 0x90, 0x02, 0x40]
|
||||
/// );
|
||||
/// ```
|
||||
#[inline]
|
||||
#[unstable(feature = "f128", issue = "116909")]
|
||||
#[must_use = "this returns the result of the operation, without modifying the original"]
|
||||
pub fn to_le_bytes(self) -> [u8; 16] {
|
||||
self.to_bits().to_le_bytes()
|
||||
}
|
||||
|
||||
/// Return the memory representation of this floating point number as a byte array in
|
||||
/// native byte order.
|
||||
///
|
||||
/// As the target platform's native endianness is used, portable code
|
||||
/// should use [`to_be_bytes`] or [`to_le_bytes`], as appropriate, instead.
|
||||
///
|
||||
/// [`to_be_bytes`]: f128::to_be_bytes
|
||||
/// [`to_le_bytes`]: f128::to_le_bytes
|
||||
///
|
||||
/// See [`from_bits`](Self::from_bits) for some discussion of the
|
||||
/// portability of this operation (there are almost no issues).
|
||||
///
|
||||
/// # Examples
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f128)]
|
||||
///
|
||||
/// let bytes = 12.5f128.to_ne_bytes();
|
||||
/// assert_eq!(
|
||||
/// bytes,
|
||||
/// if cfg!(target_endian = "big") {
|
||||
/// [0x40, 0x02, 0x90, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
/// 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00]
|
||||
/// } else {
|
||||
/// [0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
/// 0x00, 0x00, 0x00, 0x00, 0x00, 0x90, 0x02, 0x40]
|
||||
/// }
|
||||
/// );
|
||||
/// ```
|
||||
#[inline]
|
||||
#[unstable(feature = "f128", issue = "116909")]
|
||||
#[must_use = "this returns the result of the operation, without modifying the original"]
|
||||
pub fn to_ne_bytes(self) -> [u8; 16] {
|
||||
self.to_bits().to_ne_bytes()
|
||||
}
|
||||
|
||||
/// Create a floating point value from its representation as a byte array in big endian.
|
||||
///
|
||||
/// See [`from_bits`](Self::from_bits) for some discussion of the
|
||||
/// portability of this operation (there are almost no issues).
|
||||
///
|
||||
/// # Examples
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f128)]
|
||||
/// # // FIXME(f16_f128): remove when `eqtf2` is available
|
||||
/// # #[cfg(all(target_arch = "x86_64", target_os = "linux"))] {
|
||||
///
|
||||
/// let value = f128::from_be_bytes(
|
||||
/// [0x40, 0x02, 0x90, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
/// 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00]
|
||||
/// );
|
||||
/// assert_eq!(value, 12.5);
|
||||
/// # }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[must_use]
|
||||
#[unstable(feature = "f128", issue = "116909")]
|
||||
pub fn from_be_bytes(bytes: [u8; 16]) -> Self {
|
||||
Self::from_bits(u128::from_be_bytes(bytes))
|
||||
}
|
||||
|
||||
/// Create a floating point value from its representation as a byte array in little endian.
|
||||
///
|
||||
/// See [`from_bits`](Self::from_bits) for some discussion of the
|
||||
/// portability of this operation (there are almost no issues).
|
||||
///
|
||||
/// # Examples
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f128)]
|
||||
/// # // FIXME(f16_f128): remove when `eqtf2` is available
|
||||
/// # #[cfg(all(target_arch = "x86_64", target_os = "linux"))] {
|
||||
///
|
||||
/// let value = f128::from_le_bytes(
|
||||
/// [0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
/// 0x00, 0x00, 0x00, 0x00, 0x00, 0x90, 0x02, 0x40]
|
||||
/// );
|
||||
/// assert_eq!(value, 12.5);
|
||||
/// # }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[must_use]
|
||||
#[unstable(feature = "f128", issue = "116909")]
|
||||
pub fn from_le_bytes(bytes: [u8; 16]) -> Self {
|
||||
Self::from_bits(u128::from_le_bytes(bytes))
|
||||
}
|
||||
|
||||
/// Create a floating point value from its representation as a byte array in native endian.
|
||||
///
|
||||
/// As the target platform's native endianness is used, portable code
|
||||
/// likely wants to use [`from_be_bytes`] or [`from_le_bytes`], as
|
||||
/// appropriate instead.
|
||||
///
|
||||
/// [`from_be_bytes`]: f128::from_be_bytes
|
||||
/// [`from_le_bytes`]: f128::from_le_bytes
|
||||
///
|
||||
/// See [`from_bits`](Self::from_bits) for some discussion of the
|
||||
/// portability of this operation (there are almost no issues).
|
||||
///
|
||||
/// # Examples
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f128)]
|
||||
/// # // FIXME(f16_f128): remove when `eqtf2` is available
|
||||
/// # #[cfg(all(target_arch = "x86_64", target_os = "linux"))] {
|
||||
///
|
||||
/// let value = f128::from_ne_bytes(if cfg!(target_endian = "big") {
|
||||
/// [0x40, 0x02, 0x90, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
/// 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00]
|
||||
/// } else {
|
||||
/// [0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
/// 0x00, 0x00, 0x00, 0x00, 0x00, 0x90, 0x02, 0x40]
|
||||
/// });
|
||||
/// assert_eq!(value, 12.5);
|
||||
/// # }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[must_use]
|
||||
#[unstable(feature = "f128", issue = "116909")]
|
||||
pub fn from_ne_bytes(bytes: [u8; 16]) -> Self {
|
||||
Self::from_bits(u128::from_ne_bytes(bytes))
|
||||
}
|
||||
|
||||
/// Return the ordering between `self` and `other`.
|
||||
///
|
||||
/// Unlike the standard partial comparison between floating point numbers,
|
||||
/// this comparison always produces an ordering in accordance to
|
||||
/// the `totalOrder` predicate as defined in the IEEE 754 (2008 revision)
|
||||
/// floating point standard. The values are ordered in the following sequence:
|
||||
///
|
||||
/// - negative quiet NaN
|
||||
/// - negative signaling NaN
|
||||
/// - negative infinity
|
||||
/// - negative numbers
|
||||
/// - negative subnormal numbers
|
||||
/// - negative zero
|
||||
/// - positive zero
|
||||
/// - positive subnormal numbers
|
||||
/// - positive numbers
|
||||
/// - positive infinity
|
||||
/// - positive signaling NaN
|
||||
/// - positive quiet NaN.
|
||||
///
|
||||
/// The ordering established by this function does not always agree with the
|
||||
/// [`PartialOrd`] and [`PartialEq`] implementations of `f128`. For example,
|
||||
/// they consider negative and positive zero equal, while `total_cmp`
|
||||
/// doesn't.
|
||||
///
|
||||
/// The interpretation of the signaling NaN bit follows the definition in
|
||||
/// the IEEE 754 standard, which may not match the interpretation by some of
|
||||
/// the older, non-conformant (e.g. MIPS) hardware implementations.
|
||||
///
|
||||
/// # Example
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f128)]
|
||||
///
|
||||
/// struct GoodBoy {
|
||||
/// name: &'static str,
|
||||
/// weight: f128,
|
||||
/// }
|
||||
///
|
||||
/// let mut bois = vec![
|
||||
/// GoodBoy { name: "Pucci", weight: 0.1 },
|
||||
/// GoodBoy { name: "Woofer", weight: 99.0 },
|
||||
/// GoodBoy { name: "Yapper", weight: 10.0 },
|
||||
/// GoodBoy { name: "Chonk", weight: f128::INFINITY },
|
||||
/// GoodBoy { name: "Abs. Unit", weight: f128::NAN },
|
||||
/// GoodBoy { name: "Floaty", weight: -5.0 },
|
||||
/// ];
|
||||
///
|
||||
/// bois.sort_by(|a, b| a.weight.total_cmp(&b.weight));
|
||||
///
|
||||
/// // `f128::NAN` could be positive or negative, which will affect the sort order.
|
||||
/// if f128::NAN.is_sign_negative() {
|
||||
/// bois.into_iter().map(|b| b.weight)
|
||||
/// .zip([f128::NAN, -5.0, 0.1, 10.0, 99.0, f128::INFINITY].iter())
|
||||
/// .for_each(|(a, b)| assert_eq!(a.to_bits(), b.to_bits()))
|
||||
/// } else {
|
||||
/// bois.into_iter().map(|b| b.weight)
|
||||
/// .zip([-5.0, 0.1, 10.0, 99.0, f128::INFINITY, f128::NAN].iter())
|
||||
/// .for_each(|(a, b)| assert_eq!(a.to_bits(), b.to_bits()))
|
||||
/// }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[must_use]
|
||||
#[cfg(not(bootstrap))]
|
||||
#[unstable(feature = "f128", issue = "116909")]
|
||||
pub fn total_cmp(&self, other: &Self) -> crate::cmp::Ordering {
|
||||
let mut left = self.to_bits() as i128;
|
||||
let mut right = other.to_bits() as i128;
|
||||
|
||||
// In case of negatives, flip all the bits except the sign
|
||||
// to achieve a similar layout as two's complement integers
|
||||
//
|
||||
// Why does this work? IEEE 754 floats consist of three fields:
|
||||
// Sign bit, exponent and mantissa. The set of exponent and mantissa
|
||||
// fields as a whole have the property that their bitwise order is
|
||||
// equal to the numeric magnitude where the magnitude is defined.
|
||||
// The magnitude is not normally defined on NaN values, but
|
||||
// IEEE 754 totalOrder defines the NaN values also to follow the
|
||||
// bitwise order. This leads to order explained in the doc comment.
|
||||
// However, the representation of magnitude is the same for negative
|
||||
// and positive numbers – only the sign bit is different.
|
||||
// To easily compare the floats as signed integers, we need to
|
||||
// flip the exponent and mantissa bits in case of negative numbers.
|
||||
// We effectively convert the numbers to "two's complement" form.
|
||||
//
|
||||
// To do the flipping, we construct a mask and XOR against it.
|
||||
// We branchlessly calculate an "all-ones except for the sign bit"
|
||||
// mask from negative-signed values: right shifting sign-extends
|
||||
// the integer, so we "fill" the mask with sign bits, and then
|
||||
// convert to unsigned to push one more zero bit.
|
||||
// On positive values, the mask is all zeros, so it's a no-op.
|
||||
left ^= (((left >> 127) as u128) >> 1) as i128;
|
||||
right ^= (((right >> 127) as u128) >> 1) as i128;
|
||||
|
||||
left.cmp(&right)
|
||||
}
|
||||
|
||||
/// Restrict a value to a certain interval unless it is NaN.
|
||||
///
|
||||
/// Returns `max` if `self` is greater than `max`, and `min` if `self` is
|
||||
/// less than `min`. Otherwise this returns `self`.
|
||||
///
|
||||
/// Note that this function returns NaN if the initial value was NaN as
|
||||
/// well.
|
||||
///
|
||||
/// # Panics
|
||||
///
|
||||
/// Panics if `min > max`, `min` is NaN, or `max` is NaN.
|
||||
///
|
||||
/// # Examples
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f128)]
|
||||
/// # // FIXME(f16_f128): remove when `{eq,gt,unord}tf` are available
|
||||
/// # #[cfg(all(target_arch = "x86_64", target_os = "linux"))] {
|
||||
///
|
||||
/// assert!((-3.0f128).clamp(-2.0, 1.0) == -2.0);
|
||||
/// assert!((0.0f128).clamp(-2.0, 1.0) == 0.0);
|
||||
/// assert!((2.0f128).clamp(-2.0, 1.0) == 1.0);
|
||||
/// assert!((f128::NAN).clamp(-2.0, 1.0).is_nan());
|
||||
/// # }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[cfg(not(bootstrap))]
|
||||
#[unstable(feature = "f128", issue = "116909")]
|
||||
#[must_use = "method returns a new number and does not mutate the original value"]
|
||||
pub fn clamp(mut self, min: f128, max: f128) -> f128 {
|
||||
assert!(min <= max, "min > max, or either was NaN. min = {min:?}, max = {max:?}");
|
||||
if self < min {
|
||||
self = min;
|
||||
}
|
||||
if self > max {
|
||||
self = max;
|
||||
}
|
||||
self
|
||||
}
|
||||
}
|
||||
|
|
|
|||
|
|
@ -11,6 +11,7 @@
|
|||
|
||||
#![unstable(feature = "f16", issue = "116909")]
|
||||
|
||||
use crate::convert::FloatToInt;
|
||||
use crate::mem;
|
||||
|
||||
/// Basic mathematical constants.
|
||||
|
|
@ -215,21 +216,140 @@ impl f16 {
|
|||
#[unstable(feature = "f16", issue = "116909")]
|
||||
pub const MAX_10_EXP: i32 = 4;
|
||||
|
||||
/// Not a Number (NaN).
|
||||
///
|
||||
/// Note that IEEE 754 doesn't define just a single NaN value;
|
||||
/// a plethora of bit patterns are considered to be NaN.
|
||||
/// Furthermore, the standard makes a difference
|
||||
/// between a "signaling" and a "quiet" NaN,
|
||||
/// and allows inspecting its "payload" (the unspecified bits in the bit pattern).
|
||||
/// This constant isn't guaranteed to equal to any specific NaN bitpattern,
|
||||
/// and the stability of its representation over Rust versions
|
||||
/// and target platforms isn't guaranteed.
|
||||
#[cfg(not(bootstrap))]
|
||||
#[allow(clippy::eq_op)]
|
||||
#[rustc_diagnostic_item = "f16_nan"]
|
||||
#[unstable(feature = "f16", issue = "116909")]
|
||||
pub const NAN: f16 = 0.0_f16 / 0.0_f16;
|
||||
|
||||
/// Infinity (∞).
|
||||
#[cfg(not(bootstrap))]
|
||||
#[unstable(feature = "f16", issue = "116909")]
|
||||
pub const INFINITY: f16 = 1.0_f16 / 0.0_f16;
|
||||
|
||||
/// Negative infinity (−∞).
|
||||
#[cfg(not(bootstrap))]
|
||||
#[unstable(feature = "f16", issue = "116909")]
|
||||
pub const NEG_INFINITY: f16 = -1.0_f16 / 0.0_f16;
|
||||
|
||||
/// Sign bit
|
||||
#[cfg(not(bootstrap))]
|
||||
const SIGN_MASK: u16 = 0x8000;
|
||||
|
||||
/// Minimum representable positive value (min subnormal)
|
||||
#[cfg(not(bootstrap))]
|
||||
const TINY_BITS: u16 = 0x1;
|
||||
|
||||
/// Minimum representable negative value (min negative subnormal)
|
||||
#[cfg(not(bootstrap))]
|
||||
const NEG_TINY_BITS: u16 = Self::TINY_BITS | Self::SIGN_MASK;
|
||||
|
||||
/// Returns `true` if this value is NaN.
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f16)]
|
||||
/// # #[cfg(target_arch = "aarch64")] { // FIXME(f16_F128): rust-lang/rust#123885
|
||||
///
|
||||
/// let nan = f16::NAN;
|
||||
/// let f = 7.0_f16;
|
||||
///
|
||||
/// assert!(nan.is_nan());
|
||||
/// assert!(!f.is_nan());
|
||||
/// # }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[must_use]
|
||||
#[cfg(not(bootstrap))]
|
||||
#[unstable(feature = "f16", issue = "116909")]
|
||||
#[allow(clippy::eq_op)] // > if you intended to check if the operand is NaN, use `.is_nan()` instead :)
|
||||
pub const fn is_nan(self) -> bool {
|
||||
self != self
|
||||
}
|
||||
|
||||
// FIXMxE(#50145): `abs` is publicly unavailable in core due to
|
||||
// concerns about portability, so this implementation is for
|
||||
// private use internally.
|
||||
#[inline]
|
||||
#[cfg(not(bootstrap))]
|
||||
#[rustc_const_unstable(feature = "const_float_classify", issue = "72505")]
|
||||
pub(crate) const fn abs_private(self) -> f16 {
|
||||
// SAFETY: This transmutation is fine. Probably. For the reasons std is using it.
|
||||
unsafe { mem::transmute::<u16, f16>(mem::transmute::<f16, u16>(self) & !Self::SIGN_MASK) }
|
||||
}
|
||||
|
||||
/// Returns `true` if this value is positive infinity or negative infinity, and
|
||||
/// `false` otherwise.
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f16)]
|
||||
/// # #[cfg(target_arch = "aarch64")] { // FIXME(f16_F128): rust-lang/rust#123885
|
||||
///
|
||||
/// let f = 7.0f16;
|
||||
/// let inf = f16::INFINITY;
|
||||
/// let neg_inf = f16::NEG_INFINITY;
|
||||
/// let nan = f16::NAN;
|
||||
///
|
||||
/// assert!(!f.is_infinite());
|
||||
/// assert!(!nan.is_infinite());
|
||||
///
|
||||
/// assert!(inf.is_infinite());
|
||||
/// assert!(neg_inf.is_infinite());
|
||||
/// # }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[must_use]
|
||||
#[cfg(not(bootstrap))]
|
||||
#[unstable(feature = "f16", issue = "116909")]
|
||||
#[rustc_const_unstable(feature = "const_float_classify", issue = "72505")]
|
||||
pub const fn is_infinite(self) -> bool {
|
||||
(self == f16::INFINITY) | (self == f16::NEG_INFINITY)
|
||||
}
|
||||
|
||||
/// Returns `true` if this number is neither infinite nor NaN.
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f16)]
|
||||
/// # #[cfg(target_arch = "aarch64")] { // FIXME(f16_F128): rust-lang/rust#123885
|
||||
///
|
||||
/// let f = 7.0f16;
|
||||
/// let inf: f16 = f16::INFINITY;
|
||||
/// let neg_inf: f16 = f16::NEG_INFINITY;
|
||||
/// let nan: f16 = f16::NAN;
|
||||
///
|
||||
/// assert!(f.is_finite());
|
||||
///
|
||||
/// assert!(!nan.is_finite());
|
||||
/// assert!(!inf.is_finite());
|
||||
/// assert!(!neg_inf.is_finite());
|
||||
/// # }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[must_use]
|
||||
#[cfg(not(bootstrap))]
|
||||
#[unstable(feature = "f16", issue = "116909")]
|
||||
#[rustc_const_unstable(feature = "const_float_classify", issue = "72505")]
|
||||
pub const fn is_finite(self) -> bool {
|
||||
// There's no need to handle NaN separately: if self is NaN,
|
||||
// the comparison is not true, exactly as desired.
|
||||
self.abs_private() < Self::INFINITY
|
||||
}
|
||||
|
||||
/// Returns `true` if `self` has a positive sign, including `+0.0`, NaNs with
|
||||
/// positive sign bit and positive infinity. Note that IEEE 754 doesn't assign any
|
||||
/// meaning to the sign bit in case of a NaN, and as Rust doesn't guarantee that
|
||||
/// the bit pattern of NaNs are conserved over arithmetic operations, the result of
|
||||
/// `is_sign_positive` on a NaN might produce an unexpected result in some cases.
|
||||
/// See [explanation of NaN as a special value](f32) for more info.
|
||||
/// See [explanation of NaN as a special value](f16) for more info.
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f16)]
|
||||
|
|
@ -252,7 +372,7 @@ impl f16 {
|
|||
/// meaning to the sign bit in case of a NaN, and as Rust doesn't guarantee that
|
||||
/// the bit pattern of NaNs are conserved over arithmetic operations, the result of
|
||||
/// `is_sign_negative` on a NaN might produce an unexpected result in some cases.
|
||||
/// See [explanation of NaN as a special value](f32) for more info.
|
||||
/// See [explanation of NaN as a special value](f16) for more info.
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f16)]
|
||||
|
|
@ -273,6 +393,220 @@ impl f16 {
|
|||
(self.to_bits() & (1 << 15)) != 0
|
||||
}
|
||||
|
||||
/// Returns the least number greater than `self`.
|
||||
///
|
||||
/// Let `TINY` be the smallest representable positive `f16`. Then,
|
||||
/// - if `self.is_nan()`, this returns `self`;
|
||||
/// - if `self` is [`NEG_INFINITY`], this returns [`MIN`];
|
||||
/// - if `self` is `-TINY`, this returns -0.0;
|
||||
/// - if `self` is -0.0 or +0.0, this returns `TINY`;
|
||||
/// - if `self` is [`MAX`] or [`INFINITY`], this returns [`INFINITY`];
|
||||
/// - otherwise the unique least value greater than `self` is returned.
|
||||
///
|
||||
/// The identity `x.next_up() == -(-x).next_down()` holds for all non-NaN `x`. When `x`
|
||||
/// is finite `x == x.next_up().next_down()` also holds.
|
||||
///
|
||||
/// ```rust
|
||||
/// #![feature(f16)]
|
||||
/// #![feature(float_next_up_down)]
|
||||
/// # // FIXME(f16_f128): ABI issues on MSVC
|
||||
/// # #[cfg(all(target_arch = "x86_64", target_os = "linux"))] {
|
||||
///
|
||||
/// // f16::EPSILON is the difference between 1.0 and the next number up.
|
||||
/// assert_eq!(1.0f16.next_up(), 1.0 + f16::EPSILON);
|
||||
/// // But not for most numbers.
|
||||
/// assert!(0.1f16.next_up() < 0.1 + f16::EPSILON);
|
||||
/// assert_eq!(4356f16.next_up(), 4360.0);
|
||||
/// # }
|
||||
/// ```
|
||||
///
|
||||
/// [`NEG_INFINITY`]: Self::NEG_INFINITY
|
||||
/// [`INFINITY`]: Self::INFINITY
|
||||
/// [`MIN`]: Self::MIN
|
||||
/// [`MAX`]: Self::MAX
|
||||
#[inline]
|
||||
#[cfg(not(bootstrap))]
|
||||
#[unstable(feature = "f16", issue = "116909")]
|
||||
// #[unstable(feature = "float_next_up_down", issue = "91399")]
|
||||
pub fn next_up(self) -> Self {
|
||||
// Some targets violate Rust's assumption of IEEE semantics, e.g. by flushing
|
||||
// denormals to zero. This is in general unsound and unsupported, but here
|
||||
// we do our best to still produce the correct result on such targets.
|
||||
let bits = self.to_bits();
|
||||
if self.is_nan() || bits == Self::INFINITY.to_bits() {
|
||||
return self;
|
||||
}
|
||||
|
||||
let abs = bits & !Self::SIGN_MASK;
|
||||
let next_bits = if abs == 0 {
|
||||
Self::TINY_BITS
|
||||
} else if bits == abs {
|
||||
bits + 1
|
||||
} else {
|
||||
bits - 1
|
||||
};
|
||||
Self::from_bits(next_bits)
|
||||
}
|
||||
|
||||
/// Returns the greatest number less than `self`.
|
||||
///
|
||||
/// Let `TINY` be the smallest representable positive `f16`. Then,
|
||||
/// - if `self.is_nan()`, this returns `self`;
|
||||
/// - if `self` is [`INFINITY`], this returns [`MAX`];
|
||||
/// - if `self` is `TINY`, this returns 0.0;
|
||||
/// - if `self` is -0.0 or +0.0, this returns `-TINY`;
|
||||
/// - if `self` is [`MIN`] or [`NEG_INFINITY`], this returns [`NEG_INFINITY`];
|
||||
/// - otherwise the unique greatest value less than `self` is returned.
|
||||
///
|
||||
/// The identity `x.next_down() == -(-x).next_up()` holds for all non-NaN `x`. When `x`
|
||||
/// is finite `x == x.next_down().next_up()` also holds.
|
||||
///
|
||||
/// ```rust
|
||||
/// #![feature(f16)]
|
||||
/// #![feature(float_next_up_down)]
|
||||
/// # // FIXME(f16_f128): ABI issues on MSVC
|
||||
/// # #[cfg(all(target_arch = "x86_64", target_os = "linux"))] {
|
||||
///
|
||||
/// let x = 1.0f16;
|
||||
/// // Clamp value into range [0, 1).
|
||||
/// let clamped = x.clamp(0.0, 1.0f16.next_down());
|
||||
/// assert!(clamped < 1.0);
|
||||
/// assert_eq!(clamped.next_up(), 1.0);
|
||||
/// # }
|
||||
/// ```
|
||||
///
|
||||
/// [`NEG_INFINITY`]: Self::NEG_INFINITY
|
||||
/// [`INFINITY`]: Self::INFINITY
|
||||
/// [`MIN`]: Self::MIN
|
||||
/// [`MAX`]: Self::MAX
|
||||
#[inline]
|
||||
#[cfg(not(bootstrap))]
|
||||
#[unstable(feature = "f16", issue = "116909")]
|
||||
// #[unstable(feature = "float_next_up_down", issue = "91399")]
|
||||
pub fn next_down(self) -> Self {
|
||||
// Some targets violate Rust's assumption of IEEE semantics, e.g. by flushing
|
||||
// denormals to zero. This is in general unsound and unsupported, but here
|
||||
// we do our best to still produce the correct result on such targets.
|
||||
let bits = self.to_bits();
|
||||
if self.is_nan() || bits == Self::NEG_INFINITY.to_bits() {
|
||||
return self;
|
||||
}
|
||||
|
||||
let abs = bits & !Self::SIGN_MASK;
|
||||
let next_bits = if abs == 0 {
|
||||
Self::NEG_TINY_BITS
|
||||
} else if bits == abs {
|
||||
bits - 1
|
||||
} else {
|
||||
bits + 1
|
||||
};
|
||||
Self::from_bits(next_bits)
|
||||
}
|
||||
|
||||
/// Takes the reciprocal (inverse) of a number, `1/x`.
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f16)]
|
||||
/// # // FIXME(f16_f128): remove when `extendhfsf2` and `truncsfhf2` are available
|
||||
/// # #[cfg(target_os = "linux")] {
|
||||
///
|
||||
/// let x = 2.0_f16;
|
||||
/// let abs_difference = (x.recip() - (1.0 / x)).abs();
|
||||
///
|
||||
/// assert!(abs_difference <= f16::EPSILON);
|
||||
/// # }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[cfg(not(bootstrap))]
|
||||
#[unstable(feature = "f16", issue = "116909")]
|
||||
#[must_use = "this returns the result of the operation, without modifying the original"]
|
||||
pub fn recip(self) -> Self {
|
||||
1.0 / self
|
||||
}
|
||||
|
||||
/// Converts radians to degrees.
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f16)]
|
||||
/// # // FIXME(f16_f128): remove when `extendhfsf2` and `truncsfhf2` are available
|
||||
/// # #[cfg(target_os = "linux")] {
|
||||
///
|
||||
/// let angle = std::f16::consts::PI;
|
||||
///
|
||||
/// let abs_difference = (angle.to_degrees() - 180.0).abs();
|
||||
/// assert!(abs_difference <= 0.5);
|
||||
/// # }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[cfg(not(bootstrap))]
|
||||
#[unstable(feature = "f16", issue = "116909")]
|
||||
#[must_use = "this returns the result of the operation, without modifying the original"]
|
||||
pub fn to_degrees(self) -> Self {
|
||||
// Use a literal for better precision.
|
||||
const PIS_IN_180: f16 = 57.2957795130823208767981548141051703_f16;
|
||||
self * PIS_IN_180
|
||||
}
|
||||
|
||||
/// Converts degrees to radians.
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f16)]
|
||||
/// # // FIXME(f16_f128): remove when `extendhfsf2` and `truncsfhf2` are available
|
||||
/// # #[cfg(target_os = "linux")] {
|
||||
///
|
||||
/// let angle = 180.0f16;
|
||||
///
|
||||
/// let abs_difference = (angle.to_radians() - std::f16::consts::PI).abs();
|
||||
///
|
||||
/// assert!(abs_difference <= 0.01);
|
||||
/// # }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[cfg(not(bootstrap))]
|
||||
#[unstable(feature = "f16", issue = "116909")]
|
||||
#[must_use = "this returns the result of the operation, without modifying the original"]
|
||||
pub fn to_radians(self) -> f16 {
|
||||
// Use a literal for better precision.
|
||||
const RADS_PER_DEG: f16 = 0.017453292519943295769236907684886_f16;
|
||||
self * RADS_PER_DEG
|
||||
}
|
||||
|
||||
/// Rounds toward zero and converts to any primitive integer type,
|
||||
/// assuming that the value is finite and fits in that type.
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f16)]
|
||||
/// # #[cfg(target_arch = "aarch64")] { // FIXME(f16_F128): rust-lang/rust#123885
|
||||
///
|
||||
/// let value = 4.6_f16;
|
||||
/// let rounded = unsafe { value.to_int_unchecked::<u16>() };
|
||||
/// assert_eq!(rounded, 4);
|
||||
///
|
||||
/// let value = -128.9_f16;
|
||||
/// let rounded = unsafe { value.to_int_unchecked::<i8>() };
|
||||
/// assert_eq!(rounded, i8::MIN);
|
||||
/// # }
|
||||
/// ```
|
||||
///
|
||||
/// # Safety
|
||||
///
|
||||
/// The value must:
|
||||
///
|
||||
/// * Not be `NaN`
|
||||
/// * Not be infinite
|
||||
/// * Be representable in the return type `Int`, after truncating off its fractional part
|
||||
#[inline]
|
||||
#[unstable(feature = "f16", issue = "116909")]
|
||||
#[must_use = "this returns the result of the operation, without modifying the original"]
|
||||
pub unsafe fn to_int_unchecked<Int>(self) -> Int
|
||||
where
|
||||
Self: FloatToInt<Int>,
|
||||
{
|
||||
// SAFETY: the caller must uphold the safety contract for
|
||||
// `FloatToInt::to_int_unchecked`.
|
||||
unsafe { FloatToInt::<Int>::to_int_unchecked(self) }
|
||||
}
|
||||
|
||||
/// Raw transmutation to `u16`.
|
||||
///
|
||||
/// This is currently identical to `transmute::<f16, u16>(self)` on all platforms.
|
||||
|
|
@ -282,6 +616,16 @@ impl f16 {
|
|||
///
|
||||
/// Note that this function is distinct from `as` casting, which attempts to
|
||||
/// preserve the *numeric* value, and not the bitwise value.
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f16)]
|
||||
/// # #[cfg(target_arch = "aarch64")] { // FIXME(f16_F128): rust-lang/rust#123885
|
||||
///
|
||||
/// # // FIXME(f16_f128): enable this once const casting works
|
||||
/// # // assert_ne!((1f16).to_bits(), 1f16 as u128); // to_bits() is not casting!
|
||||
/// assert_eq!((12.5f16).to_bits(), 0x4a40);
|
||||
/// # }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[unstable(feature = "f16", issue = "116909")]
|
||||
#[must_use = "this returns the result of the operation, without modifying the original"]
|
||||
|
|
@ -321,6 +665,15 @@ impl f16 {
|
|||
///
|
||||
/// Note that this function is distinct from `as` casting, which attempts to
|
||||
/// preserve the *numeric* value, and not the bitwise value.
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f16)]
|
||||
/// # #[cfg(target_arch = "aarch64")] { // FIXME(f16_F128): rust-lang/rust#123885
|
||||
///
|
||||
/// let v = f16::from_bits(0x4a40);
|
||||
/// assert_eq!(v, 12.5);
|
||||
/// # }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[must_use]
|
||||
#[unstable(feature = "f16", issue = "116909")]
|
||||
|
|
@ -330,4 +683,293 @@ impl f16 {
|
|||
// Stability concerns.
|
||||
unsafe { mem::transmute(v) }
|
||||
}
|
||||
|
||||
/// Return the memory representation of this floating point number as a byte array in
|
||||
/// big-endian (network) byte order.
|
||||
///
|
||||
/// See [`from_bits`](Self::from_bits) for some discussion of the
|
||||
/// portability of this operation (there are almost no issues).
|
||||
///
|
||||
/// # Examples
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f16)]
|
||||
///
|
||||
/// let bytes = 12.5f16.to_be_bytes();
|
||||
/// assert_eq!(bytes, [0x4a, 0x40]);
|
||||
/// ```
|
||||
#[inline]
|
||||
#[unstable(feature = "f16", issue = "116909")]
|
||||
#[must_use = "this returns the result of the operation, without modifying the original"]
|
||||
pub fn to_be_bytes(self) -> [u8; 2] {
|
||||
self.to_bits().to_be_bytes()
|
||||
}
|
||||
|
||||
/// Return the memory representation of this floating point number as a byte array in
|
||||
/// little-endian byte order.
|
||||
///
|
||||
/// See [`from_bits`](Self::from_bits) for some discussion of the
|
||||
/// portability of this operation (there are almost no issues).
|
||||
///
|
||||
/// # Examples
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f16)]
|
||||
///
|
||||
/// let bytes = 12.5f16.to_le_bytes();
|
||||
/// assert_eq!(bytes, [0x40, 0x4a]);
|
||||
/// ```
|
||||
#[inline]
|
||||
#[unstable(feature = "f16", issue = "116909")]
|
||||
#[must_use = "this returns the result of the operation, without modifying the original"]
|
||||
pub fn to_le_bytes(self) -> [u8; 2] {
|
||||
self.to_bits().to_le_bytes()
|
||||
}
|
||||
|
||||
/// Return the memory representation of this floating point number as a byte array in
|
||||
/// native byte order.
|
||||
///
|
||||
/// As the target platform's native endianness is used, portable code
|
||||
/// should use [`to_be_bytes`] or [`to_le_bytes`], as appropriate, instead.
|
||||
///
|
||||
/// [`to_be_bytes`]: f16::to_be_bytes
|
||||
/// [`to_le_bytes`]: f16::to_le_bytes
|
||||
///
|
||||
/// See [`from_bits`](Self::from_bits) for some discussion of the
|
||||
/// portability of this operation (there are almost no issues).
|
||||
///
|
||||
/// # Examples
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f16)]
|
||||
///
|
||||
/// let bytes = 12.5f16.to_ne_bytes();
|
||||
/// assert_eq!(
|
||||
/// bytes,
|
||||
/// if cfg!(target_endian = "big") {
|
||||
/// [0x4a, 0x40]
|
||||
/// } else {
|
||||
/// [0x40, 0x4a]
|
||||
/// }
|
||||
/// );
|
||||
/// ```
|
||||
#[inline]
|
||||
#[unstable(feature = "f16", issue = "116909")]
|
||||
#[must_use = "this returns the result of the operation, without modifying the original"]
|
||||
pub fn to_ne_bytes(self) -> [u8; 2] {
|
||||
self.to_bits().to_ne_bytes()
|
||||
}
|
||||
|
||||
/// Create a floating point value from its representation as a byte array in big endian.
|
||||
///
|
||||
/// See [`from_bits`](Self::from_bits) for some discussion of the
|
||||
/// portability of this operation (there are almost no issues).
|
||||
///
|
||||
/// # Examples
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f16)]
|
||||
/// # #[cfg(target_arch = "aarch64")] { // FIXME(f16_F128): rust-lang/rust#123885
|
||||
///
|
||||
/// let value = f16::from_be_bytes([0x4a, 0x40]);
|
||||
/// assert_eq!(value, 12.5);
|
||||
/// # }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[must_use]
|
||||
#[unstable(feature = "f16", issue = "116909")]
|
||||
pub fn from_be_bytes(bytes: [u8; 2]) -> Self {
|
||||
Self::from_bits(u16::from_be_bytes(bytes))
|
||||
}
|
||||
|
||||
/// Create a floating point value from its representation as a byte array in little endian.
|
||||
///
|
||||
/// See [`from_bits`](Self::from_bits) for some discussion of the
|
||||
/// portability of this operation (there are almost no issues).
|
||||
///
|
||||
/// # Examples
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f16)]
|
||||
/// # #[cfg(target_arch = "aarch64")] { // FIXME(f16_F128): rust-lang/rust#123885
|
||||
///
|
||||
/// let value = f16::from_le_bytes([0x40, 0x4a]);
|
||||
/// assert_eq!(value, 12.5);
|
||||
/// # }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[must_use]
|
||||
#[unstable(feature = "f16", issue = "116909")]
|
||||
pub fn from_le_bytes(bytes: [u8; 2]) -> Self {
|
||||
Self::from_bits(u16::from_le_bytes(bytes))
|
||||
}
|
||||
|
||||
/// Create a floating point value from its representation as a byte array in native endian.
|
||||
///
|
||||
/// As the target platform's native endianness is used, portable code
|
||||
/// likely wants to use [`from_be_bytes`] or [`from_le_bytes`], as
|
||||
/// appropriate instead.
|
||||
///
|
||||
/// [`from_be_bytes`]: f16::from_be_bytes
|
||||
/// [`from_le_bytes`]: f16::from_le_bytes
|
||||
///
|
||||
/// See [`from_bits`](Self::from_bits) for some discussion of the
|
||||
/// portability of this operation (there are almost no issues).
|
||||
///
|
||||
/// # Examples
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f16)]
|
||||
/// # #[cfg(target_arch = "aarch64")] { // FIXME(f16_F128): rust-lang/rust#123885
|
||||
///
|
||||
/// let value = f16::from_ne_bytes(if cfg!(target_endian = "big") {
|
||||
/// [0x4a, 0x40]
|
||||
/// } else {
|
||||
/// [0x40, 0x4a]
|
||||
/// });
|
||||
/// assert_eq!(value, 12.5);
|
||||
/// # }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[must_use]
|
||||
#[unstable(feature = "f16", issue = "116909")]
|
||||
pub fn from_ne_bytes(bytes: [u8; 2]) -> Self {
|
||||
Self::from_bits(u16::from_ne_bytes(bytes))
|
||||
}
|
||||
|
||||
/// Return the ordering between `self` and `other`.
|
||||
///
|
||||
/// Unlike the standard partial comparison between floating point numbers,
|
||||
/// this comparison always produces an ordering in accordance to
|
||||
/// the `totalOrder` predicate as defined in the IEEE 754 (2008 revision)
|
||||
/// floating point standard. The values are ordered in the following sequence:
|
||||
///
|
||||
/// - negative quiet NaN
|
||||
/// - negative signaling NaN
|
||||
/// - negative infinity
|
||||
/// - negative numbers
|
||||
/// - negative subnormal numbers
|
||||
/// - negative zero
|
||||
/// - positive zero
|
||||
/// - positive subnormal numbers
|
||||
/// - positive numbers
|
||||
/// - positive infinity
|
||||
/// - positive signaling NaN
|
||||
/// - positive quiet NaN.
|
||||
///
|
||||
/// The ordering established by this function does not always agree with the
|
||||
/// [`PartialOrd`] and [`PartialEq`] implementations of `f16`. For example,
|
||||
/// they consider negative and positive zero equal, while `total_cmp`
|
||||
/// doesn't.
|
||||
///
|
||||
/// The interpretation of the signaling NaN bit follows the definition in
|
||||
/// the IEEE 754 standard, which may not match the interpretation by some of
|
||||
/// the older, non-conformant (e.g. MIPS) hardware implementations.
|
||||
///
|
||||
/// # Example
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f16)]
|
||||
///
|
||||
/// struct GoodBoy {
|
||||
/// name: &'static str,
|
||||
/// weight: f16,
|
||||
/// }
|
||||
///
|
||||
/// let mut bois = vec![
|
||||
/// GoodBoy { name: "Pucci", weight: 0.1 },
|
||||
/// GoodBoy { name: "Woofer", weight: 99.0 },
|
||||
/// GoodBoy { name: "Yapper", weight: 10.0 },
|
||||
/// GoodBoy { name: "Chonk", weight: f16::INFINITY },
|
||||
/// GoodBoy { name: "Abs. Unit", weight: f16::NAN },
|
||||
/// GoodBoy { name: "Floaty", weight: -5.0 },
|
||||
/// ];
|
||||
///
|
||||
/// bois.sort_by(|a, b| a.weight.total_cmp(&b.weight));
|
||||
///
|
||||
/// // `f16::NAN` could be positive or negative, which will affect the sort order.
|
||||
/// if f16::NAN.is_sign_negative() {
|
||||
/// bois.into_iter().map(|b| b.weight)
|
||||
/// .zip([f16::NAN, -5.0, 0.1, 10.0, 99.0, f16::INFINITY].iter())
|
||||
/// .for_each(|(a, b)| assert_eq!(a.to_bits(), b.to_bits()))
|
||||
/// } else {
|
||||
/// bois.into_iter().map(|b| b.weight)
|
||||
/// .zip([-5.0, 0.1, 10.0, 99.0, f16::INFINITY, f16::NAN].iter())
|
||||
/// .for_each(|(a, b)| assert_eq!(a.to_bits(), b.to_bits()))
|
||||
/// }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[must_use]
|
||||
#[cfg(not(bootstrap))]
|
||||
#[unstable(feature = "f16", issue = "116909")]
|
||||
pub fn total_cmp(&self, other: &Self) -> crate::cmp::Ordering {
|
||||
let mut left = self.to_bits() as i16;
|
||||
let mut right = other.to_bits() as i16;
|
||||
|
||||
// In case of negatives, flip all the bits except the sign
|
||||
// to achieve a similar layout as two's complement integers
|
||||
//
|
||||
// Why does this work? IEEE 754 floats consist of three fields:
|
||||
// Sign bit, exponent and mantissa. The set of exponent and mantissa
|
||||
// fields as a whole have the property that their bitwise order is
|
||||
// equal to the numeric magnitude where the magnitude is defined.
|
||||
// The magnitude is not normally defined on NaN values, but
|
||||
// IEEE 754 totalOrder defines the NaN values also to follow the
|
||||
// bitwise order. This leads to order explained in the doc comment.
|
||||
// However, the representation of magnitude is the same for negative
|
||||
// and positive numbers – only the sign bit is different.
|
||||
// To easily compare the floats as signed integers, we need to
|
||||
// flip the exponent and mantissa bits in case of negative numbers.
|
||||
// We effectively convert the numbers to "two's complement" form.
|
||||
//
|
||||
// To do the flipping, we construct a mask and XOR against it.
|
||||
// We branchlessly calculate an "all-ones except for the sign bit"
|
||||
// mask from negative-signed values: right shifting sign-extends
|
||||
// the integer, so we "fill" the mask with sign bits, and then
|
||||
// convert to unsigned to push one more zero bit.
|
||||
// On positive values, the mask is all zeros, so it's a no-op.
|
||||
left ^= (((left >> 15) as u16) >> 1) as i16;
|
||||
right ^= (((right >> 15) as u16) >> 1) as i16;
|
||||
|
||||
left.cmp(&right)
|
||||
}
|
||||
|
||||
/// Restrict a value to a certain interval unless it is NaN.
|
||||
///
|
||||
/// Returns `max` if `self` is greater than `max`, and `min` if `self` is
|
||||
/// less than `min`. Otherwise this returns `self`.
|
||||
///
|
||||
/// Note that this function returns NaN if the initial value was NaN as
|
||||
/// well.
|
||||
///
|
||||
/// # Panics
|
||||
///
|
||||
/// Panics if `min > max`, `min` is NaN, or `max` is NaN.
|
||||
///
|
||||
/// # Examples
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f16)]
|
||||
/// # #[cfg(target_arch = "aarch64")] { // FIXME(f16_F128): rust-lang/rust#123885
|
||||
///
|
||||
/// assert!((-3.0f16).clamp(-2.0, 1.0) == -2.0);
|
||||
/// assert!((0.0f16).clamp(-2.0, 1.0) == 0.0);
|
||||
/// assert!((2.0f16).clamp(-2.0, 1.0) == 1.0);
|
||||
/// assert!((f16::NAN).clamp(-2.0, 1.0).is_nan());
|
||||
/// # }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[cfg(not(bootstrap))]
|
||||
#[unstable(feature = "f16", issue = "116909")]
|
||||
#[must_use = "method returns a new number and does not mutate the original value"]
|
||||
pub fn clamp(mut self, min: f16, max: f16) -> f16 {
|
||||
assert!(min <= max, "min > max, or either was NaN. min = {min:?}, max = {max:?}");
|
||||
if self < min {
|
||||
self = min;
|
||||
}
|
||||
if self > max {
|
||||
self = max;
|
||||
}
|
||||
self
|
||||
}
|
||||
}
|
||||
|
|
|
|||
|
|
@ -490,6 +490,21 @@ impl f32 {
|
|||
#[stable(feature = "assoc_int_consts", since = "1.43.0")]
|
||||
pub const NEG_INFINITY: f32 = -1.0_f32 / 0.0_f32;
|
||||
|
||||
/// Sign bit
|
||||
const SIGN_MASK: u32 = 0x8000_0000;
|
||||
|
||||
/// Exponent mask
|
||||
const EXP_MASK: u32 = 0x7f80_0000;
|
||||
|
||||
/// Mantissa mask
|
||||
const MAN_MASK: u32 = 0x007f_ffff;
|
||||
|
||||
/// Minimum representable positive value (min subnormal)
|
||||
const TINY_BITS: u32 = 0x1;
|
||||
|
||||
/// Minimum representable negative value (min negative subnormal)
|
||||
const NEG_TINY_BITS: u32 = Self::TINY_BITS | Self::SIGN_MASK;
|
||||
|
||||
/// Returns `true` if this value is NaN.
|
||||
///
|
||||
/// ```
|
||||
|
|
@ -515,7 +530,7 @@ impl f32 {
|
|||
#[rustc_const_unstable(feature = "const_float_classify", issue = "72505")]
|
||||
pub(crate) const fn abs_private(self) -> f32 {
|
||||
// SAFETY: This transmutation is fine. Probably. For the reasons std is using it.
|
||||
unsafe { mem::transmute::<u32, f32>(mem::transmute::<f32, u32>(self) & 0x7fff_ffff) }
|
||||
unsafe { mem::transmute::<u32, f32>(mem::transmute::<f32, u32>(self) & !Self::SIGN_MASK) }
|
||||
}
|
||||
|
||||
/// Returns `true` if this value is positive infinity or negative infinity, and
|
||||
|
|
@ -682,12 +697,9 @@ impl f32 {
|
|||
// runtime-deviating logic which may or may not be acceptable.
|
||||
#[rustc_const_unstable(feature = "const_float_classify", issue = "72505")]
|
||||
const unsafe fn partial_classify(self) -> FpCategory {
|
||||
const EXP_MASK: u32 = 0x7f800000;
|
||||
const MAN_MASK: u32 = 0x007fffff;
|
||||
|
||||
// SAFETY: The caller is not asking questions for which this will tell lies.
|
||||
let b = unsafe { mem::transmute::<f32, u32>(self) };
|
||||
match (b & MAN_MASK, b & EXP_MASK) {
|
||||
match (b & Self::MAN_MASK, b & Self::EXP_MASK) {
|
||||
(0, 0) => FpCategory::Zero,
|
||||
(_, 0) => FpCategory::Subnormal,
|
||||
_ => FpCategory::Normal,
|
||||
|
|
@ -699,12 +711,9 @@ impl f32 {
|
|||
// plus a transmute. We do not live in a just world, but we can make it more so.
|
||||
#[rustc_const_unstable(feature = "const_float_classify", issue = "72505")]
|
||||
const fn classify_bits(b: u32) -> FpCategory {
|
||||
const EXP_MASK: u32 = 0x7f800000;
|
||||
const MAN_MASK: u32 = 0x007fffff;
|
||||
|
||||
match (b & MAN_MASK, b & EXP_MASK) {
|
||||
(0, EXP_MASK) => FpCategory::Infinite,
|
||||
(_, EXP_MASK) => FpCategory::Nan,
|
||||
match (b & Self::MAN_MASK, b & Self::EXP_MASK) {
|
||||
(0, Self::EXP_MASK) => FpCategory::Infinite,
|
||||
(_, Self::EXP_MASK) => FpCategory::Nan,
|
||||
(0, 0) => FpCategory::Zero,
|
||||
(_, 0) => FpCategory::Subnormal,
|
||||
_ => FpCategory::Normal,
|
||||
|
|
@ -787,19 +796,17 @@ impl f32 {
|
|||
#[unstable(feature = "float_next_up_down", issue = "91399")]
|
||||
#[rustc_const_unstable(feature = "float_next_up_down", issue = "91399")]
|
||||
pub const fn next_up(self) -> Self {
|
||||
// We must use strictly integer arithmetic to prevent denormals from
|
||||
// flushing to zero after an arithmetic operation on some platforms.
|
||||
const TINY_BITS: u32 = 0x1; // Smallest positive f32.
|
||||
const CLEAR_SIGN_MASK: u32 = 0x7fff_ffff;
|
||||
|
||||
// Some targets violate Rust's assumption of IEEE semantics, e.g. by flushing
|
||||
// denormals to zero. This is in general unsound and unsupported, but here
|
||||
// we do our best to still produce the correct result on such targets.
|
||||
let bits = self.to_bits();
|
||||
if self.is_nan() || bits == Self::INFINITY.to_bits() {
|
||||
return self;
|
||||
}
|
||||
|
||||
let abs = bits & CLEAR_SIGN_MASK;
|
||||
let abs = bits & !Self::SIGN_MASK;
|
||||
let next_bits = if abs == 0 {
|
||||
TINY_BITS
|
||||
Self::TINY_BITS
|
||||
} else if bits == abs {
|
||||
bits + 1
|
||||
} else {
|
||||
|
|
@ -837,19 +844,17 @@ impl f32 {
|
|||
#[unstable(feature = "float_next_up_down", issue = "91399")]
|
||||
#[rustc_const_unstable(feature = "float_next_up_down", issue = "91399")]
|
||||
pub const fn next_down(self) -> Self {
|
||||
// We must use strictly integer arithmetic to prevent denormals from
|
||||
// flushing to zero after an arithmetic operation on some platforms.
|
||||
const NEG_TINY_BITS: u32 = 0x8000_0001; // Smallest (in magnitude) negative f32.
|
||||
const CLEAR_SIGN_MASK: u32 = 0x7fff_ffff;
|
||||
|
||||
// Some targets violate Rust's assumption of IEEE semantics, e.g. by flushing
|
||||
// denormals to zero. This is in general unsound and unsupported, but here
|
||||
// we do our best to still produce the correct result on such targets.
|
||||
let bits = self.to_bits();
|
||||
if self.is_nan() || bits == Self::NEG_INFINITY.to_bits() {
|
||||
return self;
|
||||
}
|
||||
|
||||
let abs = bits & CLEAR_SIGN_MASK;
|
||||
let abs = bits & !Self::SIGN_MASK;
|
||||
let next_bits = if abs == 0 {
|
||||
NEG_TINY_BITS
|
||||
Self::NEG_TINY_BITS
|
||||
} else if bits == abs {
|
||||
bits - 1
|
||||
} else {
|
||||
|
|
|
|||
|
|
@ -489,6 +489,21 @@ impl f64 {
|
|||
#[stable(feature = "assoc_int_consts", since = "1.43.0")]
|
||||
pub const NEG_INFINITY: f64 = -1.0_f64 / 0.0_f64;
|
||||
|
||||
/// Sign bit
|
||||
const SIGN_MASK: u64 = 0x8000_0000_0000_0000;
|
||||
|
||||
/// Exponent mask
|
||||
const EXP_MASK: u64 = 0x7ff0_0000_0000_0000;
|
||||
|
||||
/// Mantissa mask
|
||||
const MAN_MASK: u64 = 0x000f_ffff_ffff_ffff;
|
||||
|
||||
/// Minimum representable positive value (min subnormal)
|
||||
const TINY_BITS: u64 = 0x1;
|
||||
|
||||
/// Minimum representable negative value (min negative subnormal)
|
||||
const NEG_TINY_BITS: u64 = Self::TINY_BITS | Self::SIGN_MASK;
|
||||
|
||||
/// Returns `true` if this value is NaN.
|
||||
///
|
||||
/// ```
|
||||
|
|
@ -514,9 +529,7 @@ impl f64 {
|
|||
#[rustc_const_unstable(feature = "const_float_classify", issue = "72505")]
|
||||
pub(crate) const fn abs_private(self) -> f64 {
|
||||
// SAFETY: This transmutation is fine. Probably. For the reasons std is using it.
|
||||
unsafe {
|
||||
mem::transmute::<u64, f64>(mem::transmute::<f64, u64>(self) & 0x7fff_ffff_ffff_ffff)
|
||||
}
|
||||
unsafe { mem::transmute::<u64, f64>(mem::transmute::<f64, u64>(self) & !Self::SIGN_MASK) }
|
||||
}
|
||||
|
||||
/// Returns `true` if this value is positive infinity or negative infinity, and
|
||||
|
|
@ -673,13 +686,10 @@ impl f64 {
|
|||
// and some normal floating point numbers truncated from an x87 FPU.
|
||||
#[rustc_const_unstable(feature = "const_float_classify", issue = "72505")]
|
||||
const unsafe fn partial_classify(self) -> FpCategory {
|
||||
const EXP_MASK: u64 = 0x7ff0000000000000;
|
||||
const MAN_MASK: u64 = 0x000fffffffffffff;
|
||||
|
||||
// SAFETY: The caller is not asking questions for which this will tell lies.
|
||||
let b = unsafe { mem::transmute::<f64, u64>(self) };
|
||||
match (b & MAN_MASK, b & EXP_MASK) {
|
||||
(0, EXP_MASK) => FpCategory::Infinite,
|
||||
match (b & Self::MAN_MASK, b & Self::EXP_MASK) {
|
||||
(0, Self::EXP_MASK) => FpCategory::Infinite,
|
||||
(0, 0) => FpCategory::Zero,
|
||||
(_, 0) => FpCategory::Subnormal,
|
||||
_ => FpCategory::Normal,
|
||||
|
|
@ -691,12 +701,9 @@ impl f64 {
|
|||
// plus a transmute. We do not live in a just world, but we can make it more so.
|
||||
#[rustc_const_unstable(feature = "const_float_classify", issue = "72505")]
|
||||
const fn classify_bits(b: u64) -> FpCategory {
|
||||
const EXP_MASK: u64 = 0x7ff0000000000000;
|
||||
const MAN_MASK: u64 = 0x000fffffffffffff;
|
||||
|
||||
match (b & MAN_MASK, b & EXP_MASK) {
|
||||
(0, EXP_MASK) => FpCategory::Infinite,
|
||||
(_, EXP_MASK) => FpCategory::Nan,
|
||||
match (b & Self::MAN_MASK, b & Self::EXP_MASK) {
|
||||
(0, Self::EXP_MASK) => FpCategory::Infinite,
|
||||
(_, Self::EXP_MASK) => FpCategory::Nan,
|
||||
(0, 0) => FpCategory::Zero,
|
||||
(_, 0) => FpCategory::Subnormal,
|
||||
_ => FpCategory::Normal,
|
||||
|
|
@ -756,7 +763,7 @@ impl f64 {
|
|||
// IEEE754 says: isSignMinus(x) is true if and only if x has negative sign. isSignMinus
|
||||
// applies to zeros and NaNs as well.
|
||||
// SAFETY: This is just transmuting to get the sign bit, it's fine.
|
||||
unsafe { mem::transmute::<f64, u64>(self) & 0x8000_0000_0000_0000 != 0 }
|
||||
unsafe { mem::transmute::<f64, u64>(self) & Self::SIGN_MASK != 0 }
|
||||
}
|
||||
|
||||
#[must_use]
|
||||
|
|
@ -797,19 +804,17 @@ impl f64 {
|
|||
#[unstable(feature = "float_next_up_down", issue = "91399")]
|
||||
#[rustc_const_unstable(feature = "float_next_up_down", issue = "91399")]
|
||||
pub const fn next_up(self) -> Self {
|
||||
// We must use strictly integer arithmetic to prevent denormals from
|
||||
// flushing to zero after an arithmetic operation on some platforms.
|
||||
const TINY_BITS: u64 = 0x1; // Smallest positive f64.
|
||||
const CLEAR_SIGN_MASK: u64 = 0x7fff_ffff_ffff_ffff;
|
||||
|
||||
// Some targets violate Rust's assumption of IEEE semantics, e.g. by flushing
|
||||
// denormals to zero. This is in general unsound and unsupported, but here
|
||||
// we do our best to still produce the correct result on such targets.
|
||||
let bits = self.to_bits();
|
||||
if self.is_nan() || bits == Self::INFINITY.to_bits() {
|
||||
return self;
|
||||
}
|
||||
|
||||
let abs = bits & CLEAR_SIGN_MASK;
|
||||
let abs = bits & !Self::SIGN_MASK;
|
||||
let next_bits = if abs == 0 {
|
||||
TINY_BITS
|
||||
Self::TINY_BITS
|
||||
} else if bits == abs {
|
||||
bits + 1
|
||||
} else {
|
||||
|
|
@ -847,19 +852,17 @@ impl f64 {
|
|||
#[unstable(feature = "float_next_up_down", issue = "91399")]
|
||||
#[rustc_const_unstable(feature = "float_next_up_down", issue = "91399")]
|
||||
pub const fn next_down(self) -> Self {
|
||||
// We must use strictly integer arithmetic to prevent denormals from
|
||||
// flushing to zero after an arithmetic operation on some platforms.
|
||||
const NEG_TINY_BITS: u64 = 0x8000_0000_0000_0001; // Smallest (in magnitude) negative f64.
|
||||
const CLEAR_SIGN_MASK: u64 = 0x7fff_ffff_ffff_ffff;
|
||||
|
||||
// Some targets violate Rust's assumption of IEEE semantics, e.g. by flushing
|
||||
// denormals to zero. This is in general unsound and unsupported, but here
|
||||
// we do our best to still produce the correct result on such targets.
|
||||
let bits = self.to_bits();
|
||||
if self.is_nan() || bits == Self::NEG_INFINITY.to_bits() {
|
||||
return self;
|
||||
}
|
||||
|
||||
let abs = bits & CLEAR_SIGN_MASK;
|
||||
let abs = bits & !Self::SIGN_MASK;
|
||||
let next_bits = if abs == 0 {
|
||||
NEG_TINY_BITS
|
||||
Self::NEG_TINY_BITS
|
||||
} else if bits == abs {
|
||||
bits - 1
|
||||
} else {
|
||||
|
|
|
|||
|
|
@ -7,6 +7,10 @@ fn main() {
|
|||
let target_vendor =
|
||||
env::var("CARGO_CFG_TARGET_VENDOR").expect("CARGO_CFG_TARGET_VENDOR was not set");
|
||||
let target_env = env::var("CARGO_CFG_TARGET_ENV").expect("CARGO_CFG_TARGET_ENV was not set");
|
||||
let target_pointer_width: u32 = env::var("CARGO_CFG_TARGET_POINTER_WIDTH")
|
||||
.expect("CARGO_CFG_TARGET_POINTER_WIDTH was not set")
|
||||
.parse()
|
||||
.unwrap();
|
||||
|
||||
println!("cargo:rustc-check-cfg=cfg(netbsd10)");
|
||||
if target_os == "netbsd" && env::var("RUSTC_STD_NETBSD10").is_ok() {
|
||||
|
|
@ -70,4 +74,62 @@ fn main() {
|
|||
println!("cargo:rustc-cfg=backtrace_in_libstd");
|
||||
|
||||
println!("cargo:rustc-env=STD_ENV_ARCH={}", env::var("CARGO_CFG_TARGET_ARCH").unwrap());
|
||||
|
||||
// Emit these on platforms that have no known ABI bugs, LLVM selection bugs, lowering bugs,
|
||||
// missing symbols, or other problems, to determine when tests get run.
|
||||
// If more broken platforms are found, please update the tracking issue at
|
||||
// <https://github.com/rust-lang/rust/issues/116909>
|
||||
//
|
||||
// Some of these match arms are redundant; the goal is to separate reasons that the type is
|
||||
// unreliable, even when multiple reasons might fail the same platform.
|
||||
println!("cargo:rustc-check-cfg=cfg(reliable_f16)");
|
||||
println!("cargo:rustc-check-cfg=cfg(reliable_f128)");
|
||||
|
||||
let has_reliable_f16 = match (target_arch.as_str(), target_os.as_str()) {
|
||||
// Selection failure until recent LLVM <https://github.com/llvm/llvm-project/issues/93894>
|
||||
// FIXME(llvm19): can probably be removed at the version bump
|
||||
("loongarch64", _) => false,
|
||||
// Selection failure <https://github.com/llvm/llvm-project/issues/50374>
|
||||
("s390x", _) => false,
|
||||
// Unsupported <https://github.com/llvm/llvm-project/issues/94434>
|
||||
("arm64ec", _) => false,
|
||||
// MinGW ABI bugs <https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115054>
|
||||
("x86", "windows") => false,
|
||||
// x86 has ABI bugs that show up with optimizations. This should be partially fixed with
|
||||
// the compiler-builtins update. <https://github.com/rust-lang/rust/issues/123885>
|
||||
("x86" | "x86_64", _) => false,
|
||||
// Missing `__gnu_h2f_ieee` and `__gnu_f2h_ieee`
|
||||
("powerpc" | "powerpc64" | "powerpc64le", _) => false,
|
||||
// Missing `__extendhfsf` and `__truncsfhf`
|
||||
("riscv32" | "riscv64", _) => false,
|
||||
// Most OSs are missing `__extendhfsf` and `__truncsfhf`
|
||||
(_, "linux" | "macos") => true,
|
||||
// Almost all OSs besides Linux and MacOS are missing symbols until compiler-builtins can
|
||||
// be updated. <https://github.com/rust-lang/rust/pull/125016> will get some of these, the
|
||||
// next CB update should get the rest.
|
||||
_ => false,
|
||||
};
|
||||
|
||||
let has_reliable_f128 = match (target_arch.as_str(), target_os.as_str()) {
|
||||
// Unsupported <https://github.com/llvm/llvm-project/issues/94434>
|
||||
("arm64ec", _) => false,
|
||||
// ABI and precision bugs <https://github.com/rust-lang/rust/issues/125109>
|
||||
// <https://github.com/rust-lang/rust/issues/125102>
|
||||
("powerpc" | "powerpc64", _) => false,
|
||||
// Selection bug <https://github.com/llvm/llvm-project/issues/95471>
|
||||
("nvptx64", _) => false,
|
||||
// ABI unsupported <https://github.com/llvm/llvm-project/issues/41838>
|
||||
("sparc", _) => false,
|
||||
// 64-bit Linux is about the only platform to have f128 symbols by default
|
||||
(_, "linux") if target_pointer_width == 64 => true,
|
||||
// Same as for f16, except MacOS is also missing f128 symbols.
|
||||
_ => false,
|
||||
};
|
||||
|
||||
if has_reliable_f16 {
|
||||
println!("cargo:rustc-cfg=reliable_f16");
|
||||
}
|
||||
if has_reliable_f128 {
|
||||
println!("cargo:rustc-cfg=reliable_f128");
|
||||
}
|
||||
}
|
||||
|
|
|
|||
|
|
@ -32,4 +32,34 @@ impl f128 {
|
|||
pub fn powi(self, n: i32) -> f128 {
|
||||
unsafe { intrinsics::powif128(self, n) }
|
||||
}
|
||||
|
||||
/// Computes the absolute value of `self`.
|
||||
///
|
||||
/// This function always returns the precise result.
|
||||
///
|
||||
/// # Examples
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f128)]
|
||||
/// # #[cfg(reliable_f128)] { // FIXME(f16_f128): reliable_f128
|
||||
///
|
||||
/// let x = 3.5_f128;
|
||||
/// let y = -3.5_f128;
|
||||
///
|
||||
/// assert_eq!(x.abs(), x);
|
||||
/// assert_eq!(y.abs(), -y);
|
||||
///
|
||||
/// assert!(f128::NAN.abs().is_nan());
|
||||
/// # }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[cfg(not(bootstrap))]
|
||||
#[rustc_allow_incoherent_impl]
|
||||
#[unstable(feature = "f128", issue = "116909")]
|
||||
#[must_use = "method returns a new number and does not mutate the original value"]
|
||||
pub fn abs(self) -> Self {
|
||||
// FIXME(f16_f128): replace with `intrinsics::fabsf128` when available
|
||||
// We don't do this now because LLVM has lowering bugs for f128 math.
|
||||
Self::from_bits(self.to_bits() & !(1 << 127))
|
||||
}
|
||||
}
|
||||
|
|
|
|||
|
|
@ -1,29 +1,31 @@
|
|||
#![allow(dead_code)] // FIXME(f16_f128): remove once constants are used
|
||||
#![cfg(not(bootstrap))]
|
||||
// FIXME(f16_f128): only tested on platforms that have symbols and aren't buggy
|
||||
#![cfg(reliable_f128)]
|
||||
|
||||
use crate::f128::consts;
|
||||
use crate::num::*;
|
||||
|
||||
/// Smallest number
|
||||
const TINY_BITS: u128 = 0x1;
|
||||
|
||||
/// Next smallest number
|
||||
const TINY_UP_BITS: u128 = 0x2;
|
||||
|
||||
/// Exponent = 0b11...10, Sifnificand 0b1111..10. Min val > 0
|
||||
const MAX_DOWN_BITS: u128 = 0x7ffeffffffffffffffffffffffffffff;
|
||||
const MAX_DOWN_BITS: u128 = 0x7ffefffffffffffffffffffffffffffe;
|
||||
|
||||
/// Zeroed exponent, full significant
|
||||
const LARGEST_SUBNORMAL_BITS: u128 = 0x0000ffffffffffffffffffffffffffff;
|
||||
|
||||
/// Exponent = 0b1, zeroed significand
|
||||
const SMALLEST_NORMAL_BITS: u128 = 0x00010000000000000000000000000000;
|
||||
|
||||
/// First pattern over the mantissa
|
||||
const NAN_MASK1: u128 = 0x0000aaaaaaaaaaaaaaaaaaaaaaaaaaaa;
|
||||
|
||||
/// Second pattern over the mantissa
|
||||
const NAN_MASK2: u128 = 0x00005555555555555555555555555555;
|
||||
|
||||
/// Compare by value
|
||||
#[allow(unused_macros)]
|
||||
macro_rules! assert_f128_eq {
|
||||
($a:expr, $b:expr) => {
|
||||
let (l, r): (&f128, &f128) = (&$a, &$b);
|
||||
assert_eq!(*l, *r, "\na: {:#0130x}\nb: {:#0130x}", l.to_bits(), r.to_bits())
|
||||
};
|
||||
}
|
||||
|
||||
/// Compare by representation
|
||||
#[allow(unused_macros)]
|
||||
macro_rules! assert_f128_biteq {
|
||||
|
|
@ -31,10 +33,503 @@ macro_rules! assert_f128_biteq {
|
|||
let (l, r): (&f128, &f128) = (&$a, &$b);
|
||||
let lb = l.to_bits();
|
||||
let rb = r.to_bits();
|
||||
assert_eq!(
|
||||
lb, rb,
|
||||
"float {:?} is not bitequal to {:?}.\na: {:#0130x}\nb: {:#0130x}",
|
||||
*l, *r, lb, rb
|
||||
);
|
||||
assert_eq!(lb, rb, "float {l:?} is not bitequal to {r:?}.\na: {lb:#034x}\nb: {rb:#034x}");
|
||||
};
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_num_f128() {
|
||||
test_num(10f128, 2f128);
|
||||
}
|
||||
|
||||
// FIXME(f16_f128): add min and max tests when available
|
||||
|
||||
#[test]
|
||||
fn test_nan() {
|
||||
let nan: f128 = f128::NAN;
|
||||
assert!(nan.is_nan());
|
||||
assert!(!nan.is_infinite());
|
||||
assert!(!nan.is_finite());
|
||||
assert!(nan.is_sign_positive());
|
||||
assert!(!nan.is_sign_negative());
|
||||
// FIXME(f16_f128): classify
|
||||
// assert!(!nan.is_normal());
|
||||
// assert_eq!(Fp::Nan, nan.classify());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_infinity() {
|
||||
let inf: f128 = f128::INFINITY;
|
||||
assert!(inf.is_infinite());
|
||||
assert!(!inf.is_finite());
|
||||
assert!(inf.is_sign_positive());
|
||||
assert!(!inf.is_sign_negative());
|
||||
assert!(!inf.is_nan());
|
||||
// FIXME(f16_f128): classify
|
||||
// assert!(!inf.is_normal());
|
||||
// assert_eq!(Fp::Infinite, inf.classify());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_neg_infinity() {
|
||||
let neg_inf: f128 = f128::NEG_INFINITY;
|
||||
assert!(neg_inf.is_infinite());
|
||||
assert!(!neg_inf.is_finite());
|
||||
assert!(!neg_inf.is_sign_positive());
|
||||
assert!(neg_inf.is_sign_negative());
|
||||
assert!(!neg_inf.is_nan());
|
||||
// FIXME(f16_f128): classify
|
||||
// assert!(!neg_inf.is_normal());
|
||||
// assert_eq!(Fp::Infinite, neg_inf.classify());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_zero() {
|
||||
let zero: f128 = 0.0f128;
|
||||
assert_eq!(0.0, zero);
|
||||
assert!(!zero.is_infinite());
|
||||
assert!(zero.is_finite());
|
||||
assert!(zero.is_sign_positive());
|
||||
assert!(!zero.is_sign_negative());
|
||||
assert!(!zero.is_nan());
|
||||
// FIXME(f16_f128): classify
|
||||
// assert!(!zero.is_normal());
|
||||
// assert_eq!(Fp::Zero, zero.classify());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_neg_zero() {
|
||||
let neg_zero: f128 = -0.0;
|
||||
assert_eq!(0.0, neg_zero);
|
||||
assert!(!neg_zero.is_infinite());
|
||||
assert!(neg_zero.is_finite());
|
||||
assert!(!neg_zero.is_sign_positive());
|
||||
assert!(neg_zero.is_sign_negative());
|
||||
assert!(!neg_zero.is_nan());
|
||||
// FIXME(f16_f128): classify
|
||||
// assert!(!neg_zero.is_normal());
|
||||
// assert_eq!(Fp::Zero, neg_zero.classify());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_one() {
|
||||
let one: f128 = 1.0f128;
|
||||
assert_eq!(1.0, one);
|
||||
assert!(!one.is_infinite());
|
||||
assert!(one.is_finite());
|
||||
assert!(one.is_sign_positive());
|
||||
assert!(!one.is_sign_negative());
|
||||
assert!(!one.is_nan());
|
||||
// FIXME(f16_f128): classify
|
||||
// assert!(one.is_normal());
|
||||
// assert_eq!(Fp::Normal, one.classify());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_is_nan() {
|
||||
let nan: f128 = f128::NAN;
|
||||
let inf: f128 = f128::INFINITY;
|
||||
let neg_inf: f128 = f128::NEG_INFINITY;
|
||||
assert!(nan.is_nan());
|
||||
assert!(!0.0f128.is_nan());
|
||||
assert!(!5.3f128.is_nan());
|
||||
assert!(!(-10.732f128).is_nan());
|
||||
assert!(!inf.is_nan());
|
||||
assert!(!neg_inf.is_nan());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_is_infinite() {
|
||||
let nan: f128 = f128::NAN;
|
||||
let inf: f128 = f128::INFINITY;
|
||||
let neg_inf: f128 = f128::NEG_INFINITY;
|
||||
assert!(!nan.is_infinite());
|
||||
assert!(inf.is_infinite());
|
||||
assert!(neg_inf.is_infinite());
|
||||
assert!(!0.0f128.is_infinite());
|
||||
assert!(!42.8f128.is_infinite());
|
||||
assert!(!(-109.2f128).is_infinite());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_is_finite() {
|
||||
let nan: f128 = f128::NAN;
|
||||
let inf: f128 = f128::INFINITY;
|
||||
let neg_inf: f128 = f128::NEG_INFINITY;
|
||||
assert!(!nan.is_finite());
|
||||
assert!(!inf.is_finite());
|
||||
assert!(!neg_inf.is_finite());
|
||||
assert!(0.0f128.is_finite());
|
||||
assert!(42.8f128.is_finite());
|
||||
assert!((-109.2f128).is_finite());
|
||||
}
|
||||
|
||||
// FIXME(f16_f128): add `test_is_normal` and `test_classify` when classify is working
|
||||
// FIXME(f16_f128): add missing math functions when available
|
||||
|
||||
#[test]
|
||||
fn test_abs() {
|
||||
assert_eq!(f128::INFINITY.abs(), f128::INFINITY);
|
||||
assert_eq!(1f128.abs(), 1f128);
|
||||
assert_eq!(0f128.abs(), 0f128);
|
||||
assert_eq!((-0f128).abs(), 0f128);
|
||||
assert_eq!((-1f128).abs(), 1f128);
|
||||
assert_eq!(f128::NEG_INFINITY.abs(), f128::INFINITY);
|
||||
assert_eq!((1f128 / f128::NEG_INFINITY).abs(), 0f128);
|
||||
assert!(f128::NAN.abs().is_nan());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_is_sign_positive() {
|
||||
assert!(f128::INFINITY.is_sign_positive());
|
||||
assert!(1f128.is_sign_positive());
|
||||
assert!(0f128.is_sign_positive());
|
||||
assert!(!(-0f128).is_sign_positive());
|
||||
assert!(!(-1f128).is_sign_positive());
|
||||
assert!(!f128::NEG_INFINITY.is_sign_positive());
|
||||
assert!(!(1f128 / f128::NEG_INFINITY).is_sign_positive());
|
||||
assert!(f128::NAN.is_sign_positive());
|
||||
assert!(!(-f128::NAN).is_sign_positive());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_is_sign_negative() {
|
||||
assert!(!f128::INFINITY.is_sign_negative());
|
||||
assert!(!1f128.is_sign_negative());
|
||||
assert!(!0f128.is_sign_negative());
|
||||
assert!((-0f128).is_sign_negative());
|
||||
assert!((-1f128).is_sign_negative());
|
||||
assert!(f128::NEG_INFINITY.is_sign_negative());
|
||||
assert!((1f128 / f128::NEG_INFINITY).is_sign_negative());
|
||||
assert!(!f128::NAN.is_sign_negative());
|
||||
assert!((-f128::NAN).is_sign_negative());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_next_up() {
|
||||
let tiny = f128::from_bits(TINY_BITS);
|
||||
let tiny_up = f128::from_bits(TINY_UP_BITS);
|
||||
let max_down = f128::from_bits(MAX_DOWN_BITS);
|
||||
let largest_subnormal = f128::from_bits(LARGEST_SUBNORMAL_BITS);
|
||||
let smallest_normal = f128::from_bits(SMALLEST_NORMAL_BITS);
|
||||
assert_f128_biteq!(f128::NEG_INFINITY.next_up(), f128::MIN);
|
||||
assert_f128_biteq!(f128::MIN.next_up(), -max_down);
|
||||
assert_f128_biteq!((-1.0 - f128::EPSILON).next_up(), -1.0);
|
||||
assert_f128_biteq!((-smallest_normal).next_up(), -largest_subnormal);
|
||||
assert_f128_biteq!((-tiny_up).next_up(), -tiny);
|
||||
assert_f128_biteq!((-tiny).next_up(), -0.0f128);
|
||||
assert_f128_biteq!((-0.0f128).next_up(), tiny);
|
||||
assert_f128_biteq!(0.0f128.next_up(), tiny);
|
||||
assert_f128_biteq!(tiny.next_up(), tiny_up);
|
||||
assert_f128_biteq!(largest_subnormal.next_up(), smallest_normal);
|
||||
assert_f128_biteq!(1.0f128.next_up(), 1.0 + f128::EPSILON);
|
||||
assert_f128_biteq!(f128::MAX.next_up(), f128::INFINITY);
|
||||
assert_f128_biteq!(f128::INFINITY.next_up(), f128::INFINITY);
|
||||
|
||||
// Check that NaNs roundtrip.
|
||||
let nan0 = f128::NAN;
|
||||
let nan1 = f128::from_bits(f128::NAN.to_bits() ^ 0x002a_aaaa);
|
||||
let nan2 = f128::from_bits(f128::NAN.to_bits() ^ 0x0055_5555);
|
||||
assert_f128_biteq!(nan0.next_up(), nan0);
|
||||
assert_f128_biteq!(nan1.next_up(), nan1);
|
||||
assert_f128_biteq!(nan2.next_up(), nan2);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_next_down() {
|
||||
let tiny = f128::from_bits(TINY_BITS);
|
||||
let tiny_up = f128::from_bits(TINY_UP_BITS);
|
||||
let max_down = f128::from_bits(MAX_DOWN_BITS);
|
||||
let largest_subnormal = f128::from_bits(LARGEST_SUBNORMAL_BITS);
|
||||
let smallest_normal = f128::from_bits(SMALLEST_NORMAL_BITS);
|
||||
assert_f128_biteq!(f128::NEG_INFINITY.next_down(), f128::NEG_INFINITY);
|
||||
assert_f128_biteq!(f128::MIN.next_down(), f128::NEG_INFINITY);
|
||||
assert_f128_biteq!((-max_down).next_down(), f128::MIN);
|
||||
assert_f128_biteq!((-1.0f128).next_down(), -1.0 - f128::EPSILON);
|
||||
assert_f128_biteq!((-largest_subnormal).next_down(), -smallest_normal);
|
||||
assert_f128_biteq!((-tiny).next_down(), -tiny_up);
|
||||
assert_f128_biteq!((-0.0f128).next_down(), -tiny);
|
||||
assert_f128_biteq!((0.0f128).next_down(), -tiny);
|
||||
assert_f128_biteq!(tiny.next_down(), 0.0f128);
|
||||
assert_f128_biteq!(tiny_up.next_down(), tiny);
|
||||
assert_f128_biteq!(smallest_normal.next_down(), largest_subnormal);
|
||||
assert_f128_biteq!((1.0 + f128::EPSILON).next_down(), 1.0f128);
|
||||
assert_f128_biteq!(f128::MAX.next_down(), max_down);
|
||||
assert_f128_biteq!(f128::INFINITY.next_down(), f128::MAX);
|
||||
|
||||
// Check that NaNs roundtrip.
|
||||
let nan0 = f128::NAN;
|
||||
let nan1 = f128::from_bits(f128::NAN.to_bits() ^ 0x002a_aaaa);
|
||||
let nan2 = f128::from_bits(f128::NAN.to_bits() ^ 0x0055_5555);
|
||||
assert_f128_biteq!(nan0.next_down(), nan0);
|
||||
assert_f128_biteq!(nan1.next_down(), nan1);
|
||||
assert_f128_biteq!(nan2.next_down(), nan2);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_recip() {
|
||||
let nan: f128 = f128::NAN;
|
||||
let inf: f128 = f128::INFINITY;
|
||||
let neg_inf: f128 = f128::NEG_INFINITY;
|
||||
assert_eq!(1.0f128.recip(), 1.0);
|
||||
assert_eq!(2.0f128.recip(), 0.5);
|
||||
assert_eq!((-0.4f128).recip(), -2.5);
|
||||
assert_eq!(0.0f128.recip(), inf);
|
||||
assert!(nan.recip().is_nan());
|
||||
assert_eq!(inf.recip(), 0.0);
|
||||
assert_eq!(neg_inf.recip(), 0.0);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_to_degrees() {
|
||||
let pi: f128 = consts::PI;
|
||||
let nan: f128 = f128::NAN;
|
||||
let inf: f128 = f128::INFINITY;
|
||||
let neg_inf: f128 = f128::NEG_INFINITY;
|
||||
assert_eq!(0.0f128.to_degrees(), 0.0);
|
||||
assert_approx_eq!((-5.8f128).to_degrees(), -332.315521);
|
||||
assert_eq!(pi.to_degrees(), 180.0);
|
||||
assert!(nan.to_degrees().is_nan());
|
||||
assert_eq!(inf.to_degrees(), inf);
|
||||
assert_eq!(neg_inf.to_degrees(), neg_inf);
|
||||
assert_eq!(1_f128.to_degrees(), 57.2957795130823208767981548141051703);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_to_radians() {
|
||||
let pi: f128 = consts::PI;
|
||||
let nan: f128 = f128::NAN;
|
||||
let inf: f128 = f128::INFINITY;
|
||||
let neg_inf: f128 = f128::NEG_INFINITY;
|
||||
assert_eq!(0.0f128.to_radians(), 0.0);
|
||||
assert_approx_eq!(154.6f128.to_radians(), 2.698279);
|
||||
assert_approx_eq!((-332.31f128).to_radians(), -5.799903);
|
||||
// check approx rather than exact because round trip for pi doesn't fall on an exactly
|
||||
// representable value (unlike `f32` and `f64`).
|
||||
assert_approx_eq!(180.0f128.to_radians(), pi);
|
||||
assert!(nan.to_radians().is_nan());
|
||||
assert_eq!(inf.to_radians(), inf);
|
||||
assert_eq!(neg_inf.to_radians(), neg_inf);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_real_consts() {
|
||||
// FIXME(f16_f128): add math tests when available
|
||||
use super::consts;
|
||||
|
||||
let pi: f128 = consts::PI;
|
||||
let frac_pi_2: f128 = consts::FRAC_PI_2;
|
||||
let frac_pi_3: f128 = consts::FRAC_PI_3;
|
||||
let frac_pi_4: f128 = consts::FRAC_PI_4;
|
||||
let frac_pi_6: f128 = consts::FRAC_PI_6;
|
||||
let frac_pi_8: f128 = consts::FRAC_PI_8;
|
||||
let frac_1_pi: f128 = consts::FRAC_1_PI;
|
||||
let frac_2_pi: f128 = consts::FRAC_2_PI;
|
||||
// let frac_2_sqrtpi: f128 = consts::FRAC_2_SQRT_PI;
|
||||
// let sqrt2: f128 = consts::SQRT_2;
|
||||
// let frac_1_sqrt2: f128 = consts::FRAC_1_SQRT_2;
|
||||
// let e: f128 = consts::E;
|
||||
// let log2_e: f128 = consts::LOG2_E;
|
||||
// let log10_e: f128 = consts::LOG10_E;
|
||||
// let ln_2: f128 = consts::LN_2;
|
||||
// let ln_10: f128 = consts::LN_10;
|
||||
|
||||
assert_approx_eq!(frac_pi_2, pi / 2f128);
|
||||
assert_approx_eq!(frac_pi_3, pi / 3f128);
|
||||
assert_approx_eq!(frac_pi_4, pi / 4f128);
|
||||
assert_approx_eq!(frac_pi_6, pi / 6f128);
|
||||
assert_approx_eq!(frac_pi_8, pi / 8f128);
|
||||
assert_approx_eq!(frac_1_pi, 1f128 / pi);
|
||||
assert_approx_eq!(frac_2_pi, 2f128 / pi);
|
||||
// assert_approx_eq!(frac_2_sqrtpi, 2f128 / pi.sqrt());
|
||||
// assert_approx_eq!(sqrt2, 2f128.sqrt());
|
||||
// assert_approx_eq!(frac_1_sqrt2, 1f128 / 2f128.sqrt());
|
||||
// assert_approx_eq!(log2_e, e.log2());
|
||||
// assert_approx_eq!(log10_e, e.log10());
|
||||
// assert_approx_eq!(ln_2, 2f128.ln());
|
||||
// assert_approx_eq!(ln_10, 10f128.ln());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_float_bits_conv() {
|
||||
assert_eq!((1f128).to_bits(), 0x3fff0000000000000000000000000000);
|
||||
assert_eq!((12.5f128).to_bits(), 0x40029000000000000000000000000000);
|
||||
assert_eq!((1337f128).to_bits(), 0x40094e40000000000000000000000000);
|
||||
assert_eq!((-14.25f128).to_bits(), 0xc002c800000000000000000000000000);
|
||||
assert_approx_eq!(f128::from_bits(0x3fff0000000000000000000000000000), 1.0);
|
||||
assert_approx_eq!(f128::from_bits(0x40029000000000000000000000000000), 12.5);
|
||||
assert_approx_eq!(f128::from_bits(0x40094e40000000000000000000000000), 1337.0);
|
||||
assert_approx_eq!(f128::from_bits(0xc002c800000000000000000000000000), -14.25);
|
||||
|
||||
// Check that NaNs roundtrip their bits regardless of signaling-ness
|
||||
// 0xA is 0b1010; 0x5 is 0b0101 -- so these two together clobbers all the mantissa bits
|
||||
let masked_nan1 = f128::NAN.to_bits() ^ NAN_MASK1;
|
||||
let masked_nan2 = f128::NAN.to_bits() ^ NAN_MASK2;
|
||||
assert!(f128::from_bits(masked_nan1).is_nan());
|
||||
assert!(f128::from_bits(masked_nan2).is_nan());
|
||||
|
||||
assert_eq!(f128::from_bits(masked_nan1).to_bits(), masked_nan1);
|
||||
assert_eq!(f128::from_bits(masked_nan2).to_bits(), masked_nan2);
|
||||
}
|
||||
|
||||
#[test]
|
||||
#[should_panic]
|
||||
fn test_clamp_min_greater_than_max() {
|
||||
let _ = 1.0f128.clamp(3.0, 1.0);
|
||||
}
|
||||
|
||||
#[test]
|
||||
#[should_panic]
|
||||
fn test_clamp_min_is_nan() {
|
||||
let _ = 1.0f128.clamp(f128::NAN, 1.0);
|
||||
}
|
||||
|
||||
#[test]
|
||||
#[should_panic]
|
||||
fn test_clamp_max_is_nan() {
|
||||
let _ = 1.0f128.clamp(3.0, f128::NAN);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_total_cmp() {
|
||||
use core::cmp::Ordering;
|
||||
|
||||
fn quiet_bit_mask() -> u128 {
|
||||
1 << (f128::MANTISSA_DIGITS - 2)
|
||||
}
|
||||
|
||||
// FIXME(f16_f128): test subnormals when powf is available
|
||||
// fn min_subnorm() -> f128 {
|
||||
// f128::MIN_POSITIVE / f128::powf(2.0, f128::MANTISSA_DIGITS as f128 - 1.0)
|
||||
// }
|
||||
|
||||
// fn max_subnorm() -> f128 {
|
||||
// f128::MIN_POSITIVE - min_subnorm()
|
||||
// }
|
||||
|
||||
fn q_nan() -> f128 {
|
||||
f128::from_bits(f128::NAN.to_bits() | quiet_bit_mask())
|
||||
}
|
||||
|
||||
fn s_nan() -> f128 {
|
||||
f128::from_bits((f128::NAN.to_bits() & !quiet_bit_mask()) + 42)
|
||||
}
|
||||
|
||||
assert_eq!(Ordering::Equal, (-q_nan()).total_cmp(&-q_nan()));
|
||||
assert_eq!(Ordering::Equal, (-s_nan()).total_cmp(&-s_nan()));
|
||||
assert_eq!(Ordering::Equal, (-f128::INFINITY).total_cmp(&-f128::INFINITY));
|
||||
assert_eq!(Ordering::Equal, (-f128::MAX).total_cmp(&-f128::MAX));
|
||||
assert_eq!(Ordering::Equal, (-2.5_f128).total_cmp(&-2.5));
|
||||
assert_eq!(Ordering::Equal, (-1.0_f128).total_cmp(&-1.0));
|
||||
assert_eq!(Ordering::Equal, (-1.5_f128).total_cmp(&-1.5));
|
||||
assert_eq!(Ordering::Equal, (-0.5_f128).total_cmp(&-0.5));
|
||||
assert_eq!(Ordering::Equal, (-f128::MIN_POSITIVE).total_cmp(&-f128::MIN_POSITIVE));
|
||||
// assert_eq!(Ordering::Equal, (-max_subnorm()).total_cmp(&-max_subnorm()));
|
||||
// assert_eq!(Ordering::Equal, (-min_subnorm()).total_cmp(&-min_subnorm()));
|
||||
assert_eq!(Ordering::Equal, (-0.0_f128).total_cmp(&-0.0));
|
||||
assert_eq!(Ordering::Equal, 0.0_f128.total_cmp(&0.0));
|
||||
// assert_eq!(Ordering::Equal, min_subnorm().total_cmp(&min_subnorm()));
|
||||
// assert_eq!(Ordering::Equal, max_subnorm().total_cmp(&max_subnorm()));
|
||||
assert_eq!(Ordering::Equal, f128::MIN_POSITIVE.total_cmp(&f128::MIN_POSITIVE));
|
||||
assert_eq!(Ordering::Equal, 0.5_f128.total_cmp(&0.5));
|
||||
assert_eq!(Ordering::Equal, 1.0_f128.total_cmp(&1.0));
|
||||
assert_eq!(Ordering::Equal, 1.5_f128.total_cmp(&1.5));
|
||||
assert_eq!(Ordering::Equal, 2.5_f128.total_cmp(&2.5));
|
||||
assert_eq!(Ordering::Equal, f128::MAX.total_cmp(&f128::MAX));
|
||||
assert_eq!(Ordering::Equal, f128::INFINITY.total_cmp(&f128::INFINITY));
|
||||
assert_eq!(Ordering::Equal, s_nan().total_cmp(&s_nan()));
|
||||
assert_eq!(Ordering::Equal, q_nan().total_cmp(&q_nan()));
|
||||
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&-s_nan()));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&-f128::INFINITY));
|
||||
assert_eq!(Ordering::Less, (-f128::INFINITY).total_cmp(&-f128::MAX));
|
||||
assert_eq!(Ordering::Less, (-f128::MAX).total_cmp(&-2.5));
|
||||
assert_eq!(Ordering::Less, (-2.5_f128).total_cmp(&-1.5));
|
||||
assert_eq!(Ordering::Less, (-1.5_f128).total_cmp(&-1.0));
|
||||
assert_eq!(Ordering::Less, (-1.0_f128).total_cmp(&-0.5));
|
||||
assert_eq!(Ordering::Less, (-0.5_f128).total_cmp(&-f128::MIN_POSITIVE));
|
||||
// assert_eq!(Ordering::Less, (-f128::MIN_POSITIVE).total_cmp(&-max_subnorm()));
|
||||
// assert_eq!(Ordering::Less, (-max_subnorm()).total_cmp(&-min_subnorm()));
|
||||
// assert_eq!(Ordering::Less, (-min_subnorm()).total_cmp(&-0.0));
|
||||
assert_eq!(Ordering::Less, (-0.0_f128).total_cmp(&0.0));
|
||||
// assert_eq!(Ordering::Less, 0.0_f128.total_cmp(&min_subnorm()));
|
||||
// assert_eq!(Ordering::Less, min_subnorm().total_cmp(&max_subnorm()));
|
||||
// assert_eq!(Ordering::Less, max_subnorm().total_cmp(&f128::MIN_POSITIVE));
|
||||
assert_eq!(Ordering::Less, f128::MIN_POSITIVE.total_cmp(&0.5));
|
||||
assert_eq!(Ordering::Less, 0.5_f128.total_cmp(&1.0));
|
||||
assert_eq!(Ordering::Less, 1.0_f128.total_cmp(&1.5));
|
||||
assert_eq!(Ordering::Less, 1.5_f128.total_cmp(&2.5));
|
||||
assert_eq!(Ordering::Less, 2.5_f128.total_cmp(&f128::MAX));
|
||||
assert_eq!(Ordering::Less, f128::MAX.total_cmp(&f128::INFINITY));
|
||||
assert_eq!(Ordering::Less, f128::INFINITY.total_cmp(&s_nan()));
|
||||
assert_eq!(Ordering::Less, s_nan().total_cmp(&q_nan()));
|
||||
|
||||
assert_eq!(Ordering::Greater, (-s_nan()).total_cmp(&-q_nan()));
|
||||
assert_eq!(Ordering::Greater, (-f128::INFINITY).total_cmp(&-s_nan()));
|
||||
assert_eq!(Ordering::Greater, (-f128::MAX).total_cmp(&-f128::INFINITY));
|
||||
assert_eq!(Ordering::Greater, (-2.5_f128).total_cmp(&-f128::MAX));
|
||||
assert_eq!(Ordering::Greater, (-1.5_f128).total_cmp(&-2.5));
|
||||
assert_eq!(Ordering::Greater, (-1.0_f128).total_cmp(&-1.5));
|
||||
assert_eq!(Ordering::Greater, (-0.5_f128).total_cmp(&-1.0));
|
||||
assert_eq!(Ordering::Greater, (-f128::MIN_POSITIVE).total_cmp(&-0.5));
|
||||
// assert_eq!(Ordering::Greater, (-max_subnorm()).total_cmp(&-f128::MIN_POSITIVE));
|
||||
// assert_eq!(Ordering::Greater, (-min_subnorm()).total_cmp(&-max_subnorm()));
|
||||
// assert_eq!(Ordering::Greater, (-0.0_f128).total_cmp(&-min_subnorm()));
|
||||
assert_eq!(Ordering::Greater, 0.0_f128.total_cmp(&-0.0));
|
||||
// assert_eq!(Ordering::Greater, min_subnorm().total_cmp(&0.0));
|
||||
// assert_eq!(Ordering::Greater, max_subnorm().total_cmp(&min_subnorm()));
|
||||
// assert_eq!(Ordering::Greater, f128::MIN_POSITIVE.total_cmp(&max_subnorm()));
|
||||
assert_eq!(Ordering::Greater, 0.5_f128.total_cmp(&f128::MIN_POSITIVE));
|
||||
assert_eq!(Ordering::Greater, 1.0_f128.total_cmp(&0.5));
|
||||
assert_eq!(Ordering::Greater, 1.5_f128.total_cmp(&1.0));
|
||||
assert_eq!(Ordering::Greater, 2.5_f128.total_cmp(&1.5));
|
||||
assert_eq!(Ordering::Greater, f128::MAX.total_cmp(&2.5));
|
||||
assert_eq!(Ordering::Greater, f128::INFINITY.total_cmp(&f128::MAX));
|
||||
assert_eq!(Ordering::Greater, s_nan().total_cmp(&f128::INFINITY));
|
||||
assert_eq!(Ordering::Greater, q_nan().total_cmp(&s_nan()));
|
||||
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&-s_nan()));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&-f128::INFINITY));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&-f128::MAX));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&-2.5));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&-1.5));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&-1.0));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&-0.5));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&-f128::MIN_POSITIVE));
|
||||
// assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&-max_subnorm()));
|
||||
// assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&-min_subnorm()));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&-0.0));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&0.0));
|
||||
// assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&min_subnorm()));
|
||||
// assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&max_subnorm()));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&f128::MIN_POSITIVE));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&0.5));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&1.0));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&1.5));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&2.5));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&f128::MAX));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&f128::INFINITY));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&s_nan()));
|
||||
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&-f128::INFINITY));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&-f128::MAX));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&-2.5));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&-1.5));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&-1.0));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&-0.5));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&-f128::MIN_POSITIVE));
|
||||
// assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&-max_subnorm()));
|
||||
// assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&-min_subnorm()));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&-0.0));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&0.0));
|
||||
// assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&min_subnorm()));
|
||||
// assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&max_subnorm()));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&f128::MIN_POSITIVE));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&0.5));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&1.0));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&1.5));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&2.5));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&f128::MAX));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&f128::INFINITY));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&s_nan()));
|
||||
}
|
||||
|
|
|
|||
|
|
@ -32,4 +32,33 @@ impl f16 {
|
|||
pub fn powi(self, n: i32) -> f16 {
|
||||
unsafe { intrinsics::powif16(self, n) }
|
||||
}
|
||||
|
||||
/// Computes the absolute value of `self`.
|
||||
///
|
||||
/// This function always returns the precise result.
|
||||
///
|
||||
/// # Examples
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(f16)]
|
||||
/// # #[cfg(reliable_f16)] {
|
||||
///
|
||||
/// let x = 3.5_f16;
|
||||
/// let y = -3.5_f16;
|
||||
///
|
||||
/// assert_eq!(x.abs(), x);
|
||||
/// assert_eq!(y.abs(), -y);
|
||||
///
|
||||
/// assert!(f16::NAN.abs().is_nan());
|
||||
/// # }
|
||||
/// ```
|
||||
#[inline]
|
||||
#[cfg(not(bootstrap))]
|
||||
#[rustc_allow_incoherent_impl]
|
||||
#[unstable(feature = "f16", issue = "116909")]
|
||||
#[must_use = "method returns a new number and does not mutate the original value"]
|
||||
pub fn abs(self) -> Self {
|
||||
// FIXME(f16_f128): replace with `intrinsics::fabsf16` when available
|
||||
Self::from_bits(self.to_bits() & !(1 << 15))
|
||||
}
|
||||
}
|
||||
|
|
|
|||
|
|
@ -1,35 +1,37 @@
|
|||
#![allow(dead_code)] // FIXME(f16_f128): remove once constants are used
|
||||
#![cfg(not(bootstrap))]
|
||||
// FIXME(f16_f128): only tested on platforms that have symbols and aren't buggy
|
||||
#![cfg(reliable_f16)]
|
||||
|
||||
use crate::f16::consts;
|
||||
use crate::num::*;
|
||||
|
||||
// We run out of precision pretty quickly with f16
|
||||
const F16_APPROX_L1: f16 = 0.001;
|
||||
// const F16_APPROX_L1: f16 = 0.001;
|
||||
const F16_APPROX_L2: f16 = 0.01;
|
||||
const F16_APPROX_L3: f16 = 0.1;
|
||||
// const F16_APPROX_L3: f16 = 0.1;
|
||||
const F16_APPROX_L4: f16 = 0.5;
|
||||
|
||||
/// Smallest number
|
||||
const TINY_BITS: u16 = 0x1;
|
||||
|
||||
/// Next smallest number
|
||||
const TINY_UP_BITS: u16 = 0x2;
|
||||
|
||||
/// Exponent = 0b11...10, Sifnificand 0b1111..10. Min val > 0
|
||||
const MAX_DOWN_BITS: u16 = 0x7bfe;
|
||||
|
||||
/// Zeroed exponent, full significant
|
||||
const LARGEST_SUBNORMAL_BITS: u16 = 0x03ff;
|
||||
|
||||
/// Exponent = 0b1, zeroed significand
|
||||
const SMALLEST_NORMAL_BITS: u16 = 0x0400;
|
||||
|
||||
/// First pattern over the mantissa
|
||||
const NAN_MASK1: u16 = 0x02aa;
|
||||
|
||||
/// Second pattern over the mantissa
|
||||
const NAN_MASK2: u16 = 0x0155;
|
||||
|
||||
/// Compare by value
|
||||
#[allow(unused_macros)]
|
||||
macro_rules! assert_f16_eq {
|
||||
($a:expr, $b:expr) => {
|
||||
let (l, r): (&f16, &f16) = (&$a, &$b);
|
||||
assert_eq!(*l, *r, "\na: {:#018x}\nb: {:#018x}", l.to_bits(), r.to_bits())
|
||||
};
|
||||
}
|
||||
|
||||
/// Compare by representation
|
||||
#[allow(unused_macros)]
|
||||
macro_rules! assert_f16_biteq {
|
||||
|
|
@ -37,10 +39,500 @@ macro_rules! assert_f16_biteq {
|
|||
let (l, r): (&f16, &f16) = (&$a, &$b);
|
||||
let lb = l.to_bits();
|
||||
let rb = r.to_bits();
|
||||
assert_eq!(
|
||||
lb, rb,
|
||||
"float {:?} is not bitequal to {:?}.\na: {:#018x}\nb: {:#018x}",
|
||||
*l, *r, lb, rb
|
||||
);
|
||||
assert_eq!(lb, rb, "float {l:?} ({lb:#04x}) is not bitequal to {r:?} ({rb:#04x})");
|
||||
};
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_num_f16() {
|
||||
test_num(10f16, 2f16);
|
||||
}
|
||||
|
||||
// FIXME(f16_f128): add min and max tests when available
|
||||
|
||||
#[test]
|
||||
fn test_nan() {
|
||||
let nan: f16 = f16::NAN;
|
||||
assert!(nan.is_nan());
|
||||
assert!(!nan.is_infinite());
|
||||
assert!(!nan.is_finite());
|
||||
assert!(nan.is_sign_positive());
|
||||
assert!(!nan.is_sign_negative());
|
||||
// FIXME(f16_f128): classify
|
||||
// assert!(!nan.is_normal());
|
||||
// assert_eq!(Fp::Nan, nan.classify());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_infinity() {
|
||||
let inf: f16 = f16::INFINITY;
|
||||
assert!(inf.is_infinite());
|
||||
assert!(!inf.is_finite());
|
||||
assert!(inf.is_sign_positive());
|
||||
assert!(!inf.is_sign_negative());
|
||||
assert!(!inf.is_nan());
|
||||
// FIXME(f16_f128): classify
|
||||
// assert!(!inf.is_normal());
|
||||
// assert_eq!(Fp::Infinite, inf.classify());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_neg_infinity() {
|
||||
let neg_inf: f16 = f16::NEG_INFINITY;
|
||||
assert!(neg_inf.is_infinite());
|
||||
assert!(!neg_inf.is_finite());
|
||||
assert!(!neg_inf.is_sign_positive());
|
||||
assert!(neg_inf.is_sign_negative());
|
||||
assert!(!neg_inf.is_nan());
|
||||
// FIXME(f16_f128): classify
|
||||
// assert!(!neg_inf.is_normal());
|
||||
// assert_eq!(Fp::Infinite, neg_inf.classify());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_zero() {
|
||||
let zero: f16 = 0.0f16;
|
||||
assert_eq!(0.0, zero);
|
||||
assert!(!zero.is_infinite());
|
||||
assert!(zero.is_finite());
|
||||
assert!(zero.is_sign_positive());
|
||||
assert!(!zero.is_sign_negative());
|
||||
assert!(!zero.is_nan());
|
||||
// FIXME(f16_f128): classify
|
||||
// assert!(!zero.is_normal());
|
||||
// assert_eq!(Fp::Zero, zero.classify());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_neg_zero() {
|
||||
let neg_zero: f16 = -0.0;
|
||||
assert_eq!(0.0, neg_zero);
|
||||
assert!(!neg_zero.is_infinite());
|
||||
assert!(neg_zero.is_finite());
|
||||
assert!(!neg_zero.is_sign_positive());
|
||||
assert!(neg_zero.is_sign_negative());
|
||||
assert!(!neg_zero.is_nan());
|
||||
// FIXME(f16_f128): classify
|
||||
// assert!(!neg_zero.is_normal());
|
||||
// assert_eq!(Fp::Zero, neg_zero.classify());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_one() {
|
||||
let one: f16 = 1.0f16;
|
||||
assert_eq!(1.0, one);
|
||||
assert!(!one.is_infinite());
|
||||
assert!(one.is_finite());
|
||||
assert!(one.is_sign_positive());
|
||||
assert!(!one.is_sign_negative());
|
||||
assert!(!one.is_nan());
|
||||
// FIXME(f16_f128): classify
|
||||
// assert!(one.is_normal());
|
||||
// assert_eq!(Fp::Normal, one.classify());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_is_nan() {
|
||||
let nan: f16 = f16::NAN;
|
||||
let inf: f16 = f16::INFINITY;
|
||||
let neg_inf: f16 = f16::NEG_INFINITY;
|
||||
assert!(nan.is_nan());
|
||||
assert!(!0.0f16.is_nan());
|
||||
assert!(!5.3f16.is_nan());
|
||||
assert!(!(-10.732f16).is_nan());
|
||||
assert!(!inf.is_nan());
|
||||
assert!(!neg_inf.is_nan());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_is_infinite() {
|
||||
let nan: f16 = f16::NAN;
|
||||
let inf: f16 = f16::INFINITY;
|
||||
let neg_inf: f16 = f16::NEG_INFINITY;
|
||||
assert!(!nan.is_infinite());
|
||||
assert!(inf.is_infinite());
|
||||
assert!(neg_inf.is_infinite());
|
||||
assert!(!0.0f16.is_infinite());
|
||||
assert!(!42.8f16.is_infinite());
|
||||
assert!(!(-109.2f16).is_infinite());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_is_finite() {
|
||||
let nan: f16 = f16::NAN;
|
||||
let inf: f16 = f16::INFINITY;
|
||||
let neg_inf: f16 = f16::NEG_INFINITY;
|
||||
assert!(!nan.is_finite());
|
||||
assert!(!inf.is_finite());
|
||||
assert!(!neg_inf.is_finite());
|
||||
assert!(0.0f16.is_finite());
|
||||
assert!(42.8f16.is_finite());
|
||||
assert!((-109.2f16).is_finite());
|
||||
}
|
||||
|
||||
// FIXME(f16_f128): add `test_is_normal` and `test_classify` when classify is working
|
||||
// FIXME(f16_f128): add missing math functions when available
|
||||
|
||||
#[test]
|
||||
fn test_abs() {
|
||||
assert_eq!(f16::INFINITY.abs(), f16::INFINITY);
|
||||
assert_eq!(1f16.abs(), 1f16);
|
||||
assert_eq!(0f16.abs(), 0f16);
|
||||
assert_eq!((-0f16).abs(), 0f16);
|
||||
assert_eq!((-1f16).abs(), 1f16);
|
||||
assert_eq!(f16::NEG_INFINITY.abs(), f16::INFINITY);
|
||||
assert_eq!((1f16 / f16::NEG_INFINITY).abs(), 0f16);
|
||||
assert!(f16::NAN.abs().is_nan());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_is_sign_positive() {
|
||||
assert!(f16::INFINITY.is_sign_positive());
|
||||
assert!(1f16.is_sign_positive());
|
||||
assert!(0f16.is_sign_positive());
|
||||
assert!(!(-0f16).is_sign_positive());
|
||||
assert!(!(-1f16).is_sign_positive());
|
||||
assert!(!f16::NEG_INFINITY.is_sign_positive());
|
||||
assert!(!(1f16 / f16::NEG_INFINITY).is_sign_positive());
|
||||
assert!(f16::NAN.is_sign_positive());
|
||||
assert!(!(-f16::NAN).is_sign_positive());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_is_sign_negative() {
|
||||
assert!(!f16::INFINITY.is_sign_negative());
|
||||
assert!(!1f16.is_sign_negative());
|
||||
assert!(!0f16.is_sign_negative());
|
||||
assert!((-0f16).is_sign_negative());
|
||||
assert!((-1f16).is_sign_negative());
|
||||
assert!(f16::NEG_INFINITY.is_sign_negative());
|
||||
assert!((1f16 / f16::NEG_INFINITY).is_sign_negative());
|
||||
assert!(!f16::NAN.is_sign_negative());
|
||||
assert!((-f16::NAN).is_sign_negative());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_next_up() {
|
||||
let tiny = f16::from_bits(TINY_BITS);
|
||||
let tiny_up = f16::from_bits(TINY_UP_BITS);
|
||||
let max_down = f16::from_bits(MAX_DOWN_BITS);
|
||||
let largest_subnormal = f16::from_bits(LARGEST_SUBNORMAL_BITS);
|
||||
let smallest_normal = f16::from_bits(SMALLEST_NORMAL_BITS);
|
||||
assert_f16_biteq!(f16::NEG_INFINITY.next_up(), f16::MIN);
|
||||
assert_f16_biteq!(f16::MIN.next_up(), -max_down);
|
||||
assert_f16_biteq!((-1.0 - f16::EPSILON).next_up(), -1.0);
|
||||
assert_f16_biteq!((-smallest_normal).next_up(), -largest_subnormal);
|
||||
assert_f16_biteq!((-tiny_up).next_up(), -tiny);
|
||||
assert_f16_biteq!((-tiny).next_up(), -0.0f16);
|
||||
assert_f16_biteq!((-0.0f16).next_up(), tiny);
|
||||
assert_f16_biteq!(0.0f16.next_up(), tiny);
|
||||
assert_f16_biteq!(tiny.next_up(), tiny_up);
|
||||
assert_f16_biteq!(largest_subnormal.next_up(), smallest_normal);
|
||||
assert_f16_biteq!(1.0f16.next_up(), 1.0 + f16::EPSILON);
|
||||
assert_f16_biteq!(f16::MAX.next_up(), f16::INFINITY);
|
||||
assert_f16_biteq!(f16::INFINITY.next_up(), f16::INFINITY);
|
||||
|
||||
// Check that NaNs roundtrip.
|
||||
let nan0 = f16::NAN;
|
||||
let nan1 = f16::from_bits(f16::NAN.to_bits() ^ NAN_MASK1);
|
||||
let nan2 = f16::from_bits(f16::NAN.to_bits() ^ NAN_MASK2);
|
||||
assert_f16_biteq!(nan0.next_up(), nan0);
|
||||
assert_f16_biteq!(nan1.next_up(), nan1);
|
||||
assert_f16_biteq!(nan2.next_up(), nan2);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_next_down() {
|
||||
let tiny = f16::from_bits(TINY_BITS);
|
||||
let tiny_up = f16::from_bits(TINY_UP_BITS);
|
||||
let max_down = f16::from_bits(MAX_DOWN_BITS);
|
||||
let largest_subnormal = f16::from_bits(LARGEST_SUBNORMAL_BITS);
|
||||
let smallest_normal = f16::from_bits(SMALLEST_NORMAL_BITS);
|
||||
assert_f16_biteq!(f16::NEG_INFINITY.next_down(), f16::NEG_INFINITY);
|
||||
assert_f16_biteq!(f16::MIN.next_down(), f16::NEG_INFINITY);
|
||||
assert_f16_biteq!((-max_down).next_down(), f16::MIN);
|
||||
assert_f16_biteq!((-1.0f16).next_down(), -1.0 - f16::EPSILON);
|
||||
assert_f16_biteq!((-largest_subnormal).next_down(), -smallest_normal);
|
||||
assert_f16_biteq!((-tiny).next_down(), -tiny_up);
|
||||
assert_f16_biteq!((-0.0f16).next_down(), -tiny);
|
||||
assert_f16_biteq!((0.0f16).next_down(), -tiny);
|
||||
assert_f16_biteq!(tiny.next_down(), 0.0f16);
|
||||
assert_f16_biteq!(tiny_up.next_down(), tiny);
|
||||
assert_f16_biteq!(smallest_normal.next_down(), largest_subnormal);
|
||||
assert_f16_biteq!((1.0 + f16::EPSILON).next_down(), 1.0f16);
|
||||
assert_f16_biteq!(f16::MAX.next_down(), max_down);
|
||||
assert_f16_biteq!(f16::INFINITY.next_down(), f16::MAX);
|
||||
|
||||
// Check that NaNs roundtrip.
|
||||
let nan0 = f16::NAN;
|
||||
let nan1 = f16::from_bits(f16::NAN.to_bits() ^ NAN_MASK1);
|
||||
let nan2 = f16::from_bits(f16::NAN.to_bits() ^ NAN_MASK2);
|
||||
assert_f16_biteq!(nan0.next_down(), nan0);
|
||||
assert_f16_biteq!(nan1.next_down(), nan1);
|
||||
assert_f16_biteq!(nan2.next_down(), nan2);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_recip() {
|
||||
let nan: f16 = f16::NAN;
|
||||
let inf: f16 = f16::INFINITY;
|
||||
let neg_inf: f16 = f16::NEG_INFINITY;
|
||||
assert_eq!(1.0f16.recip(), 1.0);
|
||||
assert_eq!(2.0f16.recip(), 0.5);
|
||||
assert_eq!((-0.4f16).recip(), -2.5);
|
||||
assert_eq!(0.0f16.recip(), inf);
|
||||
assert!(nan.recip().is_nan());
|
||||
assert_eq!(inf.recip(), 0.0);
|
||||
assert_eq!(neg_inf.recip(), 0.0);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_to_degrees() {
|
||||
let pi: f16 = consts::PI;
|
||||
let nan: f16 = f16::NAN;
|
||||
let inf: f16 = f16::INFINITY;
|
||||
let neg_inf: f16 = f16::NEG_INFINITY;
|
||||
assert_eq!(0.0f16.to_degrees(), 0.0);
|
||||
assert_approx_eq!((-5.8f16).to_degrees(), -332.315521);
|
||||
assert_approx_eq!(pi.to_degrees(), 180.0, F16_APPROX_L4);
|
||||
assert!(nan.to_degrees().is_nan());
|
||||
assert_eq!(inf.to_degrees(), inf);
|
||||
assert_eq!(neg_inf.to_degrees(), neg_inf);
|
||||
assert_eq!(1_f16.to_degrees(), 57.2957795130823208767981548141051703);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_to_radians() {
|
||||
let pi: f16 = consts::PI;
|
||||
let nan: f16 = f16::NAN;
|
||||
let inf: f16 = f16::INFINITY;
|
||||
let neg_inf: f16 = f16::NEG_INFINITY;
|
||||
assert_eq!(0.0f16.to_radians(), 0.0);
|
||||
assert_approx_eq!(154.6f16.to_radians(), 2.698279);
|
||||
assert_approx_eq!((-332.31f16).to_radians(), -5.799903);
|
||||
assert_approx_eq!(180.0f16.to_radians(), pi, F16_APPROX_L2);
|
||||
assert!(nan.to_radians().is_nan());
|
||||
assert_eq!(inf.to_radians(), inf);
|
||||
assert_eq!(neg_inf.to_radians(), neg_inf);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_real_consts() {
|
||||
// FIXME(f16_f128): add math tests when available
|
||||
use super::consts;
|
||||
|
||||
let pi: f16 = consts::PI;
|
||||
let frac_pi_2: f16 = consts::FRAC_PI_2;
|
||||
let frac_pi_3: f16 = consts::FRAC_PI_3;
|
||||
let frac_pi_4: f16 = consts::FRAC_PI_4;
|
||||
let frac_pi_6: f16 = consts::FRAC_PI_6;
|
||||
let frac_pi_8: f16 = consts::FRAC_PI_8;
|
||||
let frac_1_pi: f16 = consts::FRAC_1_PI;
|
||||
let frac_2_pi: f16 = consts::FRAC_2_PI;
|
||||
// let frac_2_sqrtpi: f16 = consts::FRAC_2_SQRT_PI;
|
||||
// let sqrt2: f16 = consts::SQRT_2;
|
||||
// let frac_1_sqrt2: f16 = consts::FRAC_1_SQRT_2;
|
||||
// let e: f16 = consts::E;
|
||||
// let log2_e: f16 = consts::LOG2_E;
|
||||
// let log10_e: f16 = consts::LOG10_E;
|
||||
// let ln_2: f16 = consts::LN_2;
|
||||
// let ln_10: f16 = consts::LN_10;
|
||||
|
||||
assert_approx_eq!(frac_pi_2, pi / 2f16);
|
||||
assert_approx_eq!(frac_pi_3, pi / 3f16);
|
||||
assert_approx_eq!(frac_pi_4, pi / 4f16);
|
||||
assert_approx_eq!(frac_pi_6, pi / 6f16);
|
||||
assert_approx_eq!(frac_pi_8, pi / 8f16);
|
||||
assert_approx_eq!(frac_1_pi, 1f16 / pi);
|
||||
assert_approx_eq!(frac_2_pi, 2f16 / pi);
|
||||
// assert_approx_eq!(frac_2_sqrtpi, 2f16 / pi.sqrt());
|
||||
// assert_approx_eq!(sqrt2, 2f16.sqrt());
|
||||
// assert_approx_eq!(frac_1_sqrt2, 1f16 / 2f16.sqrt());
|
||||
// assert_approx_eq!(log2_e, e.log2());
|
||||
// assert_approx_eq!(log10_e, e.log10());
|
||||
// assert_approx_eq!(ln_2, 2f16.ln());
|
||||
// assert_approx_eq!(ln_10, 10f16.ln());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_float_bits_conv() {
|
||||
assert_eq!((1f16).to_bits(), 0x3c00);
|
||||
assert_eq!((12.5f16).to_bits(), 0x4a40);
|
||||
assert_eq!((1337f16).to_bits(), 0x6539);
|
||||
assert_eq!((-14.25f16).to_bits(), 0xcb20);
|
||||
assert_approx_eq!(f16::from_bits(0x3c00), 1.0);
|
||||
assert_approx_eq!(f16::from_bits(0x4a40), 12.5);
|
||||
assert_approx_eq!(f16::from_bits(0x6539), 1337.0);
|
||||
assert_approx_eq!(f16::from_bits(0xcb20), -14.25);
|
||||
|
||||
// Check that NaNs roundtrip their bits regardless of signaling-ness
|
||||
let masked_nan1 = f16::NAN.to_bits() ^ NAN_MASK1;
|
||||
let masked_nan2 = f16::NAN.to_bits() ^ NAN_MASK2;
|
||||
assert!(f16::from_bits(masked_nan1).is_nan());
|
||||
assert!(f16::from_bits(masked_nan2).is_nan());
|
||||
|
||||
assert_eq!(f16::from_bits(masked_nan1).to_bits(), masked_nan1);
|
||||
assert_eq!(f16::from_bits(masked_nan2).to_bits(), masked_nan2);
|
||||
}
|
||||
|
||||
#[test]
|
||||
#[should_panic]
|
||||
fn test_clamp_min_greater_than_max() {
|
||||
let _ = 1.0f16.clamp(3.0, 1.0);
|
||||
}
|
||||
|
||||
#[test]
|
||||
#[should_panic]
|
||||
fn test_clamp_min_is_nan() {
|
||||
let _ = 1.0f16.clamp(f16::NAN, 1.0);
|
||||
}
|
||||
|
||||
#[test]
|
||||
#[should_panic]
|
||||
fn test_clamp_max_is_nan() {
|
||||
let _ = 1.0f16.clamp(3.0, f16::NAN);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_total_cmp() {
|
||||
use core::cmp::Ordering;
|
||||
|
||||
fn quiet_bit_mask() -> u16 {
|
||||
1 << (f16::MANTISSA_DIGITS - 2)
|
||||
}
|
||||
|
||||
// FIXME(f16_f128): test subnormals when powf is available
|
||||
// fn min_subnorm() -> f16 {
|
||||
// f16::MIN_POSITIVE / f16::powf(2.0, f16::MANTISSA_DIGITS as f16 - 1.0)
|
||||
// }
|
||||
|
||||
// fn max_subnorm() -> f16 {
|
||||
// f16::MIN_POSITIVE - min_subnorm()
|
||||
// }
|
||||
|
||||
fn q_nan() -> f16 {
|
||||
f16::from_bits(f16::NAN.to_bits() | quiet_bit_mask())
|
||||
}
|
||||
|
||||
fn s_nan() -> f16 {
|
||||
f16::from_bits((f16::NAN.to_bits() & !quiet_bit_mask()) + 42)
|
||||
}
|
||||
|
||||
assert_eq!(Ordering::Equal, (-q_nan()).total_cmp(&-q_nan()));
|
||||
assert_eq!(Ordering::Equal, (-s_nan()).total_cmp(&-s_nan()));
|
||||
assert_eq!(Ordering::Equal, (-f16::INFINITY).total_cmp(&-f16::INFINITY));
|
||||
assert_eq!(Ordering::Equal, (-f16::MAX).total_cmp(&-f16::MAX));
|
||||
assert_eq!(Ordering::Equal, (-2.5_f16).total_cmp(&-2.5));
|
||||
assert_eq!(Ordering::Equal, (-1.0_f16).total_cmp(&-1.0));
|
||||
assert_eq!(Ordering::Equal, (-1.5_f16).total_cmp(&-1.5));
|
||||
assert_eq!(Ordering::Equal, (-0.5_f16).total_cmp(&-0.5));
|
||||
assert_eq!(Ordering::Equal, (-f16::MIN_POSITIVE).total_cmp(&-f16::MIN_POSITIVE));
|
||||
// assert_eq!(Ordering::Equal, (-max_subnorm()).total_cmp(&-max_subnorm()));
|
||||
// assert_eq!(Ordering::Equal, (-min_subnorm()).total_cmp(&-min_subnorm()));
|
||||
assert_eq!(Ordering::Equal, (-0.0_f16).total_cmp(&-0.0));
|
||||
assert_eq!(Ordering::Equal, 0.0_f16.total_cmp(&0.0));
|
||||
// assert_eq!(Ordering::Equal, min_subnorm().total_cmp(&min_subnorm()));
|
||||
// assert_eq!(Ordering::Equal, max_subnorm().total_cmp(&max_subnorm()));
|
||||
assert_eq!(Ordering::Equal, f16::MIN_POSITIVE.total_cmp(&f16::MIN_POSITIVE));
|
||||
assert_eq!(Ordering::Equal, 0.5_f16.total_cmp(&0.5));
|
||||
assert_eq!(Ordering::Equal, 1.0_f16.total_cmp(&1.0));
|
||||
assert_eq!(Ordering::Equal, 1.5_f16.total_cmp(&1.5));
|
||||
assert_eq!(Ordering::Equal, 2.5_f16.total_cmp(&2.5));
|
||||
assert_eq!(Ordering::Equal, f16::MAX.total_cmp(&f16::MAX));
|
||||
assert_eq!(Ordering::Equal, f16::INFINITY.total_cmp(&f16::INFINITY));
|
||||
assert_eq!(Ordering::Equal, s_nan().total_cmp(&s_nan()));
|
||||
assert_eq!(Ordering::Equal, q_nan().total_cmp(&q_nan()));
|
||||
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&-s_nan()));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&-f16::INFINITY));
|
||||
assert_eq!(Ordering::Less, (-f16::INFINITY).total_cmp(&-f16::MAX));
|
||||
assert_eq!(Ordering::Less, (-f16::MAX).total_cmp(&-2.5));
|
||||
assert_eq!(Ordering::Less, (-2.5_f16).total_cmp(&-1.5));
|
||||
assert_eq!(Ordering::Less, (-1.5_f16).total_cmp(&-1.0));
|
||||
assert_eq!(Ordering::Less, (-1.0_f16).total_cmp(&-0.5));
|
||||
assert_eq!(Ordering::Less, (-0.5_f16).total_cmp(&-f16::MIN_POSITIVE));
|
||||
// assert_eq!(Ordering::Less, (-f16::MIN_POSITIVE).total_cmp(&-max_subnorm()));
|
||||
// assert_eq!(Ordering::Less, (-max_subnorm()).total_cmp(&-min_subnorm()));
|
||||
// assert_eq!(Ordering::Less, (-min_subnorm()).total_cmp(&-0.0));
|
||||
assert_eq!(Ordering::Less, (-0.0_f16).total_cmp(&0.0));
|
||||
// assert_eq!(Ordering::Less, 0.0_f16.total_cmp(&min_subnorm()));
|
||||
// assert_eq!(Ordering::Less, min_subnorm().total_cmp(&max_subnorm()));
|
||||
// assert_eq!(Ordering::Less, max_subnorm().total_cmp(&f16::MIN_POSITIVE));
|
||||
assert_eq!(Ordering::Less, f16::MIN_POSITIVE.total_cmp(&0.5));
|
||||
assert_eq!(Ordering::Less, 0.5_f16.total_cmp(&1.0));
|
||||
assert_eq!(Ordering::Less, 1.0_f16.total_cmp(&1.5));
|
||||
assert_eq!(Ordering::Less, 1.5_f16.total_cmp(&2.5));
|
||||
assert_eq!(Ordering::Less, 2.5_f16.total_cmp(&f16::MAX));
|
||||
assert_eq!(Ordering::Less, f16::MAX.total_cmp(&f16::INFINITY));
|
||||
assert_eq!(Ordering::Less, f16::INFINITY.total_cmp(&s_nan()));
|
||||
assert_eq!(Ordering::Less, s_nan().total_cmp(&q_nan()));
|
||||
|
||||
assert_eq!(Ordering::Greater, (-s_nan()).total_cmp(&-q_nan()));
|
||||
assert_eq!(Ordering::Greater, (-f16::INFINITY).total_cmp(&-s_nan()));
|
||||
assert_eq!(Ordering::Greater, (-f16::MAX).total_cmp(&-f16::INFINITY));
|
||||
assert_eq!(Ordering::Greater, (-2.5_f16).total_cmp(&-f16::MAX));
|
||||
assert_eq!(Ordering::Greater, (-1.5_f16).total_cmp(&-2.5));
|
||||
assert_eq!(Ordering::Greater, (-1.0_f16).total_cmp(&-1.5));
|
||||
assert_eq!(Ordering::Greater, (-0.5_f16).total_cmp(&-1.0));
|
||||
assert_eq!(Ordering::Greater, (-f16::MIN_POSITIVE).total_cmp(&-0.5));
|
||||
// assert_eq!(Ordering::Greater, (-max_subnorm()).total_cmp(&-f16::MIN_POSITIVE));
|
||||
// assert_eq!(Ordering::Greater, (-min_subnorm()).total_cmp(&-max_subnorm()));
|
||||
// assert_eq!(Ordering::Greater, (-0.0_f16).total_cmp(&-min_subnorm()));
|
||||
assert_eq!(Ordering::Greater, 0.0_f16.total_cmp(&-0.0));
|
||||
// assert_eq!(Ordering::Greater, min_subnorm().total_cmp(&0.0));
|
||||
// assert_eq!(Ordering::Greater, max_subnorm().total_cmp(&min_subnorm()));
|
||||
// assert_eq!(Ordering::Greater, f16::MIN_POSITIVE.total_cmp(&max_subnorm()));
|
||||
assert_eq!(Ordering::Greater, 0.5_f16.total_cmp(&f16::MIN_POSITIVE));
|
||||
assert_eq!(Ordering::Greater, 1.0_f16.total_cmp(&0.5));
|
||||
assert_eq!(Ordering::Greater, 1.5_f16.total_cmp(&1.0));
|
||||
assert_eq!(Ordering::Greater, 2.5_f16.total_cmp(&1.5));
|
||||
assert_eq!(Ordering::Greater, f16::MAX.total_cmp(&2.5));
|
||||
assert_eq!(Ordering::Greater, f16::INFINITY.total_cmp(&f16::MAX));
|
||||
assert_eq!(Ordering::Greater, s_nan().total_cmp(&f16::INFINITY));
|
||||
assert_eq!(Ordering::Greater, q_nan().total_cmp(&s_nan()));
|
||||
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&-s_nan()));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&-f16::INFINITY));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&-f16::MAX));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&-2.5));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&-1.5));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&-1.0));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&-0.5));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&-f16::MIN_POSITIVE));
|
||||
// assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&-max_subnorm()));
|
||||
// assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&-min_subnorm()));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&-0.0));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&0.0));
|
||||
// assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&min_subnorm()));
|
||||
// assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&max_subnorm()));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&f16::MIN_POSITIVE));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&0.5));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&1.0));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&1.5));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&2.5));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&f16::MAX));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&f16::INFINITY));
|
||||
assert_eq!(Ordering::Less, (-q_nan()).total_cmp(&s_nan()));
|
||||
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&-f16::INFINITY));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&-f16::MAX));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&-2.5));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&-1.5));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&-1.0));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&-0.5));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&-f16::MIN_POSITIVE));
|
||||
// assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&-max_subnorm()));
|
||||
// assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&-min_subnorm()));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&-0.0));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&0.0));
|
||||
// assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&min_subnorm()));
|
||||
// assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&max_subnorm()));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&f16::MIN_POSITIVE));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&0.5));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&1.0));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&1.5));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&2.5));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&f16::MAX));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&f16::INFINITY));
|
||||
assert_eq!(Ordering::Less, (-s_nan()).total_cmp(&s_nan()));
|
||||
}
|
||||
|
|
|
|||
|
|
@ -2,6 +2,45 @@ use crate::f32::consts;
|
|||
use crate::num::FpCategory as Fp;
|
||||
use crate::num::*;
|
||||
|
||||
/// Smallest number
|
||||
#[allow(dead_code)] // unused on x86
|
||||
const TINY_BITS: u32 = 0x1;
|
||||
|
||||
/// Next smallest number
|
||||
#[allow(dead_code)] // unused on x86
|
||||
const TINY_UP_BITS: u32 = 0x2;
|
||||
|
||||
/// Exponent = 0b11...10, Sifnificand 0b1111..10. Min val > 0
|
||||
#[allow(dead_code)] // unused on x86
|
||||
const MAX_DOWN_BITS: u32 = 0x7f7f_fffe;
|
||||
|
||||
/// Zeroed exponent, full significant
|
||||
#[allow(dead_code)] // unused on x86
|
||||
const LARGEST_SUBNORMAL_BITS: u32 = 0x007f_ffff;
|
||||
|
||||
/// Exponent = 0b1, zeroed significand
|
||||
#[allow(dead_code)] // unused on x86
|
||||
const SMALLEST_NORMAL_BITS: u32 = 0x0080_0000;
|
||||
|
||||
/// First pattern over the mantissa
|
||||
#[allow(dead_code)] // unused on x86
|
||||
const NAN_MASK1: u32 = 0x002a_aaaa;
|
||||
|
||||
/// Second pattern over the mantissa
|
||||
#[allow(dead_code)] // unused on x86
|
||||
const NAN_MASK2: u32 = 0x0055_5555;
|
||||
|
||||
#[allow(unused_macros)]
|
||||
macro_rules! assert_f32_biteq {
|
||||
($left : expr, $right : expr) => {
|
||||
let l: &f32 = &$left;
|
||||
let r: &f32 = &$right;
|
||||
let lb = l.to_bits();
|
||||
let rb = r.to_bits();
|
||||
assert_eq!(lb, rb, "float {l} ({lb:#010x}) is not bitequal to {r} ({rb:#010x})");
|
||||
};
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_num_f32() {
|
||||
test_num(10f32, 2f32);
|
||||
|
|
@ -315,27 +354,16 @@ fn test_is_sign_negative() {
|
|||
assert!((-f32::NAN).is_sign_negative());
|
||||
}
|
||||
|
||||
#[allow(unused_macros)]
|
||||
macro_rules! assert_f32_biteq {
|
||||
($left : expr, $right : expr) => {
|
||||
let l: &f32 = &$left;
|
||||
let r: &f32 = &$right;
|
||||
let lb = l.to_bits();
|
||||
let rb = r.to_bits();
|
||||
assert_eq!(lb, rb, "float {} ({:#x}) is not equal to {} ({:#x})", *l, lb, *r, rb);
|
||||
};
|
||||
}
|
||||
|
||||
// Ignore test on x87 floating point, these platforms do not guarantee NaN
|
||||
// payloads are preserved and flush denormals to zero, failing the tests.
|
||||
#[cfg(not(target_arch = "x86"))]
|
||||
#[test]
|
||||
fn test_next_up() {
|
||||
let tiny = f32::from_bits(1);
|
||||
let tiny_up = f32::from_bits(2);
|
||||
let max_down = f32::from_bits(0x7f7f_fffe);
|
||||
let largest_subnormal = f32::from_bits(0x007f_ffff);
|
||||
let smallest_normal = f32::from_bits(0x0080_0000);
|
||||
let tiny = f32::from_bits(TINY_BITS);
|
||||
let tiny_up = f32::from_bits(TINY_UP_BITS);
|
||||
let max_down = f32::from_bits(MAX_DOWN_BITS);
|
||||
let largest_subnormal = f32::from_bits(LARGEST_SUBNORMAL_BITS);
|
||||
let smallest_normal = f32::from_bits(SMALLEST_NORMAL_BITS);
|
||||
assert_f32_biteq!(f32::NEG_INFINITY.next_up(), f32::MIN);
|
||||
assert_f32_biteq!(f32::MIN.next_up(), -max_down);
|
||||
assert_f32_biteq!((-1.0 - f32::EPSILON).next_up(), -1.0);
|
||||
|
|
@ -352,8 +380,8 @@ fn test_next_up() {
|
|||
|
||||
// Check that NaNs roundtrip.
|
||||
let nan0 = f32::NAN;
|
||||
let nan1 = f32::from_bits(f32::NAN.to_bits() ^ 0x002a_aaaa);
|
||||
let nan2 = f32::from_bits(f32::NAN.to_bits() ^ 0x0055_5555);
|
||||
let nan1 = f32::from_bits(f32::NAN.to_bits() ^ NAN_MASK1);
|
||||
let nan2 = f32::from_bits(f32::NAN.to_bits() ^ NAN_MASK2);
|
||||
assert_f32_biteq!(nan0.next_up(), nan0);
|
||||
assert_f32_biteq!(nan1.next_up(), nan1);
|
||||
assert_f32_biteq!(nan2.next_up(), nan2);
|
||||
|
|
@ -364,11 +392,11 @@ fn test_next_up() {
|
|||
#[cfg(not(target_arch = "x86"))]
|
||||
#[test]
|
||||
fn test_next_down() {
|
||||
let tiny = f32::from_bits(1);
|
||||
let tiny_up = f32::from_bits(2);
|
||||
let max_down = f32::from_bits(0x7f7f_fffe);
|
||||
let largest_subnormal = f32::from_bits(0x007f_ffff);
|
||||
let smallest_normal = f32::from_bits(0x0080_0000);
|
||||
let tiny = f32::from_bits(TINY_BITS);
|
||||
let tiny_up = f32::from_bits(TINY_UP_BITS);
|
||||
let max_down = f32::from_bits(MAX_DOWN_BITS);
|
||||
let largest_subnormal = f32::from_bits(LARGEST_SUBNORMAL_BITS);
|
||||
let smallest_normal = f32::from_bits(SMALLEST_NORMAL_BITS);
|
||||
assert_f32_biteq!(f32::NEG_INFINITY.next_down(), f32::NEG_INFINITY);
|
||||
assert_f32_biteq!(f32::MIN.next_down(), f32::NEG_INFINITY);
|
||||
assert_f32_biteq!((-max_down).next_down(), f32::MIN);
|
||||
|
|
@ -386,8 +414,8 @@ fn test_next_down() {
|
|||
|
||||
// Check that NaNs roundtrip.
|
||||
let nan0 = f32::NAN;
|
||||
let nan1 = f32::from_bits(f32::NAN.to_bits() ^ 0x002a_aaaa);
|
||||
let nan2 = f32::from_bits(f32::NAN.to_bits() ^ 0x0055_5555);
|
||||
let nan1 = f32::from_bits(f32::NAN.to_bits() ^ NAN_MASK1);
|
||||
let nan2 = f32::from_bits(f32::NAN.to_bits() ^ NAN_MASK2);
|
||||
assert_f32_biteq!(nan0.next_down(), nan0);
|
||||
assert_f32_biteq!(nan1.next_down(), nan1);
|
||||
assert_f32_biteq!(nan2.next_down(), nan2);
|
||||
|
|
@ -734,8 +762,8 @@ fn test_float_bits_conv() {
|
|||
|
||||
// Check that NaNs roundtrip their bits regardless of signaling-ness
|
||||
// 0xA is 0b1010; 0x5 is 0b0101 -- so these two together clobbers all the mantissa bits
|
||||
let masked_nan1 = f32::NAN.to_bits() ^ 0x002A_AAAA;
|
||||
let masked_nan2 = f32::NAN.to_bits() ^ 0x0055_5555;
|
||||
let masked_nan1 = f32::NAN.to_bits() ^ NAN_MASK1;
|
||||
let masked_nan2 = f32::NAN.to_bits() ^ NAN_MASK2;
|
||||
assert!(f32::from_bits(masked_nan1).is_nan());
|
||||
assert!(f32::from_bits(masked_nan2).is_nan());
|
||||
|
||||
|
|
|
|||
|
|
@ -2,6 +2,45 @@ use crate::f64::consts;
|
|||
use crate::num::FpCategory as Fp;
|
||||
use crate::num::*;
|
||||
|
||||
/// Smallest number
|
||||
#[allow(dead_code)] // unused on x86
|
||||
const TINY_BITS: u64 = 0x1;
|
||||
|
||||
/// Next smallest number
|
||||
#[allow(dead_code)] // unused on x86
|
||||
const TINY_UP_BITS: u64 = 0x2;
|
||||
|
||||
/// Exponent = 0b11...10, Sifnificand 0b1111..10. Min val > 0
|
||||
#[allow(dead_code)] // unused on x86
|
||||
const MAX_DOWN_BITS: u64 = 0x7fef_ffff_ffff_fffe;
|
||||
|
||||
/// Zeroed exponent, full significant
|
||||
#[allow(dead_code)] // unused on x86
|
||||
const LARGEST_SUBNORMAL_BITS: u64 = 0x000f_ffff_ffff_ffff;
|
||||
|
||||
/// Exponent = 0b1, zeroed significand
|
||||
#[allow(dead_code)] // unused on x86
|
||||
const SMALLEST_NORMAL_BITS: u64 = 0x0010_0000_0000_0000;
|
||||
|
||||
/// First pattern over the mantissa
|
||||
#[allow(dead_code)] // unused on x86
|
||||
const NAN_MASK1: u64 = 0x000a_aaaa_aaaa_aaaa;
|
||||
|
||||
/// Second pattern over the mantissa
|
||||
#[allow(dead_code)] // unused on x86
|
||||
const NAN_MASK2: u64 = 0x0005_5555_5555_5555;
|
||||
|
||||
#[allow(unused_macros)]
|
||||
macro_rules! assert_f64_biteq {
|
||||
($left : expr, $right : expr) => {
|
||||
let l: &f64 = &$left;
|
||||
let r: &f64 = &$right;
|
||||
let lb = l.to_bits();
|
||||
let rb = r.to_bits();
|
||||
assert_eq!(lb, rb, "float {l} ({lb:#018x}) is not bitequal to {r} ({rb:#018x})");
|
||||
};
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_num_f64() {
|
||||
test_num(10f64, 2f64);
|
||||
|
|
@ -305,27 +344,16 @@ fn test_is_sign_negative() {
|
|||
assert!((-f64::NAN).is_sign_negative());
|
||||
}
|
||||
|
||||
#[allow(unused_macros)]
|
||||
macro_rules! assert_f64_biteq {
|
||||
($left : expr, $right : expr) => {
|
||||
let l: &f64 = &$left;
|
||||
let r: &f64 = &$right;
|
||||
let lb = l.to_bits();
|
||||
let rb = r.to_bits();
|
||||
assert_eq!(lb, rb, "float {} ({:#x}) is not equal to {} ({:#x})", *l, lb, *r, rb);
|
||||
};
|
||||
}
|
||||
|
||||
// Ignore test on x87 floating point, these platforms do not guarantee NaN
|
||||
// payloads are preserved and flush denormals to zero, failing the tests.
|
||||
#[cfg(not(target_arch = "x86"))]
|
||||
#[test]
|
||||
fn test_next_up() {
|
||||
let tiny = f64::from_bits(1);
|
||||
let tiny_up = f64::from_bits(2);
|
||||
let max_down = f64::from_bits(0x7fef_ffff_ffff_fffe);
|
||||
let largest_subnormal = f64::from_bits(0x000f_ffff_ffff_ffff);
|
||||
let smallest_normal = f64::from_bits(0x0010_0000_0000_0000);
|
||||
let tiny = f64::from_bits(TINY_BITS);
|
||||
let tiny_up = f64::from_bits(TINY_UP_BITS);
|
||||
let max_down = f64::from_bits(MAX_DOWN_BITS);
|
||||
let largest_subnormal = f64::from_bits(LARGEST_SUBNORMAL_BITS);
|
||||
let smallest_normal = f64::from_bits(SMALLEST_NORMAL_BITS);
|
||||
assert_f64_biteq!(f64::NEG_INFINITY.next_up(), f64::MIN);
|
||||
assert_f64_biteq!(f64::MIN.next_up(), -max_down);
|
||||
assert_f64_biteq!((-1.0 - f64::EPSILON).next_up(), -1.0);
|
||||
|
|
@ -341,8 +369,8 @@ fn test_next_up() {
|
|||
assert_f64_biteq!(f64::INFINITY.next_up(), f64::INFINITY);
|
||||
|
||||
let nan0 = f64::NAN;
|
||||
let nan1 = f64::from_bits(f64::NAN.to_bits() ^ 0x000a_aaaa_aaaa_aaaa);
|
||||
let nan2 = f64::from_bits(f64::NAN.to_bits() ^ 0x0005_5555_5555_5555);
|
||||
let nan1 = f64::from_bits(f64::NAN.to_bits() ^ NAN_MASK1);
|
||||
let nan2 = f64::from_bits(f64::NAN.to_bits() ^ NAN_MASK2);
|
||||
assert_f64_biteq!(nan0.next_up(), nan0);
|
||||
assert_f64_biteq!(nan1.next_up(), nan1);
|
||||
assert_f64_biteq!(nan2.next_up(), nan2);
|
||||
|
|
@ -353,11 +381,11 @@ fn test_next_up() {
|
|||
#[cfg(not(target_arch = "x86"))]
|
||||
#[test]
|
||||
fn test_next_down() {
|
||||
let tiny = f64::from_bits(1);
|
||||
let tiny_up = f64::from_bits(2);
|
||||
let max_down = f64::from_bits(0x7fef_ffff_ffff_fffe);
|
||||
let largest_subnormal = f64::from_bits(0x000f_ffff_ffff_ffff);
|
||||
let smallest_normal = f64::from_bits(0x0010_0000_0000_0000);
|
||||
let tiny = f64::from_bits(TINY_BITS);
|
||||
let tiny_up = f64::from_bits(TINY_UP_BITS);
|
||||
let max_down = f64::from_bits(MAX_DOWN_BITS);
|
||||
let largest_subnormal = f64::from_bits(LARGEST_SUBNORMAL_BITS);
|
||||
let smallest_normal = f64::from_bits(SMALLEST_NORMAL_BITS);
|
||||
assert_f64_biteq!(f64::NEG_INFINITY.next_down(), f64::NEG_INFINITY);
|
||||
assert_f64_biteq!(f64::MIN.next_down(), f64::NEG_INFINITY);
|
||||
assert_f64_biteq!((-max_down).next_down(), f64::MIN);
|
||||
|
|
@ -374,8 +402,8 @@ fn test_next_down() {
|
|||
assert_f64_biteq!(f64::INFINITY.next_down(), f64::MAX);
|
||||
|
||||
let nan0 = f64::NAN;
|
||||
let nan1 = f64::from_bits(f64::NAN.to_bits() ^ 0x000a_aaaa_aaaa_aaaa);
|
||||
let nan2 = f64::from_bits(f64::NAN.to_bits() ^ 0x0005_5555_5555_5555);
|
||||
let nan1 = f64::from_bits(f64::NAN.to_bits() ^ NAN_MASK1);
|
||||
let nan2 = f64::from_bits(f64::NAN.to_bits() ^ NAN_MASK2);
|
||||
assert_f64_biteq!(nan0.next_down(), nan0);
|
||||
assert_f64_biteq!(nan1.next_down(), nan1);
|
||||
assert_f64_biteq!(nan2.next_down(), nan2);
|
||||
|
|
@ -715,9 +743,8 @@ fn test_float_bits_conv() {
|
|||
assert_approx_eq!(f64::from_bits(0xc02c800000000000), -14.25);
|
||||
|
||||
// Check that NaNs roundtrip their bits regardless of signaling-ness
|
||||
// 0xA is 0b1010; 0x5 is 0b0101 -- so these two together clobbers all the mantissa bits
|
||||
let masked_nan1 = f64::NAN.to_bits() ^ 0x000A_AAAA_AAAA_AAAA;
|
||||
let masked_nan2 = f64::NAN.to_bits() ^ 0x0005_5555_5555_5555;
|
||||
let masked_nan1 = f64::NAN.to_bits() ^ NAN_MASK1;
|
||||
let masked_nan2 = f64::NAN.to_bits() ^ NAN_MASK2;
|
||||
assert!(f64::from_bits(masked_nan1).is_nan());
|
||||
assert!(f64::from_bits(masked_nan2).is_nan());
|
||||
|
||||
|
|
|
|||
|
|
@ -373,10 +373,17 @@ macro_rules! dbg {
|
|||
};
|
||||
}
|
||||
|
||||
/// Verify that floats are within a tolerance of each other, 1.0e-6 by default.
|
||||
#[cfg(test)]
|
||||
macro_rules! assert_approx_eq {
|
||||
($a:expr, $b:expr) => {{
|
||||
($a:expr, $b:expr) => {{ assert_approx_eq!($a, $b, 1.0e-6) }};
|
||||
($a:expr, $b:expr, $lim:expr) => {{
|
||||
let (a, b) = (&$a, &$b);
|
||||
assert!((*a - *b).abs() < 1.0e-6, "{} is not approximately equal to {}", *a, *b);
|
||||
let diff = (*a - *b).abs();
|
||||
assert!(
|
||||
diff < $lim,
|
||||
"{a:?} is not approximately equal to {b:?} (threshold {lim:?}, actual {diff:?})",
|
||||
lim = $lim
|
||||
);
|
||||
}};
|
||||
}
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue