pub trait Float: Sized + Copy + PartialEq + PartialOrd + Send + Sync + Add<Output = Self> + AddAssign + Div<Output = Self> + DivAssign + Mul<Output = Self> + MulAssign + Rem<Output = Self> + RemAssign + Sub<Output = Self> + SubAssign + Neg<Output = Self> {
Show 21 associated constants and 8 methods const MAX_DIGITS: usize; const SIGN_MASK: u64; const EXPONENT_MASK: u64; const HIDDEN_BIT_MASK: u64; const MANTISSA_MASK: u64; const MANTISSA_SIZE: i32; const EXPONENT_BIAS: i32; const DENORMAL_EXPONENT: i32; const MAX_EXPONENT: i32; const CARRY_MASK: u64; const INVALID_FP: i32; const MAX_MANTISSA_FAST_PATH: u64; const INFINITE_POWER: i32; const MIN_EXPONENT_ROUND_TO_EVEN: i32; const MAX_EXPONENT_ROUND_TO_EVEN: i32; const MINIMUM_EXPONENT: i32; const SMALLEST_POWER_OF_TEN: i32; const LARGEST_POWER_OF_TEN: i32; const MIN_EXPONENT_FAST_PATH: i32; const MAX_EXPONENT_FAST_PATH: i32; const MAX_EXPONENT_DISGUISED_FAST_PATH: i32; fn from_u64(u: u64) -> Self;
fn from_bits(u: u64) -> Self;
fn to_bits(self) -> u64;
unsafe fn pow_fast_path(exponent: usize) -> Self; unsafe fn int_pow_fast_path(exponent: usize, radix: u32) -> u64 { ... }
fn is_denormal(self) -> bool { ... }
fn exponent(self) -> i32 { ... }
fn mantissa(self) -> u64 { ... }
}
Expand description

Generic floating-point type, to be used in generic code for parsing.

Although the trait is part of the public API, the trait provides methods and constants that are effectively non-public: they may be removed at any time without any breaking changes.

Associated Constants

Maximum number of digits that can contribute in the mantissa.

We can exactly represent a float in radix b from radix 2 if b is divisible by 2. This function calculates the exact number of digits required to exactly represent that float.

According to the “Handbook of Floating Point Arithmetic”, for IEEE754, with emin being the min exponent, p2 being the precision, and b being the radix, the number of digits follows as:

−emin + p2 + ⌊(emin + 1) log(2, b) − log(1 − 2^(−p2), b)⌋

For f32, this follows as: emin = -126 p2 = 24

For f64, this follows as: emin = -1022 p2 = 53

In Python: -emin + p2 + math.floor((emin+1)*math.log(2, b) - math.log(1-2**(-p2), b))

This was used to calculate the maximum number of digits for [2, 36].

Bitmask for the sign bit.

Bitmask for the exponent, including the hidden bit.

Bitmask for the hidden bit in exponent, which is an implicit 1 in the fraction.

Bitmask for the mantissa (fraction), excluding the hidden bit.

Size of the significand (mantissa) without hidden bit.

Bias of the exponet

Exponent portion of a denormal float.

Maximum exponent value in float.

Mask to determine if a full-carry occurred (1 in bit above hidden bit).

Bias for marking an invalid extended float.

Minimum normal exponent value -(1 << (EXPONENT_SIZE - 1)) + 1.

Smallest decimal exponent for a non-zero value.

Largest decimal exponent for a non-infinite value.

Minimum exponent that for a fast path case, or -⌊(MANTISSA_SIZE+1)/log2(10)⌋

Maximum exponent that for a fast path case, or ⌊(MANTISSA_SIZE+1)/log2(5)⌋

Maximum exponent that can be represented for a disguised-fast path case. This is MAX_EXPONENT_FAST_PATH + ⌊(MANTISSA_SIZE+1)/log2(10)⌋

Required methods

Convert 64-bit integer to float.

Get a small power-of-radix for fast-path multiplication.

Safety

Safe as long as the exponent is smaller than the table size.

Provided methods

Get a small, integral power-of-radix for fast-path multiplication.

Safety

Safe as long as the exponent is smaller than the table size.

Returns true if the float is a denormal.

Get exponent component from the float.

Get mantissa (significand) component from float.

Implementations on Foreign Types

Implementors