Trait FloatFormat 

pub trait FloatFormat: Copy {
    const EXP_BITS: u32;
    const MAN_BITS: u32;
    const MAX_NORMAL: f32;
    const NAME: &'static str;
    const BITS: u32 = _;

    // Provided method
    fn quantize(x: f32) -> f32 { ... }
}

A compile-time-known binary float format.

Required Associated Constants

const EXP_BITS: u32

Exponent bits.

const MAN_BITS: u32

Explicit mantissa bits.

const MAX_NORMAL: f32

Largest finite magnitude (format-specific: e.g. 448 for E4M3, 6 for E2M1).

const NAME: &'static str

Canonical spec string ("f8e4m3", "nvf4", …).

Provided Associated Constants

const BITS: u32 = _

Total bits (1 + EXP_BITS + MAN_BITS).

Provided Methods

fn quantize(x: f32) -> f32

Round x to this format’s grid.

Dyn Compatibility

This trait is not dyn compatible.

In older versions of Rust, dyn compatibility was called "object safety".

Implementors

impl FloatFormat for Bf16

const EXP_BITS: u32 = 8

const MAN_BITS: u32 = 7

const MAX_NORMAL: f32 = 3.3895314e38

const NAME: &'static str = "bf16"

impl FloatFormat for F6E3M2

const EXP_BITS: u32 = 3

const MAN_BITS: u32 = 2

const MAX_NORMAL: f32 = 28.0

const NAME: &'static str = "f6e3m2"

impl FloatFormat for F8E4M3

const EXP_BITS: u32 = 4

const MAN_BITS: u32 = 3

const MAX_NORMAL: f32 = 448.0

const NAME: &'static str = "f8e4m3"

impl FloatFormat for F8E5M2

const EXP_BITS: u32 = 5

const MAN_BITS: u32 = 2

const MAX_NORMAL: f32 = 57344.0

const NAME: &'static str = "bf8"

impl FloatFormat for F16

const EXP_BITS: u32 = 5

const MAN_BITS: u32 = 10

const MAX_NORMAL: f32 = 65504.0

const NAME: &'static str = "f16"

impl FloatFormat for Nvf4

const EXP_BITS: u32 = 2

const MAN_BITS: u32 = 1

const MAX_NORMAL: f32 = 6.0

const NAME: &'static str = "nvf4"