| 1 | //! A crate that provides support for half-precision 16-bit floating point types. |
| 2 | //! |
| 3 | //! This crate provides the [`f16`] type, which is an implementation of the IEEE 754-2008 standard |
| 4 | //! [`binary16`] a.k.a `half` floating point type. This 16-bit floating point type is intended for |
| 5 | //! efficient storage where the full range and precision of a larger floating point value is not |
| 6 | //! required. This is especially useful for image storage formats. |
| 7 | //! |
| 8 | //! This crate also provides a [`bf16`] type, an alternative 16-bit floating point format. The |
| 9 | //! [`bfloat16`] format is a truncated IEEE 754 standard `binary32` float that preserves the |
| 10 | //! exponent to allow the same range as [`f32`] but with only 8 bits of precision (instead of 11 |
| 11 | //! bits for [`f16`]). See the [`bf16`] type for details. |
| 12 | //! |
| 13 | //! Because [`f16`] and [`bf16`] are primarily for efficient storage, floating point operations such |
| 14 | //! as addition, multiplication, etc. are not implemented by hardware. While this crate does provide |
| 15 | //! the appropriate trait implementations for basic operations, they each convert the value to |
| 16 | //! [`f32`] before performing the operation and then back afterward. When performing complex |
| 17 | //! arithmetic, manually convert to and from [`f32`] before and after to reduce repeated conversions |
| 18 | //! for each operation. |
| 19 | //! |
| 20 | //! This crate also provides a [`slice`][mod@slice] module for zero-copy in-place conversions of |
| 21 | //! [`u16`] slices to both [`f16`] and [`bf16`], as well as efficient vectorized conversions of |
| 22 | //! larger buffers of floating point values to and from these half formats. |
| 23 | //! |
| 24 | //! The crate uses `#[no_std]` by default, so can be used in embedded environments without using the |
| 25 | //! Rust [`std`] library. A `std` feature to enable support for the standard library is available, |
| 26 | //! see the [Cargo Features](#cargo-features) section below. |
| 27 | //! |
| 28 | //! A [`prelude`] module is provided for easy importing of available utility traits. |
| 29 | //! |
| 30 | //! # Cargo Features |
| 31 | //! |
| 32 | //! This crate supports a number of optional cargo features. None of these features are enabled by |
| 33 | //! default, even `std`. |
| 34 | //! |
| 35 | //! - **`use-intrinsics`** -- Use [`core::arch`] hardware intrinsics for `f16` and `bf16` conversions |
| 36 | //! if available on the compiler target. This feature currently only works on nightly Rust |
| 37 | //! until the corresponding intrinsics are stabilized. |
| 38 | //! |
| 39 | //! When this feature is enabled and the hardware supports it, the functions and traits in the |
| 40 | //! [`slice`][mod@slice] module will use vectorized SIMD intructions for increased efficiency. |
| 41 | //! |
| 42 | //! By default, without this feature, conversions are done only in software, which will also be |
| 43 | //! the fallback if the target does not have hardware support. Note that without the `std` |
| 44 | //! feature enabled, no runtime CPU feature detection is used, so the hardware support is only |
| 45 | //! compiled if the compiler target supports the CPU feature. |
| 46 | //! |
| 47 | //! - **`alloc`** -- Enable use of the [`alloc`] crate when not using the `std` library. |
| 48 | //! |
| 49 | //! Among other functions, this enables the [`vec`] module, which contains zero-copy |
| 50 | //! conversions for the [`Vec`] type. This allows fast conversion between raw `Vec<u16>` bits and |
| 51 | //! `Vec<f16>` or `Vec<bf16>` arrays, and vice versa. |
| 52 | //! |
| 53 | //! - **`std`** -- Enable features that depend on the Rust [`std`] library. This also enables the |
| 54 | //! `alloc` feature automatically. |
| 55 | //! |
| 56 | //! Enabling the `std` feature also enables runtime CPU feature detection when the |
| 57 | //! `use-intrsincis` feature is also enabled. Without this feature detection, intrinsics are only |
| 58 | //! used when compiler target supports the target feature. |
| 59 | //! |
| 60 | //! - **`serde`** -- Adds support for the [`serde`] crate by implementing [`Serialize`] and |
| 61 | //! [`Deserialize`] traits for both [`f16`] and [`bf16`]. |
| 62 | //! |
| 63 | //! - **`num-traits`** -- Adds support for the [`num-traits`] crate by implementing [`ToPrimitive`], |
| 64 | //! [`FromPrimitive`], [`AsPrimitive`], [`Num`], [`Float`], [`FloatCore`], and [`Bounded`] traits |
| 65 | //! for both [`f16`] and [`bf16`]. |
| 66 | //! |
| 67 | //! - **`bytemuck`** -- Adds support for the [`bytemuck`] crate by implementing [`Zeroable`] and |
| 68 | //! [`Pod`] traits for both [`f16`] and [`bf16`]. |
| 69 | //! |
| 70 | //! - **`zerocopy`** -- Adds support for the [`zerocopy`] crate by implementing [`AsBytes`] and |
| 71 | //! [`FromBytes`] traits for both [`f16`] and [`bf16`]. |
| 72 | //! |
| 73 | //! [`alloc`]: https://doc.rust-lang.org/alloc/ |
| 74 | //! [`std`]: https://doc.rust-lang.org/std/ |
| 75 | //! [`binary16`]: https://en.wikipedia.org/wiki/Half-precision_floating-point_format |
| 76 | //! [`bfloat16`]: https://en.wikipedia.org/wiki/Bfloat16_floating-point_format |
| 77 | //! [`serde`]: https://crates.io/crates/serde |
| 78 | //! [`bytemuck`]: https://crates.io/crates/bytemuck |
| 79 | //! [`num-traits`]: https://crates.io/crates/num-traits |
| 80 | //! [`zerocopy`]: https://crates.io/crates/zerocopy |
| 81 | #![cfg_attr ( |
| 82 | feature = "alloc" , |
| 83 | doc = " |
| 84 | [`vec`]: mod@vec" |
| 85 | )] |
| 86 | #![cfg_attr ( |
| 87 | not(feature = "alloc" ), |
| 88 | doc = " |
| 89 | [`vec`]: # |
| 90 | [`Vec`]: https://docs.rust-lang.org/stable/alloc/vec/struct.Vec.html" |
| 91 | )] |
| 92 | #![cfg_attr ( |
| 93 | feature = "serde" , |
| 94 | doc = " |
| 95 | [`Serialize`]: serde::Serialize |
| 96 | [`Deserialize`]: serde::Deserialize" |
| 97 | )] |
| 98 | #![cfg_attr ( |
| 99 | not(feature = "serde" ), |
| 100 | doc = " |
| 101 | [`Serialize`]: https://docs.rs/serde/*/serde/trait.Serialize.html |
| 102 | [`Deserialize`]: https://docs.rs/serde/*/serde/trait.Deserialize.html" |
| 103 | )] |
| 104 | #![cfg_attr ( |
| 105 | feature = "num-traits" , |
| 106 | doc = " |
| 107 | [`ToPrimitive`]: ::num_traits::ToPrimitive |
| 108 | [`FromPrimitive`]: ::num_traits::FromPrimitive |
| 109 | [`AsPrimitive`]: ::num_traits::AsPrimitive |
| 110 | [`Num`]: ::num_traits::Num |
| 111 | [`Float`]: ::num_traits::Float |
| 112 | [`FloatCore`]: ::num_traits::float::FloatCore |
| 113 | [`Bounded`]: ::num_traits::Bounded" |
| 114 | )] |
| 115 | #![cfg_attr ( |
| 116 | not(feature = "num-traits" ), |
| 117 | doc = " |
| 118 | [`ToPrimitive`]: https://docs.rs/num-traits/*/num_traits/cast/trait.ToPrimitive.html |
| 119 | [`FromPrimitive`]: https://docs.rs/num-traits/*/num_traits/cast/trait.FromPrimitive.html |
| 120 | [`AsPrimitive`]: https://docs.rs/num-traits/*/num_traits/cast/trait.AsPrimitive.html |
| 121 | [`Num`]: https://docs.rs/num-traits/*/num_traits/trait.Num.html |
| 122 | [`Float`]: https://docs.rs/num-traits/*/num_traits/float/trait.Float.html |
| 123 | [`FloatCore`]: https://docs.rs/num-traits/*/num_traits/float/trait.FloatCore.html |
| 124 | [`Bounded`]: https://docs.rs/num-traits/*/num_traits/bounds/trait.Bounded.html" |
| 125 | )] |
| 126 | #![cfg_attr ( |
| 127 | feature = "bytemuck" , |
| 128 | doc = " |
| 129 | [`Zeroable`]: bytemuck::Zeroable |
| 130 | [`Pod`]: bytemuck::Pod" |
| 131 | )] |
| 132 | #![cfg_attr ( |
| 133 | not(feature = "bytemuck" ), |
| 134 | doc = " |
| 135 | [`Zeroable`]: https://docs.rs/bytemuck/*/bytemuck/trait.Zeroable.html |
| 136 | [`Pod`]: https://docs.rs/bytemuck/*bytemuck/trait.Pod.html" |
| 137 | )] |
| 138 | #![cfg_attr ( |
| 139 | feature = "zerocopy" , |
| 140 | doc = " |
| 141 | [`AsBytes`]: zerocopy::AsBytes |
| 142 | [`FromBytes`]: zerocopy::FromBytes" |
| 143 | )] |
| 144 | #![cfg_attr ( |
| 145 | not(feature = "zerocopy" ), |
| 146 | doc = " |
| 147 | [`AsBytes`]: https://docs.rs/zerocopy/*/zerocopy/trait.AsBytes.html |
| 148 | [`FromBytes`]: https://docs.rs/zerocopy/*/zerocopy/trait.FromBytes.html" |
| 149 | )] |
| 150 | #![warn ( |
| 151 | missing_docs, |
| 152 | missing_copy_implementations, |
| 153 | missing_debug_implementations, |
| 154 | trivial_numeric_casts, |
| 155 | future_incompatible |
| 156 | )] |
| 157 | #![allow (clippy::verbose_bit_mask, clippy::cast_lossless)] |
| 158 | #![cfg_attr (not(feature = "std" ), no_std)] |
| 159 | #![cfg_attr ( |
| 160 | all( |
| 161 | feature = "use-intrinsics" , |
| 162 | any(target_arch = "x86" , target_arch = "x86_64" ) |
| 163 | ), |
| 164 | feature(stdsimd, f16c_target_feature) |
| 165 | )] |
| 166 | #![doc (html_root_url = "https://docs.rs/half/1.8.2" )] |
| 167 | #![doc (test(attr(deny(warnings), allow(unused))))] |
| 168 | #![cfg_attr (docsrs, feature(doc_cfg))] |
| 169 | |
| 170 | #[cfg (feature = "alloc" )] |
| 171 | extern crate alloc; |
| 172 | |
| 173 | mod bfloat; |
| 174 | mod binary16; |
| 175 | #[cfg (feature = "num-traits" )] |
| 176 | mod num_traits; |
| 177 | |
| 178 | pub mod slice; |
| 179 | #[cfg (feature = "alloc" )] |
| 180 | #[cfg_attr (docsrs, doc(cfg(feature = "alloc" )))] |
| 181 | pub mod vec; |
| 182 | |
| 183 | pub use bfloat::bf16; |
| 184 | #[doc (hidden)] |
| 185 | #[allow (deprecated)] |
| 186 | pub use binary16::consts; |
| 187 | pub use binary16::f16; |
| 188 | |
| 189 | /// A collection of the most used items and traits in this crate for easy importing. |
| 190 | /// |
| 191 | /// # Examples |
| 192 | /// |
| 193 | /// ```rust |
| 194 | /// use half::prelude::*; |
| 195 | /// ``` |
| 196 | pub mod prelude { |
| 197 | #[doc (no_inline)] |
| 198 | pub use crate::{ |
| 199 | bf16, f16, |
| 200 | slice::{HalfBitsSliceExt, HalfFloatSliceExt}, |
| 201 | }; |
| 202 | |
| 203 | #[cfg (feature = "alloc" )] |
| 204 | #[doc (no_inline)] |
| 205 | #[cfg_attr (docsrs, doc(cfg(feature = "alloc" )))] |
| 206 | pub use crate::vec::{HalfBitsVecExt, HalfFloatVecExt}; |
| 207 | } |
| 208 | |
| 209 | // Keep this module private to crate |
| 210 | mod private { |
| 211 | use crate::{bf16, f16}; |
| 212 | |
| 213 | pub trait SealedHalf {} |
| 214 | |
| 215 | impl SealedHalf for f16 {} |
| 216 | impl SealedHalf for bf16 {} |
| 217 | } |
| 218 | |