Uh oh!
There was an error while loading.Please reload this page.
- Notifications
You must be signed in to change notification settings - Fork71
Half-precision floating point types f16 and bf16 for Rust.
License
Apache-2.0, MIT licenses found
Licenses found
VoidStarKat/half-rs
Folders and files
| Name | Name | Last commit message | Last commit date | |
|---|---|---|---|---|
Repository files navigation
This crate implements a half-precision floating pointf16 type for Rust implementing the IEEE754-2008 standardbinary16a.k.a "half" format, as well as abf16 type implementing thebfloat16 format.
Thef16 andbf16 types attempt to match existing Rust floating point type functionality where possible, and provides both conversion operations (such as to/fromf32 andf64) and basicarithmetic operations. Hardware support for these operations will be used whenever hardware supportis available—either through instrinsics or targeted assembly—although a nightly Rust toolchain maybe required for some hardware.
This crate providesno_std support so can easily be used in embedded code where a smaller float format is most useful.
Requires Rust 1.81 or greater. If you need support for older versions of Rust, use previousversions of this crate.
See thecrate documentation for more details.
alloc— Enable use of thealloccrate when not usingthestdlibrary.This enables the
vecmodule, which contains zero-copy conversions for theVectype. Thisallows fast conversion between rawVec<u16>bits andVec<f16>orVec<bf16>arrays, and viceversa.std— Enable features that depend on the Ruststdlibrary, including everything in theallocfeature.Enabling the
stdfeature enables runtime CPU feature detection of hardware support.Without this feature detection, harware is only used when compiler target supports them.serde- ImplementSerializeandDeserializetraits forf16andbf16. This adds adependency on theserdecrate.num-traits— EnableToPrimitive,FromPrimitive,ToBytes,FromBytes,Num,Float,FloatCore,Signed, andBoundedtrait implementations from thenum-traitscrate.bytemuck— EnableZeroableandPodtrait implementations from thebytemuckcrate.rand_distr— Enable sampling from distributions likeStandardUniformandStandardNormalfrom therand_distrcrate.rkyv-- Enable zero-copy deserializtion withrkyvcrate.aribtrary-- Enable fuzzing support witharbitrarycrate by implementingArbitrarytrait.nightly-- Enable nightly-only features (currentlyloongarch64intrinsics).
The following list details hardware support for floating point types in this crate. When usingstdlibrary, runtime CPU target detection will be used. To get the most performance benefits, compilefor specific CPU features which avoids the runtime overhead and works in ano_std environment.
| Architecture | CPU Target Feature | Notes |
|---|---|---|
x86/x86_64 | f16c | This supports conversion to/fromf16 only (including vector SIMD) and does not support anybf16 or arithmetic operations. |
aarch64 | fp16 | This supports all operations onf16 only. |
loongarch64 | lsx | (nightly feature only) This supports conversion to/fromf16 only (including vector SIMD) and does not support anybf16 or arithmetic operations. |
All files in this library are dual-licensed and distributed under the terms of either of:
- MIT License(http://opensource.org/licenses/MIT)
- Apache License, Version 2.0(http://www.apache.org/licenses/LICENSE-2.0)
at your option.
Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in thework by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without anyadditional terms or conditions.
About
Half-precision floating point types f16 and bf16 for Rust.
Topics
Resources
License
Apache-2.0, MIT licenses found
Licenses found
Uh oh!
There was an error while loading.Please reload this page.
Stars
Watchers
Forks
Sponsor this project
Uh oh!
There was an error while loading.Please reload this page.
Uh oh!
There was an error while loading.Please reload this page.