rust/library/portable-simd
2024-04-24 13:12:25 +01:00
..
.github Merge commit 'cff979eec1ac0473fc4960ee6cde462c6aeda824' into sync-portable-simd-2024-03-22 2024-03-22 16:58:39 -07:00
crates Stabilise inline_const 2024-04-24 13:12:25 +01:00
.gitignore Merge commit 'e0e9a4517f9fc021283514da387e70a56061bd3e' into sync-portable-simd-2023-11-19 2023-11-26 08:50:06 -05:00
beginners-guide.md
Cargo.lock Merge commit 'cff979eec1ac0473fc4960ee6cde462c6aeda824' into sync-portable-simd-2024-03-22 2024-03-22 16:58:39 -07:00
Cargo.toml
CONTRIBUTING.md
LICENSE-APACHE
LICENSE-MIT
README.md

The Rust standard library's portable SIMD API

Build Status

Code repository for the Portable SIMD Project Group. Please refer to CONTRIBUTING.md for our contributing guidelines.

The docs for this crate are published from the main branch. You can read them here.

If you have questions about SIMD, we have begun writing a guide. We can also be found on Zulip.

If you are interested in support for a specific architecture, you may want stdarch instead.

Hello World

Now we're gonna dip our toes into this world with a small SIMD "Hello, World!" example. Make sure your compiler is up to date and using nightly. We can do that by running

rustup update -- nightly

or by setting up rustup default nightly or else with cargo +nightly {build,test,run}. After updating, run

cargo new hellosimd

to create a new crate. Finally write this in src/main.rs:

#![feature(portable_simd)]
use std::simd::f32x4;
fn main() {
    let a = f32x4::splat(10.0);
    let b = f32x4::from_array([1.0, 2.0, 3.0, 4.0]);
    println!("{:?}", a + b);
}

Explanation: We construct our SIMD vectors with methods like splat or from_array. Next, we can use operators like + on them, and the appropriate SIMD instructions will be carried out. When we run cargo run you should get [11.0, 12.0, 13.0, 14.0].

Supported vectors

Currently, vectors may have up to 64 elements, but aliases are provided only up to 512-bit vectors.

Depending on the size of the primitive type, the number of lanes the vector will have varies. For example, 128-bit vectors have four f32 lanes and two f64 lanes.

The supported element types are as follows:

  • Floating Point: f32, f64
  • Signed Integers: i8, i16, i32, i64, isize (i128 excluded)
  • Unsigned Integers: u8, u16, u32, u64, usize (u128 excluded)
  • Pointers: *const T and *mut T (zero-sized metadata only)
  • Masks: 8-bit, 16-bit, 32-bit, 64-bit, and usize-sized masks

Floating point, signed integers, unsigned integers, and pointers are the primitive types you're already used to. The mask types have elements that are "truthy" values, like bool, but have an unspecified layout because different architectures prefer different layouts for mask types.