Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

add SIMD swap_bytes/to_le #1

Closed
wants to merge 2 commits into from
Closed

add SIMD swap_bytes/to_le #1

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
83366b6
add SIMD swap_bytes/to_le
Jun 27, 2018
c96a138
comments, consistency
Jun 29, 2018
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
31 changes: 19 additions & 12 deletions src/distributions/integer.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,10 @@ use {Rng};
use distributions::{Distribution, Standard};
#[cfg(feature="simd_support")]
use core::simd::*;
#[cfg(feature="simd_support")]
use core::mem;
#[cfg(feature="simd_support")]
use distributions::utils::ToLittleEndian;

impl Distribution<u8> for Standard {
#[inline]
Expand Down Expand Up @@ -88,37 +92,40 @@ impl_int_from_uint! { isize, usize }

#[cfg(feature="simd_support")]
macro_rules! simd_impl {
($bits:expr,) => {};
($bits:expr, $ty:ty, $($ty_more:ty,)*) => {
simd_impl!($bits, $($ty_more,)*);
() => {};
($ty:ty, $($ty_more:ty,)*) => {
simd_impl!($($ty_more,)*);

impl Distribution<$ty> for Standard {
#[inline]
fn sample<R: Rng + ?Sized>(&self, rng: &mut R) -> $ty {
let mut vec = Default::default();
let mut vec: $ty = Default::default();
unsafe {
let ptr = &mut vec;
let b_ptr = &mut *(ptr as *mut $ty as *mut [u8; $bits/8]);
let b_ptr = &mut *(ptr as *mut $ty as *mut [u8; mem::size_of::<$ty>()]);
rng.fill_bytes(b_ptr);
}
vec
// We could also use `slice::reverse`. `swap_bytes` of `to_le`
// might be faster because it can assume memory length. Assembly
// inspection shows `swap_bytes` produces fewer instructions.
vec.to_le()
}
}
}
}

#[cfg(feature="simd_support")]
simd_impl!(16, u8x2, i8x2,);
simd_impl!(u8x2, i8x2,);
#[cfg(feature="simd_support")]
simd_impl!(32, u8x4, i8x4, u16x2, i16x2,);
simd_impl!(u8x4, i8x4, u16x2, i16x2,);
#[cfg(feature="simd_support")]
simd_impl!(64, u8x8, i8x8, u16x4, i16x4, u32x2, i32x2,);
simd_impl!(u8x8, i8x8, u16x4, i16x4, u32x2, i32x2,);
#[cfg(feature="simd_support")]
simd_impl!(128, u8x16, i8x16, u16x8, i16x8, u32x4, i32x4, u64x2, i64x2,);
simd_impl!(u8x16, i8x16, u16x8, i16x8, u32x4, i32x4, u64x2, i64x2,);
#[cfg(feature="simd_support")]
simd_impl!(256, u8x32, i8x32, u16x16, i16x16, u32x8, i32x8, u64x4, i64x4,);
simd_impl!(u8x32, i8x32, u16x16, i16x16, u32x8, i32x8, u64x4, i64x4,);
#[cfg(feature="simd_support")]
simd_impl!(512, u8x64, i8x64, u16x32, i16x32, u32x16, i32x16, u64x8, i64x8,);
simd_impl!(u8x64, i8x64, u16x32, i16x32, u32x16, i32x16, u64x8, i64x8,);

#[cfg(test)]
mod tests {
Expand Down
195 changes: 195 additions & 0 deletions src/distributions/utils.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -162,3 +162,198 @@ macro_rules! simd_less_then {
#[cfg(feature="simd_support")] simd_less_then! { f64x2 }
#[cfg(feature="simd_support")] simd_less_then! { f64x4 }
#[cfg(feature="simd_support")] simd_less_then! { f64x8 }


// Until portable shuffles land in stdsimd, we expose and use the shuffle
// intrinsics directly.
//
// https://github.com/rust-lang-nursery/stdsimd/issues/388
#[cfg(feature = "simd_support")]
extern "platform-intrinsic" {
fn simd_shuffle2<T, U>(a: T, b: T, indices: [u32; 2]) -> U;
fn simd_shuffle4<T, U>(a: T, b: T, indices: [u32; 4]) -> U;
fn simd_shuffle8<T, U>(a: T, b: T, indices: [u32; 8]) -> U;
fn simd_shuffle16<T, U>(a: T, b: T, indices: [u32; 16]) -> U;
fn simd_shuffle32<T, U>(a: T, b: T, indices: [u32; 32]) -> U;
fn simd_shuffle64<T, U>(a: T, b: T, indices: [u32; 64]) -> U;
}

/// Implement byte swapping for SIMD vectors
#[cfg(feature = "simd_support")]
trait SwapBytes {
/// `swap_bytes` for a vector (horizontally)
fn swap_bytes(self) -> Self;
}

// `simd_shuffleX` require constant indices, making this a small pain to implement
#[cfg(feature = "simd_support")]
macro_rules! impl_swap_bytes {
($vec8:ident, $shuf:ident, $indices:expr, $ty:ident) => (
impl SwapBytes for $ty {
fn swap_bytes(self) -> Self {
let vec8 = $vec8::from_bits(self);
let shuffled: $vec8 = unsafe { $shuf(vec8, vec8, $indices) };
$ty::from_bits(shuffled)
}
}
);

// bulk impl for a shuffle intrinsic/vector width
($vec8:ident, $shuf:ident, $indices:expr, $($ty:ident,)+) => ($(
impl_swap_bytes! { $ty, $vec8, $shuf, $indices }
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Changing the argument order here is confusing. Keep the type at the end?

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It makes the macro invitation prettier but yeah I see how it could trip people up

)+);
}

#[cfg(feature = "simd_support")]
impl_swap_bytes! {
u8x2,
simd_shuffle2,
[1, 0],
u8x2, i8x2,
}

#[cfg(feature = "simd_support")]
impl_swap_bytes! {
u8x4,
simd_shuffle4,
[3, 2, 1, 0],
u8x4, i8x4,
u16x2, i16x2,
}

#[cfg(feature = "simd_support")]
impl_swap_bytes! {
u8x8,
simd_shuffle8,
[7, 6, 5, 4, 3, 2, 1, 0],
u8x8, i8x8,
u16x4, i16x4,
u32x2, i32x2,
}

#[cfg(feature = "simd_support")]
impl_swap_bytes! {
u8x16,
simd_shuffle16,
[15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0],
u8x16, i8x16,
u16x8, i16x8,
u32x4, i32x4,
u64x2, i64x2,
}

#[cfg(feature = "simd_support")]
impl_swap_bytes! {
u8x32,
simd_shuffle32,
[31, 30, 29, 28, 27, 26, 25, 24, 23, 22, 21, 20, 19, 18, 17, 16, 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0],
u8x32, i8x32,
u16x16, i16x16,
u32x8, i32x8,
u64x4, i64x4,
}

#[cfg(feature = "simd_support")]
impl_swap_bytes! {
u8x64,
simd_shuffle64,
[63, 62, 61, 60, 59, 58, 57, 56, 55, 54, 53, 52, 51, 50, 49, 48, 47, 46, 45, 44, 43, 42, 41, 40, 39, 38, 37, 36, 35, 34, 33, 32, 31, 30, 29, 28, 27, 26, 25, 24, 23, 22, 21, 20, 19, 18, 17, 16, 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0],
u8x64, i8x64,
u16x32, i16x32,
u32x16, i32x16,
u64x8, i64x8,
}

/// Endian byte swapping for SIMD vectors
pub trait ToLittleEndian {
/// Converts self to little endian from the target's endianness.
///
/// On little endian this is a no-op. On big endian the bytes are swapped.
fn to_le(self) -> Self;
}

#[cfg(feature = "simd_support")]
macro_rules! impl_to_le {
($($ty:ty,)+) => ($(
impl ToLittleEndian for $ty {
fn to_le(self) -> Self {
#[cfg(target_endian = "little")]
{
self
}
#[cfg(not(target_endian = "little"))]
{
self.swap_bytes()
}
}
}
)+)
}

#[cfg(feature = "simd_support")]
impl_to_le! {
u8x2, u8x4, u8x8, u8x16, u8x32, u8x64,
i8x2, i8x4, i8x8, i8x16, i8x32, i8x64,
u16x2, u16x4, u16x8, u16x16, u16x32,
i16x2, i16x4, i16x8, i16x16, i16x32,
u32x2, u32x4, u32x8, u32x16,
i32x2, i32x4, i32x8, i32x16,
u64x2, u64x4, u64x8,
i64x2, i64x4, i64x8,
}

#[cfg(test)]
mod tests {
use super::*;
use core::mem;

// testing larger vectors is less simple
#[test]
#[cfg(feature = "simd_support")]
fn swap_bytes_128() {
let x: u128 = 0x2d99787926d46932a4c1f32680f70c55;
let expected = x.swap_bytes();

let vec: u8x16 = unsafe { mem::transmute(x) };
let actual = unsafe { mem::transmute(vec.swap_bytes()) };

assert_eq!(expected, actual);
}

#[test]
#[cfg(feature = "simd_support")]
fn swap_bytes_64() {
let x: u64 = 0x2d99787926d46932;
let expected = x.swap_bytes();

let vec: u8x8 = unsafe { mem::transmute(x) };
let actual = unsafe { mem::transmute(vec.swap_bytes()) };

assert_eq!(expected, actual);
}

#[test]
#[cfg(feature = "simd_support")]
fn swap_bytes_32() {
let x: u32 = 0x2d997872;
let expected = x.swap_bytes();

let vec: u8x4 = unsafe { mem::transmute(x) };
let actual = unsafe { mem::transmute(vec.swap_bytes()) };

assert_eq!(expected, actual);
}

#[test]
#[cfg(feature = "simd_support")]
fn swap_bytes_16() {
let x: u16 = 0x2d99;
let expected = x.swap_bytes();

let vec: u8x2 = unsafe { mem::transmute(x) };
let actual = unsafe { mem::transmute(vec.swap_bytes()) };

assert_eq!(expected, actual);
}
}

1 change: 1 addition & 0 deletions src/lib.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -234,6 +234,7 @@
#![cfg_attr(all(feature="i128_support", feature="nightly"), allow(stable_features))] // stable since 2018-03-27
#![cfg_attr(all(feature="i128_support", feature="nightly"), feature(i128_type, i128))]
#![cfg_attr(all(feature="simd_support", feature="nightly"), feature(stdsimd))]
#![cfg_attr(all(feature="simd_support", feature="nightly"), feature(platform_intrinsics))]
#![cfg_attr(feature = "stdweb", recursion_limit="128")]

#[cfg(feature="std")] extern crate std as core;
Expand Down