Crate simd_aligned
source ·Expand description
NOTE - Do not use this crate for now. It has been reactivated to make FFSVM compile again, but needs some architectural work.
In One Sentence
You want to use std::simd
but realized there is no simple, safe and fast way to align your f32x8
(and friends) in memory and treat them as regular f32
slices for easy loading and manipulation; simd_aligned
to the rescue.
Highlights
- built on top of
std::simd
for easy data handling - supports everything from
u8x2
tof64x8
- think in flat slices (
&[f32]
), but get performance of properly aligned SIMD vectors (&[f32x16]
) - defines
u8s
, …,f36s
as “best guess” for current platform (WIP) - provides N-dimensional
VectorD
and NxM-dimensionalMatrixD
.
Note: Right now this is an experimental crate. Features might be added or removed depending on how std::simd
evolves. At the end of the day it’s just about being able to load and manipulate data without much fuzz.
Examples
Produces a vector that can hold 10
elements of type f64
. Might internally
allocate 5
elements of type f64x2
, or 3
of type f64x4
, depending on the platform.
All elements are guaranteed to be properly aligned for fast access.
#![feature(portable_simd)]
use std::simd::*;
use simd_aligned::*;
// Create vectors of `10` f64 elements with value `0.0`.
let mut v1 = VectorD::<f64s>::with(0.0, 10);
let mut v2 = VectorD::<f64s>::with(0.0, 10);
// Get "flat", mutable view of the vector, and set individual elements:
let v1_m = v1.flat_mut();
let v2_m = v2.flat_mut();
// Set some elements on v1
v1_m[0] = 0.0;
v1_m[4] = 4.0;
v1_m[8] = 8.0;
// Set some others on v2
v2_m[1] = 0.0;
v2_m[5] = 5.0;
v2_m[9] = 9.0;
let mut sum = f64s::splat(0.0);
// Eventually, do something with the actual SIMD types. Does
// `std::simd` vector math, e.g., f64x8 + f64x8 in one operation:
sum = v1[0] + v2[0];
Benchmarks
There is no performance penalty for using simd_aligned
, while retaining all the
simplicity of handling flat arrays.
test vectors::packed ... bench: 77 ns/iter (+/- 4)
test vectors::scalar ... bench: 1,177 ns/iter (+/- 464)
test vectors::simd_aligned ... bench: 71 ns/iter (+/- 5)
FAQ
How does it relate to faster and std::simd
?
-
simd_aligned
builds on top ofstd::simd
. At aims to provide common, SIMD-aligned data structure that support simple and safe scalar access patterns. -
faster
(as of today) is really good if you already have exiting flat slices in your code and want operate them “full SIMD ahead”. However, in particular when dealing with multiple slices at the same time (e.g., kernel computations) the performance impact of unaligned arrays can become a bit more noticeable (e.g., in the case of ffsvm up to 10% - 20%).
Re-exports
pub use crate::arch::current::*;
Modules
- Contains vector definitions with a fixed bit width.
- Unified views on SIMD types.
Macros
- simd_swizzleExperimentalConstructs a new SIMD vector by copying elements from selected lanes in other vectors.
Structs
- LaneCountExperimentalSpecifies the number of lanes in a SIMD vector as a type.
- MaskExperimentalA SIMD vector mask for
LANES
elements of width specified byElement
. - A dynamic (heap allocated) matrix with one axis aligned for fast and safe SIMD access that also provides a flat view on its data.
- Produced by
MatrixD::flat
, this allow for flat matrix access. - Provided by
MatrixD::flat_mut
, this allow for flat, mutable matrix access. - SimdExperimentalA SIMD vector of
LANES
elements of typeT
.Simd<T, N>
has the same shape as[T; N]
, but operates likeT
. - A dynamic (heap allocated) vector aligned for fast and safe SIMD access that also provides a flat view on its data.
Enums
- WhichExperimentalSpecifies a lane index into one of two SIMD vectors.
Traits
- MaskElementExperimentalMarker trait for types that may be used as SIMD mask elements.
- SimdElementExperimentalMarker trait for types that may be used as SIMD vector elements.
- SimdFloatExperimentalOperations on SIMD vectors of floats.
- SimdIntExperimentalOperations on SIMD vectors of signed integers.
- SimdOrdExperimentalParallel
Ord
. - SimdPartialEqExperimentalParallel
PartialEq
. - SimdPartialOrdExperimentalParallel
PartialOrd
. - SimdUintExperimentalOperations on SIMD vectors of unsigned integers.
- StdFloatExperimentalThis trait provides a possibly-temporary implementation of float functions that may, in the absence of hardware support, canonicalize to calling an operating system’s
math.h
dynamically-loaded library (also known as a shared object). As these conditionally require runtime support, they should only appear in binaries built assuming OS support:std
. - SupportedLaneCountExperimentalStatically guarantees that a lane count is marked as supported.
- SwizzleExperimentalCreate a vector from the elements of another vector.
- Swizzle2ExperimentalCreate a vector from the elements of two other vectors.
- ToBitMaskExperimentalConverts masks to and from integer bitmasks.
Functions
- Converts an slice of SIMD vectors into a flat slice of elements.
- Converts a mutable slice of SIMD vectors into a flat slice of elements.
Type Definitions
- f32x2ExperimentalA 64-bit SIMD vector with two elements of type
f32
. - f32x4ExperimentalA 128-bit SIMD vector with four elements of type
f32
. - f32x8ExperimentalA 256-bit SIMD vector with eight elements of type
f32
. - f32x16ExperimentalA 512-bit SIMD vector with 16 elements of type
f32
. - f64x2ExperimentalA 128-bit SIMD vector with two elements of type
f64
. - f64x4ExperimentalA 256-bit SIMD vector with four elements of type
f64
. - f64x8ExperimentalA 512-bit SIMD vector with eight elements of type
f64
. - i8x4ExperimentalA 32-bit SIMD vector with four elements of type
i8
. - i8x8ExperimentalA 64-bit SIMD vector with eight elements of type
i8
. - i8x16ExperimentalA 128-bit SIMD vector with 16 elements of type
i8
. - i8x32ExperimentalA 256-bit SIMD vector with 32 elements of type
i8
. - i8x64ExperimentalA 512-bit SIMD vector with 64 elements of type
i8
. - i16x2ExperimentalA 32-bit SIMD vector with two elements of type
i16
. - i16x4ExperimentalA 64-bit SIMD vector with four elements of type
i16
. - i16x8ExperimentalA 128-bit SIMD vector with eight elements of type
i16
. - i16x16ExperimentalA 256-bit SIMD vector with 16 elements of type
i16
. - i16x32ExperimentalA 512-bit SIMD vector with 32 elements of type
i16
. - i32x2ExperimentalA 64-bit SIMD vector with two elements of type
i32
. - i32x4ExperimentalA 128-bit SIMD vector with four elements of type
i32
. - i32x8ExperimentalA 256-bit SIMD vector with eight elements of type
i32
. - i32x16ExperimentalA 512-bit SIMD vector with 16 elements of type
i32
. - i64x2ExperimentalA 128-bit SIMD vector with two elements of type
i64
. - i64x4ExperimentalA 256-bit SIMD vector with four elements of type
i64
. - i64x8ExperimentalA 512-bit SIMD vector with eight elements of type
i64
. - isizex2ExperimentalA SIMD vector with two elements of type
isize
. - isizex4ExperimentalA SIMD vector with four elements of type
isize
. - isizex8ExperimentalA SIMD vector with eight elements of type
isize
. - mask8x8ExperimentalA mask for SIMD vectors with eight elements of 8 bits.
- mask8x16ExperimentalA mask for SIMD vectors with 16 elements of 8 bits.
- mask8x32ExperimentalA mask for SIMD vectors with 32 elements of 8 bits.
- mask8x64ExperimentalA mask for SIMD vectors with 64 elements of 8 bits.
- mask16x4ExperimentalA mask for SIMD vectors with four elements of 16 bits.
- mask16x8ExperimentalA mask for SIMD vectors with eight elements of 16 bits.
- mask16x16ExperimentalA mask for SIMD vectors with 16 elements of 16 bits.
- mask16x32ExperimentalA mask for SIMD vectors with 32 elements of 16 bits.
- mask32x2ExperimentalA mask for SIMD vectors with two elements of 32 bits.
- mask32x4ExperimentalA mask for SIMD vectors with four elements of 32 bits.
- mask32x8ExperimentalA mask for SIMD vectors with eight elements of 32 bits.
- mask32x16ExperimentalA mask for SIMD vectors with 16 elements of 32 bits.
- mask64x2ExperimentalA mask for SIMD vectors with two elements of 64 bits.
- mask64x4ExperimentalA mask for SIMD vectors with four elements of 64 bits.
- mask64x8ExperimentalA mask for SIMD vectors with eight elements of 64 bits.
- masksizex2ExperimentalA mask for SIMD vectors with two elements of pointer width.
- masksizex4ExperimentalA mask for SIMD vectors with four elements of pointer width.
- masksizex8ExperimentalA mask for SIMD vectors with eight elements of pointer width.
- u8x4ExperimentalA 32-bit SIMD vector with four elements of type
u8
. - u8x8ExperimentalA 64-bit SIMD vector with eight elements of type
u8
. - u8x16ExperimentalA 128-bit SIMD vector with 16 elements of type
u8
. - u8x32ExperimentalA 256-bit SIMD vector with 32 elements of type
u8
. - u8x64ExperimentalA 512-bit SIMD vector with 64 elements of type
u8
. - u16x2ExperimentalA 32-bit SIMD vector with two elements of type
u16
. - u16x4ExperimentalA 64-bit SIMD vector with four elements of type
u16
. - u16x8ExperimentalA 128-bit SIMD vector with eight elements of type
u16
. - u16x16ExperimentalA 256-bit SIMD vector with 16 elements of type
u16
. - u16x32ExperimentalA 512-bit SIMD vector with 32 elements of type
u16
. - u32x2ExperimentalA 64-bit SIMD vector with two elements of type
u32
. - u32x4ExperimentalA 128-bit SIMD vector with four elements of type
u32
. - u32x8ExperimentalA 256-bit SIMD vector with eight elements of type
u32
. - u32x16ExperimentalA 512-bit SIMD vector with 16 elements of type
u32
. - u64x2ExperimentalA 128-bit SIMD vector with two elements of type
u64
. - u64x4ExperimentalA 256-bit SIMD vector with four elements of type
u64
. - u64x8ExperimentalA 512-bit SIMD vector with eight elements of type
u64
. - usizex2ExperimentalA SIMD vector with two elements of type
usize
. - usizex4ExperimentalA SIMD vector with four elements of type
usize
. - usizex8ExperimentalA SIMD vector with eight elements of type
usize
.