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

Change to import std::simd in the global namespace. #32

Merged
merged 1 commit into from
Sep 28, 2023
Merged

Change to import std::simd in the global namespace. #32

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
21 changes: 10 additions & 11 deletions src/coding.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,6 +17,7 @@
use std::cell::RefCell;
#[cfg(feature = "experimental")]
use std::collections::BTreeSet;
use std::simd;
#[cfg(feature = "experimental")]
use std::sync::Arc;

Expand Down Expand Up @@ -93,15 +94,13 @@ pub fn encode_residual(config: &config::Prc, errors: &[i32], warmup_length: usiz
}

/// Pack scalars into `Vec` of `Simd`s.
pub fn pack_into_simd_vec<T, const LANES: usize>(
src: &[T],
dest: &mut Vec<std::simd::Simd<T, LANES>>,
) where
T: std::simd::SimdElement + From<i8>,
std::simd::LaneCount<LANES>: std::simd::SupportedLaneCount,
pub fn pack_into_simd_vec<T, const LANES: usize>(src: &[T], dest: &mut Vec<simd::Simd<T, LANES>>)
where
T: simd::SimdElement + From<i8>,
simd::LaneCount<LANES>: simd::SupportedLaneCount,
{
dest.clear();
let mut v = std::simd::Simd::<T, LANES>::splat(0i8.into());
let mut v = simd::Simd::<T, LANES>::splat(0i8.into());
for slice in src.chunks(LANES) {
if slice.len() < LANES {
v.as_mut_array()[0..slice.len()].copy_from_slice(slice);
Expand All @@ -114,10 +113,10 @@ pub fn pack_into_simd_vec<T, const LANES: usize>(
}

/// Unpack slice of `Simd` into `Vec` of elements.
pub fn unpack_simds<T, const LANES: usize>(src: &[std::simd::Simd<T, LANES>], dest: &mut Vec<T>)
pub fn unpack_simds<T, const LANES: usize>(src: &[simd::Simd<T, LANES>], dest: &mut Vec<T>)
where
T: std::simd::SimdElement + From<i8>,
std::simd::LaneCount<LANES>: std::simd::SupportedLaneCount,
T: simd::SimdElement + From<i8>,
simd::LaneCount<LANES>: simd::SupportedLaneCount,
{
dest.resize(src.len() * LANES, 0i8.into());
let mut offset = 0;
Expand All @@ -130,7 +129,7 @@ where

/// Helper struct holding working memory for fixed LPC.
struct FixedLpcEncoder {
errors: Vec<std::simd::i32x16>,
errors: Vec<simd::i32x16>,
/// Length of errors in the number of samples (scalars).
error_len_in_samples: usize,
/// Temporary buffer for unpacked error signal.
Expand Down
27 changes: 15 additions & 12 deletions src/lpc.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,7 +17,6 @@
use std::cell::RefCell;
use std::collections::BTreeMap;
use std::rc::Rc;
use std::simd::SimdInt;

use serde::Deserialize;
use serde::Serialize;
Expand All @@ -26,6 +25,10 @@ use super::constant::MAX_LPC_ORDER;
use super::constant::QLPC_MAX_SHIFT;
use super::constant::QLPC_MIN_SHIFT;

use std::simd;

use simd::SimdInt;

/// Analysis window descriptor.
///
/// This enum is `Serializable` and `Deserializable` because this will be
Expand Down Expand Up @@ -136,17 +139,17 @@ fn dequantize_parameter(coef: i16, shift: i8) -> f32 {
const QLPC_SIMD_LANES: usize = 16usize;
const MAX_COEF_VECTORS: usize = (MAX_LPC_ORDER + (QLPC_SIMD_LANES - 1)) / QLPC_SIMD_LANES;

const LOW_WORD_MASK: std::simd::i32x16 = std::simd::i32x16::from_array([0x0000_FFFFi32; 16]);
const LOW_WORD_DENOM: std::simd::i32x16 = std::simd::i32x16::from_array([0x0001_0000i32; 16]);
const LOW_WORD_MASK: simd::i32x16 = simd::i32x16::from_array([0x0000_FFFFi32; 16]);
const LOW_WORD_DENOM: simd::i32x16 = simd::i32x16::from_array([0x0001_0000i32; 16]);
#[allow(dead_code)]
const HIGH_WORD_SHIFT: std::simd::i32x16 = std::simd::i32x16::from_array([16i32; 16]);
const HIGH_WORD_SHIFT: simd::i32x16 = simd::i32x16::from_array([16i32; 16]);

/// Shifts elements in a vector of `T` represented as a slice of `Simd<T, N>`.
#[inline]
fn shift_lanes_right<T, const N: usize>(val: T, vecs: &mut [std::simd::Simd<T, N>])
fn shift_lanes_right<T, const N: usize>(val: T, vecs: &mut [simd::Simd<T, N>])
where
T: std::simd::SimdElement,
std::simd::LaneCount<N>: std::simd::SupportedLaneCount,
T: simd::SimdElement,
simd::LaneCount<N>: simd::SupportedLaneCount,
{
let mut carry = val;
for v in vecs {
Expand All @@ -159,7 +162,7 @@ where
/// Quantized LPC coefficients.
#[derive(Clone, Debug)]
pub struct QuantizedParameters {
coefs: heapless::Vec<std::simd::i32x16, MAX_COEF_VECTORS>,
coefs: heapless::Vec<simd::i32x16, MAX_COEF_VECTORS>,
order: usize,
shift: i8,
precision: usize,
Expand Down Expand Up @@ -198,7 +201,7 @@ impl QuantizedParameters {

let mut coefs_v = heapless::Vec::new();
for arr in q_coefs.chunks(QLPC_SIMD_LANES) {
let mut v = std::simd::i32x16::splat(0);
let mut v = simd::i32x16::splat(0);
v.as_mut_array()[0..arr.len()].copy_from_slice(arr);
coefs_v
.push(v)
Expand Down Expand Up @@ -228,15 +231,15 @@ impl QuantizedParameters {
for p in errors.iter_mut().take(self.order()) {
*p = 0;
}
let mut window_h = heapless::Vec::<std::simd::i32x16, MAX_COEF_VECTORS>::new();
let mut window_l = heapless::Vec::<std::simd::i32x16, MAX_COEF_VECTORS>::new();
let mut window_h = heapless::Vec::<simd::i32x16, MAX_COEF_VECTORS>::new();
let mut window_l = heapless::Vec::<simd::i32x16, MAX_COEF_VECTORS>::new();

for i in 0..MAX_COEF_VECTORS {
let tau: isize = (self.order() as isize - 1) - (i * QLPC_SIMD_LANES) as isize;
if tau < 0 {
break;
}
let mut v = std::simd::i32x16::splat(0);
let mut v = simd::i32x16::splat(0);
for j in 0..QLPC_SIMD_LANES {
let j = j as isize;
if tau - j < 0 {
Expand Down
41 changes: 22 additions & 19 deletions src/rice.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,42 +15,45 @@
//! Functions for partitioned rice coding (PRC).

use std::cell::RefCell;
use std::simd::SimdPartialEq;
use std::simd::SimdPartialOrd;
use std::simd::SimdUint;

use super::constant::MAX_RICE_PARAMETER;
use super::constant::MAX_RICE_PARTITIONS;
use super::constant::MAX_RICE_PARTITION_ORDER;
use super::constant::MIN_RICE_PARTITION_SIZE;

use std::simd;

use simd::SimdPartialEq;
use simd::SimdPartialOrd;
use simd::SimdUint;

/// Table that contains the numbers of bits needed for a partition.
#[derive(Clone, Debug, PartialEq, PartialOrd)]
struct PrcBitTable {
p_to_bits: std::simd::u32x16,
mask: std::simd::Mask<<u32 as std::simd::SimdElement>::Mask, 16>,
p_to_bits: simd::u32x16,
mask: simd::Mask<<u32 as simd::SimdElement>::Mask, 16>,
}

static ZEROS: std::simd::u32x16 = std::simd::u32x16::from_array([0u32; 16]);
static INDEX: std::simd::u32x16 =
std::simd::u32x16::from_array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]);
static INDEX1: std::simd::u32x16 =
std::simd::u32x16::from_array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16]);
static MAXES: std::simd::u32x16 = std::simd::u32x16::from_array([u32::MAX; 16]);
static ZEROS: simd::u32x16 = simd::u32x16::from_array([0u32; 16]);
static INDEX: simd::u32x16 =
simd::u32x16::from_array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]);
static INDEX1: simd::u32x16 =
simd::u32x16::from_array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16]);
static MAXES: simd::u32x16 = simd::u32x16::from_array([u32::MAX; 16]);

// max value of p_to_bits is chosen so that the estimates doesn't overflow
// after added 2^4 = 16 times at maximum.
// The current version exploits the fact that `MAX_P_TO_BITS` is actually a bit mask, i.e.
// can be written as 2^N - 1 for faster processing. Do not use arbitrary value here.
static MAX_P_TO_BITS: u32 = (1 << 28) - 1;
static MAX_P_TO_BITS_VEC: std::simd::u32x16 = std::simd::u32x16::from_array([MAX_P_TO_BITS; 16]);
static MAX_P_TO_BITS_VEC: simd::u32x16 = simd::u32x16::from_array([MAX_P_TO_BITS; 16]);

impl PrcBitTable {
pub fn zero(max_p: usize) -> Self {
debug_assert!(max_p <= MAX_RICE_PARAMETER);
Self {
p_to_bits: ZEROS,
mask: INDEX.simd_le(std::simd::u32x16::splat(max_p as u32)),
mask: INDEX.simd_le(simd::u32x16::splat(max_p as u32)),
}
}

Expand All @@ -63,18 +66,18 @@ impl PrcBitTable {
/// Initializes PRC bit table from the error signal.
#[allow(unused_assignments, clippy::identity_op)]
fn init_with_errors(&mut self, errors: &[u32], offset: usize) {
let mut p_to_bits = std::simd::u32x16::splat(offset as u32)
+ std::simd::u32x16::splat(errors.len() as u32) * INDEX1;
let mut p_to_bits =
simd::u32x16::splat(offset as u32) + simd::u32x16::splat(errors.len() as u32) * INDEX1;

for v in errors {
// Below is faster than doing:
// vs = splat(*v) >> INDEX;
// or
// vs = std::simd::u32x16::from_array(std::array::from_fn(
// vs = simd::u32x16::from_array(std::array::from_fn(
// |i| v >> i));
// Perhaps due to smaller memory footprint by avoiding `splat`?
let v = *v;
let vs = std::simd::u32x16::from_array([
let vs = simd::u32x16::from_array([
v,
v >> 1,
v >> 2,
Expand Down Expand Up @@ -107,7 +110,7 @@ impl PrcBitTable {
let ret_bits = self.mask.select(self.p_to_bits, MAXES).reduce_min();
let ret_p = self
.p_to_bits
.simd_eq(std::simd::u32x16::splat(ret_bits))
.simd_eq(simd::u32x16::splat(ret_bits))
.select(INDEX, ZEROS)
.reduce_max();

Expand All @@ -117,7 +120,7 @@ impl PrcBitTable {
#[allow(unused_comparisons)]
#[inline]
pub fn merge(&self, other: &Self, offset: usize) -> Self {
let offset = std::simd::u32x16::splat(offset as u32);
let offset = simd::u32x16::splat(offset as u32);
let offset = self.mask.select(offset, ZEROS);
Self {
p_to_bits: self.p_to_bits + other.p_to_bits - offset,
Expand Down
Loading