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.

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

Unroll position_simd #13

Merged
merged 4 commits into from
Jul 28, 2024
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
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
1 change: 1 addition & 0 deletions src/lib.rs
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,7 @@
#![feature(is_sorted)]
#![feature(sort_floats)]

pub const UNROLL_FACTOR: usize = 4;
pub const SIMD_LEN: usize = 32;

mod all_equal;
Expand Down
43 changes: 33 additions & 10 deletions src/position.rs
Original file line number Diff line number Diff line change
@@ -1,4 +1,5 @@
use crate::SIMD_LEN;
use crate::UNROLL_FACTOR;
use std::simd::cmp::SimdPartialEq;
use std::simd::Mask;
use std::simd::{Simd, SimdElement};
Expand Down Expand Up @@ -26,16 +27,38 @@ where
}
// SIMD
let simd_needle = Simd::splat(needle);
for (chunk_idx, chunk) in simd_data.iter().enumerate() {
let mut unrolled_loops = 0;
// Unrolled loops
let mut chunks_iter = simd_data.chunks_exact(UNROLL_FACTOR);
for chunks in chunks_iter.by_ref() {
let mut mask = Mask::default();
for chunk in chunks {
mask |= chunk.simd_eq(simd_needle);
}
if mask.any() {
for (mask_idx, c) in chunks.iter().enumerate() {
let mask = c.simd_eq(simd_needle);
if mask.any() {
return Some(
prefix.len()
+ (unrolled_loops * (SIMD_LEN * UNROLL_FACTOR)) // Full outer loops
+ mask_idx * SIMD_LEN // nth inner loop
+ mask.to_bitmask().trailing_zeros() as usize, // nth element in matching mask
);
}
}
}
unrolled_loops += 1;
}
// Remaining simd loops that where not divisible by UNROLL_FACTOR
for (idx, chunk) in chunks_iter.remainder().iter().enumerate() {
let mask = chunk.simd_eq(simd_needle).to_bitmask();
if mask != 0 {
// Example:
// needle = 10
// prefix = [1,2,3]
// SIMD = [[4,5,6,7], [8,9,10,11]]
// 3 + (1 * 4) + (trailing_zeros(0b0010) == 2) = 9
return Some(
prefix.len() + (chunk_idx * SIMD_LEN) + (mask.trailing_zeros() as usize),
prefix.len()
+ (unrolled_loops * UNROLL_FACTOR * SIMD_LEN)
+ (idx * SIMD_LEN)
+ (mask.trailing_zeros() as usize),
);
}
}
Expand Down Expand Up @@ -67,14 +90,13 @@ mod tests {
Simd<T, SIMD_LEN>: SimdPartialEq<Mask = Mask<T::Mask, SIMD_LEN>>,
Standard: Distribution<T>,
{
for len in 0..100 {
for _ in 0..5 {
for len in 0..500 {
for _ in 0..200 {
let mut v: Vec<T> = vec![T::default(); len];
let mut rng = rand::thread_rng();
for x in v.iter_mut() {
*x = rng.gen()
}

let needle = match rng.gen_bool(0.5) {
true => v.choose(&mut rng).cloned().unwrap_or(T::default()),
false => loop {
Expand All @@ -84,6 +106,7 @@ mod tests {
}
},
};

let ans = v.iter().position_simd(needle);
let correct = v.iter().position(|x| *x == needle);

Expand Down