Skip to content

Use interleave() instead of zip_low()/zip_high() for better perf on AVX2 #116

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

Draft
Shnatsel wants to merge 3 commits into
base: main
Choose a base branch
from
Draft

Use interleave() instead of zip_low()/zip_high() for better perf on AVX2 #116

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
b6e9ab8
Revert "Polyfill interleave() until the upstream fearless_simd PR is …
Shnatsel Apr 18, 2026
c69b5f6
Repoint to fearless_simd main now that my PR is merged
Shnatsel Apr 18, 2026
b53f448
Use interleave() instead of zip_low()/zip_high() in BRAVO
Shnatsel Apr 18, 2026
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
3 changes: 1 addition & 2 deletions Cargo.lock
View file
Open in desktop

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

5 changes: 4 additions & 1 deletion Cargo.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -61,4 +61,7 @@ debug = true
[package.metadata.docs.rs]
all-features = true

[lints.rust]
[lints.rust]

[patch.crates-io]
fearless_simd = {git = "https://github.com/linebender/fearless_simd"}
6 changes: 2 additions & 4 deletions benches/bit_reversal.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -89,8 +89,7 @@ mod old_bravo {
let idx1 = i + offset + stride;
let vec0 = chunks_a[idx0];
let vec1 = chunks_a[idx1];
chunks_a[idx0] = vec0.zip_low(vec1);
chunks_a[idx1] = vec0.zip_high(vec1);
(chunks_a[idx0], chunks_a[idx1]) = vec0.interleave(vec1);
}
i += stride * 2;
}
Expand All @@ -117,8 +116,7 @@ mod old_bravo {
let idx1 = i + offset + stride;
let vec0 = chunks_b[idx0];
let vec1 = chunks_b[idx1];
chunks_b[idx0] = vec0.zip_low(vec1);
chunks_b[idx1] = vec0.zip_high(vec1);
(chunks_b[idx0], chunks_b[idx1]) = vec0.interleave(vec1);
}
i += stride * 2;
}
Expand Down
6 changes: 2 additions & 4 deletions src/algorithms/bravo.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -137,8 +137,7 @@ macro_rules! impl_bit_rev_bravo {
let idx1 = i + offset + stride;
let vec0 = chunks_a[idx0];
let vec1 = chunks_a[idx1];
chunks_a[idx0] = vec0.zip_low(vec1);
chunks_a[idx1] = vec0.zip_high(vec1);
(chunks_a[idx0], chunks_a[idx1]) = vec0.interleave(vec1);
}
i += stride * 2;
}
Expand Down Expand Up @@ -167,8 +166,7 @@ macro_rules! impl_bit_rev_bravo {
let idx1 = i + offset + stride;
let vec0 = chunks_b[idx0];
let vec1 = chunks_b[idx1];
chunks_b[idx0] = vec0.zip_low(vec1);
chunks_b[idx1] = vec0.zip_high(vec1);
(chunks_b[idx0], chunks_b[idx1]) = vec0.interleave(vec1);
}
i += stride * 2;
}
Expand Down
30 changes: 8 additions & 22 deletions src/kernels/codelets.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,20 +7,6 @@ use fearless_simd::{
f32x4, f32x8, f64x4, Simd, SimdBase, SimdCombine, SimdFloat, SimdFrom, SimdSplit,
};

/// Equivalent to `a.interleave(b)` — returns `(a.zip_low(b), a.zip_high(b))`.
/// Slow polyfill for <https://github.com/linebender/fearless_simd/pull/206>
#[inline(always)]
fn interleave_f64x4<S: Simd>(a: f64x4<S>, b: f64x4<S>) -> (f64x4<S>, f64x4<S>) {
(a.zip_low(b), a.zip_high(b))
}

/// Equivalent to `a.interleave(b)` — returns `(a.zip_low(b), a.zip_high(b))`.
/// Slow polyfill for <https://github.com/linebender/fearless_simd/pull/206>
#[inline(always)]
fn interleave_f32x4<S: Simd>(a: f32x4<S>, b: f32x4<S>) -> (f32x4<S>, f32x4<S>) {
(a.zip_low(b), a.zip_high(b))
}

/// FFT-16 codelet for `f64`: executes stages 0-3 (chunk_size 2 through 16) in a single function.
///
/// Register-resident implementation: all 16 complex values are loaded into f64x4 vectors,
Expand All @@ -47,10 +33,10 @@ fn fft_dit_codelet_16_simd_f64<S: Simd>(simd: S, reals: &mut [f64], imags: &mut
for (re, im) in reals.chunks_exact_mut(16).zip(imags.chunks_exact_mut(16)) {
macro_rules! transpose4x4_f64 {
($g0:expr, $g1:expr, $g2:expr, $g3:expr) => {{
let (t0, t1) = interleave_f64x4($g0, $g2);
let (t2, t3) = interleave_f64x4($g1, $g3);
let (r0, r1) = interleave_f64x4(t0, t2);
let (r2, r3) = interleave_f64x4(t1, t3);
let (t0, t1) = $g0.interleave($g2);
let (t2, t3) = $g1.interleave($g3);
let (r0, r1) = t0.interleave(t2);
let (r2, r3) = t1.interleave(t3);
(r0, r1, r2, r3)
}};
}
Expand Down Expand Up @@ -250,10 +236,10 @@ fn fft_dit_codelet_32_simd_f32<S: Simd>(simd: S, reals: &mut [f32], imags: &mut
{
macro_rules! transpose4x4 {
($g0:expr, $g1:expr, $g2:expr, $g3:expr) => {{
let (t0, t1) = interleave_f32x4($g0, $g2);
let (t2, t3) = interleave_f32x4($g1, $g3);
let (r0, r1) = interleave_f32x4(t0, t2);
let (r2, r3) = interleave_f32x4(t1, t3);
let (t0, t1) = $g0.interleave($g2);
let (t2, t3) = $g1.interleave($g3);
let (r0, r1) = t0.interleave(t2);
let (r2, r3) = t1.interleave(t3);
(r0, r1, r2, r3)
}};
}
Expand Down
Loading