Skip to content

[diskann-wide] Add byte-vector reinterpretations for u32x8 and u64x8 #1445

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

Merged
juchen-ms (partychen) merged 5 commits into from
Sep 30, 2026
Merged

[diskann-wide] Add byte-vector reinterpretations for u32x8 and u64x8 #1445

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
85b9d63
[diskann-wide] Add word-to-byte SIMD reinterpretations
partychen Sep 29, 2026
b35d438
[diskann-wide] Simplify reinterpretation tests
partychen Sep 29, 2026
fce399f
[diskann-wide] Follow existing conversion test conventions
partychen Sep 29, 2026
256b507
[diskann-wide] Match existing V4 conversion test gating
partychen Sep 29, 2026
839cb7e
[diskann-wide] Test all existing SIMD reinterpretations
partychen Sep 30, 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
32 changes: 31 additions & 1 deletion diskann-wide/src/arch/x86_64/v3/conversion.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,9 @@
// x86 intrinsics
use std::arch::x86_64::*;

use super::{f16x8, f16x16, f32x8, f32x16, i8x16, i16x8, i16x16, i32x8, u8x16, u32x8};
use super::{
f16x8, f16x16, f32x8, f32x16, i8x16, i8x32, i16x8, i16x16, i32x8, u8x16, u8x32, u32x8,
};
use crate::{LoHi, SIMDCast, SIMDReinterpret, SplitJoin, helpers};

/////////////////
Expand Down Expand Up @@ -100,6 +102,18 @@ impl SIMDReinterpret<i16x16> for u32x8 {
}
}

impl SIMDReinterpret<u8x32> for u32x8 {
fn reinterpret_simd(self) -> u8x32 {
u8x32(self.0)
}
}

impl SIMDReinterpret<i8x32> for u32x8 {
fn reinterpret_simd(self) -> i8x32 {
i8x32(self.0)
}
}

impl SIMDReinterpret<u8x16> for i16x8 {
fn reinterpret_simd(self) -> u8x16 {
u8x16(self.0)
Expand All @@ -121,6 +135,22 @@ mod test_x86_conversions {
use super::*;
use crate::{arch::x86_64::V3, test_utils};

test_utils::ops::test_reinterpret!(
u32x8 => i16x16, 0x763582a914ec0bfd, V3::new_checked_uncached()
);
test_utils::ops::test_reinterpret!(
u32x8 => u8x32, 0x69b42dafe0917358, V3::new_checked_uncached()
);
test_utils::ops::test_reinterpret!(
u32x8 => i8x32, 0x69b42dafe0917358, V3::new_checked_uncached()
);
test_utils::ops::test_reinterpret!(
i16x8 => u8x16, 0x4b96d02e85f713ca, V3::new_checked_uncached()
);
test_utils::ops::test_reinterpret!(
u8x16 => i16x8, 0xd5302e6b194a87fc, V3::new_checked_uncached()
);

// Lossless Conversions
#[cfg(not(miri))]
test_utils::ops::test_lossless_convert!(
Expand Down
45 changes: 44 additions & 1 deletion diskann-wide/src/arch/x86_64/v4/conversion.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,7 +9,7 @@ use std::arch::x86_64::*;
use super::{
f16x8_::f16x8, f16x16_::f16x16, f32x8_::f32x8, f32x16_::f32x16, i8x16_::i8x16, i8x32_::i8x32,
i8x64_::i8x64, i16x8_::i16x8, i16x16_::i16x16, i16x32_::i16x32, i32x8_::i32x8, u8x16_::u8x16,
u8x32_::u8x32, u8x64_::u8x64, u32x8_::u32x8, u32x16_::u32x16,
u8x32_::u8x32, u8x64_::u8x64, u32x8_::u32x8, u32x16_::u32x16, u64x8_::u64x8,
};
use crate::{SIMDCast, SIMDReinterpret, SIMDVector, arch::x86_64::v3, helpers};

Expand Down Expand Up @@ -112,6 +112,18 @@ impl SIMDReinterpret<i8x64> for u32x16 {
}
}

impl SIMDReinterpret<u8x64> for u64x8 {
fn reinterpret_simd(self) -> u8x64 {
u8x64(self.0)
}
}

impl SIMDReinterpret<i8x64> for u64x8 {
fn reinterpret_simd(self) -> i8x64 {
i8x64(self.0)
}
}

impl SIMDReinterpret<u32x16> for u8x64 {
fn reinterpret_simd(self) -> u32x16 {
u32x16(self.0)
Expand Down Expand Up @@ -145,6 +157,37 @@ mod test_x86_conversions {
use super::*;
use crate::{arch::x86_64::V4, test_utils};

test_utils::ops::test_reinterpret!(
u32x8 => i16x16, 0x763582a914ec0bfd, V4::new_checked_uncached()
);
test_utils::ops::test_reinterpret!(
u32x16 => i16x32, 0xe96b52d0743af18c, V4::new_checked_uncached()
);
test_utils::ops::test_reinterpret!(
u32x16 => u8x64, 0x372ab891f06e5cd4, V4::new_checked_uncached()
);
test_utils::ops::test_reinterpret!(
u32x16 => i8x64, 0x951d80aefc476b23, V4::new_checked_uncached()
);
test_utils::ops::test_reinterpret!(
u64x8 => u8x64, 0x891d3c75ef2064ba, V4::new_checked_uncached()
);
test_utils::ops::test_reinterpret!(
u64x8 => i8x64, 0x891d3c75ef2064ba, V4::new_checked_uncached()
);
test_utils::ops::test_reinterpret!(
u8x64 => u32x16, 0x28fc7d6140a953be, V4::new_checked_uncached()
);
test_utils::ops::test_reinterpret!(
i8x64 => u32x16, 0xbf6308d2e471a59c, V4::new_checked_uncached()
);
test_utils::ops::test_reinterpret!(
i16x8 => u8x16, 0x4b96d02e85f713ca, V4::new_checked_uncached()
);
test_utils::ops::test_reinterpret!(
u8x16 => i16x8, 0xd5302e6b194a87fc, V4::new_checked_uncached()
);

// Lossless Conversions
#[cfg(not(miri))]
test_utils::ops::test_lossless_convert!(
Expand Down
54 changes: 54 additions & 0 deletions diskann-wide/src/emulated.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -631,6 +631,11 @@ impl_little_endian_transmute_cast!(<u8, 16> => <i16, 8>);
impl_little_endian_transmute_cast!(<u32, 16> => <u8, 64>);
impl_little_endian_transmute_cast!(<u32, 16> => <i8, 64>);

impl_little_endian_transmute_cast!(<u64, 8> => <u8, 64>);
impl_little_endian_transmute_cast!(<u64, 8> => <i8, 64>);
impl_little_endian_transmute_cast!(<u32, 8> => <u8, 32>);
impl_little_endian_transmute_cast!(<u32, 8> => <i8, 32>);

impl_little_endian_transmute_cast!(<u8, 64> => <u32, 16>);
impl_little_endian_transmute_cast!(<i8, 64> => <u32, 16>);

Expand Down Expand Up @@ -1110,6 +1115,55 @@ mod test_emulated {
Emulated<u8, 4> => Emulated<i32, 4>, 0x75929494c5d333d0, SC
);

#[cfg(target_endian = "little")]
Comment thread
partychen marked this conversation as resolved.
test_utils::ops::test_reinterpret!(
Emulated<u32, 8> => Emulated<i16, 16>, 0x763582a914ec0bfd, SC
);
#[cfg(target_endian = "little")]
test_utils::ops::test_reinterpret!(
Emulated<u32, 16> => Emulated<i16, 32>, 0xe96b52d0743af18c, SC
);
#[cfg(target_endian = "little")]
test_utils::ops::test_reinterpret!(
Emulated<i16, 8> => Emulated<u8, 16>, 0x4b96d02e85f713ca, SC
);
#[cfg(target_endian = "little")]
test_utils::ops::test_reinterpret!(
Emulated<u8, 16> => Emulated<i16, 8>, 0xd5302e6b194a87fc, SC
);
#[cfg(target_endian = "little")]
test_utils::ops::test_reinterpret!(
Emulated<u32, 16> => Emulated<u8, 64>, 0x372ab891f06e5cd4, SC
);
#[cfg(target_endian = "little")]
test_utils::ops::test_reinterpret!(
Emulated<u32, 16> => Emulated<i8, 64>, 0x951d80aefc476b23, SC
);
#[cfg(target_endian = "little")]
test_utils::ops::test_reinterpret!(
Emulated<u32, 8> => Emulated<u8, 32>, 0x69b42dafe0917358, SC
);
#[cfg(target_endian = "little")]
test_utils::ops::test_reinterpret!(
Emulated<u32, 8> => Emulated<i8, 32>, 0x69b42dafe0917358, SC
);
#[cfg(target_endian = "little")]
test_utils::ops::test_reinterpret!(
Emulated<u64, 8> => Emulated<u8, 64>, 0x891d3c75ef2064ba, SC
);
#[cfg(target_endian = "little")]
test_utils::ops::test_reinterpret!(
Emulated<u64, 8> => Emulated<i8, 64>, 0x891d3c75ef2064ba, SC
);
#[cfg(target_endian = "little")]
test_utils::ops::test_reinterpret!(
Emulated<u8, 64> => Emulated<u32, 16>, 0x28fc7d6140a953be, SC
);
#[cfg(target_endian = "little")]
test_utils::ops::test_reinterpret!(
Emulated<i8, 64> => Emulated<u32, 16>, 0xbf6308d2e471a59c, SC
);

///////////
// Casts //
///////////
Expand Down
43 changes: 43 additions & 0 deletions diskann-wide/src/test_utils/ops.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -532,6 +532,48 @@ macro_rules! test_cast {
}
}

macro_rules! test_reinterpret {
(
$from:ident $(< $($fs:tt),+ >)? => $to:ident $(< $($ts:tt),+ >)?,
$seed:literal,
$arch:expr
) => {
paste::paste! {
#[test]
fn [<reinterpret_ $from:lower $(_$($fs )x+)? _to_ $to:lower $(_$($ts )x+)?>]() {
use $crate::{SIMDReinterpret, SIMDVector};

type From = $from $(< $($fs),+>)?;
type To = $to $(< $($ts),+>)?;

if let Some(arch) = $arch {
let f = move |input: &[<From as SIMDVector>::Scalar]| {
let got: To = From::from_array(arch, input.try_into().unwrap())
.reinterpret_simd();
assert_eq!(
bytemuck::must_cast_slice::<_, u8>(&got.to_array()),
bytemuck::must_cast_slice::<_, u8>(input),
"input: {input:?}",
);
};

// Distinct byte positions detect lane/byte reordering.
let mut input = [<From as SIMDVector>::Scalar::default(); From::LANES];
for start in [0u8, 0x7f, 0x80, 0xff] {
for (i, byte) in bytemuck::must_cast_slice_mut::<_, u8>(&mut input)
.iter_mut().enumerate()
{
*byte = start.wrapping_add(i as u8);
}
f(&input);
}
$crate::test_utils::driver::drive_unary(&f, From::LANES, $seed);
}
}
}
};
}

macro_rules! test_abs {
($wide:ident $(< $($ps:tt),+ >)?, $seed:literal, $arch:expr) => {
paste::paste! {
Expand Down Expand Up @@ -1146,6 +1188,7 @@ pub(crate) use test_lossless_convert;
pub(crate) use test_minmax;
pub(crate) use test_mul;
pub(crate) use test_popcount;
pub(crate) use test_reinterpret;
pub(crate) use test_select;
pub(crate) use test_splitjoin;
pub(crate) use test_sub;
Expand Down
Loading