Skip to content

feat(be-tier11): BE support for Gray9-16 / Ya16 / Grayf32 row kernels #85

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

Open
uqio wants to merge 2 commits into
base: main
Choose a base branch
from
Open

feat(be-tier11): BE support for Gray9-16 / Ya16 / Grayf32 row kernels #85

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
5baf9e3
feat(be-tier11): BE support for Gray9-16/Ya16/Grayf32 row kernels
uqio May 7, 2026
ebefb6f
fix(be-tier11): make scalar BE conversion target-endian aware
uqio May 7, 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
453 changes: 317 additions & 136 deletions src/row/arch/neon/gray.rs
View file
Open in desktop

Large diffs are not rendered by default.

313 changes: 204 additions & 109 deletions src/row/arch/wasm_simd128/gray.rs
View file
Open in desktop

Large diffs are not rendered by default.

353 changes: 235 additions & 118 deletions src/row/arch/x86_avx2/gray.rs
View file
Open in desktop

Large diffs are not rendered by default.

349 changes: 233 additions & 116 deletions src/row/arch/x86_avx512/gray.rs
View file
Open in desktop

Large diffs are not rendered by default.

377 changes: 261 additions & 116 deletions src/row/arch/x86_sse41/gray.rs
View file
Open in desktop

Large diffs are not rendered by default.

229 changes: 117 additions & 112 deletions src/row/dispatch/gray.rs
View file
Open in desktop

Large diffs are not rendered by default.

140 changes: 80 additions & 60 deletions src/row/dispatch/grayf32.rs
View file
Open in desktop

Large diffs are not rendered by default.

142 changes: 86 additions & 56 deletions src/row/dispatch/ya16.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,259 +27,289 @@ use crate::row::{

/// Dispatch `ya16_to_rgb_row`.
#[cfg_attr(not(tarpaulin), inline(always))]
pub(crate) fn ya16_to_rgb_row(packed: &[u16], out: &mut [u8], width: usize, use_simd: bool) {
pub(crate) fn ya16_to_rgb_row<const BE: bool>(
packed: &[u16],
out: &mut [u8],
width: usize,
use_simd: bool,
) {
assert!(packed.len() >= ya_row_elems(width), "packed too short");
assert!(out.len() >= rgb_row_bytes(width), "out too short");
if !use_simd {
return scalar::ya16_to_rgb_row(packed, out, width);
return scalar::ya16_to_rgb_row::<BE>(packed, out, width);
}
cfg_select! {
target_arch = "aarch64" => {
if neon_available() {
unsafe { arch::neon::ya16_to_rgb_row(packed, out, width); }
unsafe { arch::neon::ya16_to_rgb_row::<BE>(packed, out, width); }
return;
}
},
target_arch = "x86_64" => {
if avx512_available() {
unsafe { arch::x86_avx512::ya16_to_rgb_row(packed, out, width); }
unsafe { arch::x86_avx512::ya16_to_rgb_row::<BE>(packed, out, width); }
return;
}
if avx2_available() {
unsafe { arch::x86_avx2::ya16_to_rgb_row(packed, out, width); }
unsafe { arch::x86_avx2::ya16_to_rgb_row::<BE>(packed, out, width); }
return;
}
if sse41_available() {
unsafe { arch::x86_sse41::ya16_to_rgb_row(packed, out, width); }
unsafe { arch::x86_sse41::ya16_to_rgb_row::<BE>(packed, out, width); }
return;
}
},
target_arch = "wasm32" => {
if simd128_available() {
unsafe { arch::wasm_simd128::ya16_to_rgb_row(packed, out, width); }
unsafe { arch::wasm_simd128::ya16_to_rgb_row::<BE>(packed, out, width); }
return;
}
},
_ => {}
}
scalar::ya16_to_rgb_row(packed, out, width);
scalar::ya16_to_rgb_row::<BE>(packed, out, width);
}

// ---- ya16_to_rgba_row ---------------------------------------------------------

/// Dispatch `ya16_to_rgba_row`.
#[cfg_attr(not(tarpaulin), inline(always))]
pub(crate) fn ya16_to_rgba_row(packed: &[u16], out: &mut [u8], width: usize, use_simd: bool) {
pub(crate) fn ya16_to_rgba_row<const BE: bool>(
packed: &[u16],
out: &mut [u8],
width: usize,
use_simd: bool,
) {
assert!(packed.len() >= ya_row_elems(width), "packed too short");
assert!(out.len() >= rgba_row_bytes(width), "out too short");
if !use_simd {
return scalar::ya16_to_rgba_row(packed, out, width);
return scalar::ya16_to_rgba_row::<BE>(packed, out, width);
}
cfg_select! {
target_arch = "aarch64" => {
if neon_available() {
unsafe { arch::neon::ya16_to_rgba_row(packed, out, width); }
unsafe { arch::neon::ya16_to_rgba_row::<BE>(packed, out, width); }
return;
}
},
target_arch = "x86_64" => {
if avx512_available() {
unsafe { arch::x86_avx512::ya16_to_rgba_row(packed, out, width); }
unsafe { arch::x86_avx512::ya16_to_rgba_row::<BE>(packed, out, width); }
return;
}
if avx2_available() {
unsafe { arch::x86_avx2::ya16_to_rgba_row(packed, out, width); }
unsafe { arch::x86_avx2::ya16_to_rgba_row::<BE>(packed, out, width); }
return;
}
if sse41_available() {
unsafe { arch::x86_sse41::ya16_to_rgba_row(packed, out, width); }
unsafe { arch::x86_sse41::ya16_to_rgba_row::<BE>(packed, out, width); }
return;
}
},
target_arch = "wasm32" => {
if simd128_available() {
unsafe { arch::wasm_simd128::ya16_to_rgba_row(packed, out, width); }
unsafe { arch::wasm_simd128::ya16_to_rgba_row::<BE>(packed, out, width); }
return;
}
},
_ => {}
}
scalar::ya16_to_rgba_row(packed, out, width);
scalar::ya16_to_rgba_row::<BE>(packed, out, width);
}

// ---- ya16_to_rgb_u16_row ------------------------------------------------------

/// Dispatch `ya16_to_rgb_u16_row`.
#[cfg_attr(not(tarpaulin), inline(always))]
pub(crate) fn ya16_to_rgb_u16_row(packed: &[u16], out: &mut [u16], width: usize, use_simd: bool) {
pub(crate) fn ya16_to_rgb_u16_row<const BE: bool>(
packed: &[u16],
out: &mut [u16],
width: usize,
use_simd: bool,
) {
assert!(packed.len() >= ya_row_elems(width), "packed too short");
assert!(out.len() >= rgb_row_elems(width), "out too short");
if !use_simd {
return scalar::ya16_to_rgb_u16_row(packed, out, width);
return scalar::ya16_to_rgb_u16_row::<BE>(packed, out, width);
}
cfg_select! {
target_arch = "aarch64" => {
if neon_available() {
unsafe { arch::neon::ya16_to_rgb_u16_row(packed, out, width); }
unsafe { arch::neon::ya16_to_rgb_u16_row::<BE>(packed, out, width); }
return;
}
},
target_arch = "x86_64" => {
if avx512_available() {
unsafe { arch::x86_avx512::ya16_to_rgb_u16_row(packed, out, width); }
unsafe { arch::x86_avx512::ya16_to_rgb_u16_row::<BE>(packed, out, width); }
return;
}
if avx2_available() {
unsafe { arch::x86_avx2::ya16_to_rgb_u16_row(packed, out, width); }
unsafe { arch::x86_avx2::ya16_to_rgb_u16_row::<BE>(packed, out, width); }
return;
}
if sse41_available() {
unsafe { arch::x86_sse41::ya16_to_rgb_u16_row(packed, out, width); }
unsafe { arch::x86_sse41::ya16_to_rgb_u16_row::<BE>(packed, out, width); }
return;
}
},
target_arch = "wasm32" => {
if simd128_available() {
unsafe { arch::wasm_simd128::ya16_to_rgb_u16_row(packed, out, width); }
unsafe { arch::wasm_simd128::ya16_to_rgb_u16_row::<BE>(packed, out, width); }
return;
}
},
_ => {}
}
scalar::ya16_to_rgb_u16_row(packed, out, width);
scalar::ya16_to_rgb_u16_row::<BE>(packed, out, width);
}

// ---- ya16_to_rgba_u16_row -----------------------------------------------------

/// Dispatch `ya16_to_rgba_u16_row`.
#[cfg_attr(not(tarpaulin), inline(always))]
pub(crate) fn ya16_to_rgba_u16_row(packed: &[u16], out: &mut [u16], width: usize, use_simd: bool) {
pub(crate) fn ya16_to_rgba_u16_row<const BE: bool>(
packed: &[u16],
out: &mut [u16],
width: usize,
use_simd: bool,
) {
assert!(packed.len() >= ya_row_elems(width), "packed too short");
assert!(out.len() >= rgba_row_elems(width), "out too short");
if !use_simd {
return scalar::ya16_to_rgba_u16_row(packed, out, width);
return scalar::ya16_to_rgba_u16_row::<BE>(packed, out, width);
}
cfg_select! {
target_arch = "aarch64" => {
if neon_available() {
unsafe { arch::neon::ya16_to_rgba_u16_row(packed, out, width); }
unsafe { arch::neon::ya16_to_rgba_u16_row::<BE>(packed, out, width); }
return;
}
},
target_arch = "x86_64" => {
if avx512_available() {
unsafe { arch::x86_avx512::ya16_to_rgba_u16_row(packed, out, width); }
unsafe { arch::x86_avx512::ya16_to_rgba_u16_row::<BE>(packed, out, width); }
return;
}
if avx2_available() {
unsafe { arch::x86_avx2::ya16_to_rgba_u16_row(packed, out, width); }
unsafe { arch::x86_avx2::ya16_to_rgba_u16_row::<BE>(packed, out, width); }
return;
}
if sse41_available() {
unsafe { arch::x86_sse41::ya16_to_rgba_u16_row(packed, out, width); }
unsafe { arch::x86_sse41::ya16_to_rgba_u16_row::<BE>(packed, out, width); }
return;
}
},
target_arch = "wasm32" => {
if simd128_available() {
unsafe { arch::wasm_simd128::ya16_to_rgba_u16_row(packed, out, width); }
unsafe { arch::wasm_simd128::ya16_to_rgba_u16_row::<BE>(packed, out, width); }
return;
}
},
_ => {}
}
scalar::ya16_to_rgba_u16_row(packed, out, width);
scalar::ya16_to_rgba_u16_row::<BE>(packed, out, width);
}

// ---- ya16_to_luma_row ---------------------------------------------------------

/// Dispatch `ya16_to_luma_row`.
#[cfg_attr(not(tarpaulin), inline(always))]
pub(crate) fn ya16_to_luma_row(packed: &[u16], out: &mut [u8], width: usize, use_simd: bool) {
pub(crate) fn ya16_to_luma_row<const BE: bool>(
packed: &[u16],
out: &mut [u8],
width: usize,
use_simd: bool,
) {
assert!(packed.len() >= ya_row_elems(width), "packed too short");
assert!(out.len() >= width, "out too short");
if !use_simd {
return scalar::ya16_to_luma_row(packed, out, width);
return scalar::ya16_to_luma_row::<BE>(packed, out, width);
}
cfg_select! {
target_arch = "aarch64" => {
if neon_available() {
unsafe { arch::neon::ya16_to_luma_row(packed, out, width); }
unsafe { arch::neon::ya16_to_luma_row::<BE>(packed, out, width); }
return;
}
},
target_arch = "x86_64" => {
if avx512_available() {
unsafe { arch::x86_avx512::ya16_to_luma_row(packed, out, width); }
unsafe { arch::x86_avx512::ya16_to_luma_row::<BE>(packed, out, width); }
return;
}
if avx2_available() {
unsafe { arch::x86_avx2::ya16_to_luma_row(packed, out, width); }
unsafe { arch::x86_avx2::ya16_to_luma_row::<BE>(packed, out, width); }
return;
}
if sse41_available() {
unsafe { arch::x86_sse41::ya16_to_luma_row(packed, out, width); }
unsafe { arch::x86_sse41::ya16_to_luma_row::<BE>(packed, out, width); }
return;
}
},
target_arch = "wasm32" => {
if simd128_available() {
unsafe { arch::wasm_simd128::ya16_to_luma_row(packed, out, width); }
unsafe { arch::wasm_simd128::ya16_to_luma_row::<BE>(packed, out, width); }
return;
}
},
_ => {}
}
scalar::ya16_to_luma_row(packed, out, width);
scalar::ya16_to_luma_row::<BE>(packed, out, width);
}

// ---- ya16_to_luma_u16_row -----------------------------------------------------

/// Dispatch `ya16_to_luma_u16_row`.
#[cfg_attr(not(tarpaulin), inline(always))]
pub(crate) fn ya16_to_luma_u16_row(packed: &[u16], out: &mut [u16], width: usize, use_simd: bool) {
pub(crate) fn ya16_to_luma_u16_row<const BE: bool>(
packed: &[u16],
out: &mut [u16],
width: usize,
use_simd: bool,
) {
assert!(packed.len() >= ya_row_elems(width), "packed too short");
assert!(out.len() >= width, "out too short");
if !use_simd {
return scalar::ya16_to_luma_u16_row(packed, out, width);
return scalar::ya16_to_luma_u16_row::<BE>(packed, out, width);
}
cfg_select! {
target_arch = "aarch64" => {
if neon_available() {
unsafe { arch::neon::ya16_to_luma_u16_row(packed, out, width); }
unsafe { arch::neon::ya16_to_luma_u16_row::<BE>(packed, out, width); }
return;
}
},
target_arch = "x86_64" => {
if avx512_available() {
unsafe { arch::x86_avx512::ya16_to_luma_u16_row(packed, out, width); }
unsafe { arch::x86_avx512::ya16_to_luma_u16_row::<BE>(packed, out, width); }
return;
}
if avx2_available() {
unsafe { arch::x86_avx2::ya16_to_luma_u16_row(packed, out, width); }
unsafe { arch::x86_avx2::ya16_to_luma_u16_row::<BE>(packed, out, width); }
return;
}
if sse41_available() {
unsafe { arch::x86_sse41::ya16_to_luma_u16_row(packed, out, width); }
unsafe { arch::x86_sse41::ya16_to_luma_u16_row::<BE>(packed, out, width); }
return;
}
},
target_arch = "wasm32" => {
if simd128_available() {
unsafe { arch::wasm_simd128::ya16_to_luma_u16_row(packed, out, width); }
unsafe { arch::wasm_simd128::ya16_to_luma_u16_row::<BE>(packed, out, width); }
return;
}
},
_ => {}
}
scalar::ya16_to_luma_u16_row(packed, out, width);
scalar::ya16_to_luma_u16_row::<BE>(packed, out, width);
}

// ---- ya16_to_hsv_row ----------------------------------------------------------

/// Dispatch `ya16_to_hsv_row`.
#[cfg_attr(not(tarpaulin), inline(always))]
pub(crate) fn ya16_to_hsv_row(
pub(crate) fn ya16_to_hsv_row<const BE: bool>(
packed: &[u16],
h_out: &mut [u8],
s_out: &mut [u8],
Expand All @@ -292,36 +322,36 @@ pub(crate) fn ya16_to_hsv_row(
assert!(s_out.len() >= width, "S out too short");
assert!(v_out.len() >= width, "V out too short");
if !use_simd {
return scalar::ya16_to_hsv_row(packed, h_out, s_out, v_out, width);
return scalar::ya16_to_hsv_row::<BE>(packed, h_out, s_out, v_out, width);
}
cfg_select! {
target_arch = "aarch64" => {
if neon_available() {
unsafe { arch::neon::ya16_to_hsv_row(packed, h_out, s_out, v_out, width); }
unsafe { arch::neon::ya16_to_hsv_row::<BE>(packed, h_out, s_out, v_out, width); }
return;
}
},
target_arch = "x86_64" => {
if avx512_available() {
unsafe { arch::x86_avx512::ya16_to_hsv_row(packed, h_out, s_out, v_out, width); }
unsafe { arch::x86_avx512::ya16_to_hsv_row::<BE>(packed, h_out, s_out, v_out, width); }
return;
}
if avx2_available() {
unsafe { arch::x86_avx2::ya16_to_hsv_row(packed, h_out, s_out, v_out, width); }
unsafe { arch::x86_avx2::ya16_to_hsv_row::<BE>(packed, h_out, s_out, v_out, width); }
return;
}
if sse41_available() {
unsafe { arch::x86_sse41::ya16_to_hsv_row(packed, h_out, s_out, v_out, width); }
unsafe { arch::x86_sse41::ya16_to_hsv_row::<BE>(packed, h_out, s_out, v_out, width); }
return;
}
},
target_arch = "wasm32" => {
if simd128_available() {
unsafe { arch::wasm_simd128::ya16_to_hsv_row(packed, h_out, s_out, v_out, width); }
unsafe { arch::wasm_simd128::ya16_to_hsv_row::<BE>(packed, h_out, s_out, v_out, width); }
return;
}
},
_ => {}
}
scalar::ya16_to_hsv_row(packed, h_out, s_out, v_out, width);
scalar::ya16_to_hsv_row::<BE>(packed, h_out, s_out, v_out, width);
}
Loading
Loading