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

x86 fast checking for alpha singularity #16

Merged
merged 16 commits into from
Jan 1, 2025
Merged

x86 fast checking for alpha singularity #16

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
16 commits
Select commit Hold shift + click to select a range
6e35f99
Refactorings and improvements
awxkee Dec 29, 2024
9f7573f
x86 Fast checking if alpha pre-multiplication is needed
awxkee Dec 30, 2024
c40a982
Build fixes
awxkee Dec 30, 2024
6e19729
Build fixes
awxkee Dec 30, 2024
b62ba8e
Build fixes
awxkee Dec 30, 2024
83fec59
Build fixes
awxkee Dec 30, 2024
bacda2c
NEON rgba8 alpha scan
awxkee Dec 31, 2024
f57cde8
NEON alpha scanning, tests, improvements
awxkee Dec 31, 2024
1c0b6df
NEON alpha scanning, tests, improvements
awxkee Dec 31, 2024
eab5d80
NEON some reformat and improvements
awxkee Dec 31, 2024
979ed3d
AVX alpha pre-multiplication improvements
awxkee Dec 31, 2024
537f4a5
NEON faster alpha pre-multiplication for f16
awxkee Dec 31, 2024
3c35e95
Clippy
awxkee Dec 31, 2024
8688a74
Updates high bit depth
awxkee Jan 1, 2025
38f214e
Build fixing
awxkee Jan 1, 2025
ab50ceb
AVX, SSE refactor and improvements
awxkee Jan 1, 2025
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
11 changes: 11 additions & 0 deletions .github/workflows/build_push.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -44,6 +44,17 @@ jobs:
- uses: dtolnay/rust-toolchain@stable
- run: cargo clippy

tests:
name: Testing
strategy:
matrix:
os: [ ubuntu-latest, macos-latest ]
runs-on: ${{ matrix.os }}
steps:
- uses: actions/checkout@v4
- uses: dtolnay/rust-toolchain@nightly
- run: cargo test

fuzz_rgba_8bit:
name: Fuzzing 8bit
strategy:
Expand Down
229 changes: 140 additions & 89 deletions app/benches/resize_rgba/main.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,7 +4,8 @@ use fast_image_resize::FilterType::Lanczos3;
use fast_image_resize::{CpuExtensions, PixelType, ResizeAlg, ResizeOptions, Resizer};
use image::{GenericImageView, ImageReader};
use pic_scale::{
ImageStore, ImageStoreMut, ResamplingFunction, Scaler, Scaling, ScalingF32, ThreadingPolicy,
ImageStore, ImageStoreMut, ResamplingFunction, Scaler, Scaling, ScalingF32, ScalingU16,
ThreadingPolicy,
};

pub fn criterion_benchmark(c: &mut Criterion) {
Expand All @@ -14,114 +15,164 @@ pub fn criterion_benchmark(c: &mut Criterion) {
.unwrap();
let dimensions = img.dimensions();
let src_bytes = img.as_bytes();
c.bench_function("Pic scale RGBA with alpha: Lanczos 3", |b| {
let mut copied: Vec<u8> = Vec::from(src_bytes);
b.iter(|| {
let mut scaler = Scaler::new(ResamplingFunction::Lanczos3);
scaler.set_threading_policy(ThreadingPolicy::Single);
let store = ImageStore::<u8, 4>::from_slice(
&mut copied,
dimensions.0 as usize,
dimensions.1 as usize,
)
.unwrap();
let mut target =
ImageStoreMut::alloc(dimensions.0 as usize / 4, dimensions.1 as usize / 4);
_ = scaler.resize_rgba(&store, &mut target, true);
})
});

let f32_image: Vec<f32> = src_bytes.iter().map(|&x| x as f32 / 255f32).collect();
// c.bench_function("Pic scale RGBA with alpha: Lanczos 3", |b| {
// let mut copied: Vec<u8> = Vec::from(src_bytes);
// b.iter(|| {
// let mut scaler = Scaler::new(ResamplingFunction::Lanczos3);
// scaler.set_threading_policy(ThreadingPolicy::Single);
// let store = ImageStore::<u8, 4>::from_slice(
// &mut copied,
// dimensions.0 as usize,
// dimensions.1 as usize,
// )
// .unwrap();
// let mut target =
// ImageStoreMut::alloc(dimensions.0 as usize / 4, dimensions.1 as usize / 4);
// _ = scaler.resize_rgba(&store, &mut target, true);
// })
// });
//
// let f32_image: Vec<f32> = src_bytes.iter().map(|&x| x as f32 / 255f32).collect();
//
// c.bench_function("Pic scale RGBA with alpha f32: Lanczos 3", |b| {
// let mut copied: Vec<f32> = Vec::from(f32_image.clone());
// b.iter(|| {
// let mut scaler = Scaler::new(ResamplingFunction::Lanczos3);
// scaler.set_threading_policy(ThreadingPolicy::Single);
// let store = ImageStore::<f32, 4>::from_slice(
// &mut copied,
// dimensions.0 as usize,
// dimensions.1 as usize,
// )
// .unwrap();
// let mut target =
// ImageStoreMut::alloc(dimensions.0 as usize / 4, dimensions.1 as usize / 4);
// _ = scaler.resize_rgba_f32(&store, &mut target, false);
// })
// });
//
// c.bench_function("Fast image resize RGBA with alpha: Lanczos 3", |b| {
// let mut vc = Vec::from(img.as_bytes());
// b.iter(|| {
// let pixel_type: PixelType = PixelType::U8x4;
// let src_image =
// Image::from_slice_u8(dimensions.0, dimensions.1, &mut vc, pixel_type).unwrap();
// let mut dst_image = Image::new(dimensions.0 / 2, dimensions.1 / 2, pixel_type);
//
// let mut resizer = Resizer::new();
// #[cfg(all(target_arch = "aarch64", target_feature = "neon"))]
// unsafe {
// resizer.set_cpu_extensions(CpuExtensions::Neon);
// }
// #[cfg(any(target_arch = "x86_64", target_arch = "x86"))]
// unsafe {
// resizer.set_cpu_extensions(CpuExtensions::Avx2);
// }
// resizer
// .resize(
// &src_image,
// &mut dst_image,
// &ResizeOptions::new()
// .resize_alg(ResizeAlg::Convolution(Lanczos3))
// .use_alpha(true),
// )
// .unwrap();
// })
// });
//
// c.bench_function("Pic scale RGBA without alpha: Lanczos 3", |b| {
// let mut copied: Vec<u8> = Vec::from(src_bytes);
// b.iter(|| {
// let mut scaler = Scaler::new(ResamplingFunction::Lanczos3);
// scaler.set_threading_policy(ThreadingPolicy::Single);
// let store = ImageStore::<u8, 4>::from_slice(
// &mut copied,
// dimensions.0 as usize,
// dimensions.1 as usize,
// )
// .unwrap();
// let mut target =
// ImageStoreMut::alloc(dimensions.0 as usize / 4, dimensions.1 as usize / 4);
// _ = scaler.resize_rgba(&store, &mut target, false);
// })
// });
//
// c.bench_function("Fast image resize RGBA without alpha: Lanczos 3", |b| {
// let mut vc = Vec::from(img.as_bytes());
// b.iter(|| {
// let pixel_type: PixelType = PixelType::U8x4;
// let src_image =
// Image::from_slice_u8(dimensions.0, dimensions.1, &mut vc, pixel_type).unwrap();
// let mut dst_image = Image::new(dimensions.0 / 2, dimensions.1 / 2, pixel_type);
//
// let mut resizer = Resizer::new();
// #[cfg(all(target_arch = "aarch64", target_feature = "neon"))]
// unsafe {
// resizer.set_cpu_extensions(CpuExtensions::Neon);
// }
// #[cfg(any(target_arch = "x86_64", target_arch = "x86"))]
// unsafe {
// resizer.set_cpu_extensions(CpuExtensions::Avx2);
// }
// resizer
// .resize(
// &src_image,
// &mut dst_image,
// &ResizeOptions::new()
// .resize_alg(ResizeAlg::Convolution(Lanczos3))
// .use_alpha(false),
// )
// .unwrap();
// })
// });

c.bench_function("Pic scale RGBA with alpha f32: Lanczos 3", |b| {
let mut copied: Vec<f32> = Vec::from(f32_image.clone());
c.bench_function("Pic scale RGBA10 with alpha: Lanczos 3", |b| {
let mut copied: Vec<u16> = Vec::from(
src_bytes
.iter()
.map(|&x| ((x as u16) << 2) | ((x as u16) >> 6))
.collect::<Vec<_>>(),
);
b.iter(|| {
let mut scaler = Scaler::new(ResamplingFunction::Lanczos3);
scaler.set_threading_policy(ThreadingPolicy::Single);
let store = ImageStore::<f32, 4>::from_slice(
let store = ImageStore::<u16, 4>::from_slice(
&mut copied,
dimensions.0 as usize,
dimensions.1 as usize,
)
.unwrap();
let mut target =
ImageStoreMut::alloc(dimensions.0 as usize / 4, dimensions.1 as usize / 4);
_ = scaler.resize_rgba_f32(&store, &mut target, false);
let mut target = ImageStoreMut::alloc_with_depth(
dimensions.0 as usize / 4,
dimensions.1 as usize / 4,
10,
);
_ = scaler.resize_rgba_u16(&store, &mut target, true);
})
});

c.bench_function("Fast image resize RGBA with alpha: Lanczos 3", |b| {
let mut vc = Vec::from(img.as_bytes());
b.iter(|| {
let pixel_type: PixelType = PixelType::U8x4;
let src_image =
Image::from_slice_u8(dimensions.0, dimensions.1, &mut vc, pixel_type).unwrap();
let mut dst_image = Image::new(dimensions.0 / 2, dimensions.1 / 2, pixel_type);

let mut resizer = Resizer::new();
#[cfg(all(target_arch = "aarch64", target_feature = "neon"))]
unsafe {
resizer.set_cpu_extensions(CpuExtensions::Neon);
}
#[cfg(any(target_arch = "x86_64", target_arch = "x86"))]
unsafe {
resizer.set_cpu_extensions(CpuExtensions::Avx2);
}
resizer
.resize(
&src_image,
&mut dst_image,
&ResizeOptions::new()
.resize_alg(ResizeAlg::Convolution(Lanczos3))
.use_alpha(true),
)
.unwrap();
})
});

c.bench_function("Pic scale RGBA without alpha: Lanczos 3", |b| {
let mut copied: Vec<u8> = Vec::from(src_bytes);
c.bench_function("Pic scale RGBA10 without alpha: Lanczos 3", |b| {
let mut copied: Vec<u16> = Vec::from(
src_bytes
.iter()
.map(|&x| ((x as u16) << 2) | ((x as u16) >> 6))
.collect::<Vec<_>>(),
);
b.iter(|| {
let mut scaler = Scaler::new(ResamplingFunction::Lanczos3);
scaler.set_threading_policy(ThreadingPolicy::Single);
let store = ImageStore::<u8, 4>::from_slice(
let store = ImageStore::<u16, 4>::from_slice(
&mut copied,
dimensions.0 as usize,
dimensions.1 as usize,
)
.unwrap();
let mut target =
ImageStoreMut::alloc(dimensions.0 as usize / 4, dimensions.1 as usize / 4);
_ = scaler.resize_rgba(&store, &mut target, false);
})
});

c.bench_function("Fast image resize RGBA without alpha: Lanczos 3", |b| {
let mut vc = Vec::from(img.as_bytes());
b.iter(|| {
let pixel_type: PixelType = PixelType::U8x4;
let src_image =
Image::from_slice_u8(dimensions.0, dimensions.1, &mut vc, pixel_type).unwrap();
let mut dst_image = Image::new(dimensions.0 / 2, dimensions.1 / 2, pixel_type);

let mut resizer = Resizer::new();
#[cfg(all(target_arch = "aarch64", target_feature = "neon"))]
unsafe {
resizer.set_cpu_extensions(CpuExtensions::Neon);
}
#[cfg(any(target_arch = "x86_64", target_arch = "x86"))]
unsafe {
resizer.set_cpu_extensions(CpuExtensions::Avx2);
}
resizer
.resize(
&src_image,
&mut dst_image,
&ResizeOptions::new()
.resize_alg(ResizeAlg::Convolution(Lanczos3))
.use_alpha(false),
)
.unwrap();
let mut target = ImageStoreMut::alloc_with_depth(
dimensions.0 as usize / 4,
dimensions.1 as usize / 4,
10,
);
_ = scaler.resize_rgba_u16(&store, &mut target, false);
})
});
}
Expand Down
20 changes: 9 additions & 11 deletions app/src/main.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,17 +3,14 @@ mod split;

use std::time::Instant;

use crate::merge::merge_channels_3;
use crate::split::split_channels_3;
use fast_image_resize::images::Image;
use fast_image_resize::{
CpuExtensions, FilterType, IntoImageView, PixelType, ResizeAlg, ResizeOptions, Resizer,
};
use image::{EncodableLayout, GenericImageView, ImageReader};
use pic_scale::{
Ar30ByteOrder, ImageSize, ImageStore, ImageStoreMut, JzazbzScaler, LChScaler, LabScaler,
LinearApproxScaler, LinearScaler, LuvScaler, OklabScaler, ResamplingFunction, Scaler, Scaling,
ScalingU16, SigmoidalScaler, ThreadingPolicy, TransferFunction, XYZScaler,
ImageSize, ImageStore, ImageStoreMut, ResamplingFunction, Scaler, Scaling, ScalingU16,
ThreadingPolicy,
};

fn resize_plane(
Expand Down Expand Up @@ -53,15 +50,16 @@ fn main() {
let transient = img.to_rgba8();
let mut bytes = Vec::from(transient.as_bytes());

let mut scaler = Scaler::new(ResamplingFunction::Lanczos3);
let mut scaler = Scaler::new(ResamplingFunction::Bilinear);
scaler.set_threading_policy(ThreadingPolicy::Single);

// resize_plane(378, 257, 257, 257, ResamplingFunction::Bilinear);

let mut choke: Vec<u16> = bytes.iter().map(|&x| (x as u16) << 2).collect();

//
let store =
ImageStore::<u16, 4>::from_slice(&mut choke, dimensions.0 as usize, dimensions.1 as usize)
ImageStore::<u16, 4>::from_slice(&choke, dimensions.0 as usize, dimensions.1 as usize)
.unwrap();

let dst_size = ImageSize::new(dimensions.0 as usize / 4, dimensions.1 as usize / 4);
Expand All @@ -78,13 +76,13 @@ fn main() {
// .unwrap();

let mut dst_store = ImageStoreMut::<u16, 4>::alloc_with_depth(
dimensions.0 as usize / 3,
dimensions.1 as usize / 3,
dimensions.0 as usize,
dimensions.1 as usize / 2,
10,
);

scaler
.resize_rgba_u16(&store, &mut dst_store, false)
.resize_rgba_u16(&store, &mut dst_store, true)
.unwrap();

let elapsed_time = start_time.elapsed();
Expand Down Expand Up @@ -167,7 +165,7 @@ fn main() {
.iter()
.map(|&x| (x >> 2) as u8)
.collect();
//

// let dst = dst_store.as_bytes();
// let dst = resized;
// image::save_buffer(
Expand Down
6 changes: 4 additions & 2 deletions src/alpha_check.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -30,10 +30,12 @@
use num_traits::AsPrimitive;
use std::ops::{AddAssign, BitXor};

#[allow(dead_code)]
pub(crate) fn has_non_constant_cap_alpha_rgba8(store: &[u8], width: usize) -> bool {
has_non_constant_cap_alpha::<u8, u32, 3, 4>(store, width)
}

#[allow(dead_code)]
pub(crate) fn has_non_constant_cap_alpha_rgba16(store: &[u16], width: usize) -> bool {
has_non_constant_cap_alpha::<u16, u64, 3, 4>(store, width)
}
Expand Down Expand Up @@ -61,7 +63,7 @@ where
if store.is_empty() {
return false;
}
let first = store[0];
let first = store[ALPHA_CHANNEL_INDEX];
let mut row_sums: J = 0u32.as_();
for row in store.chunks_exact(width * CHANNELS) {
for color in row.chunks_exact(CHANNELS) {
Expand All @@ -87,7 +89,7 @@ fn has_non_constant_cap_alpha_f32_impl<const ALPHA_CHANNEL_INDEX: usize, const C
if store.is_empty() {
return false;
}
let first = store[0].to_bits();
let first = store[ALPHA_CHANNEL_INDEX].to_bits();
let mut row_sums: u64 = 0u64;
for row in store.chunks_exact(width * CHANNELS) {
for color in row.chunks_exact(CHANNELS) {
Expand Down
Loading
Loading