x86 improvements and bugfixes

app/Cargo.toml

@@ -6,7 +6,7 @@ edition = "2021"
 [dependencies]
 image = { version = "0.25.5", features = ["default"] }
 #image = { path= "../../../RustroverProjects/image", features = ["default", "avif", "avif-native"] }
-pic-scale = { path = "..", features = ["half", "nightly_i8mm"], default-features = true }
+pic-scale = { path = "..", features = ["half", "nightly_avx512"], default-features = true }
 fast_image_resize = { version = "5.0.0", features = [] }
 half = { version = "2.4.1", default-features = true }
 

app/benches/resize_rgb/main.rs

@@ -17,13 +17,10 @@ pub fn criterion_benchmark(c: &mut Criterion) {
     let src_bytes = binding.as_bytes();
 
     c.bench_function("Pic scale RGB: Lanczos 3", |b| {
-        let mut copied: Vec<u8> = Vec::from(src_bytes);
-        let store = ImageStore::<u8, 3>::from_slice(
-            &mut copied,
-            dimensions.0 as usize,
-            dimensions.1 as usize,
-        )
-        .unwrap();
+        let copied: Vec<u8> = Vec::from(src_bytes);
+        let store =
+            ImageStore::<u8, 3>::from_slice(&copied, dimensions.0 as usize, dimensions.1 as usize)
+                .unwrap();
         b.iter(|| {
             let mut scaler = Scaler::new(ResamplingFunction::Lanczos3);
             scaler.set_threading_policy(ThreadingPolicy::Single);

app/benches/resize_rgba/main.rs

@@ -16,12 +16,12 @@ pub fn criterion_benchmark(c: &mut Criterion) {
     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);
+        let 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,
+                &copied,
                 dimensions.0 as usize,
                 dimensions.1 as usize,
             )
@@ -35,12 +35,12 @@ pub fn criterion_benchmark(c: &mut Criterion) {
     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());
+        let 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,
+                &copied,
                 dimensions.0 as usize,
                 dimensions.1 as usize,
             )
@@ -81,12 +81,12 @@ pub fn criterion_benchmark(c: &mut Criterion) {
     });
 
     c.bench_function("Pic scale RGBA without alpha: Lanczos 3", |b| {
-        let mut copied: Vec<u8> = Vec::from(src_bytes);
+        let 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,
+                &copied,
                 dimensions.0 as usize,
                 dimensions.1 as usize,
             )

app/src/main.rs

@@ -42,13 +42,21 @@ fn resize_plane(
 }
 
 fn main() {
+    let mut j = [
+        3, 3, 3, 3, 7, 7, 7, 7, 11, 11, 11, 11, 15, 15, 15, 15, 19, 19, 19, 19, 23, 23, 23, 23, 27,
+        27, 27, 27, 31, 31, 31, 31, 35, 35, 35, 35, 39, 39, 39, 39, 43, 43, 43, 43, 47, 47, 47, 47,
+        51, 51, 51, 51, 55, 55, 55, 55, 59, 59, 59, 59, 63, 63, 63, 63,
+    ];
+    j.reverse();
+
+    println!("{:?}", j);
     // test_fast_image();
-    let img = ImageReader::open("./assets/asset_4.png")
+    let img = ImageReader::open("./assets/asset_middle.jpg")
         .unwrap()
         .decode()
         .unwrap();
     let dimensions = img.dimensions();
-    let transient = img.to_rgba8();
+    let transient = img.to_rgb8();
     let mut bytes = Vec::from(transient.as_bytes());
 
     let mut scaler = Scaler::new(ResamplingFunction::Lanczos3);
@@ -60,7 +68,7 @@ fn main() {
 
     //
     let store =
-        ImageStore::<u8, 4>::from_slice(&bytes, dimensions.0 as usize, dimensions.1 as usize)
+        ImageStore::<u8, 3>::from_slice(&bytes, dimensions.0 as usize, dimensions.1 as usize)
             .unwrap();
 
     let dst_size = ImageSize::new(dimensions.0 as usize / 4, dimensions.1 as usize / 4);
@@ -75,15 +83,15 @@ fn main() {
     //     )
     //     .unwrap();
 
-    let mut dst_store = ImageStoreMut::<u8, 4>::alloc_with_depth(
-        dimensions.0 as usize / 2,
-        dimensions.1 as usize / 2,
+    let mut dst_store = ImageStoreMut::<u8, 3>::alloc_with_depth(
+        dimensions.0 as usize / 3,
+        dimensions.1 as usize / 3,
         10,
     );
 
     // for i in 0..25 {
     let start_time = Instant::now();
-    scaler.resize_rgba(&store, &mut dst_store, true).unwrap();
+    scaler.resize_rgb(&store, &mut dst_store).unwrap();
 
     let elapsed_time = start_time.elapsed();
     // Print the elapsed time in milliseconds

src/avx2/mod.rs

@@ -39,8 +39,6 @@ mod rgb_u8_dot_i8;
 #[cfg(feature = "half")]
 mod rgba_f16;
 mod rgba_f32;
-#[cfg(feature = "nightly_avx512")]
-mod rgba_u8_dot_lp;
 mod rgba_u8_lb;
 pub(crate) mod utils;
 #[cfg(feature = "half")]
@@ -50,10 +48,6 @@ mod vertical_u16_lb;
 mod vertical_u8;
 mod vertical_u8_lp;
 
-#[cfg(feature = "nightly_avx512")]
-pub(crate) use crate::avx2::rgba_u8_dot_lp::{
-    convolve_horizontal_rgba_row_dot, convolve_horizontal_rgba_rows_4_dot,
-};
 #[cfg(feature = "half")]
 pub(crate) use alpha_f16::{avx_premultiply_alpha_rgba_f16, avx_unpremultiply_alpha_rgba_f16};
 pub(crate) use alpha_f32::avx_premultiply_alpha_rgba_f32;