Added AVX2, some rework and improvements

Cargo.toml

@@ -2,7 +2,7 @@ workspace = { members = ["src/app"] }
 
 [package]
 name = "colorutils-rs"
-version = "0.2.1"
+version = "0.2.2"
 edition = "2021"
 description = "Hig performance utilities for color format handling and conversion."
 readme = "README.md"

src/image_to_xyz_lab.rs

@@ -65,7 +65,7 @@ fn channels_to_xyz<const CHANNELS_CONFIGURATION: u8, const USE_ALPHA: bool, cons
 
     #[cfg(any(target_arch = "x86_64", target_arch = "x86"))]
     {
-        #[cfg(feature = "avx2")]
+        #[cfg(target_feature = "avx2")]
         if is_x86_feature_detected!("avx2") {
             _has_avx2 = true;
         }

src/linear_to_image.rs

@@ -36,19 +36,19 @@ fn linear_to_gamma_channels<const CHANNELS_CONFIGURATION: u8, const USE_ALPHA: b
     let channels = image_configuration.get_channels_count();
 
     #[cfg(target_arch = "x86_64")]
-    let mut has_sse = false;
+    let mut _has_sse = false;
 
     #[cfg(target_arch = "x86_64")]
     if is_x86_feature_detected!("sse4.1") {
-        has_sse = true;
+        _has_sse = true;
     }
 
     for _ in 0..height as usize {
         let mut cx = 0usize;
 
         #[cfg(any(target_arch = "x86_64", target_arch = "x86"))]
         unsafe {
-            if has_sse {
+            if _has_sse {
                 cx = sse_linear_to_gamma::<CHANNELS_CONFIGURATION, USE_ALPHA>(
                     cx,
                     src.as_ptr(),

src/rgb_expand.rs

@@ -24,18 +24,19 @@ pub fn rgb_to_rgba(
     let mut src_offset = 0usize;
 
     #[cfg(any(target_arch = "x86_64", target_arch = "x86"))]
-    let mut use_sse = false;
+    let mut _use_sse = false;
 
     #[cfg(any(target_arch = "x86_64", target_arch = "x86"))]
-    let mut use_avx = false;
+    let mut _use_avx = false;
 
     #[cfg(any(target_arch = "x86_64", target_arch = "x86"))]
     {
         if is_x86_feature_detected!("sse4.1") {
-            use_sse = true;
+            _use_sse = true;
         }
+        #[cfg(target_feature = "avx2")]
         if is_x86_feature_detected!("avx2") {
-            use_avx = true;
+            _use_avx = true;
         }
     }
 
@@ -47,7 +48,7 @@ pub fn rgb_to_rgba(
         unsafe {
             let src_ptr = src.as_ptr().add(src_offset);
             let dst_ptr = dst.as_mut_ptr().add(dst_offset);
-            if use_avx {
+            if _use_avx {
                 let v_alpha = _mm256_set1_epi8(default_alpha as i8);
                 while cx + 32 < width as usize {
                     let xyz_chan_ptr = src_ptr.add(cx * 3usize);
@@ -66,7 +67,7 @@ pub fn rgb_to_rgba(
                     cx += 32;
                 }
             }
-            if use_sse {
+            if _use_sse {
                 let v_alpha = _mm_set1_epi8(default_alpha as i8);
                 while cx + 16 < width as usize {
                     let xyz_chan_ptr = src_ptr.add(cx * 3usize);
@@ -105,7 +106,7 @@ pub fn rgb_to_rgba(
             dst[dst_offset + x * 4] = src[src_offset + x * 3];
             dst[dst_offset + x * 4 + 1] = src[src_offset + x * 3 + 1];
             dst[dst_offset + x * 4 + 2] = src[src_offset + x * 3 + 2];
-            dst[dst_offset + x * 4 + 3] = 255;
+            dst[dst_offset + x * 4 + 3] = default_alpha;
         }
 
         dst_offset += dst_stride as usize;

src/sse_math.rs

@@ -245,7 +245,7 @@ pub(crate) unsafe fn _mm_neg_epi32(x: __m128i) -> __m128i {
 #[inline(always)]
 #[allow(dead_code)]
 /// This is Cube Root using Pow functions,
-/// it also precise however due to of inexact nature of power 1/3 result slightly differ
+/// it is also precise however due to of inexact nature of power 1/3 result slightly differ
 /// from real cbrt with about ULP 3-4, but this is almost 2 times faster than cbrt with real ULP 3.5
 pub unsafe fn _mm_cbrt_ps(d: __m128) -> __m128 {
     _mm_pow_n_ps(d, 1f32 / 3f32)