Improve linear and u8

Cargo.toml

@@ -2,7 +2,7 @@ workspace = { members = ["src/app"] }
 
 [package]
 name = "colorutils-rs"
-version = "0.2.8"
+version = "0.2.9"
 edition = "2021"
 description = "Hig performance utilities for color format handling and conversion."
 readme = "README.md"

src/image_to_linear_u8.rs

@@ -115,9 +115,9 @@ fn channels_to_linear<const CHANNELS_CONFIGURATION: u8, const USE_ALPHA: bool>(
             let rgb_f32 = rgb.to_rgb_f32();
 
             unsafe {
-                let t_r = transfer(rgb_f32.r).min(255f32).max(0f32) as u8;
-                let t_g = transfer(rgb_f32.g).min(255f32).max(0f32) as u8;
-                let t_b = transfer(rgb_f32.b).min(255f32).max(0f32) as u8;
+                let t_r = (transfer(rgb_f32.r) * 255f32).min(255f32).max(0f32) as u8;
+                let t_g = (transfer(rgb_f32.g) * 255f32).min(255f32).max(0f32) as u8;
+                let t_b = (transfer(rgb_f32.b) * 255f32).min(255f32).max(0f32) as u8;
                 *dst_slice.get_unchecked_mut(px) = t_r;
                 *dst_slice.get_unchecked_mut(px + 1) = t_g;
                 *dst_slice.get_unchecked_mut(px + 2) = t_b;

src/linear_to_image_u8.rs

@@ -6,17 +6,17 @@ use crate::image::ImageConfiguration;
     any(target_arch = "aarch64", target_arch = "arm"),
     target_feature = "neon"
 ))]
-use crate::neon_to_linear::get_neon_linear_transfer;
+use crate::neon_linear_to_image::get_neon_gamma_transfer;
 #[cfg(all(
     any(target_arch = "aarch64", target_arch = "arm"),
     target_feature = "neon"
 ))]
 use crate::neon_to_linear_u8::neon_image_linear_to_u8;
 #[cfg(any(target_arch = "x86_64", target_arch = "x86"))]
 use crate::sse_image_to_linear_u8::sse_image_to_linear_unsigned;
-#[cfg(any(target_arch = "x86_64", target_arch = "x86"))]
-use crate::sse_to_linear::get_sse_linear_transfer;
 use crate::Rgb;
+#[cfg(any(target_arch = "x86_64", target_arch = "x86"))]
+use crate::sse_linear_to_image::get_sse_gamma_transfer;
 
 #[inline]
 fn linear_to_gamma_channels<const CHANNELS_CONFIGURATION: u8, const USE_ALPHA: bool>(
@@ -56,7 +56,7 @@ fn linear_to_gamma_channels<const CHANNELS_CONFIGURATION: u8, const USE_ALPHA: b
         #[cfg(any(target_arch = "x86_64", target_arch = "x86"))]
         unsafe {
             if _has_sse {
-                let transfer = get_sse_linear_transfer(transfer_function);
+                let transfer = get_sse_gamma_transfer(transfer_function);
                 cx = sse_image_to_linear_unsigned::sse_channels_to_linear::<
                     CHANNELS_CONFIGURATION,
                     USE_ALPHA,
@@ -77,7 +77,7 @@ fn linear_to_gamma_channels<const CHANNELS_CONFIGURATION: u8, const USE_ALPHA: b
             target_feature = "neon"
         ))]
         unsafe {
-            let transfer = get_neon_linear_transfer(transfer_function);
+            let transfer = get_neon_gamma_transfer(transfer_function);
             cx = neon_image_linear_to_u8::neon_channels_to_linear_u8::<
                 CHANNELS_CONFIGURATION,
                 USE_ALPHA,

src/neon_to_linear_u8.rs

@@ -16,17 +16,18 @@ pub mod neon_image_linear_to_u8 {
         let r_f = vmulq_n_f32(vcvtq_f32_u32(r), 1f32 / 255f32);
         let g_f = vmulq_n_f32(vcvtq_f32_u32(g), 1f32 / 255f32);
         let b_f = vmulq_n_f32(vcvtq_f32_u32(b), 1f32 / 255f32);
-        let r_linear = transfer(r_f);
-        let g_linear = transfer(g_f);
-        let b_linear = transfer(b_f);
+        let r_linear = vmulq_n_f32(transfer(r_f), 255f32);
+        let g_linear = vmulq_n_f32(transfer(g_f), 255f32);
+        let b_linear = vmulq_n_f32(transfer(b_f), 255f32);
+
         (
             vcvtaq_u32_f32(r_linear),
             vcvtaq_u32_f32(g_linear),
             vcvtaq_u32_f32(b_linear),
         )
     }
 
-    #[inline(always)]
+    #[inline]
     pub(crate) unsafe fn neon_channels_to_linear_u8<
         const CHANNELS_CONFIGURATION: u8,
         const USE_ALPHA: bool,
@@ -142,5 +143,4 @@ pub mod neon_image_linear_to_u8 {
 
         cx
     }
-
-}
+}

src/sse_image_to_linear_u8.rs

@@ -30,9 +30,10 @@ pub mod sse_image_to_linear_unsigned {
         let r_f = _mm_mul_ps(_mm_cvtepi32_ps(r), u8_scale);
         let g_f = _mm_mul_ps(_mm_cvtepi32_ps(g), u8_scale);
         let b_f = _mm_mul_ps(_mm_cvtepi32_ps(b), u8_scale);
-        let r_linear = transfer(r_f);
-        let g_linear = transfer(g_f);
-        let b_linear = transfer(b_f);
+        let u8_backwards = _mm_set1_ps(255f32);
+        let r_linear = _mm_mul_ps(transfer(r_f), u8_backwards);
+        let g_linear = _mm_mul_ps(transfer(g_f), u8_backwards);
+        let b_linear = _mm_mul_ps(transfer(b_f), u8_backwards);
         (
             _mm_cvtps_epi32(r_linear),
             _mm_cvtps_epi32(g_linear),