Merge branch 'main' into dev

# Conflicts:
#	src/alpha/u16x2/avx2.rs
#	src/alpha/u16x2/sse4.rs

CHANGELOG.md

@@ -66,6 +66,15 @@
 - Optimized convolution algorythm by deleting zero coefficients from start and
   end of bounds.
 
+## [Unreleased] - ReleaseDate
+
+### Fixed
+
+- Fixed `SSE4.1` and `AVX2` implementation fo dividing image by
+  alpha channel for images with `U16x2` pixels.
+- Fixed `NEON` implementation for dividing image by
+  alpha channel for images with `U16x2` and `U16x4` pixels.
+
 ## [4.2.2] - 2025-04-06
 
 ## Fixed

src/alpha/u16x2/avx2.rs

@@ -221,7 +221,6 @@ unsafe fn divide_alpha_8_pixels(pixels: __m256i) -> __m256i {
     let divided_luma_f32x8 = _mm256_div_ps(scaled_luma_f32x8, alpha_f32x8);
     let mut divided_luma_i32x8 = _mm256_cvtps_epi32(divided_luma_f32x8);
     // Clamp result to [0..0xffff]
-    divided_luma_i32x8 = _mm256_max_epi32(divided_luma_i32x8, _mm256_setzero_si256());
     divided_luma_i32x8 = _mm256_min_epi32(divided_luma_i32x8, luma_mask);
 
     let alpha = _mm256_and_si256(pixels, alpha_mask);

src/alpha/u16x2/sse4.rs

@@ -211,7 +211,6 @@ unsafe fn divide_alpha_4_pixels(pixels: __m128i) -> __m128i {
     let scaled_luma_f32x4 = _mm_mul_ps(luma_f32x4, alpha_max);
     let mut divided_luma_i32x4 = _mm_cvtps_epi32(_mm_div_ps(scaled_luma_f32x4, alpha_f32x4));
     // Clamp result to [0..0xffff]
-    divided_luma_i32x4 = _mm_max_epi32(divided_luma_i32x4, _mm_setzero_si128());
     divided_luma_i32x4 = _mm_min_epi32(divided_luma_i32x4, luma_mask);
 
     let alpha = _mm_and_si128(pixels, alpha_mask);