RGBA to YUV, SSE

awxkee · May 22, 2024 · 650be4d · 650be4d
1 parent 969013d
commit 650be4d
Show file tree

Hide file tree

Showing 3 changed files with 12 additions and 12 deletions.
diff --git a/src/intel_simd_support.rs b/src/intel_simd_support.rs
@@ -183,9 +183,10 @@ pub unsafe fn store_u8_rgb_avx2(
     if use_transient {
         _mm256_storeu_si256(ptr.add(72) as *mut __m256i, rgb3);
     } else {
-        let mut transient: [u8; 32] = [0u8; 32];
-        _mm256_storeu_si256(transient.as_mut_ptr() as *mut __m256i, rgb3);
-        std::ptr::copy_nonoverlapping(transient.as_ptr(), ptr.add(72), 24);
+        // let mut transient: [u8; 32] = [0u8; 32];
+        // _mm256_storeu_si256(transient.as_mut_ptr() as *mut __m256i, rgb3);
+        // std::ptr::copy_nonoverlapping(transient.as_ptr(), ptr.add(72), 24);
+        std::ptr::copy_nonoverlapping(&rgb3 as *const _ as *const u8, ptr.add(72), 24);
     }
 }
 
@@ -352,8 +353,6 @@ pub unsafe fn store_u8_rgb_sse(
     if use_transient {
         _mm_storeu_si128(ptr.add(36) as *mut __m128i, rgb3);
     } else {
-        let mut transient: [u8; 16] = [0u8; 16];
-        _mm_storeu_si128(transient.as_mut_ptr() as *mut __m128i, rgb3);
-        std::ptr::copy(transient.as_ptr(), ptr.add(36), 12);
+        std::ptr::copy(&rgb3 as *const _ as *const u8, ptr.add(36), 12);
     }
 }
diff --git a/src/rgba_to_yuv.rs b/src/rgba_to_yuv.rs
@@ -119,12 +119,12 @@ unsafe fn sse_row<const ORIGIN_CHANNELS: u8, const SAMPLING: u8>(
                 let cr_h = _mm_avg_epu8(cr, cr);
                 std::ptr::copy_nonoverlapping(
                     &cb_h as *const _ as *const u8,
-                    u_ptr.add(uv_x) as *mut u8,
+                    u_ptr.add(uv_x),
                     8,
                 );
                 std::ptr::copy_nonoverlapping(
                     &cr_h as *const _ as *const u8,
-                    v_ptr.add(uv_x) as *mut u8,
+                    v_ptr.add(uv_x),
                     8,
                 );
                 uv_x += 8;
@@ -190,7 +190,7 @@ fn rgbx_to_yuv8<const ORIGIN_CHANNELS: u8, const SAMPLING: u8>(
 
     #[cfg(target_arch = "x86_64")]
     {
-        if is_x86_feature_detected!("sse4.1") && is_x86_feature_detected!("ssse3") {
+        if is_x86_feature_detected!("sse4.1") {
             use_sse = true;
         }
     }

diff --git a/src/yuv_to_rgba.rs b/src/yuv_to_rgba.rs
@@ -638,16 +638,17 @@ fn yuv_to_rgbx<const DESTINATION_CHANNELS: u8, const SAMPLING: u8>(
             if chroma_subsampling == YuvChromaSample::YUV420
                 || chroma_subsampling == YuvChromaSample::YUV422
             {
-                if x + 1 < width as usize {
-                    let y_value = (y_plane[y_offset + x + 1] as i32 - bias_y) * y_coef;
+                let next_x = x + 1;
+                if next_x < width as usize {
+                    let y_value = (y_plane[y_offset + next_x] as i32 - bias_y) * y_coef;
 
                     let r = ((y_value + cr_coef * cr_value) >> 6).min(255).max(0);
                     let b = ((y_value + cb_coef * cb_value) >> 6).min(255).max(0);
                     let g = ((y_value - g_coef_1 * cr_value - g_coef_2 * cb_value) >> 6)
                         .min(255)
                         .max(0);
 
-                    let next_px = (x + 1) * channels;
+                    let next_px = next_x * channels;
 
                     let rgba_shift = rgba_offset + next_px;