如何修改YUV420转RGB24的Intrinsics代码以输出RGBA32

我首次使用Intrinsics做图像转换优化，参考了一个YUV420转RGB24的实现代码，但我的目标输出格式是r8g8b8a8，原代码只输出r8g8b8。我对代码做了如下修改：

const __m128 rgb1_1 = _mm_shuffle_ps(
                _mm_shuffle_ps(b1, g1, _MM_SHUFFLE(0, 0, 0, 0)),
                _mm_shuffle_ps(r1, b1, _MM_SHUFFLE(1, 1, 0, 0)),
                _MM_SHUFFLE(2, 0, 2, 0));
const __m128 rgb1_2 = _mm_shuffle_ps(
                _mm_shuffle_ps(g1, r1, _MM_SHUFFLE(1, 1, 1, 1)), 
                _mm_shuffle_ps(b1, g1, _MM_SHUFFLE(2, 2, 2, 2)), 
                _MM_SHUFFLE(2, 0, 2, 0)); 
const __m128 rgb1_3 = _mm_shuffle_ps(
                _mm_shuffle_ps(r1, b1, _MM_SHUFFLE(3, 3, 2, 2)), 
                _mm_shuffle_ps(g1, r1, _MM_SHUFFLE(3, 3, 3, 3)), 
                _MM_SHUFFLE(2, 0, 2, 0));

const __m128 rgb2_1 = _mm_shuffle_ps(
                _mm_shuffle_ps(b2, g2, _MM_SHUFFLE(0, 0, 0, 0)),
                _mm_shuffle_ps(r2, b2, _MM_SHUFFLE(1, 1, 0, 0)), 
                _MM_SHUFFLE(2, 0, 2, 0)); 
const __m128 rgb2_2 = _mm_shuffle_ps(
                _mm_shuffle_ps(g2, r2, _MM_SHUFFLE(1, 1, 1, 1)), 
                _mm_shuffle_ps(b2, g2, _MM_SHUFFLE(2, 2, 2, 2)), 
                _MM_SHUFFLE(2, 0, 2, 0)); 
const __m128 rgb2_3 = _mm_shuffle_ps(
                _mm_shuffle_ps(r2, b2, _MM_SHUFFLE(3, 3, 2, 2)), 
                _mm_shuffle_ps(g2, r2, _MM_SHUFFLE(3, 3, 3, 3)), 
                _MM_SHUFFLE(2, 0, 2, 0));

const __m128 rgb3_1 = _mm_shuffle_ps(
                _mm_shuffle_ps(b3, g3, _MM_SHUFFLE(0, 0, 0, 0)),
                _mm_shuffle_ps(r3, b3, _MM_SHUFFLE(1, 1, 0, 0)),
                _MM_SHUFFLE(2, 0, 2, 0));
const __m128 rgb3_2 = _mm_shuffle_ps(
                _mm_shuffle_ps(g3, r3, _MM_SHUFFLE(1, 1, 1, 1)),
                _mm_shuffle_ps(b3, g3, _MM_SHUFFLE(2, 2, 2, 2)),
                _MM_SHUFFLE(2, 0, 2, 0));
const __m128 rgb3_3 = _mm_shuffle_ps(
                _mm_shuffle_ps(r3, b3, _MM_SHUFFLE(3, 3, 2, 2)),
                _mm_shuffle_ps(g3, r3, _MM_SHUFFLE(3, 3, 3, 3)),
                _MM_SHUFFLE(2, 0, 2, 0));

const __m128 rgb4_1 = _mm_shuffle_ps(
                _mm_shuffle_ps(b4, g4, _MM_SHUFFLE(0, 0, 0, 0)),
                _mm_shuffle_ps(r4, b4, _MM_SHUFFLE(1, 1, 0, 0)),
                _MM_SHUFFLE(2, 0, 2, 0));
const __m128 rgb4_2 = _mm_shuffle_ps(
                _mm_shuffle_ps(g4, r4, _MM_SHUFFLE(1, 1, 1, 1)),
                _mm_shuffle_ps(b4, g4, _MM_SHUFFLE(2, 2, 2, 2)),
                _MM_SHUFFLE(2, 0, 2, 0));
const __m128 rgb4_3 = _mm_shuffle_ps(
                _mm_shuffle_ps(r4, b4, _MM_SHUFFLE(3, 3, 2, 2)),
                _mm_shuffle_ps(g4, r4, _MM_SHUFFLE(3, 3, 3, 3)),
                _MM_SHUFFLE(2, 0, 2, 0));

const __m128i pack1l = _mm_packs_epi32(_mm_cvtps_epi32(rgb1_1), _mm_cvtps_epi32(rgb1_2));
const __m128i pack1h = _mm_packs_epi32(_mm_cvtps_epi32(rgb1_3), _mm_cvtps_epi32(rgb2_1));
const __m128i pack1 = _mm_packus_epi16(pack1l, pack1h);
    
const __m128i pack2l = _mm_packs_epi32(_mm_cvtps_epi32(rgb2_2), _mm_cvtps_epi32(rgb2_3));
const __m128i pack2h = _mm_packs_epi32(_mm_cvtps_epi32(rgb3_1), _mm_cvtps_epi32(rgb3_2));
const __m128i pack2 = _mm_packus_epi16(pack2l, pack2h);
    
const __m128i pack3l = _mm_packs_epi32(_mm_cvtps_epi32(rgb3_3), _mm_cvtps_epi32(rgb4_1));
const __m128i pack3h = _mm_packs_epi32(_mm_cvtps_epi32(rgb4_2), _mm_cvtps_epi32(rgb4_3));
const __m128i pack3 = _mm_packus_epi16(pack3l, pack3h);
const __m128i packAlpha = { 0xFFFFFFFFFFFFFFFF, 0xFFFFFFFFFFFFFFFF };
    
    // and finally store in output
    
_mm_storeu_si128((__m128i*)(pixels + ((wh - width) * 4) - h * width * 4 + w * 4 + 0 * 16), pack1);
_mm_storeu_si128((__m128i*)(pixels + ((wh - width) * 4) - h * width * 4 + w * 4 + 1 * 16), pack2);
_mm_storeu_si128((__m128i*)(pixels + ((wh - width) * 4) - h * width * 4 + w * 4 + 2 * 16), pack3);
_mm_storeu_si128((__m128i*)(pixels + ((wh - width) * 4) - h * width * 4 + w * 4 + 3 * 16), packAlpha);

我新增了packAlpha相关代码，原本以为能正确对齐r8g8b8a8格式，但输出图像呈现灰度且带有垂直白线。另外，图像双轴翻转的问题在修改前就存在，后续可以通过旋转纹理解决，暂时不用关注这个问题。

内容的提问来源于stack exchange，提问作者Connor