- __m128i *pdst = (__m128i*) dst;
- const uint8_t *ysrc0 = src + y0*src_stride + x0*sizeof(uint32_t);
- unsigned int tile_stridex = src_stride*(TILE_VECTOR_HEIGHT - 1) - sizeof(uint32_t)*TILE_VECTOR_WIDTH;
- unsigned int tile_stridey = src_stride*TILE_VECTOR_HEIGHT;
-
- const __m128i shuffle00 = _mm_setr_epi8(0x02,0x06,0xff,0xff,0x0a,0x0e,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff);
- const __m128i shuffle01 = _mm_setr_epi8(0x01,0x05,0xff,0xff,0x09,0x0d,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff);
- const __m128i shuffle02 = _mm_setr_epi8(0x00,0x04,0xff,0xff,0x08,0x0c,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff);
- const __m128i shuffle03 = _mm_setr_epi8(0x03,0x07,0xff,0xff,0x0b,0x0f,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff);
-
- const __m128i shuffle10 = _mm_setr_epi8(0xff,0xff,0x02,0x06,0xff,0xff,0x0a,0x0e,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff);
- const __m128i shuffle11 = _mm_setr_epi8(0xff,0xff,0x01,0x05,0xff,0xff,0x09,0x0d,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff);
- const __m128i shuffle12 = _mm_setr_epi8(0xff,0xff,0x00,0x04,0xff,0xff,0x08,0x0c,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff);
- const __m128i shuffle13 = _mm_setr_epi8(0xff,0xff,0x03,0x07,0xff,0xff,0x0b,0x0f,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff);
-
- const __m128i shuffle20 = _mm_setr_epi8(0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0x02,0x06,0xff,0xff,0x0a,0x0e,0xff,0xff);
- const __m128i shuffle21 = _mm_setr_epi8(0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0x01,0x05,0xff,0xff,0x09,0x0d,0xff,0xff);
- const __m128i shuffle22 = _mm_setr_epi8(0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0x00,0x04,0xff,0xff,0x08,0x0c,0xff,0xff);
- const __m128i shuffle23 = _mm_setr_epi8(0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0x03,0x07,0xff,0xff,0x0b,0x0f,0xff,0xff);
-
- const __m128i shuffle30 = _mm_setr_epi8(0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0x02,0x06,0xff,0xff,0x0a,0x0e);
- const __m128i shuffle31 = _mm_setr_epi8(0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0x01,0x05,0xff,0xff,0x09,0x0d);
- const __m128i shuffle32 = _mm_setr_epi8(0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0x00,0x04,0xff,0xff,0x08,0x0c);
- const __m128i shuffle33 = _mm_setr_epi8(0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0x03,0x07,0xff,0xff,0x0b,0x0f);
-
- for (y = 0; y < TILE_SIZE; y += TILE_VECTOR_HEIGHT) {
- __m128i line0 = *(__m128i*)ysrc0;
- const uint8_t *ysrc = ysrc0 + src_stride;
- ysrc0 += tile_stridey;
-
- for (x = 0; x < TILE_SIZE; x += TILE_VECTOR_WIDTH) {
- __m128i r, g, b, a, line1;
- line1 = *(__m128i*)ysrc;
- PIPE_READ_WRITE_BARRIER();
- ysrc += src_stride;
- r = _mm_shuffle_epi8(line0, shuffle00);
- g = _mm_shuffle_epi8(line0, shuffle01);
- b = _mm_shuffle_epi8(line0, shuffle02);
- a = _mm_shuffle_epi8(line0, shuffle03);
-
- line0 = *(__m128i*)ysrc;
- PIPE_READ_WRITE_BARRIER();
- ysrc += src_stride;
- r = _mm_or_si128(r, _mm_shuffle_epi8(line1, shuffle10));
- g = _mm_or_si128(g, _mm_shuffle_epi8(line1, shuffle11));
- b = _mm_or_si128(b, _mm_shuffle_epi8(line1, shuffle12));
- a = _mm_or_si128(a, _mm_shuffle_epi8(line1, shuffle13));
-
- line1 = *(__m128i*)ysrc;
- PIPE_READ_WRITE_BARRIER();
- ysrc -= tile_stridex;
- r = _mm_or_si128(r, _mm_shuffle_epi8(line0, shuffle20));
- g = _mm_or_si128(g, _mm_shuffle_epi8(line0, shuffle21));
- b = _mm_or_si128(b, _mm_shuffle_epi8(line0, shuffle22));
- a = _mm_or_si128(a, _mm_shuffle_epi8(line0, shuffle23));
-
- if (x + 1 < TILE_SIZE) {
- line0 = *(__m128i*)ysrc;
- ysrc += src_stride;
- }
-
- PIPE_READ_WRITE_BARRIER();
- r = _mm_or_si128(r, _mm_shuffle_epi8(line1, shuffle30));
- g = _mm_or_si128(g, _mm_shuffle_epi8(line1, shuffle31));
- b = _mm_or_si128(b, _mm_shuffle_epi8(line1, shuffle32));
- a = _mm_or_si128(a, _mm_shuffle_epi8(line1, shuffle33));
-
- *pdst++ = r;
- *pdst++ = g;
- *pdst++ = b;
- *pdst++ = a;