luminance.c File Reference

Go to the source code of this file.

Functions
void	simd_luminance_ssse3 (const uint8_t *rgb, uint8_t *out, size_t width)

Function Documentation

simd_luminance_ssse3()

void simd_luminance_ssse3	(	const uint8_t *	rgb,
		uint8_t *	out,
		size_t	width )

Definition at line 5 of file ssse3/luminance.c.

                                                                          {
  const rgb_pixel_t *row = (const rgb_pixel_t *)rgb;
  size_t x = 0;
  for (; width - x >= 16; x += 16) {
    // Manual deinterleave RGB components (SSSE3 limitation vs NEON's vld3q_u8)
    uint8_t r_array[16], g_array[16], b_array[16];
    for (int j = 0; j < 16; j++) {
      r_array[j] = row[x + j].r;
      g_array[j] = row[x + j].g;
      b_array[j] = row[x + j].b;
    }
 
    // Process 16 pixels in two 8-pixel SSSE3 batches (same as SSE2 approach)
    __m128i r_vec_lo = _mm_loadl_epi64((__m128i *)(r_array + 0));
    __m128i r_vec_hi = _mm_loadl_epi64((__m128i *)(r_array + 8));
    __m128i g_vec_lo = _mm_loadl_epi64((__m128i *)(g_array + 0));
    __m128i g_vec_hi = _mm_loadl_epi64((__m128i *)(g_array + 8));
    __m128i b_vec_lo = _mm_loadl_epi64((__m128i *)(b_array + 0));
    __m128i b_vec_hi = _mm_loadl_epi64((__m128i *)(b_array + 8));
 
    // Process first 8 pixels
    __m128i r_16_lo = _mm_unpacklo_epi8(r_vec_lo, _mm_setzero_si128());
    __m128i g_16_lo = _mm_unpacklo_epi8(g_vec_lo, _mm_setzero_si128());
    __m128i b_16_lo = _mm_unpacklo_epi8(b_vec_lo, _mm_setzero_si128());
 
    __m128i luma_r_lo = _mm_mullo_epi16(r_16_lo, _mm_set1_epi16(LUMA_RED));
    __m128i luma_g_lo = _mm_mullo_epi16(g_16_lo, _mm_set1_epi16(LUMA_GREEN));
    __m128i luma_b_lo = _mm_mullo_epi16(b_16_lo, _mm_set1_epi16(LUMA_BLUE));
 
    __m128i luma_sum_lo = _mm_add_epi16(luma_r_lo, luma_g_lo);
    luma_sum_lo = _mm_add_epi16(luma_sum_lo, luma_b_lo);
    luma_sum_lo = _mm_add_epi16(luma_sum_lo, _mm_set1_epi16(LUMA_THRESHOLD));
    luma_sum_lo = _mm_srli_epi16(luma_sum_lo, 8);
 
    // Process second 8 pixels
    __m128i r_16_hi = _mm_unpacklo_epi8(r_vec_hi, _mm_setzero_si128());
    __m128i g_16_hi = _mm_unpacklo_epi8(g_vec_hi, _mm_setzero_si128());
    __m128i b_16_hi = _mm_unpacklo_epi8(b_vec_hi, _mm_setzero_si128());
 
    __m128i luma_r_hi = _mm_mullo_epi16(r_16_hi, _mm_set1_epi16(LUMA_RED));
    __m128i luma_g_hi = _mm_mullo_epi16(g_16_hi, _mm_set1_epi16(LUMA_GREEN));
    __m128i luma_b_hi = _mm_mullo_epi16(b_16_hi, _mm_set1_epi16(LUMA_BLUE));
 
    __m128i luma_sum_hi = _mm_add_epi16(luma_r_hi, luma_g_hi);
    luma_sum_hi = _mm_add_epi16(luma_sum_hi, luma_b_hi);
    luma_sum_hi = _mm_add_epi16(luma_sum_hi, _mm_set1_epi16(LUMA_THRESHOLD));
    luma_sum_hi = _mm_srli_epi16(luma_sum_hi, 8);
 
    // Pack and store
    __m128i luminance_lo = _mm_packus_epi16(luma_sum_lo, _mm_setzero_si128());
    __m128i luminance_hi = _mm_packus_epi16(luma_sum_hi, _mm_setzero_si128());
 
    uint8_t luma_array[16];
    _mm_storel_epi64((__m128i *)(luma_array + 0), luminance_lo);
    _mm_storel_epi64((__m128i *)(luma_array + 8), luminance_hi);
    memcpy(out + x, luma_array, 16);
  }
  simd_luminance_scalar((const uint8_t *)(row + x), out + x, width - x);
}

References rgb_pixel_t::b, rgb_pixel_t::g, LUMA_BLUE, LUMA_GREEN, LUMA_RED, LUMA_THRESHOLD, out, rgb_pixel_t::r, simd_luminance_scalar(), and width.