me_distortion.c

H.264编码实现

        src_line = src_tmp + x;
        for (y4 = 0; y4 < BLOCK_SIZE; y4++ )
        {
          *d++ = *src_line++ - *ref_line++ ;
          *d++ = *src_line++ - *ref_line++ ;
          *d++ = *src_line++ - *ref_line++ ;
          *d++ = *src_line++ - *ref_line++ ;

          ref_line += img_padded_size_x_m4x4;
          src_line += src_size_x;
        }
        if ((mcost += HadamardSAD4x4 (diff)) > min_mcost) return mcost;
      }
      src_tmp += src_size_mul;
    }
  }
  else
  { // 8x8 TRANSFORM
    src_size_x = (blocksize_x - BLOCK_SIZE_8x8);
    src_size_mul = blocksize_x * BLOCK_SIZE_8x8;
    for (y = cand_y; y < cand_y + (blocksize_y<<2); y += (BLOCK_SIZE_8x8_SP) )
    {
      for (x=0; x<blocksize_x; x += BLOCK_SIZE_8x8 )
      {
        d = diff;
        ref_line  = get_line[ref_access_method] (ref_pic_sub.luma, y, cand_x + (x<<2));
        src_line = src_tmp + x;
        for (y4 = 0; y4 < BLOCK_SIZE_8x8; y4++ )
        {
          *d++ = *src_line++ - *ref_line++ ;
          *d++ = *src_line++ - *ref_line++ ;
          *d++ = *src_line++ - *ref_line++ ;
          *d++ = *src_line++ - *ref_line++ ;
          *d++ = *src_line++ - *ref_line++ ;
          *d++ = *src_line++ - *ref_line++ ;
          *d++ = *src_line++ - *ref_line++ ;
          *d++ = *src_line++ - *ref_line++ ;

          ref_line += img_padded_size_x_m8x8;
          src_line += src_size_x;
        }
        if ((mcost += HadamardSAD8x8 (diff)) > min_mcost) return mcost;
      }
      src_tmp += src_size_mul;
    }
  }
  return mcost;
}

/*!
 ************************************************************************
 * \brief
 *    SAD computation of weighted samples _with_ Hadamard Transform
 ************************************************************************
*/
int computeSATDWP(imgpel* src_pic,
                int blocksize_y,
                int blocksize_x,
                int min_mcost,
                int cand_x,
                int cand_y)
{
  int mcost = 0;
  int y, x, y4, *d;
  int weighted_pel;
  int src_size_x, src_size_mul;
  imgpel *src_tmp = src_pic;

  if ( !test8x8transform )
  { // 4x4 TRANSFORM
    src_size_x = (blocksize_x - BLOCK_SIZE);
    src_size_mul = blocksize_x * BLOCK_SIZE;
    for (y = cand_y; y < cand_y + (blocksize_y<<2); y += (BLOCK_SIZE_SP))
    {
      for (x=0; x<blocksize_x; x += BLOCK_SIZE)
      {
        d    = diff;
        ref_line = get_line[ref_access_method] (ref_pic_sub.luma, y, cand_x + (x<<2));
        src_line = src_tmp + x;
        for (y4 = 0; y4 < BLOCK_SIZE; y4++ )
        {
          weighted_pel = iClip1( img->max_imgpel_value, ((weight_luma * *ref_line++  + wp_luma_round) >> luma_log_weight_denom) + offset_luma);
          *d++ = *src_line++ - weighted_pel;
          weighted_pel = iClip1( img->max_imgpel_value, ((weight_luma * *ref_line++  + wp_luma_round) >> luma_log_weight_denom) + offset_luma);
          *d++ = *src_line++ - weighted_pel;
          weighted_pel = iClip1( img->max_imgpel_value, ((weight_luma * *ref_line++  + wp_luma_round) >> luma_log_weight_denom) + offset_luma);
          *d++ = *src_line++ - weighted_pel;
          weighted_pel = iClip1( img->max_imgpel_value, ((weight_luma * *ref_line++  + wp_luma_round) >> luma_log_weight_denom) + offset_luma);
          *d++ = *src_line++ - weighted_pel;

          ref_line += img_padded_size_x_m4x4;
          src_line += src_size_x;
        }
        if ((mcost += HadamardSAD4x4 (diff)) > min_mcost) return mcost;
      }
      src_tmp += src_size_mul;
    }
  }
  else
  { // 8x8 TRANSFORM
    src_size_x = (blocksize_x - BLOCK_SIZE_8x8);
    src_size_mul = blocksize_x * BLOCK_SIZE_8x8;
    for (y = cand_y; y < cand_y + (blocksize_y<<2); y += (BLOCK_SIZE_8x8_SP) )
    {
      for (x=0; x<blocksize_x; x += BLOCK_SIZE_8x8 )
      {
        d = diff;
        ref_line  = get_line[ref_access_method] (ref_pic_sub.luma, y, cand_x + (x<<2));
        src_line = src_tmp + x;
        for (y4 = 0; y4 < BLOCK_SIZE_8x8; y4++ )
        {
          weighted_pel = iClip1( img->max_imgpel_value, ((weight_luma * *ref_line++  + wp_luma_round) >> luma_log_weight_denom) + offset_luma);
          *d++ = *src_line++ - weighted_pel;
          weighted_pel = iClip1( img->max_imgpel_value, ((weight_luma * *ref_line++  + wp_luma_round) >> luma_log_weight_denom) + offset_luma);
          *d++ = *src_line++ - weighted_pel;
          weighted_pel = iClip1( img->max_imgpel_value, ((weight_luma * *ref_line++  + wp_luma_round) >> luma_log_weight_denom) + offset_luma);
          *d++ = *src_line++ - weighted_pel;
          weighted_pel = iClip1( img->max_imgpel_value, ((weight_luma * *ref_line++  + wp_luma_round) >> luma_log_weight_denom) + offset_luma);
          *d++ = *src_line++ - weighted_pel;
          weighted_pel = iClip1( img->max_imgpel_value, ((weight_luma * *ref_line++  + wp_luma_round) >> luma_log_weight_denom) + offset_luma);
          *d++ = *src_line++ - weighted_pel;
          weighted_pel = iClip1( img->max_imgpel_value, ((weight_luma * *ref_line++  + wp_luma_round) >> luma_log_weight_denom) + offset_luma);
          *d++ = *src_line++ - weighted_pel;
          weighted_pel = iClip1( img->max_imgpel_value, ((weight_luma * *ref_line++  + wp_luma_round) >> luma_log_weight_denom) + offset_luma);
          *d++ = *src_line++ - weighted_pel;
          weighted_pel = iClip1( img->max_imgpel_value, ((weight_luma * *ref_line++  + wp_luma_round) >> luma_log_weight_denom) + offset_luma);
          *d++ = *src_line++ - weighted_pel;

          ref_line += img_padded_size_x_m8x8;
          src_line += src_size_x;
        }
        if ((mcost += HadamardSAD8x8 (diff)) > min_mcost) return mcost;
      }
      src_tmp += src_size_mul;
    }
  }
  return mcost;
}

/*!
 ************************************************************************
 * \brief
 *    BiPred (w/o weights) SATD computation
 ************************************************************************
*/
int computeBiPredSATD1(imgpel* src_pic,
                       int blocksize_y,
                       int blocksize_x,
                       int min_mcost,
                       int cand_x1, int cand_y1,
                       int cand_x2, int cand_y2)
{
  int mcost = 0;
  int y, x, y4, *d;
  int src_size_x, src_size_mul;
  imgpel *src_tmp = src_pic;

  if ( !test8x8transform )
  { // 4x4 TRANSFORM
    src_size_x = (blocksize_x - BLOCK_SIZE);
    src_size_mul = blocksize_x * BLOCK_SIZE;
    for (y=0; y<(blocksize_y<<2); y += (BLOCK_SIZE_SP))
    {
      for (x=0; x<blocksize_x; x += BLOCK_SIZE)
      {
        d    = diff;
        src_line   = src_tmp + x;
        ref2_line  = get_line[bipred2_access_method] (ref_pic2_sub.luma, cand_y2 + y, cand_x2 + (x<<2));
        ref1_line  = get_line[bipred1_access_method] (ref_pic1_sub.luma, cand_y1 + y, cand_x1 + (x<<2));
        for (y4 = 0; y4 < BLOCK_SIZE; y4++ )
        {
          *d++ = (*src_line++) - ((*ref1_line++ + *ref2_line++ + 1)>>1);
          *d++ = (*src_line++) - ((*ref1_line++ + *ref2_line++ + 1)>>1);
          *d++ = (*src_line++) - ((*ref1_line++ + *ref2_line++ + 1)>>1);
          *d++ = (*src_line++) - ((*ref1_line++ + *ref2_line++ + 1)>>1);

          ref1_line += img_padded_size_x_m4x4;
          ref2_line += img_padded_size_x_m4x4;
          src_line  += src_size_x;
        }
        if ((mcost += HadamardSAD4x4 (diff)) > min_mcost) return mcost;
      }
      src_tmp += src_size_mul;
    }
  }
  else
  { // 8x8 TRANSFORM
    src_size_x = (blocksize_x - BLOCK_SIZE_8x8);
    src_size_mul = blocksize_x * BLOCK_SIZE_8x8;
    for (y=0; y<blocksize_y; y += BLOCK_SIZE_8x8 )
    {
      int y_pos2 = cand_y2 + (y<<2);
      int y_pos1 = cand_y1 + (y<<2);
      for (x=0; x<blocksize_x; x += BLOCK_SIZE_8x8 )
      {
        d = diff;
        src_line   = src_tmp + x;
        ref2_line  = get_line[bipred2_access_method] (ref_pic2_sub.luma, y_pos2, cand_x2 + (x<<2));
        ref1_line  = get_line[bipred1_access_method] (ref_pic1_sub.luma, y_pos1, cand_x1 + (x<<2));
        for (y4 = 0; y4 < BLOCK_SIZE_8x8; y4++ )
        {
          *d++ = (*src_line++) - ((*ref1_line++ + *ref2_line++ + 1)>>1);
          *d++ = (*src_line++) - ((*ref1_line++ + *ref2_line++ + 1)>>1);
          *d++ = (*src_line++) - ((*ref1_line++ + *ref2_line++ + 1)>>1);
          *d++ = (*src_line++) - ((*ref1_line++ + *ref2_line++ + 1)>>1);
          *d++ = (*src_line++) - ((*ref1_line++ + *ref2_line++ + 1)>>1);
          *d++ = (*src_line++) - ((*ref1_line++ + *ref2_line++ + 1)>>1);
          *d++ = (*src_line++) - ((*ref1_line++ + *ref2_line++ + 1)>>1);
          *d++ = (*src_line++) - ((*ref1_line++ + *ref2_line++ + 1)>>1);

          ref1_line += img_padded_size_x_m8x8;
          ref2_line += img_padded_size_x_m8x8;
          src_line += src_size_x;
        }
        if ((mcost += HadamardSAD8x8 (diff)) > min_mcost) return mcost;
      }
      src_tmp += src_size_mul;
    }
  }
  return mcost;
}

/*!
************************************************************************
* \brief
*    BiPred (w/ weights) SATD computation
************************************************************************
*/
int computeBiPredSATD2(imgpel* src_pic,
                       int blocksize_y,
                       int blocksize_x,
                       int min_mcost,
                       int cand_x1, int cand_y1,
                       int cand_x2, int cand_y2)
{
  int mcost = 0;
  int y, x, y4, *d;
  int weighted_pel, pixel1, pixel2;
  int denom = luma_log_weight_denom + 1;
  int lround = 2 * wp_luma_round;
  int src_size_x, src_size_mul;
  imgpel *src_tmp = src_pic;

  if ( !test8x8transform )
  { // 4x4 TRANSFORM
    src_size_x = (blocksize_x - BLOCK_SIZE);
    src_size_mul = blocksize_x * BLOCK_SIZE;
    for (y=0; y<(blocksize_y<<2); y += BLOCK_SIZE_SP)
    {
      for (x=0; x<blocksize_x; x += BLOCK_SIZE)
      {
        d    = diff;
        src_line   = src_tmp + x;
        ref2_line  = get_line[bipred2_access_method] (ref_pic2_sub.luma, cand_y2 + y, cand_x2 + (x<<2));
        ref1_line  = get_line[bipred1_access_method] (ref_pic1_sub.luma, cand_y1 + y, cand_x1 + (x<<2));
        for (y4 = 0; y4 < BLOCK_SIZE; y4++ )
        {
          // 0
          pixel1 = weight1 * (*ref1_line++);
          pixel2 = weight2 * (*ref2_line++);
          weighted_pel =  iClip1( img->max_imgpel_value, ((pixel1 + pixel2 + lround) >> denom) + offsetBi);
          *d++ =  (*src_line++) - weighted_pel;
          // 1
          pixel1 = weight1 * (*ref1_line++);
          pixel2 = weight2 * (*ref2_line++);
          weighted_pel =  iClip1( img->max_imgpel_value, ((pixel1 + pixel2 + lround) >> denom) + offsetBi);
          *d++ =  (*src_line++) - weighted_pel;
          // 2
          pixel1 = weight1 * (*ref1_line++);
          pixel2 = weight2 * (*ref2_line++);
          weighted_pel =  iClip1( img->max_imgpel_value, ((pixel1 + pixel2 + lround) >> denom) + offsetBi);
          *d++ =  (*src_line++) - weighted_pel;
          // 3
          pixel1 = weight1 * (*ref1_line++);
          pixel2 = weight2 * (*ref2_line++);
          weighted_pel =  iClip1( img->max_imgpel_value, ((pixel1 + pixel2 + lround) >> denom) + offsetBi);
          *d++ =  (*src_line++) - weighted_pel;

          ref1_line += img_padded_size_x_m4x4;
          ref2_line += img_padded_size_x_m4x4;
          src_line  += src_size_x;
        }
        if ((mcost += HadamardSAD4x4 (diff)) > min_mcost) return mcost;
      }
      src_tmp += src_size_mul;
    }
  }
  else
  { // 8x8 TRANSFORM
    src_size_x = (blocksize_x - BLOCK_SIZE_8x8);
    src_size_mul = blocksize_x * BLOCK_SIZE_8x8;
    for (y=0; y<blocksize_y; y += BLOCK_SIZE_8x8 )
    {
      int y_pos2 = cand_y2 + (y<<2);
      int y_pos1 = cand_y1 + (y<<2);
      for (x=0; x<blocksize_x; x += BLOCK_SIZE_8x8 )
      {
        d = diff;
        src_line   = src_tmp + x;
        ref2_line  = get_line[bipred2_access_method] (ref_pic2_sub.luma, y_pos2, cand_x2 + (x<<2));
        ref1_line  = get_line[bipred1_access_method] (ref_pic1_sub.luma, y_pos1, cand_x1 + (x<<2));
        for (y4 = 0; y4 < BLOCK_SIZE_8x8; y4++ )
        {
          // 0
          pixel1 = weight1 * (*ref1_line++);
          pixel2 = weight2 * (*ref2_line++);
          weighted_pel =  iClip1( img->max_imgpel_value, ((pixel1 + pixel2 + lround) >> denom) + offsetBi);
          *d++ =  (*src_line++) - weighted_pel;
          // 1
          pixel1 = weight1 * (*ref1_line++);
          pixel2 = weight2 * (*ref2_line++);
          weighted_pel =  iClip1( img->max_imgpel_value, ((pixel1 + pixel2 + lround) >> denom) + offsetBi);
          *d++ =  (*src_line++) - weighted_pel;
          // 2
          pixel1 = weight1 * (*ref1_line++);
          pixel2 = weight2 * (*ref2_line++);
          weighted_pel =  iClip1( img->max_imgpel_value, ((pixel1 + pixel2 + lround) >> denom) + offsetBi);
          *d++ =  (*src_line++) - weighted_pel;
          // 3
          pixel1 = weight1 * (*ref1_line++);
          pixel2 = weight2 * (*ref2_line++);
          weighted_pel =  iClip1( img->max_imgpel_value, ((pixel1 + pixel2 + lround) >> denom) + offsetBi);
          *d++ =  (*src_line++) - weighted_pel;
          // 4
          pixel1 = weight1 * (*ref1_line++);
          pixel2 = weight2 * (*ref2_line++);
          weighted_pel =  iClip1( img->max_imgpel_value, ((pixel1 + pixel2 + lround) >> denom) + offsetBi);
          *d++ =  (*src_line++) - weighted_pel;
          // 5
          pixel1 = weight1 * (*ref1_line++);
          pixel2 = weight2 * (*ref2_line++);
          weighted_pel =  iClip1( img->max_imgpel_value, ((pixel1 + pixel2 + lround) >> denom) + offsetBi);
          *d++ =  (*src_line++) - weighted_pel;
          // 6
          pixel1 = weight1 * (*ref1_line++);
          pixel2 = weight2 * (*ref2_line++);
          weighted_pel =  iClip1( img->max_imgpel_value, ((pixel1 + pixel2 + lround) >> denom) + offsetBi);
          *d++ =  (*src_line++) - weighted_pel;
          // 7
          pixel1 = weight1 * (*ref1_line++);