transform8x8.c

the newest JM software by h.264 JVT official reference model.

    cur_pred[0][1] =
    cur_pred[2][2] =
    cur_pred[4][3] =
    cur_pred[6][4] = (imgpel) ((P_A + P_B + 1) >> 1);
    cur_pred[0][2] =
    cur_pred[2][3] =
    cur_pred[4][4] =
    cur_pred[6][5] = (imgpel) ((P_B + P_C + 1) >> 1);
    cur_pred[0][3] =
    cur_pred[2][4] =
    cur_pred[4][5] =
    cur_pred[6][6] = (imgpel) ((P_C + P_D + 1) >> 1);
    cur_pred[0][4] =
    cur_pred[2][5] =
    cur_pred[4][6] =
    cur_pred[6][7] = (imgpel) ((P_D + P_E + 1) >> 1);
    cur_pred[0][5] =
    cur_pred[2][6] =
    cur_pred[4][7] = (imgpel) ((P_E + P_F + 1) >> 1);
    cur_pred[0][6] =
    cur_pred[2][7] = (imgpel) ((P_F + P_G + 1) >> 1);
    cur_pred[0][7] = (imgpel) ((P_G + P_H + 1) >> 1);
    cur_pred[1][0] =
    cur_pred[3][1] =
    cur_pred[5][2] =
    cur_pred[7][3] = (imgpel) ((P_Q + P_A + (P_Z << 1) + 2) >> 2);
    cur_pred[1][1] =
    cur_pred[3][2] =
    cur_pred[5][3] =
    cur_pred[7][4] = (imgpel) ((P_Z + P_B + (P_A << 1) + 2) >> 2);
    cur_pred[1][2] =
    cur_pred[3][3] =
    cur_pred[5][4] =
    cur_pred[7][5] = (imgpel) ((P_A + P_C + (P_B << 1) + 2) >> 2);
    cur_pred[1][3] =
    cur_pred[3][4] =
    cur_pred[5][5] =
    cur_pred[7][6] = (imgpel) ((P_B + P_D + (P_C << 1) + 2) >> 2);
    cur_pred[1][4] =
    cur_pred[3][5] =
    cur_pred[5][6] =
    cur_pred[7][7] = (imgpel) ((P_C + P_E + (P_D << 1) + 2) >> 2);
    cur_pred[1][5] =
    cur_pred[3][6] =
    cur_pred[5][7] = (imgpel) ((P_D + P_F + (P_E << 1) + 2) >> 2);
    cur_pred[1][6] =
    cur_pred[3][7] = (imgpel) ((P_E + P_G + (P_F << 1) + 2) >> 2);
    cur_pred[1][7] = (imgpel) ((P_F + P_H + (P_G << 1) + 2) >> 2);
    cur_pred[2][0] =
    cur_pred[4][1] =
    cur_pred[6][2] = (imgpel) ((P_R + P_Z + (P_Q << 1) + 2) >> 2);
    cur_pred[3][0] =
    cur_pred[5][1] =
    cur_pred[7][2] = (imgpel) ((P_S + P_Q + (P_R << 1) + 2) >> 2);
    cur_pred[4][0] =
    cur_pred[6][1] = (imgpel) ((P_T + P_R + (P_S << 1) + 2) >> 2);
    cur_pred[5][0] =
    cur_pred[7][1] = (imgpel) ((P_U + P_S + (P_T << 1) + 2) >> 2);
    cur_pred[6][0] = (imgpel) ((P_V + P_T + (P_U << 1) + 2) >> 2);
    cur_pred[7][0] = (imgpel) ((P_W + P_U + (P_V << 1) + 2) >> 2);
}

static inline void get_hor_down(imgpel **cur_pred, imgpel *PredPel)
{
    cur_pred[0][0] =
    cur_pred[1][2] =
    cur_pred[2][4] =
    cur_pred[3][6] = (imgpel) ((P_Q + P_Z + 1) >> 1);
    cur_pred[1][0] =
    cur_pred[2][2] =
    cur_pred[3][4] =
    cur_pred[4][6] = (imgpel) ((P_R + P_Q + 1) >> 1);
    cur_pred[2][0] =
    cur_pred[3][2] =
    cur_pred[4][4] =
    cur_pred[5][6] = (imgpel) ((P_S + P_R + 1) >> 1);
    cur_pred[3][0] =
    cur_pred[4][2] =
    cur_pred[5][4] =
    cur_pred[6][6] = (imgpel) ((P_T + P_S + 1) >> 1);
    cur_pred[4][0] =
    cur_pred[5][2] =
    cur_pred[6][4] =
    cur_pred[7][6] = (imgpel) ((P_U + P_T + 1) >> 1);
    cur_pred[5][0] =
    cur_pred[6][2] =
    cur_pred[7][4] = (imgpel) ((P_V + P_U + 1) >> 1);
    cur_pred[6][0] =
    cur_pred[7][2] = (imgpel) ((P_W + P_V + 1) >> 1);
    cur_pred[7][0] = (imgpel) ((P_X + P_W + 1) >> 1);
    cur_pred[0][1] =
    cur_pred[1][3] =
    cur_pred[2][5] =
    cur_pred[3][7] = (imgpel) ((P_Q + P_A + (P_Z << 1) + 2) >> 2);
    cur_pred[1][1] =
    cur_pred[2][3] =
    cur_pred[3][5] =
    cur_pred[4][7] = (imgpel) ((P_Z + P_R + (P_Q << 1) + 2) >> 2);
    cur_pred[2][1] =
    cur_pred[3][3] =
    cur_pred[4][5] =
    cur_pred[5][7] = (imgpel) ((P_Q + P_S + (P_R << 1) + 2) >> 2);
    cur_pred[3][1] =
    cur_pred[4][3] =
    cur_pred[5][5] =
    cur_pred[6][7] = (imgpel) ((P_R + P_T + (P_S << 1) + 2) >> 2);
    cur_pred[4][1] =
    cur_pred[5][3] =
    cur_pred[6][5] =
    cur_pred[7][7] = (imgpel) ((P_S + P_U + (P_T << 1) + 2) >> 2);
    cur_pred[5][1] =
    cur_pred[6][3] =
    cur_pred[7][5] = (imgpel) ((P_T + P_V + (P_U << 1) + 2) >> 2);
    cur_pred[6][1] =
    cur_pred[7][3] = (imgpel) ((P_U + P_W + (P_V << 1) + 2) >> 2);
    cur_pred[7][1] = (imgpel) ((P_V + P_X + (P_W << 1) + 2) >> 2);
    cur_pred[0][2] =
    cur_pred[1][4] =
    cur_pred[2][6] = (imgpel) ((P_Z + P_B + (P_A << 1) + 2) >> 2);
    cur_pred[0][3] =
    cur_pred[1][5] =
    cur_pred[2][7] = (imgpel) ((P_A + P_C + (P_B << 1) + 2) >> 2);
    cur_pred[0][4] =
    cur_pred[1][6] = (imgpel) ((P_B + P_D + (P_C << 1) + 2) >> 2);
    cur_pred[0][5] =
    cur_pred[1][7] = (imgpel) ((P_C + P_E + (P_D << 1) + 2) >> 2);
    cur_pred[0][6] = (imgpel) ((P_D + P_F + (P_E << 1) + 2) >> 2);
    cur_pred[0][7] = (imgpel) ((P_E + P_G + (P_F << 1) + 2) >> 2);
}

static inline void get_hor_up(imgpel **cur_pred, imgpel *PredPel)
{
  cur_pred[0][0] = (imgpel) ((P_Q + P_R + 1) >> 1);
  cur_pred[1][0] =
    cur_pred[0][2] = (imgpel) ((P_R + P_S + 1) >> 1);
  cur_pred[2][0] =
    cur_pred[1][2] =
    cur_pred[0][4] = (imgpel) ((P_S + P_T + 1) >> 1);
  cur_pred[3][0] =
    cur_pred[2][2] =
    cur_pred[1][4] =
    cur_pred[0][6] = (imgpel) ((P_T + P_U + 1) >> 1);
  cur_pred[4][0] =
    cur_pred[3][2] =
    cur_pred[2][4] =
    cur_pred[1][6] = (imgpel) ((P_U + P_V + 1) >> 1);
  cur_pred[5][0] =
    cur_pred[4][2] =
    cur_pred[3][4] =
    cur_pred[2][6] = (imgpel) ((P_V + P_W + 1) >> 1);
  cur_pred[6][0] =
    cur_pred[5][2] =
    cur_pred[4][4] =
    cur_pred[3][6] = (imgpel) ((P_W + P_X + 1) >> 1);
  cur_pred[4][6] =
    cur_pred[4][7] =
    cur_pred[5][4] =
    cur_pred[5][5] =
    cur_pred[5][6] =
    cur_pred[5][7] =
    cur_pred[6][2] =
    cur_pred[6][3] =
    cur_pred[6][4] =
    cur_pred[6][5] =
    cur_pred[6][6] =
    cur_pred[6][7] =
    cur_pred[7][0] =
    cur_pred[7][1] =
    cur_pred[7][2] =
    cur_pred[7][3] =
    cur_pred[7][4] =
    cur_pred[7][5] =
    cur_pred[7][6] =
    cur_pred[7][7] = (imgpel) P_X;
  cur_pred[6][1] =
    cur_pred[5][3] =
    cur_pred[4][5] =
    cur_pred[3][7] = (imgpel) ((P_W + P_X + (P_X << 1) + 2) >> 2);
  cur_pred[5][1] =
    cur_pred[4][3] =
    cur_pred[3][5] =
    cur_pred[2][7] = (imgpel) ((P_X + P_V + (P_W << 1) + 2) >> 2);
  cur_pred[4][1] =
    cur_pred[3][3] =
    cur_pred[2][5] =
    cur_pred[1][7] = (imgpel) ((P_W + P_U + (P_V << 1) + 2) >> 2);
  cur_pred[3][1] =
    cur_pred[2][3] =
    cur_pred[1][5] =
    cur_pred[0][7] = (imgpel) ((P_V + P_T + (P_U << 1) + 2) >> 2);
  cur_pred[2][1] =
    cur_pred[1][3] =
    cur_pred[0][5] = (imgpel) ((P_U + P_S + (P_T << 1) + 2) >> 2);
  cur_pred[1][1] =
    cur_pred[0][3] = (imgpel) ((P_T + P_R + (P_S << 1) + 2) >> 2);
  cur_pred[0][1] = (imgpel) ((P_S + P_Q + (P_R << 1) + 2) >> 2);
}

/*!
 ************************************************************************
 * \brief
 *    Make intra 8x8 prediction according to all 9 prediction modes.
 *    The routine uses left and upper neighbouring points from
 *    previous coded blocks to do this (if available). Notice that
 *    inaccessible neighbouring points are signalled with a negative
 *    value in the predmode array .
 *
 *  \par Input:
 *     Starting point of current 8x8 block image posision
 *
 *  \par Output:
 *      none
 ************************************************************************
 */
void intrapred_8x8(Macroblock *currMB, ColorPlane pl, int img_x,int img_y, int *left_available, int *up_available, int *all_available)
{
  int i,j;
  int s0;
  static imgpel PredPel[25];  // array of predictor pels
  imgpel **img_enc = enc_picture->p_curr_img;
  imgpel *img_pel;
  static imgpel **cur_pred;
  imgpel ***curr_mpr_8x8  = img->mpr_8x8[pl];
  unsigned int dc_pred_value = img->dc_pred_value;
  int *mb_size = img->mb_size[IS_LUMA];

  int ioff = (img_x & 15);
  int joff = (img_y & 15);

  PixelPos pix_a[8];
  PixelPos pix_b, pix_c, pix_d;

  int block_available_up;
  int block_available_left;
  int block_available_up_left;
  int block_available_up_right;

  for (i=0;i<8;i++)
  {
    getNeighbour(currMB, ioff - 1, joff + i , mb_size, &pix_a[i]);
  }

  getNeighbour(currMB, ioff    , joff - 1, mb_size, &pix_b);
  getNeighbour(currMB, ioff + 8, joff - 1, mb_size, &pix_c);
  getNeighbour(currMB, ioff - 1, joff - 1, mb_size, &pix_d);

  pix_c.available = pix_c.available &&!(ioff == 8 && joff == 8);

  if (params->UseConstrainedIntraPred)
  {
    for (i=0, block_available_left=1; i<8;i++)
      block_available_left  &= pix_a[i].available ? img->intra_block[pix_a[i].mb_addr]: 0;

    block_available_up       = pix_b.available ? img->intra_block [pix_b.mb_addr] : 0;
    block_available_up_right = pix_c.available ? img->intra_block [pix_c.mb_addr] : 0;
    block_available_up_left  = pix_d.available ? img->intra_block [pix_d.mb_addr] : 0;
  }
  else
  {
    block_available_left     = pix_a[0].available;
    block_available_up       = pix_b.available;
    block_available_up_right = pix_c.available;
    block_available_up_left  = pix_d.available;
  }

  *left_available = block_available_left;
  *up_available   = block_available_up;
  *all_available  = block_available_up && block_available_left && block_available_up_left;

  // form predictor pels
  if (block_available_up)
  {
    img_pel = &img_enc[pix_b.pos_y][pix_b.pos_x];
    P_A = *(img_pel++);
    P_B = *(img_pel++);
    P_C = *(img_pel++);
    P_D = *(img_pel++);
    P_E = *(img_pel++);
    P_F = *(img_pel++);
    P_G = *(img_pel++);
    P_H = *(img_pel);
  }
  else
  {
    P_A = P_B = P_C = P_D = P_E = P_F = P_G = P_H = dc_pred_value;
  }

  if (block_available_up_right)
  {
    img_pel = &img_enc[pix_c.pos_y][pix_c.pos_x];
    P_I = *(img_pel++);
    P_J = *(img_pel++);
    P_K = *(img_pel++);
    P_L = *(img_pel++);
    P_M = *(img_pel++);
    P_N = *(img_pel++);
    P_O = *(img_pel++);
    P_P = *(img_pel);

  }
  else
  {
    P_I = P_J = P_K = P_L = P_M = P_N = P_O = P_P = P_H;
  }

  if (block_available_left)
  {
    P_Q = img_enc[pix_a[0].pos_y][pix_a[0].pos_x];
    P_R = img_enc[pix_a[1].pos_y][pix_a[1].pos_x];
    P_S = img_enc[pix_a[2].pos_y][pix_a[2].pos_x];
    P_T = img_enc[pix_a[3].pos_y][pix_a[3].pos_x];
    P_U = img_enc[pix_a[4].pos_y][pix_a[4].pos_x];
    P_V = img_enc[pix_a[5].pos_y][pix_a[5].pos_x];
    P_W = img_enc[pix_a[6].pos_y][pix_a[6].pos_x];
    P_X = img_enc[pix_a[7].pos_y][pix_a[7].pos_x];
  }
  else
  {
    P_Q = P_R = P_S = P_T = P_U = P_V = P_W = P_X = dc_pred_value;
  }

  if (block_available_up_left)
  {
    P_Z = img_enc[pix_d.pos_y][pix_d.pos_x];
  }
  else
  {
    P_Z = dc_pred_value;
  }

  for(i=0;i<9;i++)
    curr_mpr_8x8[i][0][0]=-1;

  LowPassForIntra8x8Pred(&(P_Z), block_available_up_left, block_available_up, block_available_left);

  ///////////////////////////////
  // make DC prediction
  ///////////////////////////////
  s0 = 0;
  if (block_available_up && block_available_left)
  {
    // no edge
    s0 = rshift_rnd_sf((P_A + P_B + P_C + P_D + P_E + P_F + P_G + P_H + P_Q + P_R + P_S + P_T + P_U + P_V + P_W + P_X), 4);
  }
  else if (!block_available_up && block_available_left)
  {
    // upper edge
    s0 = rshift_rnd_sf((P_Q + P_R + P_S + P_T + P_U + P_V + P_W + P_X), 3);
  }
  else if (block_available_up && !block_available_left)
  {
    // left edge
    s0 = rshift_rnd_sf((P_A + P_B + P_C + P_D + P_E + P_F + P_G + P_H), 3);
  }
  else //if (!block_available_up && !block_available_left)
  {
    // top left corner, nothing to predict from
    s0 = dc_pred_value;
  }

  // store DC prediction
  cur_pred = curr_mpr_8x8[DC_PRED];
  // First line
  for (i=0; i < BLOCK_SIZE_8x8; i++)
  {
    cur_pred[0][i] = (imgpel) s0;
  }
  // remaining lines
  for (j=0; j < BLOCK_SIZE_8x8 - 1; j++)
  {
    memcpy (cur_pred[j + 1], cur_pred[j], BLOCK_SIZE_8x8 * sizeof(imgpel));
  }

  ///////////////////////////////
  // make horiz and vert prediction
  ///////////////////////////////
  
  if(block_available_up)
  {
    cur_pred = curr_mpr_8x8[VERT_PRED];
    for (j=0; j < BLOCK_SIZE_8x8; j++)
      memcpy(cur_pred[j], &P_A, BLOCK_SIZE_8x8 * sizeof(imgpel));
  }
  
  if(block_available_left)
  {
    cur_pred = curr_mpr_8x8[HOR_PRED];
    for (i=0; i < BLOCK_SIZE_8x8; i++)
    {
      cur_pred[i][0]  =
        cur_pred[i][1]  =
        cur_pred[i][2]  =
        cur_pred[i][3]  =
        cur_pred[i][4]  =
        cur_pred[i][5]  =
        cur_pred[i][6]  =
        cur_pred[i][7]  = (imgpel) (&P_Q)[i];
    }
  }

  ///////////////////////////////////
  // make diagonal down left prediction
  ///////////////////////////////////
  if (block_available_up)
  {
    // Mode DIAG_DOWN_LEFT_PRED
    get_diag_down_left(curr_mpr_8x8[DIAG_DOWN_LEFT_PRED], PredPel);

    ///////////////////////////////////
    // make vertical left prediction
    ///////////////////////////////////
    get_vert_left(curr_mpr_8x8[VERT_LEFT_PRED], PredPel);
  }

  ///////////////////////////////////
  // make diagonal down right prediction
  ///////////////////////////////////
  if (block_available_up && block_available_left && block_available_up_left)
  {
    // Mode DIAG_DOWN_RIGHT_PRED
    get_diag_down_right(curr_mpr_8x8[DIAG_DOWN_RIGHT_PRED], PredPel);
    
    ///////////////////////////////////
    // make vertical right prediction
    ///////////////////////////////////
    get_vert_right(curr_mpr_8x8[VERT_RIGHT_PRED], PredPel);

    ///////////////////////////////////
    // make horizontal down prediction
    ///////////////////////////////////
    get_hor_down(curr_mpr_8x8[HOR_DOWN_PRED], PredPel);
  }

  ///////////////////////////////////
  // make horizontal up prediction
  ///////////////////////////////////
  if (block_available_left)
  {
    get_hor_up(curr_mpr_8x8[HOR_UP_PRED], PredPel);
  }
}

/*!
 *************************************************************************************
 * \brief
 *    Prefiltering for Intra8x8 prediction
 *************************************************************************************
 */
void LowPassForIntra8x8Pred(imgpel *PredPel, int block_up_left, int block_up, int block_left)
{
  int i;