/*! 
 *************************************************************************************
 * \file decoder.c
 *
 * \brief
 *    Contains functions that implement the "decoders in the encoder" concept for the
 *    rate-distortion optimization with losses.
 * \date
 *    October 22nd, 2001
 *
 * \author
 *    Main contributors (see contributors.h for copyright, address and 
 *    affiliation details)
 *    - Dimitrios Kontopodis                    <dkonto@eikon.tum.de>
 *************************************************************************************
 */

#include <stdlib.h>
#include <memory.h>

#include "global.h"
#include "refbuf.h"
#include "image.h"

/*! 
 *************************************************************************************
 * \brief
 *    decodes one 8x8 partition
 *
 * \note
 *    Gives the expected value in the decoder of one 8x8 block. This is done based on the 
 *    stored reconstructed residue decs->resY[][], the reconstructed values imgY[][]
 *    and the motion vectors. The decoded 8x8 block is moved to decs->decY[][].
 *************************************************************************************
 */
void decode_one_b8block (int decoder, int mbmode, int b8block, int b8mode, int b8ref)
{
  int i,j,block_y,block_x,bx,by;
  int ref_inx = (IMG_NUMBER-1)%img->num_ref_frames;

  int mv[2][BLOCK_MULTIPLE][BLOCK_MULTIPLE];
  int resY_tmp[MB_BLOCK_SIZE][MB_BLOCK_SIZE];

  int i0 = (b8block%2)<<3,   i1 = i0+8,   bx0 = i0>>2,   bx1 = bx0+2;
  int j0 = (b8block/2)<<3,   j1 = j0+8,   by0 = j0>>2,   by1 = by0+2;

  if (img->type==I_SLICE)
  {
    for(i=i0;i<i1;i++)
    for(j=j0;j<j1;j++)
    {
      decs->decY[decoder][img->pix_y+j][img->pix_x+i]=enc_picture->imgY[img->pix_y+j][img->pix_x+i];
    }
  }
  else
  {
    if (mbmode==0 && (img->type==P_SLICE || (img->type==B_SLICE && img->nal_reference_idc>0)))
    {
      for(i=i0;i<i1;i++)
      for(j=j0;j<j1;j++)
      {
        resY_tmp[j][i]=0;
      }
      for (by=by0; by<by1; by++)
      for (bx=bx0; bx<bx1; bx++)
      {
        mv[0][by][bx] = mv[1][by][bx] = 0;
      }
    }
    else
    {
      if (b8mode>=1 && b8mode<=7)
      {
        for (by=by0; by<by1; by++)
        for (bx=bx0; bx<bx1; bx++)
        {
          mv[0][by][bx] = img->all_mv[bx][by][LIST_0][b8ref][b8mode][0];
          mv[1][by][bx] = img->all_mv[bx][by][LIST_0][b8ref][b8mode][1];
        }
      }
      else
      {
        for (by=by0; by<by1; by++)
        for (bx=bx0; bx<bx1; bx++)
        {
          mv[0][by][bx] = mv[1][by][bx] = 0;
        }
      }
          
      for(i=i0;i<i1;i++)
      for(j=j0;j<j1;j++)
      {
        resY_tmp[j][i]=decs->resY[j][i];
      }
    }

    // Decode Luminance
    if ((b8mode>=1 && b8mode<=7) || (mbmode==0 && (img->type==P_SLICE || (img->type==B_SLICE && img->nal_reference_idc>0))))
    {
      for (by=by0; by<by1; by++)
      for (bx=bx0; bx<bx1; bx++)
      {
        block_x = img->block_x+bx;
        block_y = img->block_y+by;
        if (img->type == B_SLICE && enc_picture != enc_frame_picture)
          ref_inx = (IMG_NUMBER-b8ref-2)%img->num_ref_frames;

        Get_Reference_Block (decs->decref[decoder][ref_inx],
                             block_y, block_x,
                             mv[0][by][bx],
                             mv[1][by][bx],
                             decs->RefBlock);
        for (j=0; j<4; j++)
        for (i=0; i<4; i++)
        {
          /*
          if (decs->RefBlock[j][i] != UMVPelY_14 (mref[ref_inx],
                                                  (block_y*4+j)*4+mv[1][by][bx],
                                                  (block_x*4+i)*4+mv[0][by][bx]))
          ref_inx = (img->number-ref-1)%img->num_ref_frames;
          */
          decs->decY[decoder][block_y*4+j][block_x*4+i] = resY_tmp[by*4+j][bx*4+i] + decs->RefBlock[j][i];
        }
      }
    }
    else 
    {
      // Intra Refresh - Assume no spatial prediction
      for(i=i0;i<i1;i++)
      for(j=j0;j<j1;j++)
      {
        decs->decY[decoder][img->pix_y+j][img->pix_x+i] = enc_picture->imgY[img->pix_y+j][img->pix_x+i];
      }
    }
  }
}


/*! 
 *************************************************************************************
 * \brief
 *    decodes one macroblock
 *************************************************************************************
 */
void decode_one_mb (int decoder, Macroblock* currMB)
{
  /*
  decode_one_b8block (decoder, currMB->mb_type, 0, currMB->b8mode[0], refFrArr[img->block_y+0][img->block_x+0]);
  decode_one_b8block (decoder, currMB->mb_type, 1, currMB->b8mode[1], refFrArr[img->block_y+0][img->block_x+2]);
  decode_one_b8block (decoder, currMB->mb_type, 2, currMB->b8mode[2], refFrArr[img->block_y+2][img->block_x+0]);
  decode_one_b8block (decoder, currMB->mb_type, 3, currMB->b8mode[3], refFrArr[img->block_y+2][img->block_x+2]);
  */
}

/*! 
 *************************************************************************************
 * \brief
 *    Finds the reference MB given the decoded reference frame
 * \note
 *    This is based on the function UnifiedOneForthPix, only it is modified to
 *    be used at the "many decoders in the encoder" RD optimization. In this case
 *    we dont want to keep full upsampled reference frames for all decoders, so
 *    we just upsample when it is necessary.
 * \param imY
 *    The frame to be upsampled
 * \param block_y
 *    The row of the block, whose prediction we want to find
 * \param block_x
 *    The column of the block, whose prediction we want to track
 * \param mvhor
 *    Motion vector, horizontal part
 * \param mvver
 *    Motion vector, vertical part
 * \param out
 *    Output: The prediction for the block (block_y, block_x)
 *************************************************************************************
 */
void Get_Reference_Block(imgpel **imY, 
                         int block_y, 
                         int block_x, 
                         int mvhor, 
                         int mvver, 
                         imgpel **out)
{
  int i,j,y,x;

  y = block_y * BLOCK_SIZE * 4 + mvver;
  x = block_x * BLOCK_SIZE * 4 + mvhor;

  for (j=0; j<BLOCK_SIZE; j++)
    for (i=0; i<BLOCK_SIZE; i++)
      out[j][i] = Get_Reference_Pixel(imY, y+j*4, x+i*4);
}

/*! 
 *************************************************************************************
 * \brief
 *    Finds a pixel (y,x) of the upsampled reference frame
 * \note
 *    This is based on the function UnifiedOneForthPix, only it is modified to
 *    be used at the "many decoders in the encoder" RD optimization. In this case
 *    we dont want to keep full upsampled reference frames for all decoders, so
 *    we just upsample when it is necessary.
 *************************************************************************************
 */
byte Get_Reference_Pixel(imgpel **imY, int y_pos, int x_pos)
{

  int dx, x;
  int dy, y;
  int maxold_x,maxold_y;

  int result = 0, result1, result2;
  int pres_x;
  int pres_y; 

  int tmp_res[6];

  static const int COEF[6] = {
    1, -5, 20, 20, -5, 1
  };


  dx = x_pos&3;
  dy = y_pos&3;
  x_pos = (x_pos-dx)/4;
  y_pos = (y_pos-dy)/4;
  maxold_x = img->width-1;
  maxold_y = img->height-1;

  if (dx == 0 && dy == 0) { /* fullpel position */
    result = imY[max(0,min(maxold_y,y_pos))][max(0,min(maxold_x,x_pos))];
  }
  else { /* other positions */

    if (dy == 0) {

      pres_y = max(0,min(maxold_y,y_pos));
      for(x=-2;x<4;x++) {
        pres_x = max(0,min(maxold_x,x_pos+x));
        result += imY[pres_y][pres_x]*COEF[x+2];
      }

      result = max(0, min(img->max_imgpel_value, (result+16)/32));

      if (dx == 1) {
        result = (result + imY[pres_y][max(0,min(maxold_x,x_pos))])/2;
      }
      else if (dx == 3) {
        result = (result + imY[pres_y][max(0,min(maxold_x,x_pos+1))])/2;
      }
    }
    else if (dx == 0) {

      pres_x = max(0,min(maxold_x,x_pos));
      for(y=-2;y<4;y++) {
        pres_y = max(0,min(maxold_y,y_pos+y));
        result += imY[pres_y][pres_x]*COEF[y+2];
      }

      result = max(0, min(img->max_imgpel_value, (result+16)/32));

      if (dy == 1) {
        result = (result + imY[max(0,min(maxold_y,y_pos))][pres_x])/2;
      }
      else if (dy == 3) {
        result = (result + imY[max(0,min(maxold_y,y_pos+1))][pres_x])/2;
      }
    }
    else if (dx == 2) {

      for(y=-2;y<4;y++) {
        result = 0;
        pres_y = max(0,min(maxold_y,y_pos+y));
        for(x=-2;x<4;x++) {
          pres_x = max(0,min(maxold_x,x_pos+x));
          result += imY[pres_y][pres_x]*COEF[x+2];
        }
        tmp_res[y+2] = result;
      }

      result = 0;
      for(y=-2;y<4;y++) {
        result += tmp_res[y+2]*COEF[y+2];
      }

      result = max(0, min(img->max_imgpel_value, (result+512)/1024));

      if (dy == 1) {
        result = (result + max(0, min(img->max_imgpel_value, (tmp_res[2]+16)/32)))/2;
      }
      else if (dy == 3) {
        result = (result + max(0, min(img->max_imgpel_value, (tmp_res[3]+16)/32)))/2;
      }
    }
    else if (dy == 2) {

      for(x=-2;x<4;x++) {
        result = 0;
        pres_x = max(0,min(maxold_x,x_pos+x));
        for(y=-2;y<4;y++) {
          pres_y = max(0,min(maxold_y,y_pos+y));
          result += imY[pres_y][pres_x]*COEF[y+2];
        }
        tmp_res[x+2] = result;
      }

      result = 0;
      for(x=-2;x<4;x++) {
        result += tmp_res[x+2]*COEF[x+2];
      }

      result = max(0, min(img->max_imgpel_value, (result+512)/1024));

      if (dx == 1) {
        result = (result + max(0, min(img->max_imgpel_value, (tmp_res[2]+16)/32)))/2;
      }
      else {
        result = (result + max(0, min(img->max_imgpel_value, (tmp_res[3]+16)/32)))/2;
      }
    }
    else {

      result = 0;
      pres_y = dy == 1 ? y_pos : y_pos+1;
      pres_y = max(0,min(maxold_y,pres_y));