block.c

h.264/avc 视频编码程序,实现分数像素匹配功能,非原创.

/*
***********************************************************************
* COPYRIGHT AND WARRANTY INFORMATION
*
* Copyright 2001, International Telecommunications Union, Geneva
*
* DISCLAIMER OF WARRANTY
*
* These software programs are available to the user without any
* license fee or royalty on an "as is" basis. The ITU disclaims
* any and all warranties, whether express, implied, or
* statutory, including any implied warranties of merchantability
* or of fitness for a particular purpose.  In no event shall the
* contributor or the ITU be liable for any incidental, punitive, or
* consequential damages of any kind whatsoever arising from the
* use of these programs.
*
* This disclaimer of warranty extends to the user of these programs
* and user's customers, employees, agents, transferees, successors,
* and assigns.
*
* The ITU does not represent or warrant that the programs furnished
* hereunder are free of infringement of any third-party patents.
* Commercial implementations of ITU-T Recommendations, including
* shareware, may be subject to royalty fees to patent holders.
* Information regarding the ITU-T patent policy is available from
* the ITU Web site at http://www.itu.int.
*
* THIS IS NOT A GRANT OF PATENT RIGHTS - SEE THE ITU-T PATENT POLICY.
************************************************************************
*/

/*!
 *************************************************************************************
 * \file block.c
 *
 * \brief
 *    Process one block
 *
 * \author
 *    Main contributors (see contributors.h for copyright, address and affiliation details)
 *    - Inge Lille-Lang鴜               <[e-mail]>
 *    - Rickard Sjoberg                 <[e-mail]>
 *    - Stephan Wenger                  <[e-mail]>
 *    - Jani Lainema                    <[e-mail]>
 *    - Detlev Marpe                    <[e-mail]>
 *    - Thomas Wedi                     <[e-mail]>
 *    - Ragip Kurceren                  <[e-mail]>
 *************************************************************************************
 */

#include "contributors.h"


#include <math.h>
#include <stdlib.h>
#include <assert.h>

#include "block.h"
#include "refbuf.h"


#define Q_BITS          16
#define DQ_BITS         6
#define DQ_ROUND        (1<<(DQ_BITS-1))


static const int quant_coef[6][4][4] = {
  {{16384,10486,16384,10486},{10486, 6453,10486, 6453},{16384,10486,16384,10486},{10486, 6453,10486, 6453}},
  {{14564, 9118,14564, 9118},{ 9118, 5785, 9118, 5785},{14564, 9118,14564, 9118},{ 9118, 5785, 9118, 5785}},
  {{13107, 8389,13107, 8389},{ 8389, 5243, 8389, 5243},{13107, 8389,13107, 8389},{ 8389, 5243, 8389, 5243}},
  {{11916, 7490,11916, 7490},{ 7490, 4660, 7490, 4660},{11916, 7490,11916, 7490},{ 7490, 4660, 7490, 4660}},
  {{10486, 6554,10486, 6554},{ 6554, 4194, 6554, 4194},{10486, 6554,10486, 6554},{ 6554, 4194, 6554, 4194}},
  {{9362,  5825, 9362, 5825},{ 5825, 3728, 5825, 3728},{ 9362, 5825, 9362, 5825},{ 5825, 3728, 5825, 3728}}
};

static const int dequant_coef[6][4][4] = {
  {{16,20,16,20},{20,26,20,26},{16,20,16,20},{20,26,20,26}},
  {{18,23,18,23},{23,29,23,29},{18,23,18,23},{23,29,23,29}},
  {{20,25,20,25},{25,32,25,32},{20,25,20,25},{25,32,25,32}},
  {{22,28,22,28},{28,36,28,36},{22,28,22,28},{28,36,28,36}},
  {{25,32,25,32},{32,40,32,40},{25,32,25,32},{32,40,32,40}},
  {{28,36,28,36},{36,45,36,45},{28,36,28,36},{36,45,36,45}},
};


#ifndef USE_6_INTRA_MODES

// Notation for comments regarding prediction and predictors.
// The pels of the 4x4 block are labelled a..p. The predictor pels above
// are labelled A..H, from the left I..P, and from above left X, as follows:
//
//  X A B C D E F G H
//  I a b c d
//  J e f g h
//  K i j k l
//  L m n o p
//  M
//  N
//  O
//  P
//

// Predictor array index definitions
#define P_X (PredPel[0])
#define P_A (PredPel[1])
#define P_B (PredPel[2])
#define P_C (PredPel[3])
#define P_D (PredPel[4])
#define P_E (PredPel[5])
#define P_F (PredPel[6])
#define P_G (PredPel[7])
#define P_H (PredPel[8])
#define P_I (PredPel[9])
#define P_J (PredPel[10])
#define P_K (PredPel[11])
#define P_L (PredPel[12])
#define P_M (PredPel[13])
#define P_N (PredPel[14])
#define P_O (PredPel[15])
#define P_P (PredPel[16])

/*!
 ************************************************************************
 * \brief
 *    Make intra 4x4 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 .
 *
 *  \para Input:
 *     Starting point of current 4x4 block image posision
 *
 *  \para Output:
 *      none
 ************************************************************************
 */
void intrapred_luma(int img_x,int img_y)
{
  int i,j;
  int s0;
  int PredPel[17];  // array of predictor pels

  int block_available_up        = (img->ipredmode[img_x/BLOCK_SIZE+1][img_y/BLOCK_SIZE] >=0);
  int block_available_up_right  = (img->ipredmode[img_x/BLOCK_SIZE+2][img_y/BLOCK_SIZE] >=0);
  int block_available_left      = (img->ipredmode[img_x/BLOCK_SIZE][img_y/BLOCK_SIZE+1] >=0);
  int block_available_left_down = (img->ipredmode[img_x/BLOCK_SIZE][img_y/BLOCK_SIZE+2] >=0);

  i = (img_x & 15);
  j = (img_y & 15);
  if (block_available_up_right)
  {
    if ((i == 4  && j == 4) ||
        (i == 12 && j == 4) ||
        (i == 12 && j == 8) ||
        (i == 4  && j == 12) ||
        (i == 12 && j == 12))
    {
      block_available_up_right = 0;
    }
  }

  if (block_available_left_down)
  {
    if (!(i == 0 && j == 0) &&
        !(i == 8 && j == 0) &&
        !(i == 0 && j == 4) &&
        !(i == 0 && j == 8) &&
        !(i == 8 && j == 8))
    {
      block_available_left_down = 0;
    }
  }

  // form predictor pels
  if (block_available_up)
  {
    P_A = imgY[img_y-1][img_x+0];
    P_B = imgY[img_y-1][img_x+1];
    P_C = imgY[img_y-1][img_x+2];
    P_D = imgY[img_y-1][img_x+3];

    if (block_available_up_right)
    {
      P_E = imgY[img_y-1][img_x+4];
      P_F = imgY[img_y-1][img_x+5];
      P_G = imgY[img_y-1][img_x+6];
      P_H = imgY[img_y-1][img_x+7];
    }
    else
    {
      P_E = P_F = P_G = P_H = P_D;
    }
  }
  else
  {
    P_A = P_B = P_C = P_D = P_E = P_F = P_G = P_H = 128;
  }

  if (block_available_left)
  {
    P_I = imgY[img_y+0][img_x-1];
    P_J = imgY[img_y+1][img_x-1];
    P_K = imgY[img_y+2][img_x-1];
    P_L = imgY[img_y+3][img_x-1];

    if (block_available_left_down)
    {
      P_M = imgY[img_y+4][img_x-1];
      P_N = imgY[img_y+5][img_x-1];
      P_O = imgY[img_y+6][img_x-1];
      P_P = imgY[img_y+7][img_x-1];
    }
    else
    {
      P_M = P_N = P_O = P_P = P_L;
    }
  }
  else
  {
    P_I = P_J = P_K = P_L = P_M = P_N = P_O = P_P = 128;
  }

  // XXXGJC -- not quite right for this boundary
  if (block_available_up && block_available_left)
  {
    P_X = imgY[img_y-1][img_x-1];
  }
  else
  {
    P_X = 128;
  }

  ///////////////////////////////
  // make DC prediction
  ///////////////////////////////
  s0 = 0;
  if (block_available_up && block_available_left)
  {   
    // no edge
    s0 = (P_A + P_B + P_C + P_D + P_I + P_J + P_K + P_L + 4)/(2*BLOCK_SIZE);
  }
  else if (!block_available_up && block_available_left)
  {
    // upper edge
    s0 = (P_I + P_J + P_K + P_L + 2)/BLOCK_SIZE;             
  }
  else if (block_available_up && !block_available_left)
  {
    // left edge
    s0 = (P_A + P_B + P_C + P_D + 2)/BLOCK_SIZE;             
  }
  else //if (!block_available_up && !block_available_left)
  {
    // top left corner, nothing to predict from
    s0 = 128;                           
  }

  for (j=0; j < BLOCK_SIZE; j++)
  {
    for (i=0; i < BLOCK_SIZE; i++)
    {
      // store DC prediction
      img->mprr[DC_PRED][i][j] = s0;
    }
  }

  ///////////////////////////////
  // make horiz and vert prediction
  ///////////////////////////////

  for (i=0; i < BLOCK_SIZE; i++)
  {
    img->mprr[VERT_PRED][0][i] = 
    img->mprr[VERT_PRED][1][i] = 
    img->mprr[VERT_PRED][2][i] = 
    img->mprr[VERT_PRED][3][i] = (&P_A)[i];
    img->mprr[HOR_PRED][i][0]  = 
    img->mprr[HOR_PRED][i][1]  = 
    img->mprr[HOR_PRED][i][2]  = 
    img->mprr[HOR_PRED][i][3]  = (&P_I)[i];
  }

  /*  Prediction according to 'diagonal' modes */
  if (block_available_up && block_available_left)
  {
    // Mode DIAG_PRED_SE
    img->mprr[DIAG_PRED_SE][3][0] = (P_L + 2*P_K + P_J + 2) / 4; 
    img->mprr[DIAG_PRED_SE][2][0] =
    img->mprr[DIAG_PRED_SE][3][1] = (P_K + 2*P_J + P_I + 2) / 4; 
    img->mprr[DIAG_PRED_SE][1][0] =
    img->mprr[DIAG_PRED_SE][2][1] = 
    img->mprr[DIAG_PRED_SE][3][2] = (P_J + 2*P_I + P_X + 2) / 4; 
    img->mprr[DIAG_PRED_SE][0][0] =
    img->mprr[DIAG_PRED_SE][1][1] =
    img->mprr[DIAG_PRED_SE][2][2] =
    img->mprr[DIAG_PRED_SE][3][3] = (P_I + 2*P_X + P_A + 2) / 4; 
    img->mprr[DIAG_PRED_SE][0][1] =
    img->mprr[DIAG_PRED_SE][1][2] =
    img->mprr[DIAG_PRED_SE][2][3] = (P_X + 2*P_A + P_B + 2) / 4;
    img->mprr[DIAG_PRED_SE][0][2] =
    img->mprr[DIAG_PRED_SE][1][3] = (P_A + 2*P_B + P_C + 2) / 4;
    img->mprr[DIAG_PRED_SE][0][3] = (P_B + 2*P_C + P_D + 2) / 4;

    // Mode DIAG_PRED_NE
    img->mprr[DIAG_PRED_NE][0][0] = (P_A + P_C + P_I + P_K + 2*(P_B + P_J) + 4) / 8;
    img->mprr[DIAG_PRED_NE][0][1] = 
    img->mprr[DIAG_PRED_NE][1][0] = (P_B + P_D + P_J + P_L + 2*(P_C + P_K) + 4) / 8;
    img->mprr[DIAG_PRED_NE][0][2] =
    img->mprr[DIAG_PRED_NE][1][1] =
    img->mprr[DIAG_PRED_NE][2][0] = (P_C + P_E + P_K + P_M + 2*(P_D + P_L) + 4) / 8;
    img->mprr[DIAG_PRED_NE][0][3] = 
    img->mprr[DIAG_PRED_NE][1][2] = 
    img->mprr[DIAG_PRED_NE][2][1] = 
    img->mprr[DIAG_PRED_NE][3][0] = (P_D + P_F + P_L + P_N + 2*(P_E + P_M) + 4) / 8;
    img->mprr[DIAG_PRED_NE][1][3] = 
    img->mprr[DIAG_PRED_NE][2][2] = 
    img->mprr[DIAG_PRED_NE][3][1] = (P_E + P_G + P_M + P_O + 2*(P_F + P_N) + 4) / 8;
    img->mprr[DIAG_PRED_NE][2][3] = 
    img->mprr[DIAG_PRED_NE][3][2] = (P_F + P_H + P_N + P_P + 2*(P_G + P_O) + 4) / 8;
    img->mprr[DIAG_PRED_NE][3][3] = (P_G + P_O + P_H + P_P + 2) / 4;

    // Mode DIAG_PRED_SSE
    img->mprr[DIAG_PRED_SSE][0][0] = 
    img->mprr[DIAG_PRED_SSE][2][1] = (P_X + P_A + 1) / 2;
    img->mprr[DIAG_PRED_SSE][0][1] = 
    img->mprr[DIAG_PRED_SSE][2][2] = (P_A + P_B + 1) / 2;
    img->mprr[DIAG_PRED_SSE][0][2] = 
    img->mprr[DIAG_PRED_SSE][2][3] = (P_B + P_C + 1) / 2;
    img->mprr[DIAG_PRED_SSE][0][3] = (P_C + P_D + 1) / 2;
    img->mprr[DIAG_PRED_SSE][1][0] = 
    img->mprr[DIAG_PRED_SSE][3][1] = (P_I + 2*P_X + P_A + 2) / 4;
    img->mprr[DIAG_PRED_SSE][1][1] = 
    img->mprr[DIAG_PRED_SSE][3][2] = (P_X + 2*P_A + P_B + 2) / 4;
    img->mprr[DIAG_PRED_SSE][1][2] = 
    img->mprr[DIAG_PRED_SSE][3][3] = (P_A + 2*P_B + P_C + 2) / 4;
    img->mprr[DIAG_PRED_SSE][1][3] = (P_B + 2*P_C + P_D + 2) / 4;
    img->mprr[DIAG_PRED_SSE][2][0] = (P_X + 2*P_I + P_J + 2) / 4;
    img->mprr[DIAG_PRED_SSE][3][0] = (P_I + 2*P_J + P_K + 2) / 4;

    // Mode DIAG_PRED_NNE
    img->mprr[DIAG_PRED_NNE][0][0] = (2*(P_A + P_B + P_K) + P_J + P_L + 4) / 8;
    img->mprr[DIAG_PRED_NNE][0][1] = 
    img->mprr[DIAG_PRED_NNE][2][0] = (P_B + P_C + 1) / 2;
    img->mprr[DIAG_PRED_NNE][0][2] = 
    img->mprr[DIAG_PRED_NNE][2][1] = (P_C + P_D + 1) / 2;
    img->mprr[DIAG_PRED_NNE][0][3] = 
    img->mprr[DIAG_PRED_NNE][2][2] = (P_D + P_E + 1) / 2;
    img->mprr[DIAG_PRED_NNE][2][3] = (P_E + P_F + 1) / 2;
    img->mprr[DIAG_PRED_NNE][1][0] = (2*(P_B + P_L) + P_A + P_C + P_K + P_M + 4) / 8;
    img->mprr[DIAG_PRED_NNE][1][1] = 
    img->mprr[DIAG_PRED_NNE][3][0] = (P_B + 2*P_C + P_D + 2) / 4;
    img->mprr[DIAG_PRED_NNE][1][2] = 
    img->mprr[DIAG_PRED_NNE][3][1] = (P_C + 2*P_D + P_E + 2) / 4;
    img->mprr[DIAG_PRED_NNE][1][3] = 
    img->mprr[DIAG_PRED_NNE][3][2] = (P_D + 2*P_E + P_F + 2) / 4;
    img->mprr[DIAG_PRED_NNE][3][3] = (P_E + 2*P_F + P_G + 2) / 4;

    // Mode DIAG_PRED_ENE
    img->mprr[DIAG_PRED_ENE][0][0] = (2*(P_C + P_I + P_J) + P_B + P_D + 4) / 8;
    img->mprr[DIAG_PRED_ENE][0][1] = (2*(P_D + P_J) + P_C + P_E + P_I + P_K + 4) / 8;
    img->mprr[DIAG_PRED_ENE][0][2] = 
    img->mprr[DIAG_PRED_ENE][1][0] = (2*(P_E + P_J + P_K) + P_D + P_F + 4) / 8;
    img->mprr[DIAG_PRED_ENE][0][3] = 
    img->mprr[DIAG_PRED_ENE][1][1] = (2*(P_F + P_K) + P_E + P_G + P_J + P_L + 4) / 8;
    img->mprr[DIAG_PRED_ENE][1][2] = 
    img->mprr[DIAG_PRED_ENE][2][0] = (2*(P_G + P_K + P_L) + P_F + P_H + 4) / 8;
    img->mprr[DIAG_PRED_ENE][1][3] = 
    img->mprr[DIAG_PRED_ENE][2][1] = (2*(P_H + P_L) + P_G + P_H + P_K + P_L + 4) / 8;
    img->mprr[DIAG_PRED_ENE][2][3] = 
    img->mprr[DIAG_PRED_ENE][3][1] = (P_L + (P_M << 1) + P_N + 2) / 4;
    img->mprr[DIAG_PRED_ENE][3][0] = 
    img->mprr[DIAG_PRED_ENE][2][2] = (P_G + P_H + P_L + P_M + 2) / 4;
    img->mprr[DIAG_PRED_ENE][3][2] = (P_M + P_N + 1) / 2;
    img->mprr[DIAG_PRED_ENE][3][3] = (P_M + 2*P_N + P_O + 2) / 4;

    // Mode DIAG_PRED_ESE
    img->mprr[DIAG_PRED_ESE][0][0] = 
    img->mprr[DIAG_PRED_ESE][1][2] = (P_X + P_I + 1) / 2;
    img->mprr[DIAG_PRED_ESE][0][1] = 
    img->mprr[DIAG_PRED_ESE][1][3] = (P_I + 2*P_X + P_A + 2) / 4;
    img->mprr[DIAG_PRED_ESE][0][2] = (P_X + 2*P_A + P_B + 2) / 4;
    img->mprr[DIAG_PRED_ESE][0][3] = (P_A + 2*P_B + P_C + 2) / 4;
    img->mprr[DIAG_PRED_ESE][1][0] = 
    img->mprr[DIAG_PRED_ESE][2][2] = (P_I + P_J + 1) / 2;
    img->mprr[DIAG_PRED_ESE][1][1] = 
    img->mprr[DIAG_PRED_ESE][2][3] = (P_X + 2*P_I + P_J + 2) / 4;
    img->mprr[DIAG_PRED_ESE][2][0] = 
    img->mprr[DIAG_PRED_ESE][3][2] = (P_J + P_K + 1) / 2;
    img->mprr[DIAG_PRED_ESE][2][1] = 
    img->mprr[DIAG_PRED_ESE][3][3] = (P_I + 2*P_J + P_K + 2) / 4;
    img->mprr[DIAG_PRED_ESE][3][0] = (P_K + P_L + 1) / 2;
    img->mprr[DIAG_PRED_ESE][3][1] = (P_J + 2*P_K + P_L + 2) / 4;
  }
}
#else

/*!
 ************************************************************************
 * \brief
 *    Make intra 4x4 prediction according to all 6 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 i the predmode array .
 *
 *  \para Input:
 *     Starting point of current 4x4 block image posision
 *
 *  \para Output:
 *      none
 ************************************************************************
 */
void intrapred_luma(int img_x,int img_y)
{
  int i,j,s0=0,s1,s2,ia[7][3],s[4][2];

  int block_available_up = (img->ipredmode[img_x/BLOCK_SIZE+1][img_y/BLOCK_SIZE] >=0);
  int block_available_left = (img->ipredmode[img_x/BLOCK_SIZE][img_y/BLOCK_SIZE+1] >=0);

  s1=0;
  s2=0;

  // make DC prediction
  for (i=0; i < BLOCK_SIZE; i++)
  {
    if (block_available_up)
      s1 += imgY[img_y-1][img_x+i];    // sum hor pix
    if (block_available_left)
      s2 += imgY[img_y+i][img_x-1];    // sum vert pix
  }
  if (block_available_up && block_available_left)
    s0=(s1+s2+4)/(2*BLOCK_SIZE);      // no edge
  if (!block_available_up && block_available_left)
    s0=(s2+2)/BLOCK_SIZE;             // upper edge
  if (block_available_up && !block_available_left)
    s0=(s1+2)/BLOCK_SIZE;             // left edge
  if (!block_available_up && !block_available_left)
    s0=128;                           // top left corner, nothing to predict from

  for (i=0; i < BLOCK_SIZE; i++)
  {
    // vertical prediction
    if (block_available_up)
      s[i][0]=imgY[img_y-1][img_x+i];
    // horizontal prediction
    if (block_available_left)
      s[i][1]=imgY[img_y+i][img_x-1];
  }

  for (j=0; j < BLOCK_SIZE; j++)
  {
    for (i=0; i < BLOCK_SIZE; i++)
    {
      img->mprr[DC_PRED][i][j]=s0;      // store DC prediction
      img->mprr[VERT_PRED][i][j]=s[j][0]; // store vertical prediction
      img->mprr[HOR_PRED][i][j]=s[i][1]; // store horizontal prediction
    }
  }

  //  Prediction according to 'diagonal' modes
  if (block_available_up && block_available_left)
  {
    int A = imgY[img_y-1][img_x];
    int B = imgY[img_y-1][img_x+1];
    int C = imgY[img_y-1][img_x+2];
    int D = imgY[img_y-1][img_x+3];
    int E = imgY[img_y  ][img_x-1];
    int F = imgY[img_y+1][img_x-1];
    int G = imgY[img_y+2][img_x-1];
    int H = imgY[img_y+3][img_x-1];
    int I = imgY[img_y-1][img_x-1];
    ia[0][0]=(H+2*G+F+2)/4;
    ia[1][0]=(G+2*F+E+2)/4;
    ia[2][0]=(F+2*E+I+2)/4;
    ia[3][0]=(E+2*I+A+2)/4;
    ia[4][0]=(I+2*A+B+2)/4;
    ia[5][0]=(A+2*B+C+2)/4;
    ia[6][0]=(B+2*C+D+2)/4;
    for (i=0;i<4;i++)
      for (j=0;j<4;j++)
        img->mprr[DIAG_PRED_LR_45][i][j]=ia[j-i+3][0];
  }
  if (block_available_up)
  { // Do prediction 1
    int A = imgY[img_y-1][img_x+0];
    int B = imgY[img_y-1][img_x+1];
    int C = imgY[img_y-1][img_x+2];
    int D = imgY[img_y-1][img_x+3];