libavcodec/fdctref.c

   1 /**
   2  * @file libavcodec/fdctref.c
   3  * forward discrete cosine transform, double precision.
   4  */
   5
   6 /* Copyright (C) 1996, MPEG Software Simulation Group. All Rights Reserved. */
   7
   8 /*
   9  * Disclaimer of Warranty
  10  *
  11  * These software programs are available to the user without any license fee or
  12  * royalty on an "as is" basis.  The MPEG Software Simulation Group disclaims
  13  * any and all warranties, whether express, implied, or statuary, including any
  14  * implied warranties or merchantability or of fitness for a particular
  15  * purpose.  In no event shall the copyright-holder be liable for any
  16  * incidental, punitive, or consequential damages of any kind whatsoever
  17  * arising from the use of these programs.
  18  *
  19  * This disclaimer of warranty extends to the user of these programs and user's
  20  * customers, employees, agents, transferees, successors, and assigns.
  21  *
  22  * The MPEG Software Simulation Group does not represent or warrant that the
  23  * programs furnished hereunder are free of infringement of any third-party
  24  * patents.
  25  *
  26  * Commercial implementations of MPEG-1 and MPEG-2 video, including shareware,
  27  * are subject to royalty fees to patent holders.  Many of these patents are
  28  * general enough such that they are unavoidable regardless of implementation
  29  * design.
  30  */
  31
  32 #include <math.h>
  33
  34 #ifndef PI
  35 # ifdef M_PI
  36 #  define PI M_PI
  37 # else
  38 #  define PI 3.14159265358979323846
  39 # endif
  40 #endif
  41
  42 /* global declarations */
  43 void init_fdct (void);
  44 void fdct (short *block);
  45
  46 /* private data */
  47 static double c[8][8]; /* transform coefficients */
  48
  49 void init_fdct(void)
  50 {
  51   int i, j;
  52   double s;
  53
  54   for (i=0; i<8; i++)
  55   {
  56     s = (i==0) ? sqrt(0.125) : 0.5;
  57
  58     for (j=0; j<8; j++)
  59       c[i][j] = s * cos((PI/8.0)*i*(j+0.5));
  60   }
  61 }
  62
  63 void fdct(block)
  64 short *block;
  65 {
  66         register int i, j;
  67         double s;
  68         double tmp[64];
  69
  70         for(i = 0; i < 8; i++)
  71             for(j = 0; j < 8; j++)
  72             {
  73                     s = 0.0;
  74
  75 /*
  76  *                     for(k = 0; k < 8; k++)
  77  *                         s += c[j][k] * block[8 * i + k];
  78  */
  79                 s += c[j][0] * block[8 * i + 0];
  80                 s += c[j][1] * block[8 * i + 1];
  81                 s += c[j][2] * block[8 * i + 2];
  82                 s += c[j][3] * block[8 * i + 3];
  83                 s += c[j][4] * block[8 * i + 4];
  84                 s += c[j][5] * block[8 * i + 5];
  85                 s += c[j][6] * block[8 * i + 6];
  86                 s += c[j][7] * block[8 * i + 7];
  87
  88                     tmp[8 * i + j] = s;
  89             }
  90
  91         for(j = 0; j < 8; j++)
  92             for(i = 0; i < 8; i++)
  93             {
  94                     s = 0.0;
  95
  96 /*
  97  *                       for(k = 0; k < 8; k++)
  98  *                    s += c[i][k] * tmp[8 * k + j];
  99  */
 100                 s += c[i][0] * tmp[8 * 0 + j];
 101                 s += c[i][1] * tmp[8 * 1 + j];
 102                 s += c[i][2] * tmp[8 * 2 + j];
 103                 s += c[i][3] * tmp[8 * 3 + j];
 104                 s += c[i][4] * tmp[8 * 4 + j];
 105                 s += c[i][5] * tmp[8 * 5 + j];
 106                 s += c[i][6] * tmp[8 * 6 + j];
 107                 s += c[i][7] * tmp[8 * 7 + j];
 108                 s*=8.0;
 109
 110                     block[8 * i + j] = (short)floor(s + 0.499999);
 111 /*
 112  * reason for adding 0.499999 instead of 0.5:
 113  * s is quite often x.5 (at least for i and/or j = 0 or 4)
 114  * and setting the rounding threshold exactly to 0.5 leads to an
 115  * extremely high arithmetic implementation dependency of the result;
 116  * s being between x.5 and x.500001 (which is now incorrectly rounded
 117  * downwards instead of upwards) is assumed to occur less often
 118  * (if at all)
 119  */
 120       }
 121 }
 122
 123 /* perform IDCT matrix multiply for 8x8 coefficient block */
 124
 125 void idct(block)
 126 short *block;
 127 {
 128   int i, j, k, v;
 129   double partial_product;
 130   double tmp[64];
 131
 132   for (i=0; i<8; i++)
 133     for (j=0; j<8; j++)
 134     {
 135       partial_product = 0.0;
 136
 137       for (k=0; k<8; k++)
 138         partial_product+= c[k][j]*block[8*i+k];
 139
 140       tmp[8*i+j] = partial_product;
 141     }
 142
 143   /* Transpose operation is integrated into address mapping by switching
 144      loop order of i and j */
 145
 146   for (j=0; j<8; j++)
 147     for (i=0; i<8; i++)
 148     {
 149       partial_product = 0.0;
 150
 151       for (k=0; k<8; k++)
 152         partial_product+= c[k][i]*tmp[8*k+j];
 153
 154       v = (int) floor(partial_product+0.5);
 155       block[8*i+j] = v;
 156     }
 157 }