Subversion Repositories Kolibri OS

Rev

Blame | Last modification | View Log | RSS feed

  1. /*
  2.  * jfdctfst.c
  3.  *
  4.  * Copyright (C) 1994-1996, Thomas G. Lane.
  5.  * This file is part of the Independent JPEG Group's software.
  6.  * For conditions of distribution and use, see the accompanying README file.
  7.  *
  8.  * This file contains a fast, not so accurate integer implementation of the
  9.  * forward DCT (Discrete Cosine Transform).
  10.  *
  11.  * A 2-D DCT can be done by 1-D DCT on each row followed by 1-D DCT
  12.  * on each column.  Direct algorithms are also available, but they are
  13.  * much more complex and seem not to be any faster when reduced to code.
  14.  *
  15.  * This implementation is based on Arai, Agui, and Nakajima's algorithm for
  16.  * scaled DCT.  Their original paper (Trans. IEICE E-71(11):1095) is in
  17.  * Japanese, but the algorithm is described in the Pennebaker & Mitchell
  18.  * JPEG textbook (see REFERENCES section in file README).  The following code
  19.  * is based directly on figure 4-8 in P&M.
  20.  * While an 8-point DCT cannot be done in less than 11 multiplies, it is
  21.  * possible to arrange the computation so that many of the multiplies are
  22.  * simple scalings of the final outputs.  These multiplies can then be
  23.  * folded into the multiplications or divisions by the JPEG quantization
  24.  * table entries.  The AA&N method leaves only 5 multiplies and 29 adds
  25.  * to be done in the DCT itself.
  26.  * The primary disadvantage of this method is that with fixed-point math,
  27.  * accuracy is lost due to imprecise representation of the scaled
  28.  * quantization values.  The smaller the quantization table entry, the less
  29.  * precise the scaled value, so this implementation does worse with high-
  30.  * quality-setting files than with low-quality ones.
  31.  */
  32.  
  33. #define JPEG_INTERNALS
  34. #include "jinclude.h"
  35. #include "jpeglib.h"
  36. #include "jdct.h"               /* Private declarations for DCT subsystem */
  37.  
  38. #ifdef DCT_IFAST_SUPPORTED
  39.  
  40.  
  41. /*
  42.  * This module is specialized to the case DCTSIZE = 8.
  43.  */
  44.  
  45. #if DCTSIZE != 8
  46.   Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
  47. #endif
  48.  
  49.  
  50. /* Scaling decisions are generally the same as in the LL&M algorithm;
  51.  * see jfdctint.c for more details.  However, we choose to descale
  52.  * (right shift) multiplication products as soon as they are formed,
  53.  * rather than carrying additional fractional bits into subsequent additions.
  54.  * This compromises accuracy slightly, but it lets us save a few shifts.
  55.  * More importantly, 16-bit arithmetic is then adequate (for 8-bit samples)
  56.  * everywhere except in the multiplications proper; this saves a good deal
  57.  * of work on 16-bit-int machines.
  58.  *
  59.  * Again to save a few shifts, the intermediate results between pass 1 and
  60.  * pass 2 are not upscaled, but are represented only to integral precision.
  61.  *
  62.  * A final compromise is to represent the multiplicative constants to only
  63.  * 8 fractional bits, rather than 13.  This saves some shifting work on some
  64.  * machines, and may also reduce the cost of multiplication (since there
  65.  * are fewer one-bits in the constants).
  66.  */
  67.  
  68. #define CONST_BITS  8
  69.  
  70.  
  71. /* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
  72.  * causing a lot of useless floating-point operations at run time.
  73.  * To get around this we use the following pre-calculated constants.
  74.  * If you change CONST_BITS you may want to add appropriate values.
  75.  * (With a reasonable C compiler, you can just rely on the FIX() macro...)
  76.  */
  77.  
  78. #if CONST_BITS == 8
  79. #define FIX_0_382683433  ((INT32)   98)         /* FIX(0.382683433) */
  80. #define FIX_0_541196100  ((INT32)  139)         /* FIX(0.541196100) */
  81. #define FIX_0_707106781  ((INT32)  181)         /* FIX(0.707106781) */
  82. #define FIX_1_306562965  ((INT32)  334)         /* FIX(1.306562965) */
  83. #else
  84. #define FIX_0_382683433  FIX(0.382683433)
  85. #define FIX_0_541196100  FIX(0.541196100)
  86. #define FIX_0_707106781  FIX(0.707106781)
  87. #define FIX_1_306562965  FIX(1.306562965)
  88. #endif
  89.  
  90.  
  91. /* We can gain a little more speed, with a further compromise in accuracy,
  92.  * by omitting the addition in a descaling shift.  This yields an incorrectly
  93.  * rounded result half the time...
  94.  */
  95.  
  96. #ifndef USE_ACCURATE_ROUNDING
  97. #undef DESCALE
  98. #define DESCALE(x,n)  RIGHT_SHIFT(x, n)
  99. #endif
  100.  
  101.  
  102. /* Multiply a DCTELEM variable by an INT32 constant, and immediately
  103.  * descale to yield a DCTELEM result.
  104.  */
  105.  
  106. #define MULTIPLY(var,const)  ((DCTELEM) DESCALE((var) * (const), CONST_BITS))
  107.  
  108.  
  109. /*
  110.  * Perform the forward DCT on one block of samples.
  111.  */
  112.  
  113. GLOBAL(void)
  114. jpeg_fdct_ifast (DCTELEM * data)
  115. {
  116.   DCTELEM tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
  117.   DCTELEM tmp10, tmp11, tmp12, tmp13;
  118.   DCTELEM z1, z2, z3, z4, z5, z11, z13;
  119.   DCTELEM *dataptr;
  120.   int ctr;
  121.   SHIFT_TEMPS
  122.  
  123.   /* Pass 1: process rows. */
  124.  
  125.   dataptr = data;
  126.   for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
  127.     tmp0 = dataptr[0] + dataptr[7];
  128.     tmp7 = dataptr[0] - dataptr[7];
  129.     tmp1 = dataptr[1] + dataptr[6];
  130.     tmp6 = dataptr[1] - dataptr[6];
  131.     tmp2 = dataptr[2] + dataptr[5];
  132.     tmp5 = dataptr[2] - dataptr[5];
  133.     tmp3 = dataptr[3] + dataptr[4];
  134.     tmp4 = dataptr[3] - dataptr[4];
  135.    
  136.     /* Even part */
  137.    
  138.     tmp10 = tmp0 + tmp3;        /* phase 2 */
  139.     tmp13 = tmp0 - tmp3;
  140.     tmp11 = tmp1 + tmp2;
  141.     tmp12 = tmp1 - tmp2;
  142.    
  143.     dataptr[0] = tmp10 + tmp11; /* phase 3 */
  144.     dataptr[4] = tmp10 - tmp11;
  145.    
  146.     z1 = MULTIPLY(tmp12 + tmp13, FIX_0_707106781); /* c4 */
  147.     dataptr[2] = tmp13 + z1;    /* phase 5 */
  148.     dataptr[6] = tmp13 - z1;
  149.    
  150.     /* Odd part */
  151.  
  152.     tmp10 = tmp4 + tmp5;        /* phase 2 */
  153.     tmp11 = tmp5 + tmp6;
  154.     tmp12 = tmp6 + tmp7;
  155.  
  156.     /* The rotator is modified from fig 4-8 to avoid extra negations. */
  157.     z5 = MULTIPLY(tmp10 - tmp12, FIX_0_382683433); /* c6 */
  158.     z2 = MULTIPLY(tmp10, FIX_0_541196100) + z5; /* c2-c6 */
  159.     z4 = MULTIPLY(tmp12, FIX_1_306562965) + z5; /* c2+c6 */
  160.     z3 = MULTIPLY(tmp11, FIX_0_707106781); /* c4 */
  161.  
  162.     z11 = tmp7 + z3;            /* phase 5 */
  163.     z13 = tmp7 - z3;
  164.  
  165.     dataptr[5] = z13 + z2;      /* phase 6 */
  166.     dataptr[3] = z13 - z2;
  167.     dataptr[1] = z11 + z4;
  168.     dataptr[7] = z11 - z4;
  169.  
  170.     dataptr += DCTSIZE;         /* advance pointer to next row */
  171.   }
  172.  
  173.   /* Pass 2: process columns. */
  174.  
  175.   dataptr = data;
  176.   for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
  177.     tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*7];
  178.     tmp7 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*7];
  179.     tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*6];
  180.     tmp6 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*6];
  181.     tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*5];
  182.     tmp5 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*5];
  183.     tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*4];
  184.     tmp4 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*4];
  185.    
  186.     /* Even part */
  187.    
  188.     tmp10 = tmp0 + tmp3;        /* phase 2 */
  189.     tmp13 = tmp0 - tmp3;
  190.     tmp11 = tmp1 + tmp2;
  191.     tmp12 = tmp1 - tmp2;
  192.    
  193.     dataptr[DCTSIZE*0] = tmp10 + tmp11; /* phase 3 */
  194.     dataptr[DCTSIZE*4] = tmp10 - tmp11;
  195.    
  196.     z1 = MULTIPLY(tmp12 + tmp13, FIX_0_707106781); /* c4 */
  197.     dataptr[DCTSIZE*2] = tmp13 + z1; /* phase 5 */
  198.     dataptr[DCTSIZE*6] = tmp13 - z1;
  199.    
  200.     /* Odd part */
  201.  
  202.     tmp10 = tmp4 + tmp5;        /* phase 2 */
  203.     tmp11 = tmp5 + tmp6;
  204.     tmp12 = tmp6 + tmp7;
  205.  
  206.     /* The rotator is modified from fig 4-8 to avoid extra negations. */
  207.     z5 = MULTIPLY(tmp10 - tmp12, FIX_0_382683433); /* c6 */
  208.     z2 = MULTIPLY(tmp10, FIX_0_541196100) + z5; /* c2-c6 */
  209.     z4 = MULTIPLY(tmp12, FIX_1_306562965) + z5; /* c2+c6 */
  210.     z3 = MULTIPLY(tmp11, FIX_0_707106781); /* c4 */
  211.  
  212.     z11 = tmp7 + z3;            /* phase 5 */
  213.     z13 = tmp7 - z3;
  214.  
  215.     dataptr[DCTSIZE*5] = z13 + z2; /* phase 6 */
  216.     dataptr[DCTSIZE*3] = z13 - z2;
  217.     dataptr[DCTSIZE*1] = z11 + z4;
  218.     dataptr[DCTSIZE*7] = z11 - z4;
  219.  
  220.     dataptr++;                  /* advance pointer to next column */
  221.   }
  222. }
  223.  
  224. #endif /* DCT_IFAST_SUPPORTED */
  225.