jfdctflt.c 5.5 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169
  1. /*
  2. * jfdctflt.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.ijg
  7. * file.
  8. *
  9. * This file contains a floating-point implementation of the
  10. * forward DCT (Discrete Cosine Transform).
  11. *
  12. * This implementation should be more accurate than either of the integer
  13. * DCT implementations. However, it may not give the same results on all
  14. * machines because of differences in roundoff behavior. Speed will depend
  15. * on the hardware's floating point capacity.
  16. *
  17. * A 2-D DCT can be done by 1-D DCT on each row followed by 1-D DCT
  18. * on each column. Direct algorithms are also available, but they are
  19. * much more complex and seem not to be any faster when reduced to code.
  20. *
  21. * This implementation is based on Arai, Agui, and Nakajima's algorithm for
  22. * scaled DCT. Their original paper (Trans. IEICE E-71(11):1095) is in
  23. * Japanese, but the algorithm is described in the Pennebaker & Mitchell
  24. * JPEG textbook (see REFERENCES section in file README.ijg). The following
  25. * code is based directly on figure 4-8 in P&M.
  26. * While an 8-point DCT cannot be done in less than 11 multiplies, it is
  27. * possible to arrange the computation so that many of the multiplies are
  28. * simple scalings of the final outputs. These multiplies can then be
  29. * folded into the multiplications or divisions by the JPEG quantization
  30. * table entries. The AA&N method leaves only 5 multiplies and 29 adds
  31. * to be done in the DCT itself.
  32. * The primary disadvantage of this method is that with a fixed-point
  33. * implementation, accuracy is lost due to imprecise representation of the
  34. * scaled quantization values. However, that problem does not arise if
  35. * we use floating point arithmetic.
  36. */
  37. #define JPEG_INTERNALS
  38. #include "jinclude.h"
  39. #include "jpeglib.h"
  40. #include "jdct.h" /* Private declarations for DCT subsystem */
  41. #ifdef DCT_FLOAT_SUPPORTED
  42. /*
  43. * This module is specialized to the case DCTSIZE = 8.
  44. */
  45. #if DCTSIZE != 8
  46. Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
  47. #endif
  48. /*
  49. * Perform the forward DCT on one block of samples.
  50. */
  51. GLOBAL(void)
  52. jpeg_fdct_float(FAST_FLOAT *data)
  53. {
  54. FAST_FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
  55. FAST_FLOAT tmp10, tmp11, tmp12, tmp13;
  56. FAST_FLOAT z1, z2, z3, z4, z5, z11, z13;
  57. FAST_FLOAT *dataptr;
  58. int ctr;
  59. /* Pass 1: process rows. */
  60. dataptr = data;
  61. for (ctr = DCTSIZE - 1; ctr >= 0; ctr--) {
  62. tmp0 = dataptr[0] + dataptr[7];
  63. tmp7 = dataptr[0] - dataptr[7];
  64. tmp1 = dataptr[1] + dataptr[6];
  65. tmp6 = dataptr[1] - dataptr[6];
  66. tmp2 = dataptr[2] + dataptr[5];
  67. tmp5 = dataptr[2] - dataptr[5];
  68. tmp3 = dataptr[3] + dataptr[4];
  69. tmp4 = dataptr[3] - dataptr[4];
  70. /* Even part */
  71. tmp10 = tmp0 + tmp3; /* phase 2 */
  72. tmp13 = tmp0 - tmp3;
  73. tmp11 = tmp1 + tmp2;
  74. tmp12 = tmp1 - tmp2;
  75. dataptr[0] = tmp10 + tmp11; /* phase 3 */
  76. dataptr[4] = tmp10 - tmp11;
  77. z1 = (tmp12 + tmp13) * ((FAST_FLOAT)0.707106781); /* c4 */
  78. dataptr[2] = tmp13 + z1; /* phase 5 */
  79. dataptr[6] = tmp13 - z1;
  80. /* Odd part */
  81. tmp10 = tmp4 + tmp5; /* phase 2 */
  82. tmp11 = tmp5 + tmp6;
  83. tmp12 = tmp6 + tmp7;
  84. /* The rotator is modified from fig 4-8 to avoid extra negations. */
  85. z5 = (tmp10 - tmp12) * ((FAST_FLOAT)0.382683433); /* c6 */
  86. z2 = ((FAST_FLOAT)0.541196100) * tmp10 + z5; /* c2-c6 */
  87. z4 = ((FAST_FLOAT)1.306562965) * tmp12 + z5; /* c2+c6 */
  88. z3 = tmp11 * ((FAST_FLOAT)0.707106781); /* c4 */
  89. z11 = tmp7 + z3; /* phase 5 */
  90. z13 = tmp7 - z3;
  91. dataptr[5] = z13 + z2; /* phase 6 */
  92. dataptr[3] = z13 - z2;
  93. dataptr[1] = z11 + z4;
  94. dataptr[7] = z11 - z4;
  95. dataptr += DCTSIZE; /* advance pointer to next row */
  96. }
  97. /* Pass 2: process columns. */
  98. dataptr = data;
  99. for (ctr = DCTSIZE - 1; ctr >= 0; ctr--) {
  100. tmp0 = dataptr[DCTSIZE * 0] + dataptr[DCTSIZE * 7];
  101. tmp7 = dataptr[DCTSIZE * 0] - dataptr[DCTSIZE * 7];
  102. tmp1 = dataptr[DCTSIZE * 1] + dataptr[DCTSIZE * 6];
  103. tmp6 = dataptr[DCTSIZE * 1] - dataptr[DCTSIZE * 6];
  104. tmp2 = dataptr[DCTSIZE * 2] + dataptr[DCTSIZE * 5];
  105. tmp5 = dataptr[DCTSIZE * 2] - dataptr[DCTSIZE * 5];
  106. tmp3 = dataptr[DCTSIZE * 3] + dataptr[DCTSIZE * 4];
  107. tmp4 = dataptr[DCTSIZE * 3] - dataptr[DCTSIZE * 4];
  108. /* Even part */
  109. tmp10 = tmp0 + tmp3; /* phase 2 */
  110. tmp13 = tmp0 - tmp3;
  111. tmp11 = tmp1 + tmp2;
  112. tmp12 = tmp1 - tmp2;
  113. dataptr[DCTSIZE * 0] = tmp10 + tmp11; /* phase 3 */
  114. dataptr[DCTSIZE * 4] = tmp10 - tmp11;
  115. z1 = (tmp12 + tmp13) * ((FAST_FLOAT)0.707106781); /* c4 */
  116. dataptr[DCTSIZE * 2] = tmp13 + z1; /* phase 5 */
  117. dataptr[DCTSIZE * 6] = tmp13 - z1;
  118. /* Odd part */
  119. tmp10 = tmp4 + tmp5; /* phase 2 */
  120. tmp11 = tmp5 + tmp6;
  121. tmp12 = tmp6 + tmp7;
  122. /* The rotator is modified from fig 4-8 to avoid extra negations. */
  123. z5 = (tmp10 - tmp12) * ((FAST_FLOAT)0.382683433); /* c6 */
  124. z2 = ((FAST_FLOAT)0.541196100) * tmp10 + z5; /* c2-c6 */
  125. z4 = ((FAST_FLOAT)1.306562965) * tmp12 + z5; /* c2+c6 */
  126. z3 = tmp11 * ((FAST_FLOAT)0.707106781); /* c4 */
  127. z11 = tmp7 + z3; /* phase 5 */
  128. z13 = tmp7 - z3;
  129. dataptr[DCTSIZE * 5] = z13 + z2; /* phase 6 */
  130. dataptr[DCTSIZE * 3] = z13 - z2;
  131. dataptr[DCTSIZE * 1] = z11 + z4;
  132. dataptr[DCTSIZE * 7] = z11 - z4;
  133. dataptr++; /* advance pointer to next column */
  134. }
  135. }
  136. #endif /* DCT_FLOAT_SUPPORTED */