jfdctfst.c 7.6 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227
  1. /*
  2. * jfdctfst.c
  3. *
  4. * This file was part of the Independent JPEG Group's software:
  5. * Copyright (C) 1994-1996, Thomas G. Lane.
  6. * libjpeg-turbo Modifications:
  7. * Copyright (C) 2015, D. R. Commander.
  8. * For conditions of distribution and use, see the accompanying README.ijg
  9. * file.
  10. *
  11. * This file contains a fast, not so accurate integer implementation of the
  12. * forward DCT (Discrete Cosine Transform).
  13. *
  14. * A 2-D DCT can be done by 1-D DCT on each row followed by 1-D DCT
  15. * on each column. Direct algorithms are also available, but they are
  16. * much more complex and seem not to be any faster when reduced to code.
  17. *
  18. * This implementation is based on Arai, Agui, and Nakajima's algorithm for
  19. * scaled DCT. Their original paper (Trans. IEICE E-71(11):1095) is in
  20. * Japanese, but the algorithm is described in the Pennebaker & Mitchell
  21. * JPEG textbook (see REFERENCES section in file README.ijg). The following
  22. * code is based directly on figure 4-8 in P&M.
  23. * While an 8-point DCT cannot be done in less than 11 multiplies, it is
  24. * possible to arrange the computation so that many of the multiplies are
  25. * simple scalings of the final outputs. These multiplies can then be
  26. * folded into the multiplications or divisions by the JPEG quantization
  27. * table entries. The AA&N method leaves only 5 multiplies and 29 adds
  28. * to be done in the DCT itself.
  29. * The primary disadvantage of this method is that with fixed-point math,
  30. * accuracy is lost due to imprecise representation of the scaled
  31. * quantization values. The smaller the quantization table entry, the less
  32. * precise the scaled value, so this implementation does worse with high-
  33. * quality-setting files than with low-quality ones.
  34. */
  35. #define JPEG_INTERNALS
  36. #include "jinclude.h"
  37. #include "jpeglib.h"
  38. #include "jdct.h" /* Private declarations for DCT subsystem */
  39. #ifdef DCT_IFAST_SUPPORTED
  40. /*
  41. * This module is specialized to the case DCTSIZE = 8.
  42. */
  43. #if DCTSIZE != 8
  44. Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
  45. #endif
  46. /* Scaling decisions are generally the same as in the LL&M algorithm;
  47. * see jfdctint.c for more details. However, we choose to descale
  48. * (right shift) multiplication products as soon as they are formed,
  49. * rather than carrying additional fractional bits into subsequent additions.
  50. * This compromises accuracy slightly, but it lets us save a few shifts.
  51. * More importantly, 16-bit arithmetic is then adequate (for 8-bit samples)
  52. * everywhere except in the multiplications proper; this saves a good deal
  53. * of work on 16-bit-int machines.
  54. *
  55. * Again to save a few shifts, the intermediate results between pass 1 and
  56. * pass 2 are not upscaled, but are represented only to integral precision.
  57. *
  58. * A final compromise is to represent the multiplicative constants to only
  59. * 8 fractional bits, rather than 13. This saves some shifting work on some
  60. * machines, and may also reduce the cost of multiplication (since there
  61. * are fewer one-bits in the constants).
  62. */
  63. #define CONST_BITS 8
  64. /* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
  65. * causing a lot of useless floating-point operations at run time.
  66. * To get around this we use the following pre-calculated constants.
  67. * If you change CONST_BITS you may want to add appropriate values.
  68. * (With a reasonable C compiler, you can just rely on the FIX() macro...)
  69. */
  70. #if CONST_BITS == 8
  71. #define FIX_0_382683433 ((JLONG)98) /* FIX(0.382683433) */
  72. #define FIX_0_541196100 ((JLONG)139) /* FIX(0.541196100) */
  73. #define FIX_0_707106781 ((JLONG)181) /* FIX(0.707106781) */
  74. #define FIX_1_306562965 ((JLONG)334) /* FIX(1.306562965) */
  75. #else
  76. #define FIX_0_382683433 FIX(0.382683433)
  77. #define FIX_0_541196100 FIX(0.541196100)
  78. #define FIX_0_707106781 FIX(0.707106781)
  79. #define FIX_1_306562965 FIX(1.306562965)
  80. #endif
  81. /* We can gain a little more speed, with a further compromise in accuracy,
  82. * by omitting the addition in a descaling shift. This yields an incorrectly
  83. * rounded result half the time...
  84. */
  85. #ifndef USE_ACCURATE_ROUNDING
  86. #undef DESCALE
  87. #define DESCALE(x, n) RIGHT_SHIFT(x, n)
  88. #endif
  89. /* Multiply a DCTELEM variable by an JLONG constant, and immediately
  90. * descale to yield a DCTELEM result.
  91. */
  92. #define MULTIPLY(var, const) ((DCTELEM)DESCALE((var) * (const), CONST_BITS))
  93. /*
  94. * Perform the forward DCT on one block of samples.
  95. */
  96. GLOBAL(void)
  97. jpeg_fdct_ifast(DCTELEM *data)
  98. {
  99. DCTELEM tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
  100. DCTELEM tmp10, tmp11, tmp12, tmp13;
  101. DCTELEM z1, z2, z3, z4, z5, z11, z13;
  102. DCTELEM *dataptr;
  103. int ctr;
  104. SHIFT_TEMPS
  105. /* Pass 1: process rows. */
  106. dataptr = data;
  107. for (ctr = DCTSIZE - 1; ctr >= 0; ctr--) {
  108. tmp0 = dataptr[0] + dataptr[7];
  109. tmp7 = dataptr[0] - dataptr[7];
  110. tmp1 = dataptr[1] + dataptr[6];
  111. tmp6 = dataptr[1] - dataptr[6];
  112. tmp2 = dataptr[2] + dataptr[5];
  113. tmp5 = dataptr[2] - dataptr[5];
  114. tmp3 = dataptr[3] + dataptr[4];
  115. tmp4 = dataptr[3] - dataptr[4];
  116. /* Even part */
  117. tmp10 = tmp0 + tmp3; /* phase 2 */
  118. tmp13 = tmp0 - tmp3;
  119. tmp11 = tmp1 + tmp2;
  120. tmp12 = tmp1 - tmp2;
  121. dataptr[0] = tmp10 + tmp11; /* phase 3 */
  122. dataptr[4] = tmp10 - tmp11;
  123. z1 = MULTIPLY(tmp12 + tmp13, FIX_0_707106781); /* c4 */
  124. dataptr[2] = tmp13 + z1; /* phase 5 */
  125. dataptr[6] = tmp13 - z1;
  126. /* Odd part */
  127. tmp10 = tmp4 + tmp5; /* phase 2 */
  128. tmp11 = tmp5 + tmp6;
  129. tmp12 = tmp6 + tmp7;
  130. /* The rotator is modified from fig 4-8 to avoid extra negations. */
  131. z5 = MULTIPLY(tmp10 - tmp12, FIX_0_382683433); /* c6 */
  132. z2 = MULTIPLY(tmp10, FIX_0_541196100) + z5; /* c2-c6 */
  133. z4 = MULTIPLY(tmp12, FIX_1_306562965) + z5; /* c2+c6 */
  134. z3 = MULTIPLY(tmp11, FIX_0_707106781); /* c4 */
  135. z11 = tmp7 + z3; /* phase 5 */
  136. z13 = tmp7 - z3;
  137. dataptr[5] = z13 + z2; /* phase 6 */
  138. dataptr[3] = z13 - z2;
  139. dataptr[1] = z11 + z4;
  140. dataptr[7] = z11 - z4;
  141. dataptr += DCTSIZE; /* advance pointer to next row */
  142. }
  143. /* Pass 2: process columns. */
  144. dataptr = data;
  145. for (ctr = DCTSIZE - 1; ctr >= 0; ctr--) {
  146. tmp0 = dataptr[DCTSIZE * 0] + dataptr[DCTSIZE * 7];
  147. tmp7 = dataptr[DCTSIZE * 0] - dataptr[DCTSIZE * 7];
  148. tmp1 = dataptr[DCTSIZE * 1] + dataptr[DCTSIZE * 6];
  149. tmp6 = dataptr[DCTSIZE * 1] - dataptr[DCTSIZE * 6];
  150. tmp2 = dataptr[DCTSIZE * 2] + dataptr[DCTSIZE * 5];
  151. tmp5 = dataptr[DCTSIZE * 2] - dataptr[DCTSIZE * 5];
  152. tmp3 = dataptr[DCTSIZE * 3] + dataptr[DCTSIZE * 4];
  153. tmp4 = dataptr[DCTSIZE * 3] - dataptr[DCTSIZE * 4];
  154. /* Even part */
  155. tmp10 = tmp0 + tmp3; /* phase 2 */
  156. tmp13 = tmp0 - tmp3;
  157. tmp11 = tmp1 + tmp2;
  158. tmp12 = tmp1 - tmp2;
  159. dataptr[DCTSIZE * 0] = tmp10 + tmp11; /* phase 3 */
  160. dataptr[DCTSIZE * 4] = tmp10 - tmp11;
  161. z1 = MULTIPLY(tmp12 + tmp13, FIX_0_707106781); /* c4 */
  162. dataptr[DCTSIZE * 2] = tmp13 + z1; /* phase 5 */
  163. dataptr[DCTSIZE * 6] = tmp13 - z1;
  164. /* Odd part */
  165. tmp10 = tmp4 + tmp5; /* phase 2 */
  166. tmp11 = tmp5 + tmp6;
  167. tmp12 = tmp6 + tmp7;
  168. /* The rotator is modified from fig 4-8 to avoid extra negations. */
  169. z5 = MULTIPLY(tmp10 - tmp12, FIX_0_382683433); /* c6 */
  170. z2 = MULTIPLY(tmp10, FIX_0_541196100) + z5; /* c2-c6 */
  171. z4 = MULTIPLY(tmp12, FIX_1_306562965) + z5; /* c2+c6 */
  172. z3 = MULTIPLY(tmp11, FIX_0_707106781); /* c4 */
  173. z11 = tmp7 + z3; /* phase 5 */
  174. z13 = tmp7 - z3;
  175. dataptr[DCTSIZE * 5] = z13 + z2; /* phase 6 */
  176. dataptr[DCTSIZE * 3] = z13 - z2;
  177. dataptr[DCTSIZE * 1] = z11 + z4;
  178. dataptr[DCTSIZE * 7] = z11 - z4;
  179. dataptr++; /* advance pointer to next column */
  180. }
  181. }
  182. #endif /* DCT_IFAST_SUPPORTED */