## ffmpeg / libavcodec / jfdctfst.c @ bec89a84

History | View | Annotate | Download (7.55 KB)

1 | de6d9b64 | Fabrice Bellard | ```
/*
``` |
---|---|---|---|

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 | 983e3246 | Michael Niedermayer | ```
/**
``` |

34 | ```
* @file jfdctfst.c
``` |
||

35 | ```
* Independent JPEG Group's fast AAN dct.
``` |
||

36 | ```
*/
``` |
||

37 | |||

38 | de6d9b64 | Fabrice Bellard | #include <stdlib.h> |

39 | #include <stdio.h> |
||

40 | #include "common.h" |
||

41 | #include "dsputil.h" |
||

42 | |||

43 | #define DCTSIZE 8 |
||

44 | ```
#define GLOBAL(x) x
``` |
||

45 | ```
#define RIGHT_SHIFT(x, n) ((x) >> (n))
``` |
||

46 | ```
#define SHIFT_TEMPS
``` |
||

47 | |||

48 | ```
/*
``` |
||

49 | ```
* This module is specialized to the case DCTSIZE = 8.
``` |
||

50 | ```
*/
``` |
||

51 | |||

52 | #if DCTSIZE != 8 |
||

53 | Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */ |
||

54 | ```
#endif
``` |
||

55 | |||

56 | |||

57 | ```
/* Scaling decisions are generally the same as in the LL&M algorithm;
``` |
||

58 | ```
* see jfdctint.c for more details. However, we choose to descale
``` |
||

59 | ```
* (right shift) multiplication products as soon as they are formed,
``` |
||

60 | ```
* rather than carrying additional fractional bits into subsequent additions.
``` |
||

61 | ```
* This compromises accuracy slightly, but it lets us save a few shifts.
``` |
||

62 | ```
* More importantly, 16-bit arithmetic is then adequate (for 8-bit samples)
``` |
||

63 | ```
* everywhere except in the multiplications proper; this saves a good deal
``` |
||

64 | ```
* of work on 16-bit-int machines.
``` |
||

65 | ```
*
``` |
||

66 | ```
* Again to save a few shifts, the intermediate results between pass 1 and
``` |
||

67 | ```
* pass 2 are not upscaled, but are represented only to integral precision.
``` |
||

68 | ```
*
``` |
||

69 | ```
* A final compromise is to represent the multiplicative constants to only
``` |
||

70 | ```
* 8 fractional bits, rather than 13. This saves some shifting work on some
``` |
||

71 | ```
* machines, and may also reduce the cost of multiplication (since there
``` |
||

72 | ```
* are fewer one-bits in the constants).
``` |
||

73 | ```
*/
``` |
||

74 | |||

75 | #define CONST_BITS 8 |
||

76 | |||

77 | |||

78 | ```
/* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
``` |
||

79 | ```
* causing a lot of useless floating-point operations at run time.
``` |
||

80 | ```
* To get around this we use the following pre-calculated constants.
``` |
||

81 | ```
* If you change CONST_BITS you may want to add appropriate values.
``` |
||

82 | ```
* (With a reasonable C compiler, you can just rely on the FIX() macro...)
``` |
||

83 | ```
*/
``` |
||

84 | |||

85 | #if CONST_BITS == 8 |
||

86 | 0c1a9eda | Zdenek Kabelac | #define FIX_0_382683433 ((int32_t) 98) /* FIX(0.382683433) */ |

87 | #define FIX_0_541196100 ((int32_t) 139) /* FIX(0.541196100) */ |
||

88 | #define FIX_0_707106781 ((int32_t) 181) /* FIX(0.707106781) */ |
||

89 | #define FIX_1_306562965 ((int32_t) 334) /* FIX(1.306562965) */ |
||

90 | de6d9b64 | Fabrice Bellard | ```
#else
``` |

91 | #define FIX_0_382683433 FIX(0.382683433) |
||

92 | #define FIX_0_541196100 FIX(0.541196100) |
||

93 | #define FIX_0_707106781 FIX(0.707106781) |
||

94 | #define FIX_1_306562965 FIX(1.306562965) |
||

95 | ```
#endif
``` |
||

96 | |||

97 | |||

98 | ```
/* We can gain a little more speed, with a further compromise in accuracy,
``` |
||

99 | ```
* by omitting the addition in a descaling shift. This yields an incorrectly
``` |
||

100 | ```
* rounded result half the time...
``` |
||

101 | ```
*/
``` |
||

102 | |||

103 | ```
#ifndef USE_ACCURATE_ROUNDING
``` |
||

104 | ```
#undef DESCALE
``` |
||

105 | ```
#define DESCALE(x,n) RIGHT_SHIFT(x, n)
``` |
||

106 | ```
#endif
``` |
||

107 | |||

108 | |||

109 | 0c1a9eda | Zdenek Kabelac | ```
/* Multiply a DCTELEM variable by an int32_t constant, and immediately
``` |

110 | de6d9b64 | Fabrice Bellard | ```
* descale to yield a DCTELEM result.
``` |

111 | ```
*/
``` |
||

112 | |||

113 | #define MULTIPLY(var,const) ((DCTELEM) DESCALE((var) * (const), CONST_BITS)) |
||

114 | |||

115 | |||

116 | ```
/*
``` |
||

117 | ```
* Perform the forward DCT on one block of samples.
``` |
||

118 | ```
*/
``` |
||

119 | |||

120 | ```
GLOBAL(void)
``` |
||

121 | 03c94ede | Fabrice Bellard | fdct_ifast (DCTELEM * data) |

122 | de6d9b64 | Fabrice Bellard | { |

123 | DCTELEM tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; |
||

124 | DCTELEM tmp10, tmp11, tmp12, tmp13; |
||

125 | DCTELEM z1, z2, z3, z4, z5, z11, z13; |
||

126 | DCTELEM *dataptr; |
||

127 | ```
int ctr;
``` |
||

128 | SHIFT_TEMPS |
||

129 | |||

130 | ```
/* Pass 1: process rows. */
``` |
||

131 | |||

132 | dataptr = data; |
||

133 | for (ctr = DCTSIZE-1; ctr >= 0; ctr--) { |
||

134 | tmp0 = dataptr[0] + dataptr[7]; |
||

135 | tmp7 = dataptr[0] - dataptr[7]; |
||

136 | tmp1 = dataptr[1] + dataptr[6]; |
||

137 | tmp6 = dataptr[1] - dataptr[6]; |
||

138 | tmp2 = dataptr[2] + dataptr[5]; |
||

139 | tmp5 = dataptr[2] - dataptr[5]; |
||

140 | tmp3 = dataptr[3] + dataptr[4]; |
||

141 | tmp4 = dataptr[3] - dataptr[4]; |
||

142 | |||

143 | ```
/* Even part */
``` |
||

144 | |||

145 | ```
tmp10 = tmp0 + tmp3; /* phase 2 */
``` |
||

146 | tmp13 = tmp0 - tmp3; |
||

147 | tmp11 = tmp1 + tmp2; |
||

148 | tmp12 = tmp1 - tmp2; |
||

149 | |||

150 | dataptr[0] = tmp10 + tmp11; /* phase 3 */ |
||

151 | ```
dataptr[4] = tmp10 - tmp11;
``` |
||

152 | |||

153 | ```
z1 = MULTIPLY(tmp12 + tmp13, FIX_0_707106781); /* c4 */
``` |
||

154 | dataptr[2] = tmp13 + z1; /* phase 5 */ |
||

155 | ```
dataptr[6] = tmp13 - z1;
``` |
||

156 | |||

157 | ```
/* Odd part */
``` |
||

158 | |||

159 | ```
tmp10 = tmp4 + tmp5; /* phase 2 */
``` |
||

160 | tmp11 = tmp5 + tmp6; |
||

161 | tmp12 = tmp6 + tmp7; |
||

162 | |||

163 | ```
/* The rotator is modified from fig 4-8 to avoid extra negations. */
``` |
||

164 | ```
z5 = MULTIPLY(tmp10 - tmp12, FIX_0_382683433); /* c6 */
``` |
||

165 | ```
z2 = MULTIPLY(tmp10, FIX_0_541196100) + z5; /* c2-c6 */
``` |
||

166 | ```
z4 = MULTIPLY(tmp12, FIX_1_306562965) + z5; /* c2+c6 */
``` |
||

167 | ```
z3 = MULTIPLY(tmp11, FIX_0_707106781); /* c4 */
``` |
||

168 | |||

169 | ```
z11 = tmp7 + z3; /* phase 5 */
``` |
||

170 | z13 = tmp7 - z3; |
||

171 | |||

172 | dataptr[5] = z13 + z2; /* phase 6 */ |
||

173 | ```
dataptr[3] = z13 - z2;
``` |
||

174 | ```
dataptr[1] = z11 + z4;
``` |
||

175 | ```
dataptr[7] = z11 - z4;
``` |
||

176 | |||

177 | ```
dataptr += DCTSIZE; /* advance pointer to next row */
``` |
||

178 | } |
||

179 | |||

180 | ```
/* Pass 2: process columns. */
``` |
||

181 | |||

182 | dataptr = data; |
||

183 | for (ctr = DCTSIZE-1; ctr >= 0; ctr--) { |
||

184 | tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*7]; |
||

185 | tmp7 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*7]; |
||

186 | tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*6]; |
||

187 | tmp6 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*6]; |
||

188 | tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*5]; |
||

189 | tmp5 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*5]; |
||

190 | tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*4]; |
||

191 | tmp4 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*4]; |
||

192 | |||

193 | ```
/* Even part */
``` |
||

194 | |||

195 | ```
tmp10 = tmp0 + tmp3; /* phase 2 */
``` |
||

196 | tmp13 = tmp0 - tmp3; |
||

197 | tmp11 = tmp1 + tmp2; |
||

198 | tmp12 = tmp1 - tmp2; |
||

199 | |||

200 | dataptr[DCTSIZE*0] = tmp10 + tmp11; /* phase 3 */ |
||

201 | ```
dataptr[DCTSIZE*4] = tmp10 - tmp11;
``` |
||

202 | |||

203 | ```
z1 = MULTIPLY(tmp12 + tmp13, FIX_0_707106781); /* c4 */
``` |
||

204 | dataptr[DCTSIZE*2] = tmp13 + z1; /* phase 5 */ |
||

205 | ```
dataptr[DCTSIZE*6] = tmp13 - z1;
``` |
||

206 | |||

207 | ```
/* Odd part */
``` |
||

208 | |||

209 | ```
tmp10 = tmp4 + tmp5; /* phase 2 */
``` |
||

210 | tmp11 = tmp5 + tmp6; |
||

211 | tmp12 = tmp6 + tmp7; |
||

212 | |||

213 | ```
/* The rotator is modified from fig 4-8 to avoid extra negations. */
``` |
||

214 | ```
z5 = MULTIPLY(tmp10 - tmp12, FIX_0_382683433); /* c6 */
``` |
||

215 | ```
z2 = MULTIPLY(tmp10, FIX_0_541196100) + z5; /* c2-c6 */
``` |
||

216 | ```
z4 = MULTIPLY(tmp12, FIX_1_306562965) + z5; /* c2+c6 */
``` |
||

217 | ```
z3 = MULTIPLY(tmp11, FIX_0_707106781); /* c4 */
``` |
||

218 | |||

219 | ```
z11 = tmp7 + z3; /* phase 5 */
``` |
||

220 | z13 = tmp7 - z3; |
||

221 | |||

222 | dataptr[DCTSIZE*5] = z13 + z2; /* phase 6 */ |
||

223 | ```
dataptr[DCTSIZE*3] = z13 - z2;
``` |
||

224 | ```
dataptr[DCTSIZE*1] = z11 + z4;
``` |
||

225 | ```
dataptr[DCTSIZE*7] = z11 - z4;
``` |
||

226 | |||

227 | ```
dataptr++; /* advance pointer to next column */
``` |
||

228 | } |
||

229 | } |
||

230 | cd4af68a | Zdenek Kabelac | |

231 | |||

232 | ```
#undef GLOBAL
``` |
||

233 | ```
#undef CONST_BITS
``` |
||

234 | ```
#undef DESCALE
``` |
||

235 | ```
#undef FIX_0_541196100
``` |
||

236 | `#undef FIX_1_306562965` |