[svn] / branches / dev-api-4 / xvidcore / src / dct / fdct.c Repository:
ViewVC logotype

Annotation of /branches/dev-api-4/xvidcore/src/dct/fdct.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 1054 - (view) (download)

1 : edgomez 1054 /*****************************************************************************
2 :     *
3 :     * XVID MPEG-4 VIDEO CODEC
4 :     * - Forward DCT -
5 :     *
6 :     * These routines are from Independent JPEG Group's free JPEG software
7 :     * Copyright (C) 1991-1998, Thomas G. Lane (see the file README.IJG)
8 :     *
9 :     * This program is free software ; you can redistribute it and/or modify
10 :     * it under the terms of the GNU General Public License as published by
11 :     * the Free Software Foundation ; either version 2 of the License, or
12 :     * (at your option) any later version.
13 :     *
14 :     * This program is distributed in the hope that it will be useful,
15 :     * but WITHOUT ANY WARRANTY ; without even the implied warranty of
16 :     * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 :     * GNU General Public License for more details.
18 :     *
19 :     * You should have received a copy of the GNU General Public License
20 :     * along with this program ; if not, write to the Free Software
21 :     * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 :     *
23 :     * $Id: fdct.c,v 1.6.2.2 2003-06-09 13:52:41 edgomez Exp $
24 :     *
25 :     ****************************************************************************/
26 :    
27 : edgomez 851 /* Copyright (C) 1996, MPEG Software Simulation Group. All Rights Reserved. */
28 :    
29 :     /*
30 :     * Disclaimer of Warranty
31 : Isibaar 3 *
32 : edgomez 851 * These software programs are available to the user without any license fee or
33 :     * royalty on an "as is" basis. The MPEG Software Simulation Group disclaims
34 :     * any and all warranties, whether express, implied, or statuary, including any
35 :     * implied warranties or merchantability or of fitness for a particular
36 :     * purpose. In no event shall the copyright-holder be liable for any
37 :     * incidental, punitive, or consequential damages of any kind whatsoever
38 :     * arising from the use of these programs.
39 : Isibaar 3 *
40 : edgomez 851 * This disclaimer of warranty extends to the user of these programs and user's
41 :     * customers, employees, agents, transferees, successors, and assigns.
42 : Isibaar 3 *
43 : edgomez 851 * The MPEG Software Simulation Group does not represent or warrant that the
44 :     * programs furnished hereunder are free of infringement of any third-party
45 :     * patents.
46 : Isibaar 3 *
47 : edgomez 851 * Commercial implementations of MPEG-1 and MPEG-2 video, including shareware,
48 :     * are subject to royalty fees to patent holders. Many of these patents are
49 :     * general enough such that they are unavoidable regardless of implementation
50 :     * design.
51 : Isibaar 3 *
52 : edgomez 851 */
53 : Isibaar 3
54 :     /* This routine is a slow-but-accurate integer implementation of the
55 :     * forward DCT (Discrete Cosine Transform). Taken from the IJG software
56 :     *
57 :     * A 2-D DCT can be done by 1-D DCT on each row followed by 1-D DCT
58 :     * on each column. Direct algorithms are also available, but they are
59 :     * much more complex and seem not to be any faster when reduced to code.
60 :     *
61 :     * This implementation is based on an algorithm described in
62 :     * C. Loeffler, A. Ligtenberg and G. Moschytz, "Practical Fast 1-D DCT
63 :     * Algorithms with 11 Multiplications", Proc. Int'l. Conf. on Acoustics,
64 :     * Speech, and Signal Processing 1989 (ICASSP '89), pp. 988-991.
65 :     * The primary algorithm described there uses 11 multiplies and 29 adds.
66 :     * We use their alternate method with 12 multiplies and 32 adds.
67 :     * The advantage of this method is that no data path contains more than one
68 :     * multiplication; this allows a very simple and accurate implementation in
69 :     * scaled fixed-point arithmetic, with a minimal number of shifts.
70 :     *
71 :     * The poop on this scaling stuff is as follows:
72 :     *
73 :     * Each 1-D DCT step produces outputs which are a factor of sqrt(N)
74 :     * larger than the true DCT outputs. The final outputs are therefore
75 :     * a factor of N larger than desired; since N=8 this can be cured by
76 :     * a simple right shift at the end of the algorithm. The advantage of
77 :     * this arrangement is that we save two multiplications per 1-D DCT,
78 :     * because the y0 and y4 outputs need not be divided by sqrt(N).
79 :     * In the IJG code, this factor of 8 is removed by the quantization step
80 :     * (in jcdctmgr.c), here it is removed.
81 :     *
82 :     * We have to do addition and subtraction of the integer inputs, which
83 :     * is no problem, and multiplication by fractional constants, which is
84 :     * a problem to do in integer arithmetic. We multiply all the constants
85 :     * by CONST_SCALE and convert them to integer constants (thus retaining
86 :     * CONST_BITS bits of precision in the constants). After doing a
87 :     * multiplication we have to divide the product by CONST_SCALE, with proper
88 :     * rounding, to produce the correct output. This division can be done
89 :     * cheaply as a right shift of CONST_BITS bits. We postpone shifting
90 :     * as long as possible so that partial sums can be added together with
91 :     * full fractional precision.
92 :     *
93 :     * The outputs of the first pass are scaled up by PASS1_BITS bits so that
94 :     * they are represented to better-than-integral precision. These outputs
95 :     * require 8 + PASS1_BITS + 3 bits; this fits in a 16-bit word
96 :     * with the recommended scaling. (For 12-bit sample data, the intermediate
97 :     * array is INT32 anyway.)
98 :     *
99 :     * To avoid overflow of the 32-bit intermediate results in pass 2, we must
100 :     * have 8 + CONST_BITS + PASS1_BITS <= 26. Error analysis
101 :     * shows that the values given below are the most effective.
102 :     *
103 :     * We can gain a little more speed, with a further compromise in accuracy,
104 :     * by omitting the addition in a descaling shift. This yields an incorrectly
105 :     * rounded result half the time...
106 :     */
107 :    
108 :     #include "fdct.h"
109 :    
110 :     #define USE_ACCURATE_ROUNDING
111 :    
112 :     #define RIGHT_SHIFT(x, shft) ((x) >> (shft))
113 :    
114 :     #ifdef USE_ACCURATE_ROUNDING
115 :     #define ONE ((int) 1)
116 :     #define DESCALE(x, n) RIGHT_SHIFT((x) + (ONE << ((n) - 1)), n)
117 :     #else
118 :     #define DESCALE(x, n) RIGHT_SHIFT(x, n)
119 :     #endif
120 :    
121 :     #define CONST_BITS 13
122 :     #define PASS1_BITS 2
123 :    
124 :     #define FIX_0_298631336 ((int) 2446) /* FIX(0.298631336) */
125 :     #define FIX_0_390180644 ((int) 3196) /* FIX(0.390180644) */
126 :     #define FIX_0_541196100 ((int) 4433) /* FIX(0.541196100) */
127 :     #define FIX_0_765366865 ((int) 6270) /* FIX(0.765366865) */
128 :     #define FIX_0_899976223 ((int) 7373) /* FIX(0.899976223) */
129 :     #define FIX_1_175875602 ((int) 9633) /* FIX(1.175875602) */
130 :     #define FIX_1_501321110 ((int) 12299) /* FIX(1.501321110) */
131 :     #define FIX_1_847759065 ((int) 15137) /* FIX(1.847759065) */
132 :     #define FIX_1_961570560 ((int) 16069) /* FIX(1.961570560) */
133 :     #define FIX_2_053119869 ((int) 16819) /* FIX(2.053119869) */
134 :     #define FIX_2_562915447 ((int) 20995) /* FIX(2.562915447) */
135 :     #define FIX_3_072711026 ((int) 25172) /* FIX(3.072711026) */
136 :    
137 : edgomez 1053 /* function pointer */
138 : Isibaar 3 fdctFuncPtr fdct;
139 :    
140 :     /*
141 :     * Perform an integer forward DCT on one block of samples.
142 :     */
143 :    
144 : edgomez 195 void
145 :     fdct_int32(short *const block)
146 : Isibaar 3 {
147 : edgomez 195 int tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
148 :     int tmp10, tmp11, tmp12, tmp13;
149 :     int z1, z2, z3, z4, z5;
150 :     short *blkptr;
151 :     int *dataptr;
152 :     int data[64];
153 :     int i;
154 : Isibaar 3
155 : edgomez 195 /* Pass 1: process rows. */
156 :     /* Note results are scaled up by sqrt(8) compared to a true DCT; */
157 :     /* furthermore, we scale the results by 2**PASS1_BITS. */
158 : Isibaar 3
159 : edgomez 195 dataptr = data;
160 :     blkptr = block;
161 :     for (i = 0; i < 8; i++) {
162 :     tmp0 = blkptr[0] + blkptr[7];
163 :     tmp7 = blkptr[0] - blkptr[7];
164 :     tmp1 = blkptr[1] + blkptr[6];
165 :     tmp6 = blkptr[1] - blkptr[6];
166 :     tmp2 = blkptr[2] + blkptr[5];
167 :     tmp5 = blkptr[2] - blkptr[5];
168 :     tmp3 = blkptr[3] + blkptr[4];
169 :     tmp4 = blkptr[3] - blkptr[4];
170 : Isibaar 3
171 : edgomez 195 /* Even part per LL&M figure 1 --- note that published figure is faulty;
172 :     * rotator "sqrt(2)*c1" should be "sqrt(2)*c6".
173 :     */
174 : Isibaar 3
175 : edgomez 195 tmp10 = tmp0 + tmp3;
176 :     tmp13 = tmp0 - tmp3;
177 :     tmp11 = tmp1 + tmp2;
178 :     tmp12 = tmp1 - tmp2;
179 : Isibaar 3
180 : edgomez 195 dataptr[0] = (tmp10 + tmp11) << PASS1_BITS;
181 :     dataptr[4] = (tmp10 - tmp11) << PASS1_BITS;
182 : Isibaar 3
183 : edgomez 195 z1 = (tmp12 + tmp13) * FIX_0_541196100;
184 :     dataptr[2] =
185 :     DESCALE(z1 + tmp13 * FIX_0_765366865, CONST_BITS - PASS1_BITS);
186 :     dataptr[6] =
187 :     DESCALE(z1 + tmp12 * (-FIX_1_847759065), CONST_BITS - PASS1_BITS);
188 : Isibaar 3
189 : edgomez 195 /* Odd part per figure 8 --- note paper omits factor of sqrt(2).
190 :     * cK represents cos(K*pi/16).
191 :     * i0..i3 in the paper are tmp4..tmp7 here.
192 :     */
193 : Isibaar 3
194 : edgomez 195 z1 = tmp4 + tmp7;
195 :     z2 = tmp5 + tmp6;
196 :     z3 = tmp4 + tmp6;
197 :     z4 = tmp5 + tmp7;
198 :     z5 = (z3 + z4) * FIX_1_175875602; /* sqrt(2) * c3 */
199 : Isibaar 3
200 : edgomez 195 tmp4 *= FIX_0_298631336; /* sqrt(2) * (-c1+c3+c5-c7) */
201 :     tmp5 *= FIX_2_053119869; /* sqrt(2) * ( c1+c3-c5+c7) */
202 :     tmp6 *= FIX_3_072711026; /* sqrt(2) * ( c1+c3+c5-c7) */
203 :     tmp7 *= FIX_1_501321110; /* sqrt(2) * ( c1+c3-c5-c7) */
204 :     z1 *= -FIX_0_899976223; /* sqrt(2) * (c7-c3) */
205 :     z2 *= -FIX_2_562915447; /* sqrt(2) * (-c1-c3) */
206 :     z3 *= -FIX_1_961570560; /* sqrt(2) * (-c3-c5) */
207 :     z4 *= -FIX_0_390180644; /* sqrt(2) * (c5-c3) */
208 : Isibaar 3
209 : edgomez 195 z3 += z5;
210 :     z4 += z5;
211 : Isibaar 3
212 : edgomez 195 dataptr[7] = DESCALE(tmp4 + z1 + z3, CONST_BITS - PASS1_BITS);
213 :     dataptr[5] = DESCALE(tmp5 + z2 + z4, CONST_BITS - PASS1_BITS);
214 :     dataptr[3] = DESCALE(tmp6 + z2 + z3, CONST_BITS - PASS1_BITS);
215 :     dataptr[1] = DESCALE(tmp7 + z1 + z4, CONST_BITS - PASS1_BITS);
216 : Isibaar 3
217 : edgomez 195 dataptr += 8; /* advance pointer to next row */
218 :     blkptr += 8;
219 :     }
220 : Isibaar 3
221 : edgomez 195 /* Pass 2: process columns.
222 :     * We remove the PASS1_BITS scaling, but leave the results scaled up
223 :     * by an overall factor of 8.
224 :     */
225 : Isibaar 3
226 : edgomez 195 dataptr = data;
227 :     for (i = 0; i < 8; i++) {
228 :     tmp0 = dataptr[0] + dataptr[56];
229 :     tmp7 = dataptr[0] - dataptr[56];
230 :     tmp1 = dataptr[8] + dataptr[48];
231 :     tmp6 = dataptr[8] - dataptr[48];
232 :     tmp2 = dataptr[16] + dataptr[40];
233 :     tmp5 = dataptr[16] - dataptr[40];
234 :     tmp3 = dataptr[24] + dataptr[32];
235 :     tmp4 = dataptr[24] - dataptr[32];
236 : Isibaar 3
237 : edgomez 195 /* Even part per LL&M figure 1 --- note that published figure is faulty;
238 :     * rotator "sqrt(2)*c1" should be "sqrt(2)*c6".
239 :     */
240 : Isibaar 3
241 : edgomez 195 tmp10 = tmp0 + tmp3;
242 :     tmp13 = tmp0 - tmp3;
243 :     tmp11 = tmp1 + tmp2;
244 :     tmp12 = tmp1 - tmp2;
245 : Isibaar 3
246 : edgomez 195 dataptr[0] = DESCALE(tmp10 + tmp11, PASS1_BITS);
247 :     dataptr[32] = DESCALE(tmp10 - tmp11, PASS1_BITS);
248 : Isibaar 3
249 : edgomez 195 z1 = (tmp12 + tmp13) * FIX_0_541196100;
250 :     dataptr[16] =
251 :     DESCALE(z1 + tmp13 * FIX_0_765366865, CONST_BITS + PASS1_BITS);
252 :     dataptr[48] =
253 :     DESCALE(z1 + tmp12 * (-FIX_1_847759065), CONST_BITS + PASS1_BITS);
254 : Isibaar 3
255 : edgomez 195 /* Odd part per figure 8 --- note paper omits factor of sqrt(2).
256 :     * cK represents cos(K*pi/16).
257 :     * i0..i3 in the paper are tmp4..tmp7 here.
258 :     */
259 : Isibaar 3
260 : edgomez 195 z1 = tmp4 + tmp7;
261 :     z2 = tmp5 + tmp6;
262 :     z3 = tmp4 + tmp6;
263 :     z4 = tmp5 + tmp7;
264 :     z5 = (z3 + z4) * FIX_1_175875602; /* sqrt(2) * c3 */
265 : Isibaar 3
266 : edgomez 195 tmp4 *= FIX_0_298631336; /* sqrt(2) * (-c1+c3+c5-c7) */
267 :     tmp5 *= FIX_2_053119869; /* sqrt(2) * ( c1+c3-c5+c7) */
268 :     tmp6 *= FIX_3_072711026; /* sqrt(2) * ( c1+c3+c5-c7) */
269 :     tmp7 *= FIX_1_501321110; /* sqrt(2) * ( c1+c3-c5-c7) */
270 :     z1 *= -FIX_0_899976223; /* sqrt(2) * (c7-c3) */
271 :     z2 *= -FIX_2_562915447; /* sqrt(2) * (-c1-c3) */
272 :     z3 *= -FIX_1_961570560; /* sqrt(2) * (-c3-c5) */
273 :     z4 *= -FIX_0_390180644; /* sqrt(2) * (c5-c3) */
274 : Isibaar 3
275 : edgomez 195 z3 += z5;
276 :     z4 += z5;
277 : Isibaar 3
278 : edgomez 195 dataptr[56] = DESCALE(tmp4 + z1 + z3, CONST_BITS + PASS1_BITS);
279 :     dataptr[40] = DESCALE(tmp5 + z2 + z4, CONST_BITS + PASS1_BITS);
280 :     dataptr[24] = DESCALE(tmp6 + z2 + z3, CONST_BITS + PASS1_BITS);
281 :     dataptr[8] = DESCALE(tmp7 + z1 + z4, CONST_BITS + PASS1_BITS);
282 : Isibaar 3
283 : edgomez 195 dataptr++; /* advance pointer to next column */
284 :     }
285 :     /* descale */
286 :     for (i = 0; i < 64; i++)
287 :     block[i] = (short int) DESCALE(data[i], 3);
288 : Isibaar 3 }

No admin address has been configured
ViewVC Help
Powered by ViewVC 1.0.4