Parent Directory | Revision Log
Revision 1989 - (view) (download)
1 : | Isibaar | 1989 | /***************************************************************************** |
2 : | * | ||
3 : | * XVID MPEG-4 VIDEO CODEC | ||
4 : | * - Inverse DCT - | ||
5 : | * | ||
6 : | * Copyright (C) 2006-2011 Xvid Solutions GmbH | ||
7 : | * | ||
8 : | * This program is free software ; you can redistribute it and/or modify | ||
9 : | * it under the terms of the GNU General Public License as published by | ||
10 : | * the Free Software Foundation ; either version 2 of the License, or | ||
11 : | * (at your option) any later version. | ||
12 : | * | ||
13 : | * This program is distributed in the hope that it will be useful, | ||
14 : | * but WITHOUT ANY WARRANTY ; without even the implied warranty of | ||
15 : | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
16 : | * GNU General Public License for more details. | ||
17 : | * | ||
18 : | * You should have received a copy of the GNU General Public License | ||
19 : | * along with this program ; if not, write to the Free Software | ||
20 : | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
21 : | * | ||
22 : | * $Id$ | ||
23 : | * | ||
24 : | ****************************************************************************/ | ||
25 : | Isibaar | 1979 | |
26 : | /* | ||
27 : | * Authors: Skal | ||
28 : | * | ||
29 : | * Walken IDCT | ||
30 : | * Alternative idct implementations for decoding compatibility | ||
31 : | * | ||
32 : | * NOTE: this "C" version is not the original one, | ||
33 : | * but is modified to yield the same error profile | ||
34 : | * than the MMX version. | ||
35 : | * | ||
36 : | ************************************************************************/ | ||
37 : | Isibaar | 1989 | |
38 : | #include "idct.h" | ||
39 : | |||
40 : | /* function pointer */ | ||
41 : | idctFuncPtr idct; | ||
42 : | |||
43 : | Isibaar | 1979 | #define XVID_DSP_CLIP_255(x) ( ((x)&~255) ? ((-(x)) >> (8*sizeof((x))-1))&0xff : (x) ) |
44 : | Isibaar | 1989 | |
45 : | Isibaar | 1979 | #define ROW_SHIFT 11 |
46 : | #define COL_SHIFT 6 | ||
47 : | |||
48 : | // #define FIX(x) (int)((x) * (1<<ROW_SHIFT)) | ||
49 : | #define Rnd0 65536 // 1<<(COL_SHIFT+ROW_SHIFT-1); | ||
50 : | #define Rnd1 3597 // FIX (1.75683487303); | ||
51 : | #define Rnd2 2260 // FIX (1.10355339059); | ||
52 : | #define Rnd3 1203 // FIX (0.587788325588); | ||
53 : | #define Rnd4 0 | ||
54 : | #define Rnd5 120 // FIX (0.058658283817); | ||
55 : | #define Rnd6 512 // FIX (0.25); | ||
56 : | #define Rnd7 512 // FIX (0.25); | ||
57 : | #undef FIX | ||
58 : | |||
59 : | static const int Tab04[] = { 22725, 21407, 19266, 16384, 12873, 8867, 4520 }; | ||
60 : | static const int Tab17[] = { 31521, 29692, 26722, 22725, 17855, 12299, 6270 }; | ||
61 : | static const int Tab26[] = { 29692, 27969, 25172, 21407, 16819, 11585, 5906 }; | ||
62 : | static const int Tab35[] = { 26722, 25172, 22654, 19266, 15137, 10426, 5315 }; | ||
63 : | |||
64 : | static int Idct_Row(short * In, const int * const Tab, int Rnd) | ||
65 : | { | ||
66 : | const int C1 = Tab[0]; | ||
67 : | const int C2 = Tab[1]; | ||
68 : | const int C3 = Tab[2]; | ||
69 : | const int C4 = Tab[3]; | ||
70 : | const int C5 = Tab[4]; | ||
71 : | const int C6 = Tab[5]; | ||
72 : | const int C7 = Tab[6]; | ||
73 : | |||
74 : | const int Right = In[5]|In[6]|In[7]; | ||
75 : | const int Left = In[1]|In[2]|In[3]; | ||
76 : | if (!(Right | In[4])) | ||
77 : | { | ||
78 : | const int K = C4*In[0] + Rnd; | ||
79 : | if (Left) | ||
80 : | { | ||
81 : | const int a0 = K + C2*In[2]; | ||
82 : | const int a1 = K + C6*In[2]; | ||
83 : | const int a2 = K - C6*In[2]; | ||
84 : | const int a3 = K - C2*In[2]; | ||
85 : | |||
86 : | const int b0 = C1*In[1] + C3*In[3]; | ||
87 : | const int b1 = C3*In[1] - C7*In[3]; | ||
88 : | const int b2 = C5*In[1] - C1*In[3]; | ||
89 : | const int b3 = C7*In[1] - C5*In[3]; | ||
90 : | |||
91 : | In[0] = (a0 + b0) >> ROW_SHIFT; | ||
92 : | In[1] = (a1 + b1) >> ROW_SHIFT; | ||
93 : | In[2] = (a2 + b2) >> ROW_SHIFT; | ||
94 : | In[3] = (a3 + b3) >> ROW_SHIFT; | ||
95 : | In[4] = (a3 - b3) >> ROW_SHIFT; | ||
96 : | In[5] = (a2 - b2) >> ROW_SHIFT; | ||
97 : | In[6] = (a1 - b1) >> ROW_SHIFT; | ||
98 : | In[7] = (a0 - b0) >> ROW_SHIFT; | ||
99 : | } | ||
100 : | else | ||
101 : | { | ||
102 : | const int a0 = K >> ROW_SHIFT; | ||
103 : | if (a0) { | ||
104 : | In[0] = In[1] = In[2] = In[3] = | ||
105 : | In[4] = In[5] = In[6] = In[7] = a0; | ||
106 : | } | ||
107 : | else return 0; | ||
108 : | } | ||
109 : | } | ||
110 : | else if (!(Left|Right)) | ||
111 : | { | ||
112 : | const int a0 = (Rnd + C4*(In[0]+In[4])) >> ROW_SHIFT; | ||
113 : | const int a1 = (Rnd + C4*(In[0]-In[4])) >> ROW_SHIFT; | ||
114 : | |||
115 : | In[0] = a0; | ||
116 : | In[3] = a0; | ||
117 : | In[4] = a0; | ||
118 : | In[7] = a0; | ||
119 : | In[1] = a1; | ||
120 : | In[2] = a1; | ||
121 : | In[5] = a1; | ||
122 : | In[6] = a1; | ||
123 : | } | ||
124 : | else | ||
125 : | { | ||
126 : | const int K = C4*In[0] + Rnd; | ||
127 : | const int a0 = K + C2*In[2] + C4*In[4] + C6*In[6]; | ||
128 : | const int a1 = K + C6*In[2] - C4*In[4] - C2*In[6]; | ||
129 : | const int a2 = K - C6*In[2] - C4*In[4] + C2*In[6]; | ||
130 : | const int a3 = K - C2*In[2] + C4*In[4] - C6*In[6]; | ||
131 : | |||
132 : | const int b0 = C1*In[1] + C3*In[3] + C5*In[5] + C7*In[7]; | ||
133 : | const int b1 = C3*In[1] - C7*In[3] - C1*In[5] - C5*In[7]; | ||
134 : | const int b2 = C5*In[1] - C1*In[3] + C7*In[5] + C3*In[7]; | ||
135 : | const int b3 = C7*In[1] - C5*In[3] + C3*In[5] - C1*In[7]; | ||
136 : | |||
137 : | In[0] = (a0 + b0) >> ROW_SHIFT; | ||
138 : | In[1] = (a1 + b1) >> ROW_SHIFT; | ||
139 : | In[2] = (a2 + b2) >> ROW_SHIFT; | ||
140 : | In[3] = (a3 + b3) >> ROW_SHIFT; | ||
141 : | In[4] = (a3 - b3) >> ROW_SHIFT; | ||
142 : | In[5] = (a2 - b2) >> ROW_SHIFT; | ||
143 : | In[6] = (a1 - b1) >> ROW_SHIFT; | ||
144 : | In[7] = (a0 - b0) >> ROW_SHIFT; | ||
145 : | } | ||
146 : | return 1; | ||
147 : | } | ||
148 : | |||
149 : | #define Tan1 0x32ec | ||
150 : | #define Tan2 0x6a0a | ||
151 : | #define Tan3 0xab0e | ||
152 : | #define Sqrt2 0x5a82 | ||
153 : | |||
154 : | #define MULT(c,x, n) ( ((c) * (x)) >> (n) ) | ||
155 : | // 12b version => #define MULT(c,x, n) ( (((c)>>3) * (x)) >> ((n)-3) ) | ||
156 : | // 12b zero-testing version: | ||
157 : | |||
158 : | #define BUTF(a, b, tmp) \ | ||
159 : | (tmp) = (a)+(b); \ | ||
160 : | (b) = (a)-(b); \ | ||
161 : | (a) = (tmp) | ||
162 : | |||
163 : | #define LOAD_BUTF(m1, m2, a, b, tmp, S) \ | ||
164 : | (m1) = (S)[(a)] + (S)[(b)]; \ | ||
165 : | (m2) = (S)[(a)] - (S)[(b)] | ||
166 : | |||
167 : | static void Idct_Col_8(short * const In) | ||
168 : | { | ||
169 : | int mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7, Spill; | ||
170 : | |||
171 : | // odd | ||
172 : | |||
173 : | mm4 = (int)In[7*8]; | ||
174 : | mm5 = (int)In[5*8]; | ||
175 : | mm6 = (int)In[3*8]; | ||
176 : | mm7 = (int)In[1*8]; | ||
177 : | |||
178 : | mm0 = MULT(Tan1, mm4, 16) + mm7; | ||
179 : | mm1 = MULT(Tan1, mm7, 16) - mm4; | ||
180 : | mm2 = MULT(Tan3, mm5, 16) + mm6; | ||
181 : | mm3 = MULT(Tan3, mm6, 16) - mm5; | ||
182 : | |||
183 : | mm7 = mm0 + mm2; | ||
184 : | mm4 = mm1 - mm3; | ||
185 : | mm0 = mm0 - mm2; | ||
186 : | mm1 = mm1 + mm3; | ||
187 : | mm6 = mm0 + mm1; | ||
188 : | mm5 = mm0 - mm1; | ||
189 : | mm5 = 2*MULT(Sqrt2, mm5, 16); // 2*sqrt2 | ||
190 : | mm6 = 2*MULT(Sqrt2, mm6, 16); // Watch out: precision loss but done to match | ||
191 : | // the pmulhw used in mmx/sse versions | ||
192 : | |||
193 : | // even | ||
194 : | |||
195 : | mm1 = (int)In[2*8]; | ||
196 : | mm2 = (int)In[6*8]; | ||
197 : | mm3 = MULT(Tan2,mm2, 16) + mm1; | ||
198 : | mm2 = MULT(Tan2,mm1, 16) - mm2; | ||
199 : | |||
200 : | LOAD_BUTF(mm0, mm1, 0*8, 4*8, Spill, In); | ||
201 : | |||
202 : | BUTF(mm0, mm3, Spill); | ||
203 : | BUTF(mm0, mm7, Spill); | ||
204 : | In[8*0] = (int16_t) (mm0 >> COL_SHIFT); | ||
205 : | In[8*7] = (int16_t) (mm7 >> COL_SHIFT); | ||
206 : | BUTF(mm3, mm4, mm0); | ||
207 : | In[8*3] = (int16_t) (mm3 >> COL_SHIFT); | ||
208 : | In[8*4] = (int16_t) (mm4 >> COL_SHIFT); | ||
209 : | |||
210 : | BUTF(mm1, mm2, mm0); | ||
211 : | BUTF(mm1, mm6, mm0); | ||
212 : | In[8*1] = (int16_t) (mm1 >> COL_SHIFT); | ||
213 : | In[8*6] = (int16_t) (mm6 >> COL_SHIFT); | ||
214 : | BUTF(mm2, mm5, mm0); | ||
215 : | In[8*2] = (int16_t) (mm2 >> COL_SHIFT); | ||
216 : | In[8*5] = (int16_t) (mm5 >> COL_SHIFT); | ||
217 : | } | ||
218 : | |||
219 : | static void Idct_Col_4(short * const In) | ||
220 : | { | ||
221 : | int mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7, Spill; | ||
222 : | |||
223 : | // odd | ||
224 : | |||
225 : | mm0 = (int)In[1*8]; | ||
226 : | mm2 = (int)In[3*8]; | ||
227 : | |||
228 : | mm1 = MULT(Tan1, mm0, 16); | ||
229 : | mm3 = MULT(Tan3, mm2, 16); | ||
230 : | |||
231 : | mm7 = mm0 + mm2; | ||
232 : | mm4 = mm1 - mm3; | ||
233 : | mm0 = mm0 - mm2; | ||
234 : | mm1 = mm1 + mm3; | ||
235 : | mm6 = mm0 + mm1; | ||
236 : | mm5 = mm0 - mm1; | ||
237 : | mm6 = 2*MULT(Sqrt2, mm6, 16); // 2*sqrt2 | ||
238 : | mm5 = 2*MULT(Sqrt2, mm5, 16); | ||
239 : | |||
240 : | // even | ||
241 : | |||
242 : | mm0 = mm1 = (int)In[0*8]; | ||
243 : | mm3 = (int)In[2*8]; | ||
244 : | mm2 = MULT(Tan2,mm3, 16); | ||
245 : | |||
246 : | BUTF(mm0, mm3, Spill); | ||
247 : | BUTF(mm0, mm7, Spill); | ||
248 : | In[8*0] = (int16_t) (mm0 >> COL_SHIFT); | ||
249 : | In[8*7] = (int16_t) (mm7 >> COL_SHIFT); | ||
250 : | BUTF(mm3, mm4, mm0); | ||
251 : | In[8*3] = (int16_t) (mm3 >> COL_SHIFT); | ||
252 : | In[8*4] = (int16_t) (mm4 >> COL_SHIFT); | ||
253 : | |||
254 : | BUTF(mm1, mm2, mm0); | ||
255 : | BUTF(mm1, mm6, mm0); | ||
256 : | In[8*1] = (int16_t) (mm1 >> COL_SHIFT); | ||
257 : | In[8*6] = (int16_t) (mm6 >> COL_SHIFT); | ||
258 : | BUTF(mm2, mm5, mm0); | ||
259 : | In[8*2] = (int16_t) (mm2 >> COL_SHIFT); | ||
260 : | In[8*5] = (int16_t) (mm5 >> COL_SHIFT); | ||
261 : | } | ||
262 : | |||
263 : | static void Idct_Col_3(short * const In) | ||
264 : | { | ||
265 : | int mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7, Spill; | ||
266 : | |||
267 : | // odd | ||
268 : | |||
269 : | mm7 = (int)In[1*8]; | ||
270 : | mm4 = MULT(Tan1, mm7, 16); | ||
271 : | |||
272 : | mm6 = mm7 + mm4; | ||
273 : | mm5 = mm7 - mm4; | ||
274 : | mm6 = 2*MULT(Sqrt2, mm6, 16); // 2*sqrt2 | ||
275 : | mm5 = 2*MULT(Sqrt2, mm5, 16); | ||
276 : | |||
277 : | // even | ||
278 : | |||
279 : | mm0 = mm1 = (int)In[0*8]; | ||
280 : | mm3 = (int)In[2*8]; | ||
281 : | mm2 = MULT(Tan2,mm3, 16); | ||
282 : | |||
283 : | BUTF(mm0, mm3, Spill); | ||
284 : | BUTF(mm0, mm7, Spill); | ||
285 : | In[8*0] = (int16_t) (mm0 >> COL_SHIFT); | ||
286 : | In[8*7] = (int16_t) (mm7 >> COL_SHIFT); | ||
287 : | BUTF(mm3, mm4, mm0); | ||
288 : | In[8*3] = (int16_t) (mm3 >> COL_SHIFT); | ||
289 : | In[8*4] = (int16_t) (mm4 >> COL_SHIFT); | ||
290 : | |||
291 : | BUTF(mm1, mm2, mm0); | ||
292 : | BUTF(mm1, mm6, mm0); | ||
293 : | In[8*1] = (int16_t) (mm1 >> COL_SHIFT); | ||
294 : | In[8*6] = (int16_t) (mm6 >> COL_SHIFT); | ||
295 : | BUTF(mm2, mm5, mm0); | ||
296 : | In[8*2] = (int16_t) (mm2 >> COL_SHIFT); | ||
297 : | In[8*5] = (int16_t) (mm5 >> COL_SHIFT); | ||
298 : | } | ||
299 : | |||
300 : | #undef Tan1 | ||
301 : | #undef Tan2 | ||
302 : | #undef Tan3 | ||
303 : | #undef Sqrt2 | ||
304 : | |||
305 : | #undef ROW_SHIFT | ||
306 : | #undef COL_SHIFT | ||
307 : | |||
308 : | ////////////////////////////////////////////////////////// | ||
309 : | |||
310 : | void idct_int32(short *const In) | ||
311 : | { | ||
312 : | int i, Rows = 0x07; | ||
313 : | |||
314 : | Idct_Row(In + 0*8, Tab04, Rnd0); | ||
315 : | Idct_Row(In + 1*8, Tab17, Rnd1); | ||
316 : | Idct_Row(In + 2*8, Tab26, Rnd2); | ||
317 : | if (Idct_Row(In + 3*8, Tab35, Rnd3)) Rows |= 0x08; | ||
318 : | if (Idct_Row(In + 4*8, Tab04, Rnd4)) Rows |= 0x10; | ||
319 : | if (Idct_Row(In + 5*8, Tab35, Rnd5)) Rows |= 0x20; | ||
320 : | if (Idct_Row(In + 6*8, Tab26, Rnd6)) Rows |= 0x40; | ||
321 : | if (Idct_Row(In + 7*8, Tab17, Rnd7)) Rows |= 0x80; | ||
322 : | |||
323 : | if (Rows&0xf0) { | ||
324 : | for(i=0; i<8; i++) | ||
325 : | Idct_Col_8(In + i); | ||
326 : | } | ||
327 : | else if (Rows&0x08) { | ||
328 : | for(i=0; i<8; i++) | ||
329 : | Idct_Col_4(In + i); | ||
330 : | } | ||
331 : | else { | ||
332 : | for(i=0; i<8; i++) | ||
333 : | Idct_Col_3(In + i); | ||
334 : | } | ||
335 : | } |
No admin address has been configured | ViewVC Help |
Powered by ViewVC 1.0.4 |