blob: 4642a30d62e7f7a15f6b6cdca341f88558ffbb50
1 | /* |
2 | * Xvid MPEG-4 IDCT |
3 | * |
4 | * Copyright (C) 2006-2011 Xvid Solutions GmbH |
5 | * |
6 | * This file is part of FFmpeg. |
7 | * |
8 | * FFmpeg is free software; you can redistribute it and/or |
9 | * modify it under the terms of the GNU Lesser General Public |
10 | * License as published by the Free Software Foundation; either |
11 | * version 2.1 of the License, or (at your option) any later version. |
12 | * |
13 | * FFmpeg 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 GNU |
16 | * Lesser General Public License for more details. |
17 | * |
18 | * You should have received a copy of the GNU Lesser General Public |
19 | * License along with FFmpeg; if not, write to the Free Software |
20 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
21 | */ |
22 | |
23 | /** |
24 | * @file |
25 | * Walken IDCT |
26 | * Alternative IDCT implementation for decoding compatibility. |
27 | * |
28 | * @author Skal |
29 | * @note This C version is not the original IDCT, but a modified one that |
30 | * yields the same error profile as the MMX/MMXEXT/SSE2 versions. |
31 | */ |
32 | |
33 | #include "config.h" |
34 | #include "libavutil/attributes.h" |
35 | #include "avcodec.h" |
36 | #include "idctdsp.h" |
37 | #include "xvididct.h" |
38 | |
39 | #define ROW_SHIFT 11 |
40 | #define COL_SHIFT 6 |
41 | |
42 | // #define FIX(x) (int)((x) * (1 << ROW_SHIFT)) |
43 | #define RND0 65536 // 1 << (COL_SHIFT + ROW_SHIFT - 1); |
44 | #define RND1 3597 // FIX (1.75683487303); |
45 | #define RND2 2260 // FIX (1.10355339059); |
46 | #define RND3 1203 // FIX (0.587788325588); |
47 | #define RND4 0 |
48 | #define RND5 120 // FIX (0.058658283817); |
49 | #define RND6 512 // FIX (0.25); |
50 | #define RND7 512 // FIX (0.25); |
51 | |
52 | static const int TAB04[] = { 22725, 21407, 19266, 16384, 12873, 8867, 4520 }; |
53 | static const int TAB17[] = { 31521, 29692, 26722, 22725, 17855, 12299, 6270 }; |
54 | static const int TAB26[] = { 29692, 27969, 25172, 21407, 16819, 11585, 5906 }; |
55 | static const int TAB35[] = { 26722, 25172, 22654, 19266, 15137, 10426, 5315 }; |
56 | |
57 | static int idct_row(short *in, const int *const tab, int rnd) |
58 | { |
59 | const int c1 = tab[0]; |
60 | const int c2 = tab[1]; |
61 | const int c3 = tab[2]; |
62 | const int c4 = tab[3]; |
63 | const int c5 = tab[4]; |
64 | const int c6 = tab[5]; |
65 | const int c7 = tab[6]; |
66 | |
67 | const int right = in[5] | in[6] | in[7]; |
68 | const int left = in[1] | in[2] | in[3]; |
69 | if (!(right | in[4])) { |
70 | const int k = c4 * in[0] + rnd; |
71 | if (left) { |
72 | const int a0 = k + c2 * in[2]; |
73 | const int a1 = k + c6 * in[2]; |
74 | const int a2 = k - c6 * in[2]; |
75 | const int a3 = k - c2 * in[2]; |
76 | |
77 | const int b0 = c1 * in[1] + c3 * in[3]; |
78 | const int b1 = c3 * in[1] - c7 * in[3]; |
79 | const int b2 = c5 * in[1] - c1 * in[3]; |
80 | const int b3 = c7 * in[1] - c5 * in[3]; |
81 | |
82 | in[0] = (a0 + b0) >> ROW_SHIFT; |
83 | in[1] = (a1 + b1) >> ROW_SHIFT; |
84 | in[2] = (a2 + b2) >> ROW_SHIFT; |
85 | in[3] = (a3 + b3) >> ROW_SHIFT; |
86 | in[4] = (a3 - b3) >> ROW_SHIFT; |
87 | in[5] = (a2 - b2) >> ROW_SHIFT; |
88 | in[6] = (a1 - b1) >> ROW_SHIFT; |
89 | in[7] = (a0 - b0) >> ROW_SHIFT; |
90 | } else { |
91 | const int a0 = k >> ROW_SHIFT; |
92 | if (a0) { |
93 | in[0] = |
94 | in[1] = |
95 | in[2] = |
96 | in[3] = |
97 | in[4] = |
98 | in[5] = |
99 | in[6] = |
100 | in[7] = a0; |
101 | } else |
102 | return 0; |
103 | } |
104 | } else if (!(left | right)) { |
105 | const int a0 = (rnd + c4 * (in[0] + in[4])) >> ROW_SHIFT; |
106 | const int a1 = (rnd + c4 * (in[0] - in[4])) >> ROW_SHIFT; |
107 | |
108 | in[0] = a0; |
109 | in[3] = a0; |
110 | in[4] = a0; |
111 | in[7] = a0; |
112 | in[1] = a1; |
113 | in[2] = a1; |
114 | in[5] = a1; |
115 | in[6] = a1; |
116 | } else { |
117 | const int k = c4 * in[0] + rnd; |
118 | const int a0 = k + c2 * in[2] + c4 * in[4] + c6 * in[6]; |
119 | const int a1 = k + c6 * in[2] - c4 * in[4] - c2 * in[6]; |
120 | const int a2 = k - c6 * in[2] - c4 * in[4] + c2 * in[6]; |
121 | const int a3 = k - c2 * in[2] + c4 * in[4] - c6 * in[6]; |
122 | |
123 | const int b0 = c1 * in[1] + c3 * in[3] + c5 * in[5] + c7 * in[7]; |
124 | const int b1 = c3 * in[1] - c7 * in[3] - c1 * in[5] - c5 * in[7]; |
125 | const int b2 = c5 * in[1] - c1 * in[3] + c7 * in[5] + c3 * in[7]; |
126 | const int b3 = c7 * in[1] - c5 * in[3] + c3 * in[5] - c1 * in[7]; |
127 | |
128 | in[0] = (a0 + b0) >> ROW_SHIFT; |
129 | in[1] = (a1 + b1) >> ROW_SHIFT; |
130 | in[2] = (a2 + b2) >> ROW_SHIFT; |
131 | in[3] = (a3 + b3) >> ROW_SHIFT; |
132 | in[4] = (a3 - b3) >> ROW_SHIFT; |
133 | in[5] = (a2 - b2) >> ROW_SHIFT; |
134 | in[6] = (a1 - b1) >> ROW_SHIFT; |
135 | in[7] = (a0 - b0) >> ROW_SHIFT; |
136 | } |
137 | return 1; |
138 | } |
139 | |
140 | #define TAN1 0x32EC |
141 | #define TAN2 0x6A0A |
142 | #define TAN3 0xAB0E |
143 | #define SQRT2 0x5A82 |
144 | |
145 | #define MULT(c, x, n) (((c) * (x)) >> (n)) |
146 | // 12b version => #define MULT(c,x, n) ((((c) >> 3) * (x)) >> ((n) - 3)) |
147 | // 12b zero-testing version: |
148 | |
149 | #define BUTTERFLY(a, b, tmp) \ |
150 | (tmp) = (a) + (b); \ |
151 | (b) = (a) - (b); \ |
152 | (a) = (tmp) |
153 | |
154 | #define LOAD_BUTTERFLY(m1, m2, a, b, tmp, s) \ |
155 | (m1) = (s)[(a)] + (s)[(b)]; \ |
156 | (m2) = (s)[(a)] - (s)[(b)] |
157 | |
158 | static void idct_col_8(short *const in) |
159 | { |
160 | int mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7, spill; |
161 | |
162 | // odd |
163 | |
164 | mm4 = (int) in[7 * 8]; |
165 | mm5 = (int) in[5 * 8]; |
166 | mm6 = (int) in[3 * 8]; |
167 | mm7 = (int) in[1 * 8]; |
168 | |
169 | mm0 = MULT(TAN1, mm4, 16) + mm7; |
170 | mm1 = MULT(TAN1, mm7, 16) - mm4; |
171 | mm2 = MULT(TAN3, mm5, 16) + mm6; |
172 | mm3 = MULT(TAN3, mm6, 16) - mm5; |
173 | |
174 | mm7 = mm0 + mm2; |
175 | mm4 = mm1 - mm3; |
176 | mm0 = mm0 - mm2; |
177 | mm1 = mm1 + mm3; |
178 | mm6 = mm0 + mm1; |
179 | mm5 = mm0 - mm1; |
180 | mm5 = 2 * MULT(SQRT2, mm5, 16); // 2*sqrt2 |
181 | mm6 = 2 * MULT(SQRT2, mm6, 16); // Watch out: precision loss but done to match |
182 | // the pmulhw used in MMX/MMXEXT/SSE2 versions |
183 | |
184 | // even |
185 | |
186 | mm1 = (int) in[2 * 8]; |
187 | mm2 = (int) in[6 * 8]; |
188 | mm3 = MULT(TAN2, mm2, 16) + mm1; |
189 | mm2 = MULT(TAN2, mm1, 16) - mm2; |
190 | |
191 | LOAD_BUTTERFLY(mm0, mm1, 0 * 8, 4 * 8, spill, in); |
192 | |
193 | BUTTERFLY(mm0, mm3, spill); |
194 | BUTTERFLY(mm0, mm7, spill); |
195 | in[8 * 0] = (int16_t) (mm0 >> COL_SHIFT); |
196 | in[8 * 7] = (int16_t) (mm7 >> COL_SHIFT); |
197 | BUTTERFLY(mm3, mm4, mm0); |
198 | in[8 * 3] = (int16_t) (mm3 >> COL_SHIFT); |
199 | in[8 * 4] = (int16_t) (mm4 >> COL_SHIFT); |
200 | |
201 | BUTTERFLY(mm1, mm2, mm0); |
202 | BUTTERFLY(mm1, mm6, mm0); |
203 | in[8 * 1] = (int16_t) (mm1 >> COL_SHIFT); |
204 | in[8 * 6] = (int16_t) (mm6 >> COL_SHIFT); |
205 | BUTTERFLY(mm2, mm5, mm0); |
206 | in[8 * 2] = (int16_t) (mm2 >> COL_SHIFT); |
207 | in[8 * 5] = (int16_t) (mm5 >> COL_SHIFT); |
208 | } |
209 | |
210 | static void idct_col_4(short *const in) |
211 | { |
212 | int mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7, spill; |
213 | |
214 | // odd |
215 | |
216 | mm0 = (int) in[1 * 8]; |
217 | mm2 = (int) in[3 * 8]; |
218 | |
219 | mm1 = MULT(TAN1, mm0, 16); |
220 | mm3 = MULT(TAN3, mm2, 16); |
221 | |
222 | mm7 = mm0 + mm2; |
223 | mm4 = mm1 - mm3; |
224 | mm0 = mm0 - mm2; |
225 | mm1 = mm1 + mm3; |
226 | mm6 = mm0 + mm1; |
227 | mm5 = mm0 - mm1; |
228 | mm6 = 2 * MULT(SQRT2, mm6, 16); // 2*sqrt2 |
229 | mm5 = 2 * MULT(SQRT2, mm5, 16); |
230 | |
231 | // even |
232 | |
233 | mm0 = mm1 = (int) in[0 * 8]; |
234 | mm3 = (int) in[2 * 8]; |
235 | mm2 = MULT(TAN2, mm3, 16); |
236 | |
237 | BUTTERFLY(mm0, mm3, spill); |
238 | BUTTERFLY(mm0, mm7, spill); |
239 | in[8 * 0] = (int16_t) (mm0 >> COL_SHIFT); |
240 | in[8 * 7] = (int16_t) (mm7 >> COL_SHIFT); |
241 | BUTTERFLY(mm3, mm4, mm0); |
242 | in[8 * 3] = (int16_t) (mm3 >> COL_SHIFT); |
243 | in[8 * 4] = (int16_t) (mm4 >> COL_SHIFT); |
244 | |
245 | BUTTERFLY(mm1, mm2, mm0); |
246 | BUTTERFLY(mm1, mm6, mm0); |
247 | in[8 * 1] = (int16_t) (mm1 >> COL_SHIFT); |
248 | in[8 * 6] = (int16_t) (mm6 >> COL_SHIFT); |
249 | BUTTERFLY(mm2, mm5, mm0); |
250 | in[8 * 2] = (int16_t) (mm2 >> COL_SHIFT); |
251 | in[8 * 5] = (int16_t) (mm5 >> COL_SHIFT); |
252 | } |
253 | |
254 | static void idct_col_3(short *const in) |
255 | { |
256 | int mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7, spill; |
257 | |
258 | // odd |
259 | |
260 | mm7 = (int) in[1 * 8]; |
261 | mm4 = MULT(TAN1, mm7, 16); |
262 | |
263 | mm6 = mm7 + mm4; |
264 | mm5 = mm7 - mm4; |
265 | mm6 = 2 * MULT(SQRT2, mm6, 16); // 2*sqrt2 |
266 | mm5 = 2 * MULT(SQRT2, mm5, 16); |
267 | |
268 | // even |
269 | |
270 | mm0 = mm1 = (int) in[0 * 8]; |
271 | mm3 = (int) in[2 * 8]; |
272 | mm2 = MULT(TAN2, mm3, 16); |
273 | |
274 | BUTTERFLY(mm0, mm3, spill); |
275 | BUTTERFLY(mm0, mm7, spill); |
276 | in[8 * 0] = (int16_t) (mm0 >> COL_SHIFT); |
277 | in[8 * 7] = (int16_t) (mm7 >> COL_SHIFT); |
278 | BUTTERFLY(mm3, mm4, mm0); |
279 | in[8 * 3] = (int16_t) (mm3 >> COL_SHIFT); |
280 | in[8 * 4] = (int16_t) (mm4 >> COL_SHIFT); |
281 | |
282 | BUTTERFLY(mm1, mm2, mm0); |
283 | BUTTERFLY(mm1, mm6, mm0); |
284 | in[8 * 1] = (int16_t) (mm1 >> COL_SHIFT); |
285 | in[8 * 6] = (int16_t) (mm6 >> COL_SHIFT); |
286 | BUTTERFLY(mm2, mm5, mm0); |
287 | in[8 * 2] = (int16_t) (mm2 >> COL_SHIFT); |
288 | in[8 * 5] = (int16_t) (mm5 >> COL_SHIFT); |
289 | } |
290 | |
291 | void ff_xvid_idct(int16_t *const in) |
292 | { |
293 | int i, rows = 0x07; |
294 | |
295 | idct_row(in + 0 * 8, TAB04, RND0); |
296 | idct_row(in + 1 * 8, TAB17, RND1); |
297 | idct_row(in + 2 * 8, TAB26, RND2); |
298 | if (idct_row(in + 3 * 8, TAB35, RND3)) |
299 | rows |= 0x08; |
300 | if (idct_row(in + 4 * 8, TAB04, RND4)) |
301 | rows |= 0x10; |
302 | if (idct_row(in + 5 * 8, TAB35, RND5)) |
303 | rows |= 0x20; |
304 | if (idct_row(in + 6 * 8, TAB26, RND6)) |
305 | rows |= 0x40; |
306 | if (idct_row(in + 7 * 8, TAB17, RND7)) |
307 | rows |= 0x80; |
308 | |
309 | if (rows & 0xF0) { |
310 | for (i = 0; i < 8; i++) |
311 | idct_col_8(in + i); |
312 | } else if (rows & 0x08) { |
313 | for (i = 0; i < 8; i++) |
314 | idct_col_4(in + i); |
315 | } else { |
316 | for (i = 0; i < 8; i++) |
317 | idct_col_3(in + i); |
318 | } |
319 | } |
320 | |
321 | static void xvid_idct_put(uint8_t *dest, ptrdiff_t line_size, int16_t *block) |
322 | { |
323 | ff_xvid_idct(block); |
324 | ff_put_pixels_clamped(block, dest, line_size); |
325 | } |
326 | |
327 | static void xvid_idct_add(uint8_t *dest, ptrdiff_t line_size, int16_t *block) |
328 | { |
329 | ff_xvid_idct(block); |
330 | ff_add_pixels_clamped(block, dest, line_size); |
331 | } |
332 | |
333 | av_cold void ff_xvid_idct_init(IDCTDSPContext *c, AVCodecContext *avctx) |
334 | { |
335 | const unsigned high_bit_depth = avctx->bits_per_raw_sample > 8; |
336 | |
337 | if (high_bit_depth || avctx->lowres || |
338 | !(avctx->idct_algo == FF_IDCT_AUTO || |
339 | avctx->idct_algo == FF_IDCT_XVID)) |
340 | return; |
341 | |
342 | if (avctx->idct_algo == FF_IDCT_XVID) { |
343 | c->idct_put = xvid_idct_put; |
344 | c->idct_add = xvid_idct_add; |
345 | c->idct = ff_xvid_idct; |
346 | c->perm_type = FF_IDCT_PERM_NONE; |
347 | } |
348 | |
349 | if (ARCH_X86) |
350 | ff_xvid_idct_init_x86(c, avctx, high_bit_depth); |
351 | if (ARCH_MIPS) |
352 | ff_xvid_idct_init_mips(c, avctx, high_bit_depth); |
353 | |
354 | ff_init_scantable_permutation(c->idct_permutation, c->perm_type); |
355 | } |
356 |