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
35 #include "libavutil/attributes.h"
36 #include "avcodec.h"
37 #include "idctdsp.h"
38 #include "xvididct.h"
39
40 #define ROW_SHIFT 11
41 #define COL_SHIFT 6
42
43 // #define FIX(x) (int)((x) * (1 << ROW_SHIFT))
44 #define RND0 65536 // 1 << (COL_SHIFT + ROW_SHIFT - 1);
45 #define RND1 3597 // FIX (1.75683487303);
46 #define RND2 2260 // FIX (1.10355339059);
47 #define RND3 1203 // FIX (0.587788325588);
48 #define RND4 0
49 #define RND5 120 // FIX (0.058658283817);
50 #define RND6 512 // FIX (0.25);
51 #define RND7 512 // FIX (0.25);
52
53 static const int TAB04[] = { 22725, 21407, 19266, 16384, 12873, 8867, 4520 };
54 static const int TAB17[] = { 31521, 29692, 26722, 22725, 17855, 12299, 6270 };
55 static const int TAB26[] = { 29692, 27969, 25172, 21407, 16819, 11585, 5906 };
56 static const int TAB35[] = { 26722, 25172, 22654, 19266, 15137, 10426, 5315 };
57
idct_row(short * in,const int * const tab,int rnd)58 static int idct_row(short *in, const int *const tab, int rnd)
59 {
60 const int c1 = tab[0];
61 const int c2 = tab[1];
62 const int c3 = tab[2];
63 const int c4 = tab[3];
64 const int c5 = tab[4];
65 const int c6 = tab[5];
66 const int c7 = tab[6];
67
68 const int right = in[5] | in[6] | in[7];
69 const int left = in[1] | in[2] | in[3];
70 if (!(right | in[4])) {
71 const int k = c4 * in[0] + rnd;
72 if (left) {
73 const int a0 = k + c2 * in[2];
74 const int a1 = k + c6 * in[2];
75 const int a2 = k - c6 * in[2];
76 const int a3 = k - c2 * in[2];
77
78 const int b0 = c1 * in[1] + c3 * in[3];
79 const int b1 = c3 * in[1] - c7 * in[3];
80 const int b2 = c5 * in[1] - c1 * in[3];
81 const int b3 = c7 * in[1] - c5 * in[3];
82
83 in[0] = (a0 + b0) >> ROW_SHIFT;
84 in[1] = (a1 + b1) >> ROW_SHIFT;
85 in[2] = (a2 + b2) >> ROW_SHIFT;
86 in[3] = (a3 + b3) >> ROW_SHIFT;
87 in[4] = (a3 - b3) >> ROW_SHIFT;
88 in[5] = (a2 - b2) >> ROW_SHIFT;
89 in[6] = (a1 - b1) >> ROW_SHIFT;
90 in[7] = (a0 - b0) >> ROW_SHIFT;
91 } else {
92 const int a0 = k >> ROW_SHIFT;
93 if (a0) {
94 in[0] =
95 in[1] =
96 in[2] =
97 in[3] =
98 in[4] =
99 in[5] =
100 in[6] =
101 in[7] = a0;
102 } else
103 return 0;
104 }
105 } else if (!(left | right)) {
106 const int a0 = (rnd + c4 * (in[0] + in[4])) >> ROW_SHIFT;
107 const int a1 = (rnd + c4 * (in[0] - in[4])) >> ROW_SHIFT;
108
109 in[0] = a0;
110 in[3] = a0;
111 in[4] = a0;
112 in[7] = a0;
113 in[1] = a1;
114 in[2] = a1;
115 in[5] = a1;
116 in[6] = a1;
117 } else {
118 const int k = c4 * in[0] + rnd;
119 const int a0 = k + c2 * in[2] + c4 * in[4] + c6 * in[6];
120 const int a1 = k + c6 * in[2] - c4 * in[4] - c2 * in[6];
121 const int a2 = k - c6 * in[2] - c4 * in[4] + c2 * in[6];
122 const int a3 = k - c2 * in[2] + c4 * in[4] - c6 * in[6];
123
124 const int b0 = c1 * in[1] + c3 * in[3] + c5 * in[5] + c7 * in[7];
125 const int b1 = c3 * in[1] - c7 * in[3] - c1 * in[5] - c5 * in[7];
126 const int b2 = c5 * in[1] - c1 * in[3] + c7 * in[5] + c3 * in[7];
127 const int b3 = c7 * in[1] - c5 * in[3] + c3 * in[5] - c1 * in[7];
128
129 in[0] = (a0 + b0) >> ROW_SHIFT;
130 in[1] = (a1 + b1) >> ROW_SHIFT;
131 in[2] = (a2 + b2) >> ROW_SHIFT;
132 in[3] = (a3 + b3) >> ROW_SHIFT;
133 in[4] = (a3 - b3) >> ROW_SHIFT;
134 in[5] = (a2 - b2) >> ROW_SHIFT;
135 in[6] = (a1 - b1) >> ROW_SHIFT;
136 in[7] = (a0 - b0) >> ROW_SHIFT;
137 }
138 return 1;
139 }
140
141 #define TAN1 0x32EC
142 #define TAN2 0x6A0A
143 #define TAN3 0xAB0E
144 #define SQRT2 0x5A82
145
146 #define MULT(c, x, n) (((c) * (x)) >> (n))
147 // 12b version => #define MULT(c,x, n) ((((c) >> 3) * (x)) >> ((n) - 3))
148 // 12b zero-testing version:
149
150 #define BUTTERFLY(a, b, tmp) \
151 (tmp) = (a) + (b); \
152 (b) = (a) - (b); \
153 (a) = (tmp)
154
155 #define LOAD_BUTTERFLY(m1, m2, a, b, tmp, s) \
156 (m1) = (s)[(a)] + (s)[(b)]; \
157 (m2) = (s)[(a)] - (s)[(b)]
158
idct_col_8(short * const in)159 static void idct_col_8(short *const in)
160 {
161 int mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7, spill;
162
163 // odd
164
165 mm4 = (int) in[7 * 8];
166 mm5 = (int) in[5 * 8];
167 mm6 = (int) in[3 * 8];
168 mm7 = (int) in[1 * 8];
169
170 mm0 = MULT(TAN1, mm4, 16) + mm7;
171 mm1 = MULT(TAN1, mm7, 16) - mm4;
172 mm2 = MULT(TAN3, mm5, 16) + mm6;
173 mm3 = MULT(TAN3, mm6, 16) - mm5;
174
175 mm7 = mm0 + mm2;
176 mm4 = mm1 - mm3;
177 mm0 = mm0 - mm2;
178 mm1 = mm1 + mm3;
179 mm6 = mm0 + mm1;
180 mm5 = mm0 - mm1;
181 mm5 = 2 * MULT(SQRT2, mm5, 16); // 2*sqrt2
182 mm6 = 2 * MULT(SQRT2, mm6, 16); // Watch out: precision loss but done to match
183 // the pmulhw used in MMX/MMXEXT/SSE2 versions
184
185 // even
186
187 mm1 = (int) in[2 * 8];
188 mm2 = (int) in[6 * 8];
189 mm3 = MULT(TAN2, mm2, 16) + mm1;
190 mm2 = MULT(TAN2, mm1, 16) - mm2;
191
192 LOAD_BUTTERFLY(mm0, mm1, 0 * 8, 4 * 8, spill, in);
193
194 BUTTERFLY(mm0, mm3, spill);
195 BUTTERFLY(mm0, mm7, spill);
196 in[8 * 0] = (int16_t) (mm0 >> COL_SHIFT);
197 in[8 * 7] = (int16_t) (mm7 >> COL_SHIFT);
198 BUTTERFLY(mm3, mm4, mm0);
199 in[8 * 3] = (int16_t) (mm3 >> COL_SHIFT);
200 in[8 * 4] = (int16_t) (mm4 >> COL_SHIFT);
201
202 BUTTERFLY(mm1, mm2, mm0);
203 BUTTERFLY(mm1, mm6, mm0);
204 in[8 * 1] = (int16_t) (mm1 >> COL_SHIFT);
205 in[8 * 6] = (int16_t) (mm6 >> COL_SHIFT);
206 BUTTERFLY(mm2, mm5, mm0);
207 in[8 * 2] = (int16_t) (mm2 >> COL_SHIFT);
208 in[8 * 5] = (int16_t) (mm5 >> COL_SHIFT);
209 }
210
idct_col_4(short * const in)211 static void idct_col_4(short *const in)
212 {
213 int mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7, spill;
214
215 // odd
216
217 mm0 = (int) in[1 * 8];
218 mm2 = (int) in[3 * 8];
219
220 mm1 = MULT(TAN1, mm0, 16);
221 mm3 = MULT(TAN3, mm2, 16);
222
223 mm7 = mm0 + mm2;
224 mm4 = mm1 - mm3;
225 mm0 = mm0 - mm2;
226 mm1 = mm1 + mm3;
227 mm6 = mm0 + mm1;
228 mm5 = mm0 - mm1;
229 mm6 = 2 * MULT(SQRT2, mm6, 16); // 2*sqrt2
230 mm5 = 2 * MULT(SQRT2, mm5, 16);
231
232 // even
233
234 mm0 = mm1 = (int) in[0 * 8];
235 mm3 = (int) in[2 * 8];
236 mm2 = MULT(TAN2, mm3, 16);
237
238 BUTTERFLY(mm0, mm3, spill);
239 BUTTERFLY(mm0, mm7, spill);
240 in[8 * 0] = (int16_t) (mm0 >> COL_SHIFT);
241 in[8 * 7] = (int16_t) (mm7 >> COL_SHIFT);
242 BUTTERFLY(mm3, mm4, mm0);
243 in[8 * 3] = (int16_t) (mm3 >> COL_SHIFT);
244 in[8 * 4] = (int16_t) (mm4 >> COL_SHIFT);
245
246 BUTTERFLY(mm1, mm2, mm0);
247 BUTTERFLY(mm1, mm6, mm0);
248 in[8 * 1] = (int16_t) (mm1 >> COL_SHIFT);
249 in[8 * 6] = (int16_t) (mm6 >> COL_SHIFT);
250 BUTTERFLY(mm2, mm5, mm0);
251 in[8 * 2] = (int16_t) (mm2 >> COL_SHIFT);
252 in[8 * 5] = (int16_t) (mm5 >> COL_SHIFT);
253 }
254
idct_col_3(short * const in)255 static void idct_col_3(short *const in)
256 {
257 int mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7, spill;
258
259 // odd
260
261 mm7 = (int) in[1 * 8];
262 mm4 = MULT(TAN1, mm7, 16);
263
264 mm6 = mm7 + mm4;
265 mm5 = mm7 - mm4;
266 mm6 = 2 * MULT(SQRT2, mm6, 16); // 2*sqrt2
267 mm5 = 2 * MULT(SQRT2, mm5, 16);
268
269 // even
270
271 mm0 = mm1 = (int) in[0 * 8];
272 mm3 = (int) in[2 * 8];
273 mm2 = MULT(TAN2, mm3, 16);
274
275 BUTTERFLY(mm0, mm3, spill);
276 BUTTERFLY(mm0, mm7, spill);
277 in[8 * 0] = (int16_t) (mm0 >> COL_SHIFT);
278 in[8 * 7] = (int16_t) (mm7 >> COL_SHIFT);
279 BUTTERFLY(mm3, mm4, mm0);
280 in[8 * 3] = (int16_t) (mm3 >> COL_SHIFT);
281 in[8 * 4] = (int16_t) (mm4 >> COL_SHIFT);
282
283 BUTTERFLY(mm1, mm2, mm0);
284 BUTTERFLY(mm1, mm6, mm0);
285 in[8 * 1] = (int16_t) (mm1 >> COL_SHIFT);
286 in[8 * 6] = (int16_t) (mm6 >> COL_SHIFT);
287 BUTTERFLY(mm2, mm5, mm0);
288 in[8 * 2] = (int16_t) (mm2 >> COL_SHIFT);
289 in[8 * 5] = (int16_t) (mm5 >> COL_SHIFT);
290 }
291
ff_xvid_idct(int16_t * const in)292 void ff_xvid_idct(int16_t *const in)
293 {
294 int i, rows = 0x07;
295
296 idct_row(in + 0 * 8, TAB04, RND0);
297 idct_row(in + 1 * 8, TAB17, RND1);
298 idct_row(in + 2 * 8, TAB26, RND2);
299 if (idct_row(in + 3 * 8, TAB35, RND3))
300 rows |= 0x08;
301 if (idct_row(in + 4 * 8, TAB04, RND4))
302 rows |= 0x10;
303 if (idct_row(in + 5 * 8, TAB35, RND5))
304 rows |= 0x20;
305 if (idct_row(in + 6 * 8, TAB26, RND6))
306 rows |= 0x40;
307 if (idct_row(in + 7 * 8, TAB17, RND7))
308 rows |= 0x80;
309
310 if (rows & 0xF0) {
311 for (i = 0; i < 8; i++)
312 idct_col_8(in + i);
313 } else if (rows & 0x08) {
314 for (i = 0; i < 8; i++)
315 idct_col_4(in + i);
316 } else {
317 for (i = 0; i < 8; i++)
318 idct_col_3(in + i);
319 }
320 }
321
xvid_idct_put(uint8_t * dest,int line_size,int16_t * block)322 static void xvid_idct_put(uint8_t *dest, int line_size, int16_t *block)
323 {
324 ff_xvid_idct(block);
325 ff_put_pixels_clamped(block, dest, line_size);
326 }
327
xvid_idct_add(uint8_t * dest,int line_size,int16_t * block)328 static void xvid_idct_add(uint8_t *dest, int line_size, int16_t *block)
329 {
330 ff_xvid_idct(block);
331 ff_add_pixels_clamped(block, dest, line_size);
332 }
333
ff_xvid_idct_init(IDCTDSPContext * c,AVCodecContext * avctx)334 av_cold void ff_xvid_idct_init(IDCTDSPContext *c, AVCodecContext *avctx)
335 {
336 const unsigned high_bit_depth = avctx->bits_per_raw_sample > 8;
337
338 if (high_bit_depth || avctx->lowres ||
339 !(avctx->idct_algo == FF_IDCT_AUTO ||
340 avctx->idct_algo == FF_IDCT_XVID))
341 return;
342
343 if (avctx->idct_algo == FF_IDCT_XVID) {
344 c->idct_put = xvid_idct_put;
345 c->idct_add = xvid_idct_add;
346 c->idct = ff_xvid_idct;
347 c->perm_type = FF_IDCT_PERM_NONE;
348 }
349
350 if (ARCH_X86)
351 ff_xvid_idct_init_x86(c, avctx, high_bit_depth);
352
353 ff_init_scantable_permutation(c->idct_permutation, c->perm_type);
354 }
355