Commit | Line | Data |
---|---|---|
2ba45a60 DM |
1 | /* |
2 | * Copyright (c) 2003-2004 The FFmpeg Project | |
3 | * | |
4 | * This file is part of FFmpeg. | |
5 | * | |
6 | * FFmpeg is free software; you can redistribute it and/or | |
7 | * modify it under the terms of the GNU Lesser General Public | |
8 | * License as published by the Free Software Foundation; either | |
9 | * version 2.1 of the License, or (at your option) any later version. | |
10 | * | |
11 | * FFmpeg is distributed in the hope that it will be useful, | |
12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
14 | * Lesser General Public License for more details. | |
15 | * | |
16 | * You should have received a copy of the GNU Lesser General Public | |
17 | * License along with FFmpeg; if not, write to the Free Software | |
18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
19 | */ | |
20 | ||
21 | /** | |
22 | * @file | |
23 | * On2 VP3 Video Decoder | |
24 | * | |
25 | * VP3 Video Decoder by Mike Melanson (mike at multimedia.cx) | |
26 | * For more information about the VP3 coding process, visit: | |
27 | * http://wiki.multimedia.cx/index.php?title=On2_VP3 | |
28 | * | |
29 | * Theora decoder by Alex Beregszaszi | |
30 | */ | |
31 | ||
32 | #include <stdio.h> | |
33 | #include <stdlib.h> | |
34 | #include <string.h> | |
35 | ||
36 | #include "libavutil/imgutils.h" | |
37 | ||
38 | #include "avcodec.h" | |
39 | #include "get_bits.h" | |
40 | #include "hpeldsp.h" | |
41 | #include "internal.h" | |
42 | #include "mathops.h" | |
43 | #include "thread.h" | |
44 | #include "videodsp.h" | |
45 | #include "vp3data.h" | |
46 | #include "vp3dsp.h" | |
47 | #include "xiph.h" | |
48 | ||
49 | #define FRAGMENT_PIXELS 8 | |
50 | ||
51 | // FIXME split things out into their own arrays | |
52 | typedef struct Vp3Fragment { | |
53 | int16_t dc; | |
54 | uint8_t coding_method; | |
55 | uint8_t qpi; | |
56 | } Vp3Fragment; | |
57 | ||
58 | #define SB_NOT_CODED 0 | |
59 | #define SB_PARTIALLY_CODED 1 | |
60 | #define SB_FULLY_CODED 2 | |
61 | ||
62 | // This is the maximum length of a single long bit run that can be encoded | |
63 | // for superblock coding or block qps. Theora special-cases this to read a | |
64 | // bit instead of flipping the current bit to allow for runs longer than 4129. | |
65 | #define MAXIMUM_LONG_BIT_RUN 4129 | |
66 | ||
67 | #define MODE_INTER_NO_MV 0 | |
68 | #define MODE_INTRA 1 | |
69 | #define MODE_INTER_PLUS_MV 2 | |
70 | #define MODE_INTER_LAST_MV 3 | |
71 | #define MODE_INTER_PRIOR_LAST 4 | |
72 | #define MODE_USING_GOLDEN 5 | |
73 | #define MODE_GOLDEN_MV 6 | |
74 | #define MODE_INTER_FOURMV 7 | |
75 | #define CODING_MODE_COUNT 8 | |
76 | ||
77 | /* special internal mode */ | |
78 | #define MODE_COPY 8 | |
79 | ||
80 | static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb); | |
81 | static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb); | |
82 | ||
83 | ||
84 | /* There are 6 preset schemes, plus a free-form scheme */ | |
85 | static const int ModeAlphabet[6][CODING_MODE_COUNT] = { | |
86 | /* scheme 1: Last motion vector dominates */ | |
87 | { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST, | |
88 | MODE_INTER_PLUS_MV, MODE_INTER_NO_MV, | |
89 | MODE_INTRA, MODE_USING_GOLDEN, | |
90 | MODE_GOLDEN_MV, MODE_INTER_FOURMV }, | |
91 | ||
92 | /* scheme 2 */ | |
93 | { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST, | |
94 | MODE_INTER_NO_MV, MODE_INTER_PLUS_MV, | |
95 | MODE_INTRA, MODE_USING_GOLDEN, | |
96 | MODE_GOLDEN_MV, MODE_INTER_FOURMV }, | |
97 | ||
98 | /* scheme 3 */ | |
99 | { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV, | |
100 | MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV, | |
101 | MODE_INTRA, MODE_USING_GOLDEN, | |
102 | MODE_GOLDEN_MV, MODE_INTER_FOURMV }, | |
103 | ||
104 | /* scheme 4 */ | |
105 | { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV, | |
106 | MODE_INTER_NO_MV, MODE_INTER_PRIOR_LAST, | |
107 | MODE_INTRA, MODE_USING_GOLDEN, | |
108 | MODE_GOLDEN_MV, MODE_INTER_FOURMV }, | |
109 | ||
110 | /* scheme 5: No motion vector dominates */ | |
111 | { MODE_INTER_NO_MV, MODE_INTER_LAST_MV, | |
112 | MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV, | |
113 | MODE_INTRA, MODE_USING_GOLDEN, | |
114 | MODE_GOLDEN_MV, MODE_INTER_FOURMV }, | |
115 | ||
116 | /* scheme 6 */ | |
117 | { MODE_INTER_NO_MV, MODE_USING_GOLDEN, | |
118 | MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST, | |
119 | MODE_INTER_PLUS_MV, MODE_INTRA, | |
120 | MODE_GOLDEN_MV, MODE_INTER_FOURMV }, | |
121 | }; | |
122 | ||
123 | static const uint8_t hilbert_offset[16][2] = { | |
124 | { 0, 0 }, { 1, 0 }, { 1, 1 }, { 0, 1 }, | |
125 | { 0, 2 }, { 0, 3 }, { 1, 3 }, { 1, 2 }, | |
126 | { 2, 2 }, { 2, 3 }, { 3, 3 }, { 3, 2 }, | |
127 | { 3, 1 }, { 2, 1 }, { 2, 0 }, { 3, 0 } | |
128 | }; | |
129 | ||
130 | #define MIN_DEQUANT_VAL 2 | |
131 | ||
132 | typedef struct Vp3DecodeContext { | |
133 | AVCodecContext *avctx; | |
134 | int theora, theora_tables; | |
135 | int version; | |
136 | int width, height; | |
137 | int chroma_x_shift, chroma_y_shift; | |
138 | ThreadFrame golden_frame; | |
139 | ThreadFrame last_frame; | |
140 | ThreadFrame current_frame; | |
141 | int keyframe; | |
142 | uint8_t idct_permutation[64]; | |
143 | uint8_t idct_scantable[64]; | |
144 | HpelDSPContext hdsp; | |
145 | VideoDSPContext vdsp; | |
146 | VP3DSPContext vp3dsp; | |
147 | DECLARE_ALIGNED(16, int16_t, block)[64]; | |
148 | int flipped_image; | |
149 | int last_slice_end; | |
150 | int skip_loop_filter; | |
151 | ||
152 | int qps[3]; | |
153 | int nqps; | |
154 | int last_qps[3]; | |
155 | ||
156 | int superblock_count; | |
157 | int y_superblock_width; | |
158 | int y_superblock_height; | |
159 | int y_superblock_count; | |
160 | int c_superblock_width; | |
161 | int c_superblock_height; | |
162 | int c_superblock_count; | |
163 | int u_superblock_start; | |
164 | int v_superblock_start; | |
165 | unsigned char *superblock_coding; | |
166 | ||
167 | int macroblock_count; | |
168 | int macroblock_width; | |
169 | int macroblock_height; | |
170 | ||
171 | int fragment_count; | |
172 | int fragment_width[2]; | |
173 | int fragment_height[2]; | |
174 | ||
175 | Vp3Fragment *all_fragments; | |
176 | int fragment_start[3]; | |
177 | int data_offset[3]; | |
178 | ||
179 | int8_t (*motion_val[2])[2]; | |
180 | ||
181 | /* tables */ | |
182 | uint16_t coded_dc_scale_factor[64]; | |
183 | uint32_t coded_ac_scale_factor[64]; | |
184 | uint8_t base_matrix[384][64]; | |
185 | uint8_t qr_count[2][3]; | |
186 | uint8_t qr_size[2][3][64]; | |
187 | uint16_t qr_base[2][3][64]; | |
188 | ||
189 | /** | |
190 | * This is a list of all tokens in bitstream order. Reordering takes place | |
191 | * by pulling from each level during IDCT. As a consequence, IDCT must be | |
192 | * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32 | |
193 | * otherwise. The 32 different tokens with up to 12 bits of extradata are | |
194 | * collapsed into 3 types, packed as follows: | |
195 | * (from the low to high bits) | |
196 | * | |
197 | * 2 bits: type (0,1,2) | |
198 | * 0: EOB run, 14 bits for run length (12 needed) | |
199 | * 1: zero run, 7 bits for run length | |
200 | * 7 bits for the next coefficient (3 needed) | |
201 | * 2: coefficient, 14 bits (11 needed) | |
202 | * | |
203 | * Coefficients are signed, so are packed in the highest bits for automatic | |
204 | * sign extension. | |
205 | */ | |
206 | int16_t *dct_tokens[3][64]; | |
207 | int16_t *dct_tokens_base; | |
208 | #define TOKEN_EOB(eob_run) ((eob_run) << 2) | |
209 | #define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) << 9) + ((zero_run) << 2) + 1) | |
210 | #define TOKEN_COEFF(coeff) (((coeff) << 2) + 2) | |
211 | ||
212 | /** | |
213 | * number of blocks that contain DCT coefficients at | |
214 | * the given level or higher | |
215 | */ | |
216 | int num_coded_frags[3][64]; | |
217 | int total_num_coded_frags; | |
218 | ||
219 | /* this is a list of indexes into the all_fragments array indicating | |
220 | * which of the fragments are coded */ | |
221 | int *coded_fragment_list[3]; | |
222 | ||
223 | VLC dc_vlc[16]; | |
224 | VLC ac_vlc_1[16]; | |
225 | VLC ac_vlc_2[16]; | |
226 | VLC ac_vlc_3[16]; | |
227 | VLC ac_vlc_4[16]; | |
228 | ||
229 | VLC superblock_run_length_vlc; | |
230 | VLC fragment_run_length_vlc; | |
231 | VLC mode_code_vlc; | |
232 | VLC motion_vector_vlc; | |
233 | ||
234 | /* these arrays need to be on 16-byte boundaries since SSE2 operations | |
235 | * index into them */ | |
236 | DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64]; ///< qmat[qpi][is_inter][plane] | |
237 | ||
238 | /* This table contains superblock_count * 16 entries. Each set of 16 | |
239 | * numbers corresponds to the fragment indexes 0..15 of the superblock. | |
240 | * An entry will be -1 to indicate that no entry corresponds to that | |
241 | * index. */ | |
242 | int *superblock_fragments; | |
243 | ||
244 | /* This is an array that indicates how a particular macroblock | |
245 | * is coded. */ | |
246 | unsigned char *macroblock_coding; | |
247 | ||
248 | uint8_t *edge_emu_buffer; | |
249 | ||
250 | /* Huffman decode */ | |
251 | int hti; | |
252 | unsigned int hbits; | |
253 | int entries; | |
254 | int huff_code_size; | |
255 | uint32_t huffman_table[80][32][2]; | |
256 | ||
257 | uint8_t filter_limit_values[64]; | |
258 | DECLARE_ALIGNED(8, int, bounding_values_array)[256 + 2]; | |
259 | } Vp3DecodeContext; | |
260 | ||
261 | /************************************************************************ | |
262 | * VP3 specific functions | |
263 | ************************************************************************/ | |
264 | ||
265 | static av_cold void free_tables(AVCodecContext *avctx) | |
266 | { | |
267 | Vp3DecodeContext *s = avctx->priv_data; | |
268 | ||
269 | av_freep(&s->superblock_coding); | |
270 | av_freep(&s->all_fragments); | |
271 | av_freep(&s->coded_fragment_list[0]); | |
272 | av_freep(&s->dct_tokens_base); | |
273 | av_freep(&s->superblock_fragments); | |
274 | av_freep(&s->macroblock_coding); | |
275 | av_freep(&s->motion_val[0]); | |
276 | av_freep(&s->motion_val[1]); | |
277 | } | |
278 | ||
279 | static void vp3_decode_flush(AVCodecContext *avctx) | |
280 | { | |
281 | Vp3DecodeContext *s = avctx->priv_data; | |
282 | ||
283 | if (s->golden_frame.f) | |
284 | ff_thread_release_buffer(avctx, &s->golden_frame); | |
285 | if (s->last_frame.f) | |
286 | ff_thread_release_buffer(avctx, &s->last_frame); | |
287 | if (s->current_frame.f) | |
288 | ff_thread_release_buffer(avctx, &s->current_frame); | |
289 | } | |
290 | ||
291 | static av_cold int vp3_decode_end(AVCodecContext *avctx) | |
292 | { | |
293 | Vp3DecodeContext *s = avctx->priv_data; | |
294 | int i; | |
295 | ||
296 | free_tables(avctx); | |
297 | av_freep(&s->edge_emu_buffer); | |
298 | ||
299 | s->theora_tables = 0; | |
300 | ||
301 | /* release all frames */ | |
302 | vp3_decode_flush(avctx); | |
303 | av_frame_free(&s->current_frame.f); | |
304 | av_frame_free(&s->last_frame.f); | |
305 | av_frame_free(&s->golden_frame.f); | |
306 | ||
307 | if (avctx->internal->is_copy) | |
308 | return 0; | |
309 | ||
310 | for (i = 0; i < 16; i++) { | |
311 | ff_free_vlc(&s->dc_vlc[i]); | |
312 | ff_free_vlc(&s->ac_vlc_1[i]); | |
313 | ff_free_vlc(&s->ac_vlc_2[i]); | |
314 | ff_free_vlc(&s->ac_vlc_3[i]); | |
315 | ff_free_vlc(&s->ac_vlc_4[i]); | |
316 | } | |
317 | ||
318 | ff_free_vlc(&s->superblock_run_length_vlc); | |
319 | ff_free_vlc(&s->fragment_run_length_vlc); | |
320 | ff_free_vlc(&s->mode_code_vlc); | |
321 | ff_free_vlc(&s->motion_vector_vlc); | |
322 | ||
323 | return 0; | |
324 | } | |
325 | ||
326 | /** | |
327 | * This function sets up all of the various blocks mappings: | |
328 | * superblocks <-> fragments, macroblocks <-> fragments, | |
329 | * superblocks <-> macroblocks | |
330 | * | |
331 | * @return 0 is successful; returns 1 if *anything* went wrong. | |
332 | */ | |
333 | static int init_block_mapping(Vp3DecodeContext *s) | |
334 | { | |
335 | int sb_x, sb_y, plane; | |
336 | int x, y, i, j = 0; | |
337 | ||
338 | for (plane = 0; plane < 3; plane++) { | |
339 | int sb_width = plane ? s->c_superblock_width | |
340 | : s->y_superblock_width; | |
341 | int sb_height = plane ? s->c_superblock_height | |
342 | : s->y_superblock_height; | |
343 | int frag_width = s->fragment_width[!!plane]; | |
344 | int frag_height = s->fragment_height[!!plane]; | |
345 | ||
346 | for (sb_y = 0; sb_y < sb_height; sb_y++) | |
347 | for (sb_x = 0; sb_x < sb_width; sb_x++) | |
348 | for (i = 0; i < 16; i++) { | |
349 | x = 4 * sb_x + hilbert_offset[i][0]; | |
350 | y = 4 * sb_y + hilbert_offset[i][1]; | |
351 | ||
352 | if (x < frag_width && y < frag_height) | |
353 | s->superblock_fragments[j++] = s->fragment_start[plane] + | |
354 | y * frag_width + x; | |
355 | else | |
356 | s->superblock_fragments[j++] = -1; | |
357 | } | |
358 | } | |
359 | ||
360 | return 0; /* successful path out */ | |
361 | } | |
362 | ||
363 | /* | |
364 | * This function sets up the dequantization tables used for a particular | |
365 | * frame. | |
366 | */ | |
367 | static void init_dequantizer(Vp3DecodeContext *s, int qpi) | |
368 | { | |
369 | int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]]; | |
370 | int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]]; | |
371 | int i, plane, inter, qri, bmi, bmj, qistart; | |
372 | ||
373 | for (inter = 0; inter < 2; inter++) { | |
374 | for (plane = 0; plane < 3; plane++) { | |
375 | int sum = 0; | |
376 | for (qri = 0; qri < s->qr_count[inter][plane]; qri++) { | |
377 | sum += s->qr_size[inter][plane][qri]; | |
378 | if (s->qps[qpi] <= sum) | |
379 | break; | |
380 | } | |
381 | qistart = sum - s->qr_size[inter][plane][qri]; | |
382 | bmi = s->qr_base[inter][plane][qri]; | |
383 | bmj = s->qr_base[inter][plane][qri + 1]; | |
384 | for (i = 0; i < 64; i++) { | |
385 | int coeff = (2 * (sum - s->qps[qpi]) * s->base_matrix[bmi][i] - | |
386 | 2 * (qistart - s->qps[qpi]) * s->base_matrix[bmj][i] + | |
387 | s->qr_size[inter][plane][qri]) / | |
388 | (2 * s->qr_size[inter][plane][qri]); | |
389 | ||
390 | int qmin = 8 << (inter + !i); | |
391 | int qscale = i ? ac_scale_factor : dc_scale_factor; | |
392 | ||
393 | s->qmat[qpi][inter][plane][s->idct_permutation[i]] = | |
394 | av_clip((qscale * coeff) / 100 * 4, qmin, 4096); | |
395 | } | |
396 | /* all DC coefficients use the same quant so as not to interfere | |
397 | * with DC prediction */ | |
398 | s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0]; | |
399 | } | |
400 | } | |
401 | } | |
402 | ||
403 | /* | |
404 | * This function initializes the loop filter boundary limits if the frame's | |
405 | * quality index is different from the previous frame's. | |
406 | * | |
407 | * The filter_limit_values may not be larger than 127. | |
408 | */ | |
409 | static void init_loop_filter(Vp3DecodeContext *s) | |
410 | { | |
411 | int *bounding_values = s->bounding_values_array + 127; | |
412 | int filter_limit; | |
413 | int x; | |
414 | int value; | |
415 | ||
416 | filter_limit = s->filter_limit_values[s->qps[0]]; | |
417 | av_assert0(filter_limit < 128U); | |
418 | ||
419 | /* set up the bounding values */ | |
420 | memset(s->bounding_values_array, 0, 256 * sizeof(int)); | |
421 | for (x = 0; x < filter_limit; x++) { | |
422 | bounding_values[-x] = -x; | |
423 | bounding_values[x] = x; | |
424 | } | |
425 | for (x = value = filter_limit; x < 128 && value; x++, value--) { | |
426 | bounding_values[ x] = value; | |
427 | bounding_values[-x] = -value; | |
428 | } | |
429 | if (value) | |
430 | bounding_values[128] = value; | |
431 | bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202; | |
432 | } | |
433 | ||
434 | /* | |
435 | * This function unpacks all of the superblock/macroblock/fragment coding | |
436 | * information from the bitstream. | |
437 | */ | |
438 | static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb) | |
439 | { | |
440 | int superblock_starts[3] = { | |
441 | 0, s->u_superblock_start, s->v_superblock_start | |
442 | }; | |
443 | int bit = 0; | |
444 | int current_superblock = 0; | |
445 | int current_run = 0; | |
446 | int num_partial_superblocks = 0; | |
447 | ||
448 | int i, j; | |
449 | int current_fragment; | |
450 | int plane; | |
451 | ||
452 | if (s->keyframe) { | |
453 | memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count); | |
454 | } else { | |
455 | /* unpack the list of partially-coded superblocks */ | |
456 | bit = get_bits1(gb) ^ 1; | |
457 | current_run = 0; | |
458 | ||
459 | while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) { | |
460 | if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN) | |
461 | bit = get_bits1(gb); | |
462 | else | |
463 | bit ^= 1; | |
464 | ||
465 | current_run = get_vlc2(gb, s->superblock_run_length_vlc.table, | |
466 | 6, 2) + 1; | |
467 | if (current_run == 34) | |
468 | current_run += get_bits(gb, 12); | |
469 | ||
470 | if (current_superblock + current_run > s->superblock_count) { | |
471 | av_log(s->avctx, AV_LOG_ERROR, | |
472 | "Invalid partially coded superblock run length\n"); | |
473 | return -1; | |
474 | } | |
475 | ||
476 | memset(s->superblock_coding + current_superblock, bit, current_run); | |
477 | ||
478 | current_superblock += current_run; | |
479 | if (bit) | |
480 | num_partial_superblocks += current_run; | |
481 | } | |
482 | ||
483 | /* unpack the list of fully coded superblocks if any of the blocks were | |
484 | * not marked as partially coded in the previous step */ | |
485 | if (num_partial_superblocks < s->superblock_count) { | |
486 | int superblocks_decoded = 0; | |
487 | ||
488 | current_superblock = 0; | |
489 | bit = get_bits1(gb) ^ 1; | |
490 | current_run = 0; | |
491 | ||
492 | while (superblocks_decoded < s->superblock_count - num_partial_superblocks && | |
493 | get_bits_left(gb) > 0) { | |
494 | if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN) | |
495 | bit = get_bits1(gb); | |
496 | else | |
497 | bit ^= 1; | |
498 | ||
499 | current_run = get_vlc2(gb, s->superblock_run_length_vlc.table, | |
500 | 6, 2) + 1; | |
501 | if (current_run == 34) | |
502 | current_run += get_bits(gb, 12); | |
503 | ||
504 | for (j = 0; j < current_run; current_superblock++) { | |
505 | if (current_superblock >= s->superblock_count) { | |
506 | av_log(s->avctx, AV_LOG_ERROR, | |
507 | "Invalid fully coded superblock run length\n"); | |
508 | return -1; | |
509 | } | |
510 | ||
511 | /* skip any superblocks already marked as partially coded */ | |
512 | if (s->superblock_coding[current_superblock] == SB_NOT_CODED) { | |
513 | s->superblock_coding[current_superblock] = 2 * bit; | |
514 | j++; | |
515 | } | |
516 | } | |
517 | superblocks_decoded += current_run; | |
518 | } | |
519 | } | |
520 | ||
521 | /* if there were partial blocks, initialize bitstream for | |
522 | * unpacking fragment codings */ | |
523 | if (num_partial_superblocks) { | |
524 | current_run = 0; | |
525 | bit = get_bits1(gb); | |
526 | /* toggle the bit because as soon as the first run length is | |
527 | * fetched the bit will be toggled again */ | |
528 | bit ^= 1; | |
529 | } | |
530 | } | |
531 | ||
532 | /* figure out which fragments are coded; iterate through each | |
533 | * superblock (all planes) */ | |
534 | s->total_num_coded_frags = 0; | |
535 | memset(s->macroblock_coding, MODE_COPY, s->macroblock_count); | |
536 | ||
537 | for (plane = 0; plane < 3; plane++) { | |
538 | int sb_start = superblock_starts[plane]; | |
539 | int sb_end = sb_start + (plane ? s->c_superblock_count | |
540 | : s->y_superblock_count); | |
541 | int num_coded_frags = 0; | |
542 | ||
543 | for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) { | |
544 | /* iterate through all 16 fragments in a superblock */ | |
545 | for (j = 0; j < 16; j++) { | |
546 | /* if the fragment is in bounds, check its coding status */ | |
547 | current_fragment = s->superblock_fragments[i * 16 + j]; | |
548 | if (current_fragment != -1) { | |
549 | int coded = s->superblock_coding[i]; | |
550 | ||
551 | if (s->superblock_coding[i] == SB_PARTIALLY_CODED) { | |
552 | /* fragment may or may not be coded; this is the case | |
553 | * that cares about the fragment coding runs */ | |
554 | if (current_run-- == 0) { | |
555 | bit ^= 1; | |
556 | current_run = get_vlc2(gb, s->fragment_run_length_vlc.table, 5, 2); | |
557 | } | |
558 | coded = bit; | |
559 | } | |
560 | ||
561 | if (coded) { | |
562 | /* default mode; actual mode will be decoded in | |
563 | * the next phase */ | |
564 | s->all_fragments[current_fragment].coding_method = | |
565 | MODE_INTER_NO_MV; | |
566 | s->coded_fragment_list[plane][num_coded_frags++] = | |
567 | current_fragment; | |
568 | } else { | |
569 | /* not coded; copy this fragment from the prior frame */ | |
570 | s->all_fragments[current_fragment].coding_method = | |
571 | MODE_COPY; | |
572 | } | |
573 | } | |
574 | } | |
575 | } | |
576 | s->total_num_coded_frags += num_coded_frags; | |
577 | for (i = 0; i < 64; i++) | |
578 | s->num_coded_frags[plane][i] = num_coded_frags; | |
579 | if (plane < 2) | |
580 | s->coded_fragment_list[plane + 1] = s->coded_fragment_list[plane] + | |
581 | num_coded_frags; | |
582 | } | |
583 | return 0; | |
584 | } | |
585 | ||
586 | /* | |
587 | * This function unpacks all the coding mode data for individual macroblocks | |
588 | * from the bitstream. | |
589 | */ | |
590 | static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb) | |
591 | { | |
592 | int i, j, k, sb_x, sb_y; | |
593 | int scheme; | |
594 | int current_macroblock; | |
595 | int current_fragment; | |
596 | int coding_mode; | |
597 | int custom_mode_alphabet[CODING_MODE_COUNT]; | |
598 | const int *alphabet; | |
599 | Vp3Fragment *frag; | |
600 | ||
601 | if (s->keyframe) { | |
602 | for (i = 0; i < s->fragment_count; i++) | |
603 | s->all_fragments[i].coding_method = MODE_INTRA; | |
604 | } else { | |
605 | /* fetch the mode coding scheme for this frame */ | |
606 | scheme = get_bits(gb, 3); | |
607 | ||
608 | /* is it a custom coding scheme? */ | |
609 | if (scheme == 0) { | |
610 | for (i = 0; i < 8; i++) | |
611 | custom_mode_alphabet[i] = MODE_INTER_NO_MV; | |
612 | for (i = 0; i < 8; i++) | |
613 | custom_mode_alphabet[get_bits(gb, 3)] = i; | |
614 | alphabet = custom_mode_alphabet; | |
615 | } else | |
616 | alphabet = ModeAlphabet[scheme - 1]; | |
617 | ||
618 | /* iterate through all of the macroblocks that contain 1 or more | |
619 | * coded fragments */ | |
620 | for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) { | |
621 | for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) { | |
622 | if (get_bits_left(gb) <= 0) | |
623 | return -1; | |
624 | ||
625 | for (j = 0; j < 4; j++) { | |
626 | int mb_x = 2 * sb_x + (j >> 1); | |
627 | int mb_y = 2 * sb_y + (((j >> 1) + j) & 1); | |
628 | current_macroblock = mb_y * s->macroblock_width + mb_x; | |
629 | ||
630 | if (mb_x >= s->macroblock_width || | |
631 | mb_y >= s->macroblock_height) | |
632 | continue; | |
633 | ||
634 | #define BLOCK_X (2 * mb_x + (k & 1)) | |
635 | #define BLOCK_Y (2 * mb_y + (k >> 1)) | |
636 | /* coding modes are only stored if the macroblock has | |
637 | * at least one luma block coded, otherwise it must be | |
638 | * INTER_NO_MV */ | |
639 | for (k = 0; k < 4; k++) { | |
640 | current_fragment = BLOCK_Y * | |
641 | s->fragment_width[0] + BLOCK_X; | |
642 | if (s->all_fragments[current_fragment].coding_method != MODE_COPY) | |
643 | break; | |
644 | } | |
645 | if (k == 4) { | |
646 | s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV; | |
647 | continue; | |
648 | } | |
649 | ||
650 | /* mode 7 means get 3 bits for each coding mode */ | |
651 | if (scheme == 7) | |
652 | coding_mode = get_bits(gb, 3); | |
653 | else | |
654 | coding_mode = alphabet[get_vlc2(gb, s->mode_code_vlc.table, 3, 3)]; | |
655 | ||
656 | s->macroblock_coding[current_macroblock] = coding_mode; | |
657 | for (k = 0; k < 4; k++) { | |
658 | frag = s->all_fragments + BLOCK_Y * s->fragment_width[0] + BLOCK_X; | |
659 | if (frag->coding_method != MODE_COPY) | |
660 | frag->coding_method = coding_mode; | |
661 | } | |
662 | ||
663 | #define SET_CHROMA_MODES \ | |
664 | if (frag[s->fragment_start[1]].coding_method != MODE_COPY) \ | |
665 | frag[s->fragment_start[1]].coding_method = coding_mode; \ | |
666 | if (frag[s->fragment_start[2]].coding_method != MODE_COPY) \ | |
667 | frag[s->fragment_start[2]].coding_method = coding_mode; | |
668 | ||
669 | if (s->chroma_y_shift) { | |
670 | frag = s->all_fragments + mb_y * | |
671 | s->fragment_width[1] + mb_x; | |
672 | SET_CHROMA_MODES | |
673 | } else if (s->chroma_x_shift) { | |
674 | frag = s->all_fragments + | |
675 | 2 * mb_y * s->fragment_width[1] + mb_x; | |
676 | for (k = 0; k < 2; k++) { | |
677 | SET_CHROMA_MODES | |
678 | frag += s->fragment_width[1]; | |
679 | } | |
680 | } else { | |
681 | for (k = 0; k < 4; k++) { | |
682 | frag = s->all_fragments + | |
683 | BLOCK_Y * s->fragment_width[1] + BLOCK_X; | |
684 | SET_CHROMA_MODES | |
685 | } | |
686 | } | |
687 | } | |
688 | } | |
689 | } | |
690 | } | |
691 | ||
692 | return 0; | |
693 | } | |
694 | ||
695 | /* | |
696 | * This function unpacks all the motion vectors for the individual | |
697 | * macroblocks from the bitstream. | |
698 | */ | |
699 | static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb) | |
700 | { | |
701 | int j, k, sb_x, sb_y; | |
702 | int coding_mode; | |
703 | int motion_x[4]; | |
704 | int motion_y[4]; | |
705 | int last_motion_x = 0; | |
706 | int last_motion_y = 0; | |
707 | int prior_last_motion_x = 0; | |
708 | int prior_last_motion_y = 0; | |
709 | int current_macroblock; | |
710 | int current_fragment; | |
711 | int frag; | |
712 | ||
713 | if (s->keyframe) | |
714 | return 0; | |
715 | ||
716 | /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */ | |
717 | coding_mode = get_bits1(gb); | |
718 | ||
719 | /* iterate through all of the macroblocks that contain 1 or more | |
720 | * coded fragments */ | |
721 | for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) { | |
722 | for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) { | |
723 | if (get_bits_left(gb) <= 0) | |
724 | return -1; | |
725 | ||
726 | for (j = 0; j < 4; j++) { | |
727 | int mb_x = 2 * sb_x + (j >> 1); | |
728 | int mb_y = 2 * sb_y + (((j >> 1) + j) & 1); | |
729 | current_macroblock = mb_y * s->macroblock_width + mb_x; | |
730 | ||
731 | if (mb_x >= s->macroblock_width || | |
732 | mb_y >= s->macroblock_height || | |
733 | s->macroblock_coding[current_macroblock] == MODE_COPY) | |
734 | continue; | |
735 | ||
736 | switch (s->macroblock_coding[current_macroblock]) { | |
737 | case MODE_INTER_PLUS_MV: | |
738 | case MODE_GOLDEN_MV: | |
739 | /* all 6 fragments use the same motion vector */ | |
740 | if (coding_mode == 0) { | |
741 | motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)]; | |
742 | motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)]; | |
743 | } else { | |
744 | motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)]; | |
745 | motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)]; | |
746 | } | |
747 | ||
748 | /* vector maintenance, only on MODE_INTER_PLUS_MV */ | |
749 | if (s->macroblock_coding[current_macroblock] == MODE_INTER_PLUS_MV) { | |
750 | prior_last_motion_x = last_motion_x; | |
751 | prior_last_motion_y = last_motion_y; | |
752 | last_motion_x = motion_x[0]; | |
753 | last_motion_y = motion_y[0]; | |
754 | } | |
755 | break; | |
756 | ||
757 | case MODE_INTER_FOURMV: | |
758 | /* vector maintenance */ | |
759 | prior_last_motion_x = last_motion_x; | |
760 | prior_last_motion_y = last_motion_y; | |
761 | ||
762 | /* fetch 4 vectors from the bitstream, one for each | |
763 | * Y fragment, then average for the C fragment vectors */ | |
764 | for (k = 0; k < 4; k++) { | |
765 | current_fragment = BLOCK_Y * s->fragment_width[0] + BLOCK_X; | |
766 | if (s->all_fragments[current_fragment].coding_method != MODE_COPY) { | |
767 | if (coding_mode == 0) { | |
768 | motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)]; | |
769 | motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)]; | |
770 | } else { | |
771 | motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)]; | |
772 | motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)]; | |
773 | } | |
774 | last_motion_x = motion_x[k]; | |
775 | last_motion_y = motion_y[k]; | |
776 | } else { | |
777 | motion_x[k] = 0; | |
778 | motion_y[k] = 0; | |
779 | } | |
780 | } | |
781 | break; | |
782 | ||
783 | case MODE_INTER_LAST_MV: | |
784 | /* all 6 fragments use the last motion vector */ | |
785 | motion_x[0] = last_motion_x; | |
786 | motion_y[0] = last_motion_y; | |
787 | ||
788 | /* no vector maintenance (last vector remains the | |
789 | * last vector) */ | |
790 | break; | |
791 | ||
792 | case MODE_INTER_PRIOR_LAST: | |
793 | /* all 6 fragments use the motion vector prior to the | |
794 | * last motion vector */ | |
795 | motion_x[0] = prior_last_motion_x; | |
796 | motion_y[0] = prior_last_motion_y; | |
797 | ||
798 | /* vector maintenance */ | |
799 | prior_last_motion_x = last_motion_x; | |
800 | prior_last_motion_y = last_motion_y; | |
801 | last_motion_x = motion_x[0]; | |
802 | last_motion_y = motion_y[0]; | |
803 | break; | |
804 | ||
805 | default: | |
806 | /* covers intra, inter without MV, golden without MV */ | |
807 | motion_x[0] = 0; | |
808 | motion_y[0] = 0; | |
809 | ||
810 | /* no vector maintenance */ | |
811 | break; | |
812 | } | |
813 | ||
814 | /* assign the motion vectors to the correct fragments */ | |
815 | for (k = 0; k < 4; k++) { | |
816 | current_fragment = | |
817 | BLOCK_Y * s->fragment_width[0] + BLOCK_X; | |
818 | if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) { | |
819 | s->motion_val[0][current_fragment][0] = motion_x[k]; | |
820 | s->motion_val[0][current_fragment][1] = motion_y[k]; | |
821 | } else { | |
822 | s->motion_val[0][current_fragment][0] = motion_x[0]; | |
823 | s->motion_val[0][current_fragment][1] = motion_y[0]; | |
824 | } | |
825 | } | |
826 | ||
827 | if (s->chroma_y_shift) { | |
828 | if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) { | |
829 | motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] + | |
830 | motion_x[2] + motion_x[3], 2); | |
831 | motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] + | |
832 | motion_y[2] + motion_y[3], 2); | |
833 | } | |
834 | motion_x[0] = (motion_x[0] >> 1) | (motion_x[0] & 1); | |
835 | motion_y[0] = (motion_y[0] >> 1) | (motion_y[0] & 1); | |
836 | frag = mb_y * s->fragment_width[1] + mb_x; | |
837 | s->motion_val[1][frag][0] = motion_x[0]; | |
838 | s->motion_val[1][frag][1] = motion_y[0]; | |
839 | } else if (s->chroma_x_shift) { | |
840 | if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) { | |
841 | motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1); | |
842 | motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1); | |
843 | motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1); | |
844 | motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1); | |
845 | } else { | |
846 | motion_x[1] = motion_x[0]; | |
847 | motion_y[1] = motion_y[0]; | |
848 | } | |
849 | motion_x[0] = (motion_x[0] >> 1) | (motion_x[0] & 1); | |
850 | motion_x[1] = (motion_x[1] >> 1) | (motion_x[1] & 1); | |
851 | ||
852 | frag = 2 * mb_y * s->fragment_width[1] + mb_x; | |
853 | for (k = 0; k < 2; k++) { | |
854 | s->motion_val[1][frag][0] = motion_x[k]; | |
855 | s->motion_val[1][frag][1] = motion_y[k]; | |
856 | frag += s->fragment_width[1]; | |
857 | } | |
858 | } else { | |
859 | for (k = 0; k < 4; k++) { | |
860 | frag = BLOCK_Y * s->fragment_width[1] + BLOCK_X; | |
861 | if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) { | |
862 | s->motion_val[1][frag][0] = motion_x[k]; | |
863 | s->motion_val[1][frag][1] = motion_y[k]; | |
864 | } else { | |
865 | s->motion_val[1][frag][0] = motion_x[0]; | |
866 | s->motion_val[1][frag][1] = motion_y[0]; | |
867 | } | |
868 | } | |
869 | } | |
870 | } | |
871 | } | |
872 | } | |
873 | ||
874 | return 0; | |
875 | } | |
876 | ||
877 | static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb) | |
878 | { | |
879 | int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi; | |
880 | int num_blocks = s->total_num_coded_frags; | |
881 | ||
882 | for (qpi = 0; qpi < s->nqps - 1 && num_blocks > 0; qpi++) { | |
883 | i = blocks_decoded = num_blocks_at_qpi = 0; | |
884 | ||
885 | bit = get_bits1(gb) ^ 1; | |
886 | run_length = 0; | |
887 | ||
888 | do { | |
889 | if (run_length == MAXIMUM_LONG_BIT_RUN) | |
890 | bit = get_bits1(gb); | |
891 | else | |
892 | bit ^= 1; | |
893 | ||
894 | run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1; | |
895 | if (run_length == 34) | |
896 | run_length += get_bits(gb, 12); | |
897 | blocks_decoded += run_length; | |
898 | ||
899 | if (!bit) | |
900 | num_blocks_at_qpi += run_length; | |
901 | ||
902 | for (j = 0; j < run_length; i++) { | |
903 | if (i >= s->total_num_coded_frags) | |
904 | return -1; | |
905 | ||
906 | if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) { | |
907 | s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit; | |
908 | j++; | |
909 | } | |
910 | } | |
911 | } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0); | |
912 | ||
913 | num_blocks -= num_blocks_at_qpi; | |
914 | } | |
915 | ||
916 | return 0; | |
917 | } | |
918 | ||
919 | /* | |
920 | * This function is called by unpack_dct_coeffs() to extract the VLCs from | |
921 | * the bitstream. The VLCs encode tokens which are used to unpack DCT | |
922 | * data. This function unpacks all the VLCs for either the Y plane or both | |
923 | * C planes, and is called for DC coefficients or different AC coefficient | |
924 | * levels (since different coefficient types require different VLC tables. | |
925 | * | |
926 | * This function returns a residual eob run. E.g, if a particular token gave | |
927 | * instructions to EOB the next 5 fragments and there were only 2 fragments | |
928 | * left in the current fragment range, 3 would be returned so that it could | |
929 | * be passed into the next call to this same function. | |
930 | */ | |
931 | static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb, | |
932 | VLC *table, int coeff_index, | |
933 | int plane, | |
934 | int eob_run) | |
935 | { | |
936 | int i, j = 0; | |
937 | int token; | |
938 | int zero_run = 0; | |
939 | int16_t coeff = 0; | |
940 | int bits_to_get; | |
941 | int blocks_ended; | |
942 | int coeff_i = 0; | |
943 | int num_coeffs = s->num_coded_frags[plane][coeff_index]; | |
944 | int16_t *dct_tokens = s->dct_tokens[plane][coeff_index]; | |
945 | ||
946 | /* local references to structure members to avoid repeated deferences */ | |
947 | int *coded_fragment_list = s->coded_fragment_list[plane]; | |
948 | Vp3Fragment *all_fragments = s->all_fragments; | |
949 | VLC_TYPE(*vlc_table)[2] = table->table; | |
950 | ||
951 | if (num_coeffs < 0) | |
952 | av_log(s->avctx, AV_LOG_ERROR, | |
953 | "Invalid number of coefficents at level %d\n", coeff_index); | |
954 | ||
955 | if (eob_run > num_coeffs) { | |
956 | coeff_i = | |
957 | blocks_ended = num_coeffs; | |
958 | eob_run -= num_coeffs; | |
959 | } else { | |
960 | coeff_i = | |
961 | blocks_ended = eob_run; | |
962 | eob_run = 0; | |
963 | } | |
964 | ||
965 | // insert fake EOB token to cover the split between planes or zzi | |
966 | if (blocks_ended) | |
967 | dct_tokens[j++] = blocks_ended << 2; | |
968 | ||
969 | while (coeff_i < num_coeffs && get_bits_left(gb) > 0) { | |
970 | /* decode a VLC into a token */ | |
971 | token = get_vlc2(gb, vlc_table, 11, 3); | |
972 | /* use the token to get a zero run, a coefficient, and an eob run */ | |
973 | if ((unsigned) token <= 6U) { | |
974 | eob_run = eob_run_base[token]; | |
975 | if (eob_run_get_bits[token]) | |
976 | eob_run += get_bits(gb, eob_run_get_bits[token]); | |
977 | ||
978 | // record only the number of blocks ended in this plane, | |
979 | // any spill will be recorded in the next plane. | |
980 | if (eob_run > num_coeffs - coeff_i) { | |
981 | dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i); | |
982 | blocks_ended += num_coeffs - coeff_i; | |
983 | eob_run -= num_coeffs - coeff_i; | |
984 | coeff_i = num_coeffs; | |
985 | } else { | |
986 | dct_tokens[j++] = TOKEN_EOB(eob_run); | |
987 | blocks_ended += eob_run; | |
988 | coeff_i += eob_run; | |
989 | eob_run = 0; | |
990 | } | |
991 | } else if (token >= 0) { | |
992 | bits_to_get = coeff_get_bits[token]; | |
993 | if (bits_to_get) | |
994 | bits_to_get = get_bits(gb, bits_to_get); | |
995 | coeff = coeff_tables[token][bits_to_get]; | |
996 | ||
997 | zero_run = zero_run_base[token]; | |
998 | if (zero_run_get_bits[token]) | |
999 | zero_run += get_bits(gb, zero_run_get_bits[token]); | |
1000 | ||
1001 | if (zero_run) { | |
1002 | dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run); | |
1003 | } else { | |
1004 | // Save DC into the fragment structure. DC prediction is | |
1005 | // done in raster order, so the actual DC can't be in with | |
1006 | // other tokens. We still need the token in dct_tokens[] | |
1007 | // however, or else the structure collapses on itself. | |
1008 | if (!coeff_index) | |
1009 | all_fragments[coded_fragment_list[coeff_i]].dc = coeff; | |
1010 | ||
1011 | dct_tokens[j++] = TOKEN_COEFF(coeff); | |
1012 | } | |
1013 | ||
1014 | if (coeff_index + zero_run > 64) { | |
1015 | av_log(s->avctx, AV_LOG_DEBUG, | |
1016 | "Invalid zero run of %d with %d coeffs left\n", | |
1017 | zero_run, 64 - coeff_index); | |
1018 | zero_run = 64 - coeff_index; | |
1019 | } | |
1020 | ||
1021 | // zero runs code multiple coefficients, | |
1022 | // so don't try to decode coeffs for those higher levels | |
1023 | for (i = coeff_index + 1; i <= coeff_index + zero_run; i++) | |
1024 | s->num_coded_frags[plane][i]--; | |
1025 | coeff_i++; | |
1026 | } else { | |
1027 | av_log(s->avctx, AV_LOG_ERROR, "Invalid token %d\n", token); | |
1028 | return -1; | |
1029 | } | |
1030 | } | |
1031 | ||
1032 | if (blocks_ended > s->num_coded_frags[plane][coeff_index]) | |
1033 | av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n"); | |
1034 | ||
1035 | // decrement the number of blocks that have higher coefficients for each | |
1036 | // EOB run at this level | |
1037 | if (blocks_ended) | |
1038 | for (i = coeff_index + 1; i < 64; i++) | |
1039 | s->num_coded_frags[plane][i] -= blocks_ended; | |
1040 | ||
1041 | // setup the next buffer | |
1042 | if (plane < 2) | |
1043 | s->dct_tokens[plane + 1][coeff_index] = dct_tokens + j; | |
1044 | else if (coeff_index < 63) | |
1045 | s->dct_tokens[0][coeff_index + 1] = dct_tokens + j; | |
1046 | ||
1047 | return eob_run; | |
1048 | } | |
1049 | ||
1050 | static void reverse_dc_prediction(Vp3DecodeContext *s, | |
1051 | int first_fragment, | |
1052 | int fragment_width, | |
1053 | int fragment_height); | |
1054 | /* | |
1055 | * This function unpacks all of the DCT coefficient data from the | |
1056 | * bitstream. | |
1057 | */ | |
1058 | static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb) | |
1059 | { | |
1060 | int i; | |
1061 | int dc_y_table; | |
1062 | int dc_c_table; | |
1063 | int ac_y_table; | |
1064 | int ac_c_table; | |
1065 | int residual_eob_run = 0; | |
1066 | VLC *y_tables[64]; | |
1067 | VLC *c_tables[64]; | |
1068 | ||
1069 | s->dct_tokens[0][0] = s->dct_tokens_base; | |
1070 | ||
1071 | /* fetch the DC table indexes */ | |
1072 | dc_y_table = get_bits(gb, 4); | |
1073 | dc_c_table = get_bits(gb, 4); | |
1074 | ||
1075 | /* unpack the Y plane DC coefficients */ | |
1076 | residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0, | |
1077 | 0, residual_eob_run); | |
1078 | if (residual_eob_run < 0) | |
1079 | return residual_eob_run; | |
1080 | ||
1081 | /* reverse prediction of the Y-plane DC coefficients */ | |
1082 | reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]); | |
1083 | ||
1084 | /* unpack the C plane DC coefficients */ | |
1085 | residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0, | |
1086 | 1, residual_eob_run); | |
1087 | if (residual_eob_run < 0) | |
1088 | return residual_eob_run; | |
1089 | residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0, | |
1090 | 2, residual_eob_run); | |
1091 | if (residual_eob_run < 0) | |
1092 | return residual_eob_run; | |
1093 | ||
1094 | /* reverse prediction of the C-plane DC coefficients */ | |
1095 | if (!(s->avctx->flags & CODEC_FLAG_GRAY)) { | |
1096 | reverse_dc_prediction(s, s->fragment_start[1], | |
1097 | s->fragment_width[1], s->fragment_height[1]); | |
1098 | reverse_dc_prediction(s, s->fragment_start[2], | |
1099 | s->fragment_width[1], s->fragment_height[1]); | |
1100 | } | |
1101 | ||
1102 | /* fetch the AC table indexes */ | |
1103 | ac_y_table = get_bits(gb, 4); | |
1104 | ac_c_table = get_bits(gb, 4); | |
1105 | ||
1106 | /* build tables of AC VLC tables */ | |
1107 | for (i = 1; i <= 5; i++) { | |
1108 | y_tables[i] = &s->ac_vlc_1[ac_y_table]; | |
1109 | c_tables[i] = &s->ac_vlc_1[ac_c_table]; | |
1110 | } | |
1111 | for (i = 6; i <= 14; i++) { | |
1112 | y_tables[i] = &s->ac_vlc_2[ac_y_table]; | |
1113 | c_tables[i] = &s->ac_vlc_2[ac_c_table]; | |
1114 | } | |
1115 | for (i = 15; i <= 27; i++) { | |
1116 | y_tables[i] = &s->ac_vlc_3[ac_y_table]; | |
1117 | c_tables[i] = &s->ac_vlc_3[ac_c_table]; | |
1118 | } | |
1119 | for (i = 28; i <= 63; i++) { | |
1120 | y_tables[i] = &s->ac_vlc_4[ac_y_table]; | |
1121 | c_tables[i] = &s->ac_vlc_4[ac_c_table]; | |
1122 | } | |
1123 | ||
1124 | /* decode all AC coefficents */ | |
1125 | for (i = 1; i <= 63; i++) { | |
1126 | residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i, | |
1127 | 0, residual_eob_run); | |
1128 | if (residual_eob_run < 0) | |
1129 | return residual_eob_run; | |
1130 | ||
1131 | residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i, | |
1132 | 1, residual_eob_run); | |
1133 | if (residual_eob_run < 0) | |
1134 | return residual_eob_run; | |
1135 | residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i, | |
1136 | 2, residual_eob_run); | |
1137 | if (residual_eob_run < 0) | |
1138 | return residual_eob_run; | |
1139 | } | |
1140 | ||
1141 | return 0; | |
1142 | } | |
1143 | ||
1144 | /* | |
1145 | * This function reverses the DC prediction for each coded fragment in | |
1146 | * the frame. Much of this function is adapted directly from the original | |
1147 | * VP3 source code. | |
1148 | */ | |
1149 | #define COMPATIBLE_FRAME(x) \ | |
1150 | (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type) | |
1151 | #define DC_COEFF(u) s->all_fragments[u].dc | |
1152 | ||
1153 | static void reverse_dc_prediction(Vp3DecodeContext *s, | |
1154 | int first_fragment, | |
1155 | int fragment_width, | |
1156 | int fragment_height) | |
1157 | { | |
1158 | #define PUL 8 | |
1159 | #define PU 4 | |
1160 | #define PUR 2 | |
1161 | #define PL 1 | |
1162 | ||
1163 | int x, y; | |
1164 | int i = first_fragment; | |
1165 | ||
1166 | int predicted_dc; | |
1167 | ||
1168 | /* DC values for the left, up-left, up, and up-right fragments */ | |
1169 | int vl, vul, vu, vur; | |
1170 | ||
1171 | /* indexes for the left, up-left, up, and up-right fragments */ | |
1172 | int l, ul, u, ur; | |
1173 | ||
1174 | /* | |
1175 | * The 6 fields mean: | |
1176 | * 0: up-left multiplier | |
1177 | * 1: up multiplier | |
1178 | * 2: up-right multiplier | |
1179 | * 3: left multiplier | |
1180 | */ | |
1181 | static const int predictor_transform[16][4] = { | |
1182 | { 0, 0, 0, 0 }, | |
1183 | { 0, 0, 0, 128 }, // PL | |
1184 | { 0, 0, 128, 0 }, // PUR | |
1185 | { 0, 0, 53, 75 }, // PUR|PL | |
1186 | { 0, 128, 0, 0 }, // PU | |
1187 | { 0, 64, 0, 64 }, // PU |PL | |
1188 | { 0, 128, 0, 0 }, // PU |PUR | |
1189 | { 0, 0, 53, 75 }, // PU |PUR|PL | |
1190 | { 128, 0, 0, 0 }, // PUL | |
1191 | { 0, 0, 0, 128 }, // PUL|PL | |
1192 | { 64, 0, 64, 0 }, // PUL|PUR | |
1193 | { 0, 0, 53, 75 }, // PUL|PUR|PL | |
1194 | { 0, 128, 0, 0 }, // PUL|PU | |
1195 | { -104, 116, 0, 116 }, // PUL|PU |PL | |
1196 | { 24, 80, 24, 0 }, // PUL|PU |PUR | |
1197 | { -104, 116, 0, 116 } // PUL|PU |PUR|PL | |
1198 | }; | |
1199 | ||
1200 | /* This table shows which types of blocks can use other blocks for | |
1201 | * prediction. For example, INTRA is the only mode in this table to | |
1202 | * have a frame number of 0. That means INTRA blocks can only predict | |
1203 | * from other INTRA blocks. There are 2 golden frame coding types; | |
1204 | * blocks encoding in these modes can only predict from other blocks | |
1205 | * that were encoded with these 1 of these 2 modes. */ | |
1206 | static const unsigned char compatible_frame[9] = { | |
1207 | 1, /* MODE_INTER_NO_MV */ | |
1208 | 0, /* MODE_INTRA */ | |
1209 | 1, /* MODE_INTER_PLUS_MV */ | |
1210 | 1, /* MODE_INTER_LAST_MV */ | |
1211 | 1, /* MODE_INTER_PRIOR_MV */ | |
1212 | 2, /* MODE_USING_GOLDEN */ | |
1213 | 2, /* MODE_GOLDEN_MV */ | |
1214 | 1, /* MODE_INTER_FOUR_MV */ | |
1215 | 3 /* MODE_COPY */ | |
1216 | }; | |
1217 | int current_frame_type; | |
1218 | ||
1219 | /* there is a last DC predictor for each of the 3 frame types */ | |
1220 | short last_dc[3]; | |
1221 | ||
1222 | int transform = 0; | |
1223 | ||
1224 | vul = | |
1225 | vu = | |
1226 | vur = | |
1227 | vl = 0; | |
1228 | last_dc[0] = | |
1229 | last_dc[1] = | |
1230 | last_dc[2] = 0; | |
1231 | ||
1232 | /* for each fragment row... */ | |
1233 | for (y = 0; y < fragment_height; y++) { | |
1234 | /* for each fragment in a row... */ | |
1235 | for (x = 0; x < fragment_width; x++, i++) { | |
1236 | ||
1237 | /* reverse prediction if this block was coded */ | |
1238 | if (s->all_fragments[i].coding_method != MODE_COPY) { | |
1239 | current_frame_type = | |
1240 | compatible_frame[s->all_fragments[i].coding_method]; | |
1241 | ||
1242 | transform = 0; | |
1243 | if (x) { | |
1244 | l = i - 1; | |
1245 | vl = DC_COEFF(l); | |
1246 | if (COMPATIBLE_FRAME(l)) | |
1247 | transform |= PL; | |
1248 | } | |
1249 | if (y) { | |
1250 | u = i - fragment_width; | |
1251 | vu = DC_COEFF(u); | |
1252 | if (COMPATIBLE_FRAME(u)) | |
1253 | transform |= PU; | |
1254 | if (x) { | |
1255 | ul = i - fragment_width - 1; | |
1256 | vul = DC_COEFF(ul); | |
1257 | if (COMPATIBLE_FRAME(ul)) | |
1258 | transform |= PUL; | |
1259 | } | |
1260 | if (x + 1 < fragment_width) { | |
1261 | ur = i - fragment_width + 1; | |
1262 | vur = DC_COEFF(ur); | |
1263 | if (COMPATIBLE_FRAME(ur)) | |
1264 | transform |= PUR; | |
1265 | } | |
1266 | } | |
1267 | ||
1268 | if (transform == 0) { | |
1269 | /* if there were no fragments to predict from, use last | |
1270 | * DC saved */ | |
1271 | predicted_dc = last_dc[current_frame_type]; | |
1272 | } else { | |
1273 | /* apply the appropriate predictor transform */ | |
1274 | predicted_dc = | |
1275 | (predictor_transform[transform][0] * vul) + | |
1276 | (predictor_transform[transform][1] * vu) + | |
1277 | (predictor_transform[transform][2] * vur) + | |
1278 | (predictor_transform[transform][3] * vl); | |
1279 | ||
1280 | predicted_dc /= 128; | |
1281 | ||
1282 | /* check for outranging on the [ul u l] and | |
1283 | * [ul u ur l] predictors */ | |
1284 | if ((transform == 15) || (transform == 13)) { | |
1285 | if (FFABS(predicted_dc - vu) > 128) | |
1286 | predicted_dc = vu; | |
1287 | else if (FFABS(predicted_dc - vl) > 128) | |
1288 | predicted_dc = vl; | |
1289 | else if (FFABS(predicted_dc - vul) > 128) | |
1290 | predicted_dc = vul; | |
1291 | } | |
1292 | } | |
1293 | ||
1294 | /* at long last, apply the predictor */ | |
1295 | DC_COEFF(i) += predicted_dc; | |
1296 | /* save the DC */ | |
1297 | last_dc[current_frame_type] = DC_COEFF(i); | |
1298 | } | |
1299 | } | |
1300 | } | |
1301 | } | |
1302 | ||
1303 | static void apply_loop_filter(Vp3DecodeContext *s, int plane, | |
1304 | int ystart, int yend) | |
1305 | { | |
1306 | int x, y; | |
1307 | int *bounding_values = s->bounding_values_array + 127; | |
1308 | ||
1309 | int width = s->fragment_width[!!plane]; | |
1310 | int height = s->fragment_height[!!plane]; | |
1311 | int fragment = s->fragment_start[plane] + ystart * width; | |
1312 | ptrdiff_t stride = s->current_frame.f->linesize[plane]; | |
1313 | uint8_t *plane_data = s->current_frame.f->data[plane]; | |
1314 | if (!s->flipped_image) | |
1315 | stride = -stride; | |
1316 | plane_data += s->data_offset[plane] + 8 * ystart * stride; | |
1317 | ||
1318 | for (y = ystart; y < yend; y++) { | |
1319 | for (x = 0; x < width; x++) { | |
1320 | /* This code basically just deblocks on the edges of coded blocks. | |
1321 | * However, it has to be much more complicated because of the | |
1322 | * braindamaged deblock ordering used in VP3/Theora. Order matters | |
1323 | * because some pixels get filtered twice. */ | |
1324 | if (s->all_fragments[fragment].coding_method != MODE_COPY) { | |
1325 | /* do not perform left edge filter for left columns frags */ | |
1326 | if (x > 0) { | |
1327 | s->vp3dsp.h_loop_filter( | |
1328 | plane_data + 8 * x, | |
1329 | stride, bounding_values); | |
1330 | } | |
1331 | ||
1332 | /* do not perform top edge filter for top row fragments */ | |
1333 | if (y > 0) { | |
1334 | s->vp3dsp.v_loop_filter( | |
1335 | plane_data + 8 * x, | |
1336 | stride, bounding_values); | |
1337 | } | |
1338 | ||
1339 | /* do not perform right edge filter for right column | |
1340 | * fragments or if right fragment neighbor is also coded | |
1341 | * in this frame (it will be filtered in next iteration) */ | |
1342 | if ((x < width - 1) && | |
1343 | (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) { | |
1344 | s->vp3dsp.h_loop_filter( | |
1345 | plane_data + 8 * x + 8, | |
1346 | stride, bounding_values); | |
1347 | } | |
1348 | ||
1349 | /* do not perform bottom edge filter for bottom row | |
1350 | * fragments or if bottom fragment neighbor is also coded | |
1351 | * in this frame (it will be filtered in the next row) */ | |
1352 | if ((y < height - 1) && | |
1353 | (s->all_fragments[fragment + width].coding_method == MODE_COPY)) { | |
1354 | s->vp3dsp.v_loop_filter( | |
1355 | plane_data + 8 * x + 8 * stride, | |
1356 | stride, bounding_values); | |
1357 | } | |
1358 | } | |
1359 | ||
1360 | fragment++; | |
1361 | } | |
1362 | plane_data += 8 * stride; | |
1363 | } | |
1364 | } | |
1365 | ||
1366 | /** | |
1367 | * Pull DCT tokens from the 64 levels to decode and dequant the coefficients | |
1368 | * for the next block in coding order | |
1369 | */ | |
1370 | static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag, | |
1371 | int plane, int inter, int16_t block[64]) | |
1372 | { | |
1373 | int16_t *dequantizer = s->qmat[frag->qpi][inter][plane]; | |
1374 | uint8_t *perm = s->idct_scantable; | |
1375 | int i = 0; | |
1376 | ||
1377 | do { | |
1378 | int token = *s->dct_tokens[plane][i]; | |
1379 | switch (token & 3) { | |
1380 | case 0: // EOB | |
1381 | if (--token < 4) // 0-3 are token types so the EOB run must now be 0 | |
1382 | s->dct_tokens[plane][i]++; | |
1383 | else | |
1384 | *s->dct_tokens[plane][i] = token & ~3; | |
1385 | goto end; | |
1386 | case 1: // zero run | |
1387 | s->dct_tokens[plane][i]++; | |
1388 | i += (token >> 2) & 0x7f; | |
1389 | if (i > 63) { | |
1390 | av_log(s->avctx, AV_LOG_ERROR, "Coefficient index overflow\n"); | |
1391 | return i; | |
1392 | } | |
1393 | block[perm[i]] = (token >> 9) * dequantizer[perm[i]]; | |
1394 | i++; | |
1395 | break; | |
1396 | case 2: // coeff | |
1397 | block[perm[i]] = (token >> 2) * dequantizer[perm[i]]; | |
1398 | s->dct_tokens[plane][i++]++; | |
1399 | break; | |
1400 | default: // shouldn't happen | |
1401 | return i; | |
1402 | } | |
1403 | } while (i < 64); | |
1404 | // return value is expected to be a valid level | |
1405 | i--; | |
1406 | end: | |
1407 | // the actual DC+prediction is in the fragment structure | |
1408 | block[0] = frag->dc * s->qmat[0][inter][plane][0]; | |
1409 | return i; | |
1410 | } | |
1411 | ||
1412 | /** | |
1413 | * called when all pixels up to row y are complete | |
1414 | */ | |
1415 | static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y) | |
1416 | { | |
1417 | int h, cy, i; | |
1418 | int offset[AV_NUM_DATA_POINTERS]; | |
1419 | ||
1420 | if (HAVE_THREADS && s->avctx->active_thread_type & FF_THREAD_FRAME) { | |
1421 | int y_flipped = s->flipped_image ? s->avctx->height - y : y; | |
1422 | ||
1423 | /* At the end of the frame, report INT_MAX instead of the height of | |
1424 | * the frame. This makes the other threads' ff_thread_await_progress() | |
1425 | * calls cheaper, because they don't have to clip their values. */ | |
1426 | ff_thread_report_progress(&s->current_frame, | |
1427 | y_flipped == s->avctx->height ? INT_MAX | |
1428 | : y_flipped - 1, | |
1429 | 0); | |
1430 | } | |
1431 | ||
1432 | if (!s->avctx->draw_horiz_band) | |
1433 | return; | |
1434 | ||
1435 | h = y - s->last_slice_end; | |
1436 | s->last_slice_end = y; | |
1437 | y -= h; | |
1438 | ||
1439 | if (!s->flipped_image) | |
1440 | y = s->avctx->height - y - h; | |
1441 | ||
1442 | cy = y >> s->chroma_y_shift; | |
1443 | offset[0] = s->current_frame.f->linesize[0] * y; | |
1444 | offset[1] = s->current_frame.f->linesize[1] * cy; | |
1445 | offset[2] = s->current_frame.f->linesize[2] * cy; | |
1446 | for (i = 3; i < AV_NUM_DATA_POINTERS; i++) | |
1447 | offset[i] = 0; | |
1448 | ||
1449 | emms_c(); | |
1450 | s->avctx->draw_horiz_band(s->avctx, s->current_frame.f, offset, y, 3, h); | |
1451 | } | |
1452 | ||
1453 | /** | |
1454 | * Wait for the reference frame of the current fragment. | |
1455 | * The progress value is in luma pixel rows. | |
1456 | */ | |
1457 | static void await_reference_row(Vp3DecodeContext *s, Vp3Fragment *fragment, | |
1458 | int motion_y, int y) | |
1459 | { | |
1460 | ThreadFrame *ref_frame; | |
1461 | int ref_row; | |
1462 | int border = motion_y & 1; | |
1463 | ||
1464 | if (fragment->coding_method == MODE_USING_GOLDEN || | |
1465 | fragment->coding_method == MODE_GOLDEN_MV) | |
1466 | ref_frame = &s->golden_frame; | |
1467 | else | |
1468 | ref_frame = &s->last_frame; | |
1469 | ||
1470 | ref_row = y + (motion_y >> 1); | |
1471 | ref_row = FFMAX(FFABS(ref_row), ref_row + 8 + border); | |
1472 | ||
1473 | ff_thread_await_progress(ref_frame, ref_row, 0); | |
1474 | } | |
1475 | ||
1476 | /* | |
1477 | * Perform the final rendering for a particular slice of data. | |
1478 | * The slice number ranges from 0..(c_superblock_height - 1). | |
1479 | */ | |
1480 | static void render_slice(Vp3DecodeContext *s, int slice) | |
1481 | { | |
1482 | int x, y, i, j, fragment; | |
1483 | int16_t *block = s->block; | |
1484 | int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef; | |
1485 | int motion_halfpel_index; | |
1486 | uint8_t *motion_source; | |
1487 | int plane, first_pixel; | |
1488 | ||
1489 | if (slice >= s->c_superblock_height) | |
1490 | return; | |
1491 | ||
1492 | for (plane = 0; plane < 3; plane++) { | |
1493 | uint8_t *output_plane = s->current_frame.f->data[plane] + | |
1494 | s->data_offset[plane]; | |
1495 | uint8_t *last_plane = s->last_frame.f->data[plane] + | |
1496 | s->data_offset[plane]; | |
1497 | uint8_t *golden_plane = s->golden_frame.f->data[plane] + | |
1498 | s->data_offset[plane]; | |
1499 | ptrdiff_t stride = s->current_frame.f->linesize[plane]; | |
1500 | int plane_width = s->width >> (plane && s->chroma_x_shift); | |
1501 | int plane_height = s->height >> (plane && s->chroma_y_shift); | |
1502 | int8_t(*motion_val)[2] = s->motion_val[!!plane]; | |
1503 | ||
1504 | int sb_x, sb_y = slice << (!plane && s->chroma_y_shift); | |
1505 | int slice_height = sb_y + 1 + (!plane && s->chroma_y_shift); | |
1506 | int slice_width = plane ? s->c_superblock_width | |
1507 | : s->y_superblock_width; | |
1508 | ||
1509 | int fragment_width = s->fragment_width[!!plane]; | |
1510 | int fragment_height = s->fragment_height[!!plane]; | |
1511 | int fragment_start = s->fragment_start[plane]; | |
1512 | ||
1513 | int do_await = !plane && HAVE_THREADS && | |
1514 | (s->avctx->active_thread_type & FF_THREAD_FRAME); | |
1515 | ||
1516 | if (!s->flipped_image) | |
1517 | stride = -stride; | |
1518 | if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY)) | |
1519 | continue; | |
1520 | ||
1521 | /* for each superblock row in the slice (both of them)... */ | |
1522 | for (; sb_y < slice_height; sb_y++) { | |
1523 | /* for each superblock in a row... */ | |
1524 | for (sb_x = 0; sb_x < slice_width; sb_x++) { | |
1525 | /* for each block in a superblock... */ | |
1526 | for (j = 0; j < 16; j++) { | |
1527 | x = 4 * sb_x + hilbert_offset[j][0]; | |
1528 | y = 4 * sb_y + hilbert_offset[j][1]; | |
1529 | fragment = y * fragment_width + x; | |
1530 | ||
1531 | i = fragment_start + fragment; | |
1532 | ||
1533 | // bounds check | |
1534 | if (x >= fragment_width || y >= fragment_height) | |
1535 | continue; | |
1536 | ||
1537 | first_pixel = 8 * y * stride + 8 * x; | |
1538 | ||
1539 | if (do_await && | |
1540 | s->all_fragments[i].coding_method != MODE_INTRA) | |
1541 | await_reference_row(s, &s->all_fragments[i], | |
1542 | motion_val[fragment][1], | |
1543 | (16 * y) >> s->chroma_y_shift); | |
1544 | ||
1545 | /* transform if this block was coded */ | |
1546 | if (s->all_fragments[i].coding_method != MODE_COPY) { | |
1547 | if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) || | |
1548 | (s->all_fragments[i].coding_method == MODE_GOLDEN_MV)) | |
1549 | motion_source = golden_plane; | |
1550 | else | |
1551 | motion_source = last_plane; | |
1552 | ||
1553 | motion_source += first_pixel; | |
1554 | motion_halfpel_index = 0; | |
1555 | ||
1556 | /* sort out the motion vector if this fragment is coded | |
1557 | * using a motion vector method */ | |
1558 | if ((s->all_fragments[i].coding_method > MODE_INTRA) && | |
1559 | (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) { | |
1560 | int src_x, src_y; | |
1561 | motion_x = motion_val[fragment][0]; | |
1562 | motion_y = motion_val[fragment][1]; | |
1563 | ||
1564 | src_x = (motion_x >> 1) + 8 * x; | |
1565 | src_y = (motion_y >> 1) + 8 * y; | |
1566 | ||
1567 | motion_halfpel_index = motion_x & 0x01; | |
1568 | motion_source += (motion_x >> 1); | |
1569 | ||
1570 | motion_halfpel_index |= (motion_y & 0x01) << 1; | |
1571 | motion_source += ((motion_y >> 1) * stride); | |
1572 | ||
1573 | if (src_x < 0 || src_y < 0 || | |
1574 | src_x + 9 >= plane_width || | |
1575 | src_y + 9 >= plane_height) { | |
1576 | uint8_t *temp = s->edge_emu_buffer; | |
1577 | if (stride < 0) | |
1578 | temp -= 8 * stride; | |
1579 | ||
1580 | s->vdsp.emulated_edge_mc(temp, motion_source, | |
1581 | stride, stride, | |
1582 | 9, 9, src_x, src_y, | |
1583 | plane_width, | |
1584 | plane_height); | |
1585 | motion_source = temp; | |
1586 | } | |
1587 | } | |
1588 | ||
1589 | /* first, take care of copying a block from either the | |
1590 | * previous or the golden frame */ | |
1591 | if (s->all_fragments[i].coding_method != MODE_INTRA) { | |
1592 | /* Note, it is possible to implement all MC cases | |
1593 | * with put_no_rnd_pixels_l2 which would look more | |
1594 | * like the VP3 source but this would be slower as | |
1595 | * put_no_rnd_pixels_tab is better optimzed */ | |
1596 | if (motion_halfpel_index != 3) { | |
1597 | s->hdsp.put_no_rnd_pixels_tab[1][motion_halfpel_index]( | |
1598 | output_plane + first_pixel, | |
1599 | motion_source, stride, 8); | |
1600 | } else { | |
1601 | /* d is 0 if motion_x and _y have the same sign, | |
1602 | * else -1 */ | |
1603 | int d = (motion_x ^ motion_y) >> 31; | |
1604 | s->vp3dsp.put_no_rnd_pixels_l2(output_plane + first_pixel, | |
1605 | motion_source - d, | |
1606 | motion_source + stride + 1 + d, | |
1607 | stride, 8); | |
1608 | } | |
1609 | } | |
1610 | ||
1611 | /* invert DCT and place (or add) in final output */ | |
1612 | ||
1613 | if (s->all_fragments[i].coding_method == MODE_INTRA) { | |
1614 | vp3_dequant(s, s->all_fragments + i, | |
1615 | plane, 0, block); | |
1616 | s->vp3dsp.idct_put(output_plane + first_pixel, | |
1617 | stride, | |
1618 | block); | |
1619 | } else { | |
1620 | if (vp3_dequant(s, s->all_fragments + i, | |
1621 | plane, 1, block)) { | |
1622 | s->vp3dsp.idct_add(output_plane + first_pixel, | |
1623 | stride, | |
1624 | block); | |
1625 | } else { | |
1626 | s->vp3dsp.idct_dc_add(output_plane + first_pixel, | |
1627 | stride, block); | |
1628 | } | |
1629 | } | |
1630 | } else { | |
1631 | /* copy directly from the previous frame */ | |
1632 | s->hdsp.put_pixels_tab[1][0]( | |
1633 | output_plane + first_pixel, | |
1634 | last_plane + first_pixel, | |
1635 | stride, 8); | |
1636 | } | |
1637 | } | |
1638 | } | |
1639 | ||
1640 | // Filter up to the last row in the superblock row | |
1641 | if (!s->skip_loop_filter) | |
1642 | apply_loop_filter(s, plane, 4 * sb_y - !!sb_y, | |
1643 | FFMIN(4 * sb_y + 3, fragment_height - 1)); | |
1644 | } | |
1645 | } | |
1646 | ||
1647 | /* this looks like a good place for slice dispatch... */ | |
1648 | /* algorithm: | |
1649 | * if (slice == s->macroblock_height - 1) | |
1650 | * dispatch (both last slice & 2nd-to-last slice); | |
1651 | * else if (slice > 0) | |
1652 | * dispatch (slice - 1); | |
1653 | */ | |
1654 | ||
1655 | vp3_draw_horiz_band(s, FFMIN((32 << s->chroma_y_shift) * (slice + 1) - 16, | |
1656 | s->height - 16)); | |
1657 | } | |
1658 | ||
1659 | /// Allocate tables for per-frame data in Vp3DecodeContext | |
1660 | static av_cold int allocate_tables(AVCodecContext *avctx) | |
1661 | { | |
1662 | Vp3DecodeContext *s = avctx->priv_data; | |
1663 | int y_fragment_count, c_fragment_count; | |
1664 | ||
1665 | free_tables(avctx); | |
1666 | ||
1667 | y_fragment_count = s->fragment_width[0] * s->fragment_height[0]; | |
1668 | c_fragment_count = s->fragment_width[1] * s->fragment_height[1]; | |
1669 | ||
1670 | s->superblock_coding = av_mallocz(s->superblock_count); | |
1671 | s->all_fragments = av_mallocz_array(s->fragment_count, sizeof(Vp3Fragment)); | |
1672 | ||
1673 | s->coded_fragment_list[0] = av_mallocz_array(s->fragment_count, sizeof(int)); | |
1674 | ||
1675 | s->dct_tokens_base = av_mallocz_array(s->fragment_count, | |
1676 | 64 * sizeof(*s->dct_tokens_base)); | |
1677 | s->motion_val[0] = av_mallocz_array(y_fragment_count, sizeof(*s->motion_val[0])); | |
1678 | s->motion_val[1] = av_mallocz_array(c_fragment_count, sizeof(*s->motion_val[1])); | |
1679 | ||
1680 | /* work out the block mapping tables */ | |
1681 | s->superblock_fragments = av_mallocz_array(s->superblock_count, 16 * sizeof(int)); | |
1682 | s->macroblock_coding = av_mallocz(s->macroblock_count + 1); | |
1683 | ||
1684 | if (!s->superblock_coding || !s->all_fragments || | |
1685 | !s->dct_tokens_base || !s->coded_fragment_list[0] || | |
1686 | !s->superblock_fragments || !s->macroblock_coding || | |
1687 | !s->motion_val[0] || !s->motion_val[1]) { | |
1688 | vp3_decode_end(avctx); | |
1689 | return -1; | |
1690 | } | |
1691 | ||
1692 | init_block_mapping(s); | |
1693 | ||
1694 | return 0; | |
1695 | } | |
1696 | ||
1697 | static av_cold int init_frames(Vp3DecodeContext *s) | |
1698 | { | |
1699 | s->current_frame.f = av_frame_alloc(); | |
1700 | s->last_frame.f = av_frame_alloc(); | |
1701 | s->golden_frame.f = av_frame_alloc(); | |
1702 | ||
1703 | if (!s->current_frame.f || !s->last_frame.f || !s->golden_frame.f) { | |
1704 | av_frame_free(&s->current_frame.f); | |
1705 | av_frame_free(&s->last_frame.f); | |
1706 | av_frame_free(&s->golden_frame.f); | |
1707 | return AVERROR(ENOMEM); | |
1708 | } | |
1709 | ||
1710 | return 0; | |
1711 | } | |
1712 | ||
1713 | static av_cold int vp3_decode_init(AVCodecContext *avctx) | |
1714 | { | |
1715 | Vp3DecodeContext *s = avctx->priv_data; | |
1716 | int i, inter, plane, ret; | |
1717 | int c_width; | |
1718 | int c_height; | |
1719 | int y_fragment_count, c_fragment_count; | |
1720 | ||
1721 | ret = init_frames(s); | |
1722 | if (ret < 0) | |
1723 | return ret; | |
1724 | ||
1725 | avctx->internal->allocate_progress = 1; | |
1726 | ||
1727 | if (avctx->codec_tag == MKTAG('V', 'P', '3', '0')) | |
1728 | s->version = 0; | |
1729 | else | |
1730 | s->version = 1; | |
1731 | ||
1732 | s->avctx = avctx; | |
1733 | s->width = FFALIGN(avctx->width, 16); | |
1734 | s->height = FFALIGN(avctx->height, 16); | |
1735 | if (avctx->codec_id != AV_CODEC_ID_THEORA) | |
1736 | avctx->pix_fmt = AV_PIX_FMT_YUV420P; | |
1737 | avctx->chroma_sample_location = AVCHROMA_LOC_CENTER; | |
1738 | ff_hpeldsp_init(&s->hdsp, avctx->flags | CODEC_FLAG_BITEXACT); | |
1739 | ff_videodsp_init(&s->vdsp, 8); | |
1740 | ff_vp3dsp_init(&s->vp3dsp, avctx->flags); | |
1741 | ||
1742 | for (i = 0; i < 64; i++) { | |
1743 | #define TRANSPOSE(x) (((x) >> 3) | (((x) & 7) << 3)) | |
1744 | s->idct_permutation[i] = TRANSPOSE(i); | |
1745 | s->idct_scantable[i] = TRANSPOSE(ff_zigzag_direct[i]); | |
1746 | #undef TRANSPOSE | |
1747 | } | |
1748 | ||
1749 | /* initialize to an impossible value which will force a recalculation | |
1750 | * in the first frame decode */ | |
1751 | for (i = 0; i < 3; i++) | |
1752 | s->qps[i] = -1; | |
1753 | ||
1754 | avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift); | |
1755 | ||
1756 | s->y_superblock_width = (s->width + 31) / 32; | |
1757 | s->y_superblock_height = (s->height + 31) / 32; | |
1758 | s->y_superblock_count = s->y_superblock_width * s->y_superblock_height; | |
1759 | ||
1760 | /* work out the dimensions for the C planes */ | |
1761 | c_width = s->width >> s->chroma_x_shift; | |
1762 | c_height = s->height >> s->chroma_y_shift; | |
1763 | s->c_superblock_width = (c_width + 31) / 32; | |
1764 | s->c_superblock_height = (c_height + 31) / 32; | |
1765 | s->c_superblock_count = s->c_superblock_width * s->c_superblock_height; | |
1766 | ||
1767 | s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2); | |
1768 | s->u_superblock_start = s->y_superblock_count; | |
1769 | s->v_superblock_start = s->u_superblock_start + s->c_superblock_count; | |
1770 | ||
1771 | s->macroblock_width = (s->width + 15) / 16; | |
1772 | s->macroblock_height = (s->height + 15) / 16; | |
1773 | s->macroblock_count = s->macroblock_width * s->macroblock_height; | |
1774 | ||
1775 | s->fragment_width[0] = s->width / FRAGMENT_PIXELS; | |
1776 | s->fragment_height[0] = s->height / FRAGMENT_PIXELS; | |
1777 | s->fragment_width[1] = s->fragment_width[0] >> s->chroma_x_shift; | |
1778 | s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift; | |
1779 | ||
1780 | /* fragment count covers all 8x8 blocks for all 3 planes */ | |
1781 | y_fragment_count = s->fragment_width[0] * s->fragment_height[0]; | |
1782 | c_fragment_count = s->fragment_width[1] * s->fragment_height[1]; | |
1783 | s->fragment_count = y_fragment_count + 2 * c_fragment_count; | |
1784 | s->fragment_start[1] = y_fragment_count; | |
1785 | s->fragment_start[2] = y_fragment_count + c_fragment_count; | |
1786 | ||
1787 | if (!s->theora_tables) { | |
1788 | for (i = 0; i < 64; i++) { | |
1789 | s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i]; | |
1790 | s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i]; | |
1791 | s->base_matrix[0][i] = vp31_intra_y_dequant[i]; | |
1792 | s->base_matrix[1][i] = vp31_intra_c_dequant[i]; | |
1793 | s->base_matrix[2][i] = vp31_inter_dequant[i]; | |
1794 | s->filter_limit_values[i] = vp31_filter_limit_values[i]; | |
1795 | } | |
1796 | ||
1797 | for (inter = 0; inter < 2; inter++) { | |
1798 | for (plane = 0; plane < 3; plane++) { | |
1799 | s->qr_count[inter][plane] = 1; | |
1800 | s->qr_size[inter][plane][0] = 63; | |
1801 | s->qr_base[inter][plane][0] = | |
1802 | s->qr_base[inter][plane][1] = 2 * inter + (!!plane) * !inter; | |
1803 | } | |
1804 | } | |
1805 | ||
1806 | /* init VLC tables */ | |
1807 | for (i = 0; i < 16; i++) { | |
1808 | /* DC histograms */ | |
1809 | init_vlc(&s->dc_vlc[i], 11, 32, | |
1810 | &dc_bias[i][0][1], 4, 2, | |
1811 | &dc_bias[i][0][0], 4, 2, 0); | |
1812 | ||
1813 | /* group 1 AC histograms */ | |
1814 | init_vlc(&s->ac_vlc_1[i], 11, 32, | |
1815 | &ac_bias_0[i][0][1], 4, 2, | |
1816 | &ac_bias_0[i][0][0], 4, 2, 0); | |
1817 | ||
1818 | /* group 2 AC histograms */ | |
1819 | init_vlc(&s->ac_vlc_2[i], 11, 32, | |
1820 | &ac_bias_1[i][0][1], 4, 2, | |
1821 | &ac_bias_1[i][0][0], 4, 2, 0); | |
1822 | ||
1823 | /* group 3 AC histograms */ | |
1824 | init_vlc(&s->ac_vlc_3[i], 11, 32, | |
1825 | &ac_bias_2[i][0][1], 4, 2, | |
1826 | &ac_bias_2[i][0][0], 4, 2, 0); | |
1827 | ||
1828 | /* group 4 AC histograms */ | |
1829 | init_vlc(&s->ac_vlc_4[i], 11, 32, | |
1830 | &ac_bias_3[i][0][1], 4, 2, | |
1831 | &ac_bias_3[i][0][0], 4, 2, 0); | |
1832 | } | |
1833 | } else { | |
1834 | for (i = 0; i < 16; i++) { | |
1835 | /* DC histograms */ | |
1836 | if (init_vlc(&s->dc_vlc[i], 11, 32, | |
1837 | &s->huffman_table[i][0][1], 8, 4, | |
1838 | &s->huffman_table[i][0][0], 8, 4, 0) < 0) | |
1839 | goto vlc_fail; | |
1840 | ||
1841 | /* group 1 AC histograms */ | |
1842 | if (init_vlc(&s->ac_vlc_1[i], 11, 32, | |
1843 | &s->huffman_table[i + 16][0][1], 8, 4, | |
1844 | &s->huffman_table[i + 16][0][0], 8, 4, 0) < 0) | |
1845 | goto vlc_fail; | |
1846 | ||
1847 | /* group 2 AC histograms */ | |
1848 | if (init_vlc(&s->ac_vlc_2[i], 11, 32, | |
1849 | &s->huffman_table[i + 16 * 2][0][1], 8, 4, | |
1850 | &s->huffman_table[i + 16 * 2][0][0], 8, 4, 0) < 0) | |
1851 | goto vlc_fail; | |
1852 | ||
1853 | /* group 3 AC histograms */ | |
1854 | if (init_vlc(&s->ac_vlc_3[i], 11, 32, | |
1855 | &s->huffman_table[i + 16 * 3][0][1], 8, 4, | |
1856 | &s->huffman_table[i + 16 * 3][0][0], 8, 4, 0) < 0) | |
1857 | goto vlc_fail; | |
1858 | ||
1859 | /* group 4 AC histograms */ | |
1860 | if (init_vlc(&s->ac_vlc_4[i], 11, 32, | |
1861 | &s->huffman_table[i + 16 * 4][0][1], 8, 4, | |
1862 | &s->huffman_table[i + 16 * 4][0][0], 8, 4, 0) < 0) | |
1863 | goto vlc_fail; | |
1864 | } | |
1865 | } | |
1866 | ||
1867 | init_vlc(&s->superblock_run_length_vlc, 6, 34, | |
1868 | &superblock_run_length_vlc_table[0][1], 4, 2, | |
1869 | &superblock_run_length_vlc_table[0][0], 4, 2, 0); | |
1870 | ||
1871 | init_vlc(&s->fragment_run_length_vlc, 5, 30, | |
1872 | &fragment_run_length_vlc_table[0][1], 4, 2, | |
1873 | &fragment_run_length_vlc_table[0][0], 4, 2, 0); | |
1874 | ||
1875 | init_vlc(&s->mode_code_vlc, 3, 8, | |
1876 | &mode_code_vlc_table[0][1], 2, 1, | |
1877 | &mode_code_vlc_table[0][0], 2, 1, 0); | |
1878 | ||
1879 | init_vlc(&s->motion_vector_vlc, 6, 63, | |
1880 | &motion_vector_vlc_table[0][1], 2, 1, | |
1881 | &motion_vector_vlc_table[0][0], 2, 1, 0); | |
1882 | ||
1883 | return allocate_tables(avctx); | |
1884 | ||
1885 | vlc_fail: | |
1886 | av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n"); | |
1887 | return -1; | |
1888 | } | |
1889 | ||
1890 | /// Release and shuffle frames after decode finishes | |
1891 | static int update_frames(AVCodecContext *avctx) | |
1892 | { | |
1893 | Vp3DecodeContext *s = avctx->priv_data; | |
1894 | int ret = 0; | |
1895 | ||
1896 | /* shuffle frames (last = current) */ | |
1897 | ff_thread_release_buffer(avctx, &s->last_frame); | |
1898 | ret = ff_thread_ref_frame(&s->last_frame, &s->current_frame); | |
1899 | if (ret < 0) | |
1900 | goto fail; | |
1901 | ||
1902 | if (s->keyframe) { | |
1903 | ff_thread_release_buffer(avctx, &s->golden_frame); | |
1904 | ret = ff_thread_ref_frame(&s->golden_frame, &s->current_frame); | |
1905 | } | |
1906 | ||
1907 | fail: | |
1908 | ff_thread_release_buffer(avctx, &s->current_frame); | |
1909 | return ret; | |
1910 | } | |
1911 | ||
1912 | static int ref_frame(Vp3DecodeContext *s, ThreadFrame *dst, ThreadFrame *src) | |
1913 | { | |
1914 | ff_thread_release_buffer(s->avctx, dst); | |
1915 | if (src->f->data[0]) | |
1916 | return ff_thread_ref_frame(dst, src); | |
1917 | return 0; | |
1918 | } | |
1919 | ||
1920 | static int ref_frames(Vp3DecodeContext *dst, Vp3DecodeContext *src) | |
1921 | { | |
1922 | int ret; | |
1923 | if ((ret = ref_frame(dst, &dst->current_frame, &src->current_frame)) < 0 || | |
1924 | (ret = ref_frame(dst, &dst->golden_frame, &src->golden_frame)) < 0 || | |
1925 | (ret = ref_frame(dst, &dst->last_frame, &src->last_frame)) < 0) | |
1926 | return ret; | |
1927 | return 0; | |
1928 | } | |
1929 | ||
1930 | static int vp3_update_thread_context(AVCodecContext *dst, const AVCodecContext *src) | |
1931 | { | |
1932 | Vp3DecodeContext *s = dst->priv_data, *s1 = src->priv_data; | |
1933 | int qps_changed = 0, i, err; | |
1934 | ||
1935 | #define copy_fields(to, from, start_field, end_field) \ | |
1936 | memcpy(&to->start_field, &from->start_field, \ | |
1937 | (char *) &to->end_field - (char *) &to->start_field) | |
1938 | ||
1939 | if (!s1->current_frame.f->data[0] || | |
1940 | s->width != s1->width || s->height != s1->height) { | |
1941 | if (s != s1) | |
1942 | ref_frames(s, s1); | |
1943 | return -1; | |
1944 | } | |
1945 | ||
1946 | if (s != s1) { | |
1947 | // init tables if the first frame hasn't been decoded | |
1948 | if (!s->current_frame.f->data[0]) { | |
1949 | int y_fragment_count, c_fragment_count; | |
1950 | s->avctx = dst; | |
1951 | err = allocate_tables(dst); | |
1952 | if (err) | |
1953 | return err; | |
1954 | y_fragment_count = s->fragment_width[0] * s->fragment_height[0]; | |
1955 | c_fragment_count = s->fragment_width[1] * s->fragment_height[1]; | |
1956 | memcpy(s->motion_val[0], s1->motion_val[0], | |
1957 | y_fragment_count * sizeof(*s->motion_val[0])); | |
1958 | memcpy(s->motion_val[1], s1->motion_val[1], | |
1959 | c_fragment_count * sizeof(*s->motion_val[1])); | |
1960 | } | |
1961 | ||
1962 | // copy previous frame data | |
1963 | if ((err = ref_frames(s, s1)) < 0) | |
1964 | return err; | |
1965 | ||
1966 | s->keyframe = s1->keyframe; | |
1967 | ||
1968 | // copy qscale data if necessary | |
1969 | for (i = 0; i < 3; i++) { | |
1970 | if (s->qps[i] != s1->qps[1]) { | |
1971 | qps_changed = 1; | |
1972 | memcpy(&s->qmat[i], &s1->qmat[i], sizeof(s->qmat[i])); | |
1973 | } | |
1974 | } | |
1975 | ||
1976 | if (s->qps[0] != s1->qps[0]) | |
1977 | memcpy(&s->bounding_values_array, &s1->bounding_values_array, | |
1978 | sizeof(s->bounding_values_array)); | |
1979 | ||
1980 | if (qps_changed) | |
1981 | copy_fields(s, s1, qps, superblock_count); | |
1982 | #undef copy_fields | |
1983 | } | |
1984 | ||
1985 | return update_frames(dst); | |
1986 | } | |
1987 | ||
1988 | static int vp3_decode_frame(AVCodecContext *avctx, | |
1989 | void *data, int *got_frame, | |
1990 | AVPacket *avpkt) | |
1991 | { | |
1992 | const uint8_t *buf = avpkt->data; | |
1993 | int buf_size = avpkt->size; | |
1994 | Vp3DecodeContext *s = avctx->priv_data; | |
1995 | GetBitContext gb; | |
1996 | int i, ret; | |
1997 | ||
1998 | init_get_bits(&gb, buf, buf_size * 8); | |
1999 | ||
2000 | #if CONFIG_THEORA_DECODER | |
2001 | if (s->theora && get_bits1(&gb)) { | |
2002 | int type = get_bits(&gb, 7); | |
2003 | skip_bits_long(&gb, 6*8); /* "theora" */ | |
2004 | ||
2005 | if (s->avctx->active_thread_type&FF_THREAD_FRAME) { | |
2006 | av_log(avctx, AV_LOG_ERROR, "midstream reconfiguration with multithreading is unsupported, try -threads 1\n"); | |
2007 | return AVERROR_PATCHWELCOME; | |
2008 | } | |
2009 | if (type == 0) { | |
2010 | vp3_decode_end(avctx); | |
2011 | ret = theora_decode_header(avctx, &gb); | |
2012 | ||
2013 | if (ret < 0) { | |
2014 | vp3_decode_end(avctx); | |
2015 | } else | |
2016 | ret = vp3_decode_init(avctx); | |
2017 | return ret; | |
2018 | } else if (type == 2) { | |
2019 | ret = theora_decode_tables(avctx, &gb); | |
2020 | if (ret < 0) { | |
2021 | vp3_decode_end(avctx); | |
2022 | } else | |
2023 | ret = vp3_decode_init(avctx); | |
2024 | return ret; | |
2025 | } | |
2026 | ||
2027 | av_log(avctx, AV_LOG_ERROR, | |
2028 | "Header packet passed to frame decoder, skipping\n"); | |
2029 | return -1; | |
2030 | } | |
2031 | #endif | |
2032 | ||
2033 | s->keyframe = !get_bits1(&gb); | |
2034 | if (!s->all_fragments) { | |
2035 | av_log(avctx, AV_LOG_ERROR, "Data packet without prior valid headers\n"); | |
2036 | return -1; | |
2037 | } | |
2038 | if (!s->theora) | |
2039 | skip_bits(&gb, 1); | |
2040 | for (i = 0; i < 3; i++) | |
2041 | s->last_qps[i] = s->qps[i]; | |
2042 | ||
2043 | s->nqps = 0; | |
2044 | do { | |
2045 | s->qps[s->nqps++] = get_bits(&gb, 6); | |
2046 | } while (s->theora >= 0x030200 && s->nqps < 3 && get_bits1(&gb)); | |
2047 | for (i = s->nqps; i < 3; i++) | |
2048 | s->qps[i] = -1; | |
2049 | ||
2050 | if (s->avctx->debug & FF_DEBUG_PICT_INFO) | |
2051 | av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n", | |
2052 | s->keyframe ? "key" : "", avctx->frame_number + 1, s->qps[0]); | |
2053 | ||
2054 | s->skip_loop_filter = !s->filter_limit_values[s->qps[0]] || | |
2055 | avctx->skip_loop_filter >= (s->keyframe ? AVDISCARD_ALL | |
2056 | : AVDISCARD_NONKEY); | |
2057 | ||
2058 | if (s->qps[0] != s->last_qps[0]) | |
2059 | init_loop_filter(s); | |
2060 | ||
2061 | for (i = 0; i < s->nqps; i++) | |
2062 | // reinit all dequantizers if the first one changed, because | |
2063 | // the DC of the first quantizer must be used for all matrices | |
2064 | if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0]) | |
2065 | init_dequantizer(s, i); | |
2066 | ||
2067 | if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe) | |
2068 | return buf_size; | |
2069 | ||
2070 | s->current_frame.f->pict_type = s->keyframe ? AV_PICTURE_TYPE_I | |
2071 | : AV_PICTURE_TYPE_P; | |
2072 | s->current_frame.f->key_frame = s->keyframe; | |
2073 | if (ff_thread_get_buffer(avctx, &s->current_frame, AV_GET_BUFFER_FLAG_REF) < 0) | |
2074 | goto error; | |
2075 | ||
2076 | if (!s->edge_emu_buffer) | |
2077 | s->edge_emu_buffer = av_malloc(9 * FFABS(s->current_frame.f->linesize[0])); | |
2078 | ||
2079 | if (s->keyframe) { | |
2080 | if (!s->theora) { | |
2081 | skip_bits(&gb, 4); /* width code */ | |
2082 | skip_bits(&gb, 4); /* height code */ | |
2083 | if (s->version) { | |
2084 | s->version = get_bits(&gb, 5); | |
2085 | if (avctx->frame_number == 0) | |
2086 | av_log(s->avctx, AV_LOG_DEBUG, | |
2087 | "VP version: %d\n", s->version); | |
2088 | } | |
2089 | } | |
2090 | if (s->version || s->theora) { | |
2091 | if (get_bits1(&gb)) | |
2092 | av_log(s->avctx, AV_LOG_ERROR, | |
2093 | "Warning, unsupported keyframe coding type?!\n"); | |
2094 | skip_bits(&gb, 2); /* reserved? */ | |
2095 | } | |
2096 | } else { | |
2097 | if (!s->golden_frame.f->data[0]) { | |
2098 | av_log(s->avctx, AV_LOG_WARNING, | |
2099 | "vp3: first frame not a keyframe\n"); | |
2100 | ||
2101 | s->golden_frame.f->pict_type = AV_PICTURE_TYPE_I; | |
2102 | if (ff_thread_get_buffer(avctx, &s->golden_frame, | |
2103 | AV_GET_BUFFER_FLAG_REF) < 0) | |
2104 | goto error; | |
2105 | ff_thread_release_buffer(avctx, &s->last_frame); | |
2106 | if ((ret = ff_thread_ref_frame(&s->last_frame, | |
2107 | &s->golden_frame)) < 0) | |
2108 | goto error; | |
2109 | ff_thread_report_progress(&s->last_frame, INT_MAX, 0); | |
2110 | } | |
2111 | } | |
2112 | ||
2113 | memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment)); | |
2114 | ff_thread_finish_setup(avctx); | |
2115 | ||
2116 | if (unpack_superblocks(s, &gb)) { | |
2117 | av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n"); | |
2118 | goto error; | |
2119 | } | |
2120 | if (unpack_modes(s, &gb)) { | |
2121 | av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n"); | |
2122 | goto error; | |
2123 | } | |
2124 | if (unpack_vectors(s, &gb)) { | |
2125 | av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n"); | |
2126 | goto error; | |
2127 | } | |
2128 | if (unpack_block_qpis(s, &gb)) { | |
2129 | av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n"); | |
2130 | goto error; | |
2131 | } | |
2132 | if (unpack_dct_coeffs(s, &gb)) { | |
2133 | av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n"); | |
2134 | goto error; | |
2135 | } | |
2136 | ||
2137 | for (i = 0; i < 3; i++) { | |
2138 | int height = s->height >> (i && s->chroma_y_shift); | |
2139 | if (s->flipped_image) | |
2140 | s->data_offset[i] = 0; | |
2141 | else | |
2142 | s->data_offset[i] = (height - 1) * s->current_frame.f->linesize[i]; | |
2143 | } | |
2144 | ||
2145 | s->last_slice_end = 0; | |
2146 | for (i = 0; i < s->c_superblock_height; i++) | |
2147 | render_slice(s, i); | |
2148 | ||
2149 | // filter the last row | |
2150 | for (i = 0; i < 3; i++) { | |
2151 | int row = (s->height >> (3 + (i && s->chroma_y_shift))) - 1; | |
2152 | apply_loop_filter(s, i, row, row + 1); | |
2153 | } | |
2154 | vp3_draw_horiz_band(s, s->avctx->height); | |
2155 | ||
2156 | if ((ret = av_frame_ref(data, s->current_frame.f)) < 0) | |
2157 | return ret; | |
2158 | *got_frame = 1; | |
2159 | ||
2160 | if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_FRAME)) { | |
2161 | ret = update_frames(avctx); | |
2162 | if (ret < 0) | |
2163 | return ret; | |
2164 | } | |
2165 | ||
2166 | return buf_size; | |
2167 | ||
2168 | error: | |
2169 | ff_thread_report_progress(&s->current_frame, INT_MAX, 0); | |
2170 | ||
2171 | if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_FRAME)) | |
2172 | av_frame_unref(s->current_frame.f); | |
2173 | ||
2174 | return -1; | |
2175 | } | |
2176 | ||
2177 | static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb) | |
2178 | { | |
2179 | Vp3DecodeContext *s = avctx->priv_data; | |
2180 | ||
2181 | if (get_bits1(gb)) { | |
2182 | int token; | |
2183 | if (s->entries >= 32) { /* overflow */ | |
2184 | av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n"); | |
2185 | return -1; | |
2186 | } | |
2187 | token = get_bits(gb, 5); | |
2188 | av_dlog(avctx, "hti %d hbits %x token %d entry : %d size %d\n", | |
2189 | s->hti, s->hbits, token, s->entries, s->huff_code_size); | |
2190 | s->huffman_table[s->hti][token][0] = s->hbits; | |
2191 | s->huffman_table[s->hti][token][1] = s->huff_code_size; | |
2192 | s->entries++; | |
2193 | } else { | |
2194 | if (s->huff_code_size >= 32) { /* overflow */ | |
2195 | av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n"); | |
2196 | return -1; | |
2197 | } | |
2198 | s->huff_code_size++; | |
2199 | s->hbits <<= 1; | |
2200 | if (read_huffman_tree(avctx, gb)) | |
2201 | return -1; | |
2202 | s->hbits |= 1; | |
2203 | if (read_huffman_tree(avctx, gb)) | |
2204 | return -1; | |
2205 | s->hbits >>= 1; | |
2206 | s->huff_code_size--; | |
2207 | } | |
2208 | return 0; | |
2209 | } | |
2210 | ||
2211 | static int vp3_init_thread_copy(AVCodecContext *avctx) | |
2212 | { | |
2213 | Vp3DecodeContext *s = avctx->priv_data; | |
2214 | ||
2215 | s->superblock_coding = NULL; | |
2216 | s->all_fragments = NULL; | |
2217 | s->coded_fragment_list[0] = NULL; | |
2218 | s->dct_tokens_base = NULL; | |
2219 | s->superblock_fragments = NULL; | |
2220 | s->macroblock_coding = NULL; | |
2221 | s->motion_val[0] = NULL; | |
2222 | s->motion_val[1] = NULL; | |
2223 | s->edge_emu_buffer = NULL; | |
2224 | ||
2225 | return init_frames(s); | |
2226 | } | |
2227 | ||
2228 | #if CONFIG_THEORA_DECODER | |
2229 | static const enum AVPixelFormat theora_pix_fmts[4] = { | |
2230 | AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P | |
2231 | }; | |
2232 | ||
2233 | static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb) | |
2234 | { | |
2235 | Vp3DecodeContext *s = avctx->priv_data; | |
2236 | int visible_width, visible_height, colorspace; | |
2237 | int offset_x = 0, offset_y = 0; | |
2238 | int ret; | |
2239 | AVRational fps, aspect; | |
2240 | ||
2241 | s->theora = get_bits_long(gb, 24); | |
2242 | av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora); | |
2243 | ||
2244 | /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 | |
2245 | * but previous versions have the image flipped relative to vp3 */ | |
2246 | if (s->theora < 0x030200) { | |
2247 | s->flipped_image = 1; | |
2248 | av_log(avctx, AV_LOG_DEBUG, | |
2249 | "Old (<alpha3) Theora bitstream, flipped image\n"); | |
2250 | } | |
2251 | ||
2252 | visible_width = | |
2253 | s->width = get_bits(gb, 16) << 4; | |
2254 | visible_height = | |
2255 | s->height = get_bits(gb, 16) << 4; | |
2256 | ||
2257 | if (s->theora >= 0x030200) { | |
2258 | visible_width = get_bits_long(gb, 24); | |
2259 | visible_height = get_bits_long(gb, 24); | |
2260 | ||
2261 | offset_x = get_bits(gb, 8); /* offset x */ | |
2262 | offset_y = get_bits(gb, 8); /* offset y, from bottom */ | |
2263 | } | |
2264 | ||
2265 | fps.num = get_bits_long(gb, 32); | |
2266 | fps.den = get_bits_long(gb, 32); | |
2267 | if (fps.num && fps.den) { | |
2268 | if (fps.num < 0 || fps.den < 0) { | |
2269 | av_log(avctx, AV_LOG_ERROR, "Invalid framerate\n"); | |
2270 | return AVERROR_INVALIDDATA; | |
2271 | } | |
f6fa7814 | 2272 | av_reduce(&avctx->framerate.den, &avctx->framerate.num, |
2ba45a60 DM |
2273 | fps.den, fps.num, 1 << 30); |
2274 | } | |
2275 | ||
2276 | aspect.num = get_bits_long(gb, 24); | |
2277 | aspect.den = get_bits_long(gb, 24); | |
2278 | if (aspect.num && aspect.den) { | |
2279 | av_reduce(&avctx->sample_aspect_ratio.num, | |
2280 | &avctx->sample_aspect_ratio.den, | |
2281 | aspect.num, aspect.den, 1 << 30); | |
2282 | ff_set_sar(avctx, avctx->sample_aspect_ratio); | |
2283 | } | |
2284 | ||
2285 | if (s->theora < 0x030200) | |
2286 | skip_bits(gb, 5); /* keyframe frequency force */ | |
2287 | colorspace = get_bits(gb, 8); | |
2288 | skip_bits(gb, 24); /* bitrate */ | |
2289 | ||
2290 | skip_bits(gb, 6); /* quality hint */ | |
2291 | ||
2292 | if (s->theora >= 0x030200) { | |
2293 | skip_bits(gb, 5); /* keyframe frequency force */ | |
2294 | avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)]; | |
2295 | if (avctx->pix_fmt == AV_PIX_FMT_NONE) { | |
2296 | av_log(avctx, AV_LOG_ERROR, "Invalid pixel format\n"); | |
2297 | return AVERROR_INVALIDDATA; | |
2298 | } | |
2299 | skip_bits(gb, 3); /* reserved */ | |
2300 | } | |
2301 | ||
2302 | // align_get_bits(gb); | |
2303 | ||
2304 | if (visible_width <= s->width && visible_width > s->width - 16 && | |
2305 | visible_height <= s->height && visible_height > s->height - 16 && | |
2306 | !offset_x && (offset_y == s->height - visible_height)) | |
2307 | ret = ff_set_dimensions(avctx, visible_width, visible_height); | |
2308 | else | |
2309 | ret = ff_set_dimensions(avctx, s->width, s->height); | |
2310 | if (ret < 0) | |
2311 | return ret; | |
2312 | ||
2313 | if (colorspace == 1) | |
2314 | avctx->color_primaries = AVCOL_PRI_BT470M; | |
2315 | else if (colorspace == 2) | |
2316 | avctx->color_primaries = AVCOL_PRI_BT470BG; | |
2317 | ||
2318 | if (colorspace == 1 || colorspace == 2) { | |
2319 | avctx->colorspace = AVCOL_SPC_BT470BG; | |
2320 | avctx->color_trc = AVCOL_TRC_BT709; | |
2321 | } | |
2322 | ||
2323 | return 0; | |
2324 | } | |
2325 | ||
2326 | static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb) | |
2327 | { | |
2328 | Vp3DecodeContext *s = avctx->priv_data; | |
2329 | int i, n, matrices, inter, plane; | |
2330 | ||
2331 | if (s->theora >= 0x030200) { | |
2332 | n = get_bits(gb, 3); | |
2333 | /* loop filter limit values table */ | |
2334 | if (n) | |
2335 | for (i = 0; i < 64; i++) | |
2336 | s->filter_limit_values[i] = get_bits(gb, n); | |
2337 | } | |
2338 | ||
2339 | if (s->theora >= 0x030200) | |
2340 | n = get_bits(gb, 4) + 1; | |
2341 | else | |
2342 | n = 16; | |
2343 | /* quality threshold table */ | |
2344 | for (i = 0; i < 64; i++) | |
2345 | s->coded_ac_scale_factor[i] = get_bits(gb, n); | |
2346 | ||
2347 | if (s->theora >= 0x030200) | |
2348 | n = get_bits(gb, 4) + 1; | |
2349 | else | |
2350 | n = 16; | |
2351 | /* dc scale factor table */ | |
2352 | for (i = 0; i < 64; i++) | |
2353 | s->coded_dc_scale_factor[i] = get_bits(gb, n); | |
2354 | ||
2355 | if (s->theora >= 0x030200) | |
2356 | matrices = get_bits(gb, 9) + 1; | |
2357 | else | |
2358 | matrices = 3; | |
2359 | ||
2360 | if (matrices > 384) { | |
2361 | av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n"); | |
2362 | return -1; | |
2363 | } | |
2364 | ||
2365 | for (n = 0; n < matrices; n++) | |
2366 | for (i = 0; i < 64; i++) | |
2367 | s->base_matrix[n][i] = get_bits(gb, 8); | |
2368 | ||
2369 | for (inter = 0; inter <= 1; inter++) { | |
2370 | for (plane = 0; plane <= 2; plane++) { | |
2371 | int newqr = 1; | |
2372 | if (inter || plane > 0) | |
2373 | newqr = get_bits1(gb); | |
2374 | if (!newqr) { | |
2375 | int qtj, plj; | |
2376 | if (inter && get_bits1(gb)) { | |
2377 | qtj = 0; | |
2378 | plj = plane; | |
2379 | } else { | |
2380 | qtj = (3 * inter + plane - 1) / 3; | |
2381 | plj = (plane + 2) % 3; | |
2382 | } | |
2383 | s->qr_count[inter][plane] = s->qr_count[qtj][plj]; | |
2384 | memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], | |
2385 | sizeof(s->qr_size[0][0])); | |
2386 | memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], | |
2387 | sizeof(s->qr_base[0][0])); | |
2388 | } else { | |
2389 | int qri = 0; | |
2390 | int qi = 0; | |
2391 | ||
2392 | for (;;) { | |
2393 | i = get_bits(gb, av_log2(matrices - 1) + 1); | |
2394 | if (i >= matrices) { | |
2395 | av_log(avctx, AV_LOG_ERROR, | |
2396 | "invalid base matrix index\n"); | |
2397 | return -1; | |
2398 | } | |
2399 | s->qr_base[inter][plane][qri] = i; | |
2400 | if (qi >= 63) | |
2401 | break; | |
2402 | i = get_bits(gb, av_log2(63 - qi) + 1) + 1; | |
2403 | s->qr_size[inter][plane][qri++] = i; | |
2404 | qi += i; | |
2405 | } | |
2406 | ||
2407 | if (qi > 63) { | |
2408 | av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi); | |
2409 | return -1; | |
2410 | } | |
2411 | s->qr_count[inter][plane] = qri; | |
2412 | } | |
2413 | } | |
2414 | } | |
2415 | ||
2416 | /* Huffman tables */ | |
2417 | for (s->hti = 0; s->hti < 80; s->hti++) { | |
2418 | s->entries = 0; | |
2419 | s->huff_code_size = 1; | |
2420 | if (!get_bits1(gb)) { | |
2421 | s->hbits = 0; | |
2422 | if (read_huffman_tree(avctx, gb)) | |
2423 | return -1; | |
2424 | s->hbits = 1; | |
2425 | if (read_huffman_tree(avctx, gb)) | |
2426 | return -1; | |
2427 | } | |
2428 | } | |
2429 | ||
2430 | s->theora_tables = 1; | |
2431 | ||
2432 | return 0; | |
2433 | } | |
2434 | ||
2435 | static av_cold int theora_decode_init(AVCodecContext *avctx) | |
2436 | { | |
2437 | Vp3DecodeContext *s = avctx->priv_data; | |
2438 | GetBitContext gb; | |
2439 | int ptype; | |
2440 | uint8_t *header_start[3]; | |
2441 | int header_len[3]; | |
2442 | int i; | |
2443 | ||
2444 | avctx->pix_fmt = AV_PIX_FMT_YUV420P; | |
2445 | ||
2446 | s->theora = 1; | |
2447 | ||
2448 | if (!avctx->extradata_size) { | |
2449 | av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n"); | |
2450 | return -1; | |
2451 | } | |
2452 | ||
2453 | if (avpriv_split_xiph_headers(avctx->extradata, avctx->extradata_size, | |
2454 | 42, header_start, header_len) < 0) { | |
2455 | av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n"); | |
2456 | return -1; | |
2457 | } | |
2458 | ||
2459 | for (i = 0; i < 3; i++) { | |
2460 | if (header_len[i] <= 0) | |
2461 | continue; | |
2462 | init_get_bits(&gb, header_start[i], header_len[i] * 8); | |
2463 | ||
2464 | ptype = get_bits(&gb, 8); | |
2465 | ||
2466 | if (!(ptype & 0x80)) { | |
2467 | av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n"); | |
2468 | // return -1; | |
2469 | } | |
2470 | ||
2471 | // FIXME: Check for this as well. | |
2472 | skip_bits_long(&gb, 6 * 8); /* "theora" */ | |
2473 | ||
2474 | switch (ptype) { | |
2475 | case 0x80: | |
2476 | if (theora_decode_header(avctx, &gb) < 0) | |
2477 | return -1; | |
2478 | break; | |
2479 | case 0x81: | |
2480 | // FIXME: is this needed? it breaks sometimes | |
2481 | // theora_decode_comments(avctx, gb); | |
2482 | break; | |
2483 | case 0x82: | |
2484 | if (theora_decode_tables(avctx, &gb)) | |
2485 | return -1; | |
2486 | break; | |
2487 | default: | |
2488 | av_log(avctx, AV_LOG_ERROR, | |
2489 | "Unknown Theora config packet: %d\n", ptype & ~0x80); | |
2490 | break; | |
2491 | } | |
2492 | if (ptype != 0x81 && 8 * header_len[i] != get_bits_count(&gb)) | |
2493 | av_log(avctx, AV_LOG_WARNING, | |
2494 | "%d bits left in packet %X\n", | |
2495 | 8 * header_len[i] - get_bits_count(&gb), ptype); | |
2496 | if (s->theora < 0x030200) | |
2497 | break; | |
2498 | } | |
2499 | ||
2500 | return vp3_decode_init(avctx); | |
2501 | } | |
2502 | ||
2503 | AVCodec ff_theora_decoder = { | |
2504 | .name = "theora", | |
2505 | .long_name = NULL_IF_CONFIG_SMALL("Theora"), | |
2506 | .type = AVMEDIA_TYPE_VIDEO, | |
2507 | .id = AV_CODEC_ID_THEORA, | |
2508 | .priv_data_size = sizeof(Vp3DecodeContext), | |
2509 | .init = theora_decode_init, | |
2510 | .close = vp3_decode_end, | |
2511 | .decode = vp3_decode_frame, | |
2512 | .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND | | |
2513 | CODEC_CAP_FRAME_THREADS, | |
2514 | .flush = vp3_decode_flush, | |
2515 | .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy), | |
2516 | .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context) | |
2517 | }; | |
2518 | #endif | |
2519 | ||
2520 | AVCodec ff_vp3_decoder = { | |
2521 | .name = "vp3", | |
2522 | .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"), | |
2523 | .type = AVMEDIA_TYPE_VIDEO, | |
2524 | .id = AV_CODEC_ID_VP3, | |
2525 | .priv_data_size = sizeof(Vp3DecodeContext), | |
2526 | .init = vp3_decode_init, | |
2527 | .close = vp3_decode_end, | |
2528 | .decode = vp3_decode_frame, | |
2529 | .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND | | |
2530 | CODEC_CAP_FRAME_THREADS, | |
2531 | .flush = vp3_decode_flush, | |
2532 | .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy), | |
2533 | .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context), | |
2534 | }; |