blob: 86e5852e32935e8da4a21fe6333c9d11e7efa189
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, theora_header; |
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 | uint8_t offset_x; |
179 | uint8_t offset_y; |
180 | int offset_x_warned; |
181 | |
182 | int8_t (*motion_val[2])[2]; |
183 | |
184 | /* tables */ |
185 | uint16_t coded_dc_scale_factor[64]; |
186 | uint32_t coded_ac_scale_factor[64]; |
187 | uint8_t base_matrix[384][64]; |
188 | uint8_t qr_count[2][3]; |
189 | uint8_t qr_size[2][3][64]; |
190 | uint16_t qr_base[2][3][64]; |
191 | |
192 | /** |
193 | * This is a list of all tokens in bitstream order. Reordering takes place |
194 | * by pulling from each level during IDCT. As a consequence, IDCT must be |
195 | * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32 |
196 | * otherwise. The 32 different tokens with up to 12 bits of extradata are |
197 | * collapsed into 3 types, packed as follows: |
198 | * (from the low to high bits) |
199 | * |
200 | * 2 bits: type (0,1,2) |
201 | * 0: EOB run, 14 bits for run length (12 needed) |
202 | * 1: zero run, 7 bits for run length |
203 | * 7 bits for the next coefficient (3 needed) |
204 | * 2: coefficient, 14 bits (11 needed) |
205 | * |
206 | * Coefficients are signed, so are packed in the highest bits for automatic |
207 | * sign extension. |
208 | */ |
209 | int16_t *dct_tokens[3][64]; |
210 | int16_t *dct_tokens_base; |
211 | #define TOKEN_EOB(eob_run) ((eob_run) << 2) |
212 | #define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) * 512) + ((zero_run) << 2) + 1) |
213 | #define TOKEN_COEFF(coeff) (((coeff) * 4) + 2) |
214 | |
215 | /** |
216 | * number of blocks that contain DCT coefficients at |
217 | * the given level or higher |
218 | */ |
219 | int num_coded_frags[3][64]; |
220 | int total_num_coded_frags; |
221 | |
222 | /* this is a list of indexes into the all_fragments array indicating |
223 | * which of the fragments are coded */ |
224 | int *coded_fragment_list[3]; |
225 | |
226 | VLC dc_vlc[16]; |
227 | VLC ac_vlc_1[16]; |
228 | VLC ac_vlc_2[16]; |
229 | VLC ac_vlc_3[16]; |
230 | VLC ac_vlc_4[16]; |
231 | |
232 | VLC superblock_run_length_vlc; |
233 | VLC fragment_run_length_vlc; |
234 | VLC mode_code_vlc; |
235 | VLC motion_vector_vlc; |
236 | |
237 | /* these arrays need to be on 16-byte boundaries since SSE2 operations |
238 | * index into them */ |
239 | DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64]; ///< qmat[qpi][is_inter][plane] |
240 | |
241 | /* This table contains superblock_count * 16 entries. Each set of 16 |
242 | * numbers corresponds to the fragment indexes 0..15 of the superblock. |
243 | * An entry will be -1 to indicate that no entry corresponds to that |
244 | * index. */ |
245 | int *superblock_fragments; |
246 | |
247 | /* This is an array that indicates how a particular macroblock |
248 | * is coded. */ |
249 | unsigned char *macroblock_coding; |
250 | |
251 | uint8_t *edge_emu_buffer; |
252 | |
253 | /* Huffman decode */ |
254 | int hti; |
255 | unsigned int hbits; |
256 | int entries; |
257 | int huff_code_size; |
258 | uint32_t huffman_table[80][32][2]; |
259 | |
260 | uint8_t filter_limit_values[64]; |
261 | DECLARE_ALIGNED(8, int, bounding_values_array)[256 + 2]; |
262 | } Vp3DecodeContext; |
263 | |
264 | /************************************************************************ |
265 | * VP3 specific functions |
266 | ************************************************************************/ |
267 | |
268 | static av_cold void free_tables(AVCodecContext *avctx) |
269 | { |
270 | Vp3DecodeContext *s = avctx->priv_data; |
271 | |
272 | av_freep(&s->superblock_coding); |
273 | av_freep(&s->all_fragments); |
274 | av_freep(&s->coded_fragment_list[0]); |
275 | av_freep(&s->dct_tokens_base); |
276 | av_freep(&s->superblock_fragments); |
277 | av_freep(&s->macroblock_coding); |
278 | av_freep(&s->motion_val[0]); |
279 | av_freep(&s->motion_val[1]); |
280 | } |
281 | |
282 | static void vp3_decode_flush(AVCodecContext *avctx) |
283 | { |
284 | Vp3DecodeContext *s = avctx->priv_data; |
285 | |
286 | if (s->golden_frame.f) |
287 | ff_thread_release_buffer(avctx, &s->golden_frame); |
288 | if (s->last_frame.f) |
289 | ff_thread_release_buffer(avctx, &s->last_frame); |
290 | if (s->current_frame.f) |
291 | ff_thread_release_buffer(avctx, &s->current_frame); |
292 | } |
293 | |
294 | static av_cold int vp3_decode_end(AVCodecContext *avctx) |
295 | { |
296 | Vp3DecodeContext *s = avctx->priv_data; |
297 | int i; |
298 | |
299 | free_tables(avctx); |
300 | av_freep(&s->edge_emu_buffer); |
301 | |
302 | s->theora_tables = 0; |
303 | |
304 | /* release all frames */ |
305 | vp3_decode_flush(avctx); |
306 | av_frame_free(&s->current_frame.f); |
307 | av_frame_free(&s->last_frame.f); |
308 | av_frame_free(&s->golden_frame.f); |
309 | |
310 | if (avctx->internal->is_copy) |
311 | return 0; |
312 | |
313 | for (i = 0; i < 16; i++) { |
314 | ff_free_vlc(&s->dc_vlc[i]); |
315 | ff_free_vlc(&s->ac_vlc_1[i]); |
316 | ff_free_vlc(&s->ac_vlc_2[i]); |
317 | ff_free_vlc(&s->ac_vlc_3[i]); |
318 | ff_free_vlc(&s->ac_vlc_4[i]); |
319 | } |
320 | |
321 | ff_free_vlc(&s->superblock_run_length_vlc); |
322 | ff_free_vlc(&s->fragment_run_length_vlc); |
323 | ff_free_vlc(&s->mode_code_vlc); |
324 | ff_free_vlc(&s->motion_vector_vlc); |
325 | |
326 | return 0; |
327 | } |
328 | |
329 | /** |
330 | * This function sets up all of the various blocks mappings: |
331 | * superblocks <-> fragments, macroblocks <-> fragments, |
332 | * superblocks <-> macroblocks |
333 | * |
334 | * @return 0 is successful; returns 1 if *anything* went wrong. |
335 | */ |
336 | static int init_block_mapping(Vp3DecodeContext *s) |
337 | { |
338 | int sb_x, sb_y, plane; |
339 | int x, y, i, j = 0; |
340 | |
341 | for (plane = 0; plane < 3; plane++) { |
342 | int sb_width = plane ? s->c_superblock_width |
343 | : s->y_superblock_width; |
344 | int sb_height = plane ? s->c_superblock_height |
345 | : s->y_superblock_height; |
346 | int frag_width = s->fragment_width[!!plane]; |
347 | int frag_height = s->fragment_height[!!plane]; |
348 | |
349 | for (sb_y = 0; sb_y < sb_height; sb_y++) |
350 | for (sb_x = 0; sb_x < sb_width; sb_x++) |
351 | for (i = 0; i < 16; i++) { |
352 | x = 4 * sb_x + hilbert_offset[i][0]; |
353 | y = 4 * sb_y + hilbert_offset[i][1]; |
354 | |
355 | if (x < frag_width && y < frag_height) |
356 | s->superblock_fragments[j++] = s->fragment_start[plane] + |
357 | y * frag_width + x; |
358 | else |
359 | s->superblock_fragments[j++] = -1; |
360 | } |
361 | } |
362 | |
363 | return 0; /* successful path out */ |
364 | } |
365 | |
366 | /* |
367 | * This function sets up the dequantization tables used for a particular |
368 | * frame. |
369 | */ |
370 | static void init_dequantizer(Vp3DecodeContext *s, int qpi) |
371 | { |
372 | int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]]; |
373 | int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]]; |
374 | int i, plane, inter, qri, bmi, bmj, qistart; |
375 | |
376 | for (inter = 0; inter < 2; inter++) { |
377 | for (plane = 0; plane < 3; plane++) { |
378 | int sum = 0; |
379 | for (qri = 0; qri < s->qr_count[inter][plane]; qri++) { |
380 | sum += s->qr_size[inter][plane][qri]; |
381 | if (s->qps[qpi] <= sum) |
382 | break; |
383 | } |
384 | qistart = sum - s->qr_size[inter][plane][qri]; |
385 | bmi = s->qr_base[inter][plane][qri]; |
386 | bmj = s->qr_base[inter][plane][qri + 1]; |
387 | for (i = 0; i < 64; i++) { |
388 | int coeff = (2 * (sum - s->qps[qpi]) * s->base_matrix[bmi][i] - |
389 | 2 * (qistart - s->qps[qpi]) * s->base_matrix[bmj][i] + |
390 | s->qr_size[inter][plane][qri]) / |
391 | (2 * s->qr_size[inter][plane][qri]); |
392 | |
393 | int qmin = 8 << (inter + !i); |
394 | int qscale = i ? ac_scale_factor : dc_scale_factor; |
395 | |
396 | s->qmat[qpi][inter][plane][s->idct_permutation[i]] = |
397 | av_clip((qscale * coeff) / 100 * 4, qmin, 4096); |
398 | } |
399 | /* all DC coefficients use the same quant so as not to interfere |
400 | * with DC prediction */ |
401 | s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0]; |
402 | } |
403 | } |
404 | } |
405 | |
406 | /* |
407 | * This function initializes the loop filter boundary limits if the frame's |
408 | * quality index is different from the previous frame's. |
409 | * |
410 | * The filter_limit_values may not be larger than 127. |
411 | */ |
412 | static void init_loop_filter(Vp3DecodeContext *s) |
413 | { |
414 | int *bounding_values = s->bounding_values_array + 127; |
415 | int filter_limit; |
416 | int x; |
417 | int value; |
418 | |
419 | filter_limit = s->filter_limit_values[s->qps[0]]; |
420 | av_assert0(filter_limit < 128U); |
421 | |
422 | /* set up the bounding values */ |
423 | memset(s->bounding_values_array, 0, 256 * sizeof(int)); |
424 | for (x = 0; x < filter_limit; x++) { |
425 | bounding_values[-x] = -x; |
426 | bounding_values[x] = x; |
427 | } |
428 | for (x = value = filter_limit; x < 128 && value; x++, value--) { |
429 | bounding_values[ x] = value; |
430 | bounding_values[-x] = -value; |
431 | } |
432 | if (value) |
433 | bounding_values[128] = value; |
434 | bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202; |
435 | } |
436 | |
437 | /* |
438 | * This function unpacks all of the superblock/macroblock/fragment coding |
439 | * information from the bitstream. |
440 | */ |
441 | static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb) |
442 | { |
443 | int superblock_starts[3] = { |
444 | 0, s->u_superblock_start, s->v_superblock_start |
445 | }; |
446 | int bit = 0; |
447 | int current_superblock = 0; |
448 | int current_run = 0; |
449 | int num_partial_superblocks = 0; |
450 | |
451 | int i, j; |
452 | int current_fragment; |
453 | int plane; |
454 | |
455 | if (s->keyframe) { |
456 | memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count); |
457 | } else { |
458 | /* unpack the list of partially-coded superblocks */ |
459 | bit = get_bits1(gb) ^ 1; |
460 | current_run = 0; |
461 | |
462 | while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) { |
463 | if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN) |
464 | bit = get_bits1(gb); |
465 | else |
466 | bit ^= 1; |
467 | |
468 | current_run = get_vlc2(gb, s->superblock_run_length_vlc.table, |
469 | 6, 2) + 1; |
470 | if (current_run == 34) |
471 | current_run += get_bits(gb, 12); |
472 | |
473 | if (current_run > s->superblock_count - current_superblock) { |
474 | av_log(s->avctx, AV_LOG_ERROR, |
475 | "Invalid partially coded superblock run length\n"); |
476 | return -1; |
477 | } |
478 | |
479 | memset(s->superblock_coding + current_superblock, bit, current_run); |
480 | |
481 | current_superblock += current_run; |
482 | if (bit) |
483 | num_partial_superblocks += current_run; |
484 | } |
485 | |
486 | /* unpack the list of fully coded superblocks if any of the blocks were |
487 | * not marked as partially coded in the previous step */ |
488 | if (num_partial_superblocks < s->superblock_count) { |
489 | int superblocks_decoded = 0; |
490 | |
491 | current_superblock = 0; |
492 | bit = get_bits1(gb) ^ 1; |
493 | current_run = 0; |
494 | |
495 | while (superblocks_decoded < s->superblock_count - num_partial_superblocks && |
496 | get_bits_left(gb) > 0) { |
497 | if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN) |
498 | bit = get_bits1(gb); |
499 | else |
500 | bit ^= 1; |
501 | |
502 | current_run = get_vlc2(gb, s->superblock_run_length_vlc.table, |
503 | 6, 2) + 1; |
504 | if (current_run == 34) |
505 | current_run += get_bits(gb, 12); |
506 | |
507 | for (j = 0; j < current_run; current_superblock++) { |
508 | if (current_superblock >= s->superblock_count) { |
509 | av_log(s->avctx, AV_LOG_ERROR, |
510 | "Invalid fully coded superblock run length\n"); |
511 | return -1; |
512 | } |
513 | |
514 | /* skip any superblocks already marked as partially coded */ |
515 | if (s->superblock_coding[current_superblock] == SB_NOT_CODED) { |
516 | s->superblock_coding[current_superblock] = 2 * bit; |
517 | j++; |
518 | } |
519 | } |
520 | superblocks_decoded += current_run; |
521 | } |
522 | } |
523 | |
524 | /* if there were partial blocks, initialize bitstream for |
525 | * unpacking fragment codings */ |
526 | if (num_partial_superblocks) { |
527 | current_run = 0; |
528 | bit = get_bits1(gb); |
529 | /* toggle the bit because as soon as the first run length is |
530 | * fetched the bit will be toggled again */ |
531 | bit ^= 1; |
532 | } |
533 | } |
534 | |
535 | /* figure out which fragments are coded; iterate through each |
536 | * superblock (all planes) */ |
537 | s->total_num_coded_frags = 0; |
538 | memset(s->macroblock_coding, MODE_COPY, s->macroblock_count); |
539 | |
540 | for (plane = 0; plane < 3; plane++) { |
541 | int sb_start = superblock_starts[plane]; |
542 | int sb_end = sb_start + (plane ? s->c_superblock_count |
543 | : s->y_superblock_count); |
544 | int num_coded_frags = 0; |
545 | |
546 | for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) { |
547 | /* iterate through all 16 fragments in a superblock */ |
548 | for (j = 0; j < 16; j++) { |
549 | /* if the fragment is in bounds, check its coding status */ |
550 | current_fragment = s->superblock_fragments[i * 16 + j]; |
551 | if (current_fragment != -1) { |
552 | int coded = s->superblock_coding[i]; |
553 | |
554 | if (s->superblock_coding[i] == SB_PARTIALLY_CODED) { |
555 | /* fragment may or may not be coded; this is the case |
556 | * that cares about the fragment coding runs */ |
557 | if (current_run-- == 0) { |
558 | bit ^= 1; |
559 | current_run = get_vlc2(gb, s->fragment_run_length_vlc.table, 5, 2); |
560 | } |
561 | coded = bit; |
562 | } |
563 | |
564 | if (coded) { |
565 | /* default mode; actual mode will be decoded in |
566 | * the next phase */ |
567 | s->all_fragments[current_fragment].coding_method = |
568 | MODE_INTER_NO_MV; |
569 | s->coded_fragment_list[plane][num_coded_frags++] = |
570 | current_fragment; |
571 | } else { |
572 | /* not coded; copy this fragment from the prior frame */ |
573 | s->all_fragments[current_fragment].coding_method = |
574 | MODE_COPY; |
575 | } |
576 | } |
577 | } |
578 | } |
579 | s->total_num_coded_frags += num_coded_frags; |
580 | for (i = 0; i < 64; i++) |
581 | s->num_coded_frags[plane][i] = num_coded_frags; |
582 | if (plane < 2) |
583 | s->coded_fragment_list[plane + 1] = s->coded_fragment_list[plane] + |
584 | num_coded_frags; |
585 | } |
586 | return 0; |
587 | } |
588 | |
589 | /* |
590 | * This function unpacks all the coding mode data for individual macroblocks |
591 | * from the bitstream. |
592 | */ |
593 | static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb) |
594 | { |
595 | int i, j, k, sb_x, sb_y; |
596 | int scheme; |
597 | int current_macroblock; |
598 | int current_fragment; |
599 | int coding_mode; |
600 | int custom_mode_alphabet[CODING_MODE_COUNT]; |
601 | const int *alphabet; |
602 | Vp3Fragment *frag; |
603 | |
604 | if (s->keyframe) { |
605 | for (i = 0; i < s->fragment_count; i++) |
606 | s->all_fragments[i].coding_method = MODE_INTRA; |
607 | } else { |
608 | /* fetch the mode coding scheme for this frame */ |
609 | scheme = get_bits(gb, 3); |
610 | |
611 | /* is it a custom coding scheme? */ |
612 | if (scheme == 0) { |
613 | for (i = 0; i < 8; i++) |
614 | custom_mode_alphabet[i] = MODE_INTER_NO_MV; |
615 | for (i = 0; i < 8; i++) |
616 | custom_mode_alphabet[get_bits(gb, 3)] = i; |
617 | alphabet = custom_mode_alphabet; |
618 | } else |
619 | alphabet = ModeAlphabet[scheme - 1]; |
620 | |
621 | /* iterate through all of the macroblocks that contain 1 or more |
622 | * coded fragments */ |
623 | for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) { |
624 | for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) { |
625 | if (get_bits_left(gb) <= 0) |
626 | return -1; |
627 | |
628 | for (j = 0; j < 4; j++) { |
629 | int mb_x = 2 * sb_x + (j >> 1); |
630 | int mb_y = 2 * sb_y + (((j >> 1) + j) & 1); |
631 | current_macroblock = mb_y * s->macroblock_width + mb_x; |
632 | |
633 | if (mb_x >= s->macroblock_width || |
634 | mb_y >= s->macroblock_height) |
635 | continue; |
636 | |
637 | #define BLOCK_X (2 * mb_x + (k & 1)) |
638 | #define BLOCK_Y (2 * mb_y + (k >> 1)) |
639 | /* coding modes are only stored if the macroblock has |
640 | * at least one luma block coded, otherwise it must be |
641 | * INTER_NO_MV */ |
642 | for (k = 0; k < 4; k++) { |
643 | current_fragment = BLOCK_Y * |
644 | s->fragment_width[0] + BLOCK_X; |
645 | if (s->all_fragments[current_fragment].coding_method != MODE_COPY) |
646 | break; |
647 | } |
648 | if (k == 4) { |
649 | s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV; |
650 | continue; |
651 | } |
652 | |
653 | /* mode 7 means get 3 bits for each coding mode */ |
654 | if (scheme == 7) |
655 | coding_mode = get_bits(gb, 3); |
656 | else |
657 | coding_mode = alphabet[get_vlc2(gb, s->mode_code_vlc.table, 3, 3)]; |
658 | |
659 | s->macroblock_coding[current_macroblock] = coding_mode; |
660 | for (k = 0; k < 4; k++) { |
661 | frag = s->all_fragments + BLOCK_Y * s->fragment_width[0] + BLOCK_X; |
662 | if (frag->coding_method != MODE_COPY) |
663 | frag->coding_method = coding_mode; |
664 | } |
665 | |
666 | #define SET_CHROMA_MODES \ |
667 | if (frag[s->fragment_start[1]].coding_method != MODE_COPY) \ |
668 | frag[s->fragment_start[1]].coding_method = coding_mode; \ |
669 | if (frag[s->fragment_start[2]].coding_method != MODE_COPY) \ |
670 | frag[s->fragment_start[2]].coding_method = coding_mode; |
671 | |
672 | if (s->chroma_y_shift) { |
673 | frag = s->all_fragments + mb_y * |
674 | s->fragment_width[1] + mb_x; |
675 | SET_CHROMA_MODES |
676 | } else if (s->chroma_x_shift) { |
677 | frag = s->all_fragments + |
678 | 2 * mb_y * s->fragment_width[1] + mb_x; |
679 | for (k = 0; k < 2; k++) { |
680 | SET_CHROMA_MODES |
681 | frag += s->fragment_width[1]; |
682 | } |
683 | } else { |
684 | for (k = 0; k < 4; k++) { |
685 | frag = s->all_fragments + |
686 | BLOCK_Y * s->fragment_width[1] + BLOCK_X; |
687 | SET_CHROMA_MODES |
688 | } |
689 | } |
690 | } |
691 | } |
692 | } |
693 | } |
694 | |
695 | return 0; |
696 | } |
697 | |
698 | /* |
699 | * This function unpacks all the motion vectors for the individual |
700 | * macroblocks from the bitstream. |
701 | */ |
702 | static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb) |
703 | { |
704 | int j, k, sb_x, sb_y; |
705 | int coding_mode; |
706 | int motion_x[4]; |
707 | int motion_y[4]; |
708 | int last_motion_x = 0; |
709 | int last_motion_y = 0; |
710 | int prior_last_motion_x = 0; |
711 | int prior_last_motion_y = 0; |
712 | int current_macroblock; |
713 | int current_fragment; |
714 | int frag; |
715 | |
716 | if (s->keyframe) |
717 | return 0; |
718 | |
719 | /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */ |
720 | coding_mode = get_bits1(gb); |
721 | |
722 | /* iterate through all of the macroblocks that contain 1 or more |
723 | * coded fragments */ |
724 | for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) { |
725 | for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) { |
726 | if (get_bits_left(gb) <= 0) |
727 | return -1; |
728 | |
729 | for (j = 0; j < 4; j++) { |
730 | int mb_x = 2 * sb_x + (j >> 1); |
731 | int mb_y = 2 * sb_y + (((j >> 1) + j) & 1); |
732 | current_macroblock = mb_y * s->macroblock_width + mb_x; |
733 | |
734 | if (mb_x >= s->macroblock_width || |
735 | mb_y >= s->macroblock_height || |
736 | s->macroblock_coding[current_macroblock] == MODE_COPY) |
737 | continue; |
738 | |
739 | switch (s->macroblock_coding[current_macroblock]) { |
740 | case MODE_INTER_PLUS_MV: |
741 | case MODE_GOLDEN_MV: |
742 | /* all 6 fragments use the same motion vector */ |
743 | if (coding_mode == 0) { |
744 | motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)]; |
745 | motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)]; |
746 | } else { |
747 | motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)]; |
748 | motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)]; |
749 | } |
750 | |
751 | /* vector maintenance, only on MODE_INTER_PLUS_MV */ |
752 | if (s->macroblock_coding[current_macroblock] == MODE_INTER_PLUS_MV) { |
753 | prior_last_motion_x = last_motion_x; |
754 | prior_last_motion_y = last_motion_y; |
755 | last_motion_x = motion_x[0]; |
756 | last_motion_y = motion_y[0]; |
757 | } |
758 | break; |
759 | |
760 | case MODE_INTER_FOURMV: |
761 | /* vector maintenance */ |
762 | prior_last_motion_x = last_motion_x; |
763 | prior_last_motion_y = last_motion_y; |
764 | |
765 | /* fetch 4 vectors from the bitstream, one for each |
766 | * Y fragment, then average for the C fragment vectors */ |
767 | for (k = 0; k < 4; k++) { |
768 | current_fragment = BLOCK_Y * s->fragment_width[0] + BLOCK_X; |
769 | if (s->all_fragments[current_fragment].coding_method != MODE_COPY) { |
770 | if (coding_mode == 0) { |
771 | motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)]; |
772 | motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)]; |
773 | } else { |
774 | motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)]; |
775 | motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)]; |
776 | } |
777 | last_motion_x = motion_x[k]; |
778 | last_motion_y = motion_y[k]; |
779 | } else { |
780 | motion_x[k] = 0; |
781 | motion_y[k] = 0; |
782 | } |
783 | } |
784 | break; |
785 | |
786 | case MODE_INTER_LAST_MV: |
787 | /* all 6 fragments use the last motion vector */ |
788 | motion_x[0] = last_motion_x; |
789 | motion_y[0] = last_motion_y; |
790 | |
791 | /* no vector maintenance (last vector remains the |
792 | * last vector) */ |
793 | break; |
794 | |
795 | case MODE_INTER_PRIOR_LAST: |
796 | /* all 6 fragments use the motion vector prior to the |
797 | * last motion vector */ |
798 | motion_x[0] = prior_last_motion_x; |
799 | motion_y[0] = prior_last_motion_y; |
800 | |
801 | /* vector maintenance */ |
802 | prior_last_motion_x = last_motion_x; |
803 | prior_last_motion_y = last_motion_y; |
804 | last_motion_x = motion_x[0]; |
805 | last_motion_y = motion_y[0]; |
806 | break; |
807 | |
808 | default: |
809 | /* covers intra, inter without MV, golden without MV */ |
810 | motion_x[0] = 0; |
811 | motion_y[0] = 0; |
812 | |
813 | /* no vector maintenance */ |
814 | break; |
815 | } |
816 | |
817 | /* assign the motion vectors to the correct fragments */ |
818 | for (k = 0; k < 4; k++) { |
819 | current_fragment = |
820 | BLOCK_Y * s->fragment_width[0] + BLOCK_X; |
821 | if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) { |
822 | s->motion_val[0][current_fragment][0] = motion_x[k]; |
823 | s->motion_val[0][current_fragment][1] = motion_y[k]; |
824 | } else { |
825 | s->motion_val[0][current_fragment][0] = motion_x[0]; |
826 | s->motion_val[0][current_fragment][1] = motion_y[0]; |
827 | } |
828 | } |
829 | |
830 | if (s->chroma_y_shift) { |
831 | if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) { |
832 | motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] + |
833 | motion_x[2] + motion_x[3], 2); |
834 | motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] + |
835 | motion_y[2] + motion_y[3], 2); |
836 | } |
837 | motion_x[0] = (motion_x[0] >> 1) | (motion_x[0] & 1); |
838 | motion_y[0] = (motion_y[0] >> 1) | (motion_y[0] & 1); |
839 | frag = mb_y * s->fragment_width[1] + mb_x; |
840 | s->motion_val[1][frag][0] = motion_x[0]; |
841 | s->motion_val[1][frag][1] = motion_y[0]; |
842 | } else if (s->chroma_x_shift) { |
843 | if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) { |
844 | motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1); |
845 | motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1); |
846 | motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1); |
847 | motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1); |
848 | } else { |
849 | motion_x[1] = motion_x[0]; |
850 | motion_y[1] = motion_y[0]; |
851 | } |
852 | motion_x[0] = (motion_x[0] >> 1) | (motion_x[0] & 1); |
853 | motion_x[1] = (motion_x[1] >> 1) | (motion_x[1] & 1); |
854 | |
855 | frag = 2 * mb_y * s->fragment_width[1] + mb_x; |
856 | for (k = 0; k < 2; k++) { |
857 | s->motion_val[1][frag][0] = motion_x[k]; |
858 | s->motion_val[1][frag][1] = motion_y[k]; |
859 | frag += s->fragment_width[1]; |
860 | } |
861 | } else { |
862 | for (k = 0; k < 4; k++) { |
863 | frag = BLOCK_Y * s->fragment_width[1] + BLOCK_X; |
864 | if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) { |
865 | s->motion_val[1][frag][0] = motion_x[k]; |
866 | s->motion_val[1][frag][1] = motion_y[k]; |
867 | } else { |
868 | s->motion_val[1][frag][0] = motion_x[0]; |
869 | s->motion_val[1][frag][1] = motion_y[0]; |
870 | } |
871 | } |
872 | } |
873 | } |
874 | } |
875 | } |
876 | |
877 | return 0; |
878 | } |
879 | |
880 | static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb) |
881 | { |
882 | int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi; |
883 | int num_blocks = s->total_num_coded_frags; |
884 | |
885 | for (qpi = 0; qpi < s->nqps - 1 && num_blocks > 0; qpi++) { |
886 | i = blocks_decoded = num_blocks_at_qpi = 0; |
887 | |
888 | bit = get_bits1(gb) ^ 1; |
889 | run_length = 0; |
890 | |
891 | do { |
892 | if (run_length == MAXIMUM_LONG_BIT_RUN) |
893 | bit = get_bits1(gb); |
894 | else |
895 | bit ^= 1; |
896 | |
897 | run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1; |
898 | if (run_length == 34) |
899 | run_length += get_bits(gb, 12); |
900 | blocks_decoded += run_length; |
901 | |
902 | if (!bit) |
903 | num_blocks_at_qpi += run_length; |
904 | |
905 | for (j = 0; j < run_length; i++) { |
906 | if (i >= s->total_num_coded_frags) |
907 | return -1; |
908 | |
909 | if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) { |
910 | s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit; |
911 | j++; |
912 | } |
913 | } |
914 | } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0); |
915 | |
916 | num_blocks -= num_blocks_at_qpi; |
917 | } |
918 | |
919 | return 0; |
920 | } |
921 | |
922 | /* |
923 | * This function is called by unpack_dct_coeffs() to extract the VLCs from |
924 | * the bitstream. The VLCs encode tokens which are used to unpack DCT |
925 | * data. This function unpacks all the VLCs for either the Y plane or both |
926 | * C planes, and is called for DC coefficients or different AC coefficient |
927 | * levels (since different coefficient types require different VLC tables. |
928 | * |
929 | * This function returns a residual eob run. E.g, if a particular token gave |
930 | * instructions to EOB the next 5 fragments and there were only 2 fragments |
931 | * left in the current fragment range, 3 would be returned so that it could |
932 | * be passed into the next call to this same function. |
933 | */ |
934 | static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb, |
935 | VLC *table, int coeff_index, |
936 | int plane, |
937 | int eob_run) |
938 | { |
939 | int i, j = 0; |
940 | int token; |
941 | int zero_run = 0; |
942 | int16_t coeff = 0; |
943 | int bits_to_get; |
944 | int blocks_ended; |
945 | int coeff_i = 0; |
946 | int num_coeffs = s->num_coded_frags[plane][coeff_index]; |
947 | int16_t *dct_tokens = s->dct_tokens[plane][coeff_index]; |
948 | |
949 | /* local references to structure members to avoid repeated dereferences */ |
950 | int *coded_fragment_list = s->coded_fragment_list[plane]; |
951 | Vp3Fragment *all_fragments = s->all_fragments; |
952 | VLC_TYPE(*vlc_table)[2] = table->table; |
953 | |
954 | if (num_coeffs < 0) |
955 | av_log(s->avctx, AV_LOG_ERROR, |
956 | "Invalid number of coefficients at level %d\n", coeff_index); |
957 | |
958 | if (eob_run > num_coeffs) { |
959 | coeff_i = |
960 | blocks_ended = num_coeffs; |
961 | eob_run -= num_coeffs; |
962 | } else { |
963 | coeff_i = |
964 | blocks_ended = eob_run; |
965 | eob_run = 0; |
966 | } |
967 | |
968 | // insert fake EOB token to cover the split between planes or zzi |
969 | if (blocks_ended) |
970 | dct_tokens[j++] = blocks_ended << 2; |
971 | |
972 | while (coeff_i < num_coeffs && get_bits_left(gb) > 0) { |
973 | /* decode a VLC into a token */ |
974 | token = get_vlc2(gb, vlc_table, 11, 3); |
975 | /* use the token to get a zero run, a coefficient, and an eob run */ |
976 | if ((unsigned) token <= 6U) { |
977 | eob_run = eob_run_base[token]; |
978 | if (eob_run_get_bits[token]) |
979 | eob_run += get_bits(gb, eob_run_get_bits[token]); |
980 | |
981 | // record only the number of blocks ended in this plane, |
982 | // any spill will be recorded in the next plane. |
983 | if (eob_run > num_coeffs - coeff_i) { |
984 | dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i); |
985 | blocks_ended += num_coeffs - coeff_i; |
986 | eob_run -= num_coeffs - coeff_i; |
987 | coeff_i = num_coeffs; |
988 | } else { |
989 | dct_tokens[j++] = TOKEN_EOB(eob_run); |
990 | blocks_ended += eob_run; |
991 | coeff_i += eob_run; |
992 | eob_run = 0; |
993 | } |
994 | } else if (token >= 0) { |
995 | bits_to_get = coeff_get_bits[token]; |
996 | if (bits_to_get) |
997 | bits_to_get = get_bits(gb, bits_to_get); |
998 | coeff = coeff_tables[token][bits_to_get]; |
999 | |
1000 | zero_run = zero_run_base[token]; |
1001 | if (zero_run_get_bits[token]) |
1002 | zero_run += get_bits(gb, zero_run_get_bits[token]); |
1003 | |
1004 | if (zero_run) { |
1005 | dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run); |
1006 | } else { |
1007 | // Save DC into the fragment structure. DC prediction is |
1008 | // done in raster order, so the actual DC can't be in with |
1009 | // other tokens. We still need the token in dct_tokens[] |
1010 | // however, or else the structure collapses on itself. |
1011 | if (!coeff_index) |
1012 | all_fragments[coded_fragment_list[coeff_i]].dc = coeff; |
1013 | |
1014 | dct_tokens[j++] = TOKEN_COEFF(coeff); |
1015 | } |
1016 | |
1017 | if (coeff_index + zero_run > 64) { |
1018 | av_log(s->avctx, AV_LOG_DEBUG, |
1019 | "Invalid zero run of %d with %d coeffs left\n", |
1020 | zero_run, 64 - coeff_index); |
1021 | zero_run = 64 - coeff_index; |
1022 | } |
1023 | |
1024 | // zero runs code multiple coefficients, |
1025 | // so don't try to decode coeffs for those higher levels |
1026 | for (i = coeff_index + 1; i <= coeff_index + zero_run; i++) |
1027 | s->num_coded_frags[plane][i]--; |
1028 | coeff_i++; |
1029 | } else { |
1030 | av_log(s->avctx, AV_LOG_ERROR, "Invalid token %d\n", token); |
1031 | return -1; |
1032 | } |
1033 | } |
1034 | |
1035 | if (blocks_ended > s->num_coded_frags[plane][coeff_index]) |
1036 | av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n"); |
1037 | |
1038 | // decrement the number of blocks that have higher coefficients for each |
1039 | // EOB run at this level |
1040 | if (blocks_ended) |
1041 | for (i = coeff_index + 1; i < 64; i++) |
1042 | s->num_coded_frags[plane][i] -= blocks_ended; |
1043 | |
1044 | // setup the next buffer |
1045 | if (plane < 2) |
1046 | s->dct_tokens[plane + 1][coeff_index] = dct_tokens + j; |
1047 | else if (coeff_index < 63) |
1048 | s->dct_tokens[0][coeff_index + 1] = dct_tokens + j; |
1049 | |
1050 | return eob_run; |
1051 | } |
1052 | |
1053 | static void reverse_dc_prediction(Vp3DecodeContext *s, |
1054 | int first_fragment, |
1055 | int fragment_width, |
1056 | int fragment_height); |
1057 | /* |
1058 | * This function unpacks all of the DCT coefficient data from the |
1059 | * bitstream. |
1060 | */ |
1061 | static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb) |
1062 | { |
1063 | int i; |
1064 | int dc_y_table; |
1065 | int dc_c_table; |
1066 | int ac_y_table; |
1067 | int ac_c_table; |
1068 | int residual_eob_run = 0; |
1069 | VLC *y_tables[64]; |
1070 | VLC *c_tables[64]; |
1071 | |
1072 | s->dct_tokens[0][0] = s->dct_tokens_base; |
1073 | |
1074 | /* fetch the DC table indexes */ |
1075 | dc_y_table = get_bits(gb, 4); |
1076 | dc_c_table = get_bits(gb, 4); |
1077 | |
1078 | /* unpack the Y plane DC coefficients */ |
1079 | residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0, |
1080 | 0, residual_eob_run); |
1081 | if (residual_eob_run < 0) |
1082 | return residual_eob_run; |
1083 | |
1084 | /* reverse prediction of the Y-plane DC coefficients */ |
1085 | reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]); |
1086 | |
1087 | /* unpack the C plane DC coefficients */ |
1088 | residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0, |
1089 | 1, residual_eob_run); |
1090 | if (residual_eob_run < 0) |
1091 | return residual_eob_run; |
1092 | residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0, |
1093 | 2, residual_eob_run); |
1094 | if (residual_eob_run < 0) |
1095 | return residual_eob_run; |
1096 | |
1097 | /* reverse prediction of the C-plane DC coefficients */ |
1098 | if (!(s->avctx->flags & AV_CODEC_FLAG_GRAY)) { |
1099 | reverse_dc_prediction(s, s->fragment_start[1], |
1100 | s->fragment_width[1], s->fragment_height[1]); |
1101 | reverse_dc_prediction(s, s->fragment_start[2], |
1102 | s->fragment_width[1], s->fragment_height[1]); |
1103 | } |
1104 | |
1105 | /* fetch the AC table indexes */ |
1106 | ac_y_table = get_bits(gb, 4); |
1107 | ac_c_table = get_bits(gb, 4); |
1108 | |
1109 | /* build tables of AC VLC tables */ |
1110 | for (i = 1; i <= 5; i++) { |
1111 | y_tables[i] = &s->ac_vlc_1[ac_y_table]; |
1112 | c_tables[i] = &s->ac_vlc_1[ac_c_table]; |
1113 | } |
1114 | for (i = 6; i <= 14; i++) { |
1115 | y_tables[i] = &s->ac_vlc_2[ac_y_table]; |
1116 | c_tables[i] = &s->ac_vlc_2[ac_c_table]; |
1117 | } |
1118 | for (i = 15; i <= 27; i++) { |
1119 | y_tables[i] = &s->ac_vlc_3[ac_y_table]; |
1120 | c_tables[i] = &s->ac_vlc_3[ac_c_table]; |
1121 | } |
1122 | for (i = 28; i <= 63; i++) { |
1123 | y_tables[i] = &s->ac_vlc_4[ac_y_table]; |
1124 | c_tables[i] = &s->ac_vlc_4[ac_c_table]; |
1125 | } |
1126 | |
1127 | /* decode all AC coefficients */ |
1128 | for (i = 1; i <= 63; i++) { |
1129 | residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i, |
1130 | 0, residual_eob_run); |
1131 | if (residual_eob_run < 0) |
1132 | return residual_eob_run; |
1133 | |
1134 | residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i, |
1135 | 1, residual_eob_run); |
1136 | if (residual_eob_run < 0) |
1137 | return residual_eob_run; |
1138 | residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i, |
1139 | 2, residual_eob_run); |
1140 | if (residual_eob_run < 0) |
1141 | return residual_eob_run; |
1142 | } |
1143 | |
1144 | return 0; |
1145 | } |
1146 | |
1147 | /* |
1148 | * This function reverses the DC prediction for each coded fragment in |
1149 | * the frame. Much of this function is adapted directly from the original |
1150 | * VP3 source code. |
1151 | */ |
1152 | #define COMPATIBLE_FRAME(x) \ |
1153 | (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type) |
1154 | #define DC_COEFF(u) s->all_fragments[u].dc |
1155 | |
1156 | static void reverse_dc_prediction(Vp3DecodeContext *s, |
1157 | int first_fragment, |
1158 | int fragment_width, |
1159 | int fragment_height) |
1160 | { |
1161 | #define PUL 8 |
1162 | #define PU 4 |
1163 | #define PUR 2 |
1164 | #define PL 1 |
1165 | |
1166 | int x, y; |
1167 | int i = first_fragment; |
1168 | |
1169 | int predicted_dc; |
1170 | |
1171 | /* DC values for the left, up-left, up, and up-right fragments */ |
1172 | int vl, vul, vu, vur; |
1173 | |
1174 | /* indexes for the left, up-left, up, and up-right fragments */ |
1175 | int l, ul, u, ur; |
1176 | |
1177 | /* |
1178 | * The 6 fields mean: |
1179 | * 0: up-left multiplier |
1180 | * 1: up multiplier |
1181 | * 2: up-right multiplier |
1182 | * 3: left multiplier |
1183 | */ |
1184 | static const int predictor_transform[16][4] = { |
1185 | { 0, 0, 0, 0 }, |
1186 | { 0, 0, 0, 128 }, // PL |
1187 | { 0, 0, 128, 0 }, // PUR |
1188 | { 0, 0, 53, 75 }, // PUR|PL |
1189 | { 0, 128, 0, 0 }, // PU |
1190 | { 0, 64, 0, 64 }, // PU |PL |
1191 | { 0, 128, 0, 0 }, // PU |PUR |
1192 | { 0, 0, 53, 75 }, // PU |PUR|PL |
1193 | { 128, 0, 0, 0 }, // PUL |
1194 | { 0, 0, 0, 128 }, // PUL|PL |
1195 | { 64, 0, 64, 0 }, // PUL|PUR |
1196 | { 0, 0, 53, 75 }, // PUL|PUR|PL |
1197 | { 0, 128, 0, 0 }, // PUL|PU |
1198 | { -104, 116, 0, 116 }, // PUL|PU |PL |
1199 | { 24, 80, 24, 0 }, // PUL|PU |PUR |
1200 | { -104, 116, 0, 116 } // PUL|PU |PUR|PL |
1201 | }; |
1202 | |
1203 | /* This table shows which types of blocks can use other blocks for |
1204 | * prediction. For example, INTRA is the only mode in this table to |
1205 | * have a frame number of 0. That means INTRA blocks can only predict |
1206 | * from other INTRA blocks. There are 2 golden frame coding types; |
1207 | * blocks encoding in these modes can only predict from other blocks |
1208 | * that were encoded with these 1 of these 2 modes. */ |
1209 | static const unsigned char compatible_frame[9] = { |
1210 | 1, /* MODE_INTER_NO_MV */ |
1211 | 0, /* MODE_INTRA */ |
1212 | 1, /* MODE_INTER_PLUS_MV */ |
1213 | 1, /* MODE_INTER_LAST_MV */ |
1214 | 1, /* MODE_INTER_PRIOR_MV */ |
1215 | 2, /* MODE_USING_GOLDEN */ |
1216 | 2, /* MODE_GOLDEN_MV */ |
1217 | 1, /* MODE_INTER_FOUR_MV */ |
1218 | 3 /* MODE_COPY */ |
1219 | }; |
1220 | int current_frame_type; |
1221 | |
1222 | /* there is a last DC predictor for each of the 3 frame types */ |
1223 | short last_dc[3]; |
1224 | |
1225 | int transform = 0; |
1226 | |
1227 | vul = |
1228 | vu = |
1229 | vur = |
1230 | vl = 0; |
1231 | last_dc[0] = |
1232 | last_dc[1] = |
1233 | last_dc[2] = 0; |
1234 | |
1235 | /* for each fragment row... */ |
1236 | for (y = 0; y < fragment_height; y++) { |
1237 | /* for each fragment in a row... */ |
1238 | for (x = 0; x < fragment_width; x++, i++) { |
1239 | |
1240 | /* reverse prediction if this block was coded */ |
1241 | if (s->all_fragments[i].coding_method != MODE_COPY) { |
1242 | current_frame_type = |
1243 | compatible_frame[s->all_fragments[i].coding_method]; |
1244 | |
1245 | transform = 0; |
1246 | if (x) { |
1247 | l = i - 1; |
1248 | vl = DC_COEFF(l); |
1249 | if (COMPATIBLE_FRAME(l)) |
1250 | transform |= PL; |
1251 | } |
1252 | if (y) { |
1253 | u = i - fragment_width; |
1254 | vu = DC_COEFF(u); |
1255 | if (COMPATIBLE_FRAME(u)) |
1256 | transform |= PU; |
1257 | if (x) { |
1258 | ul = i - fragment_width - 1; |
1259 | vul = DC_COEFF(ul); |
1260 | if (COMPATIBLE_FRAME(ul)) |
1261 | transform |= PUL; |
1262 | } |
1263 | if (x + 1 < fragment_width) { |
1264 | ur = i - fragment_width + 1; |
1265 | vur = DC_COEFF(ur); |
1266 | if (COMPATIBLE_FRAME(ur)) |
1267 | transform |= PUR; |
1268 | } |
1269 | } |
1270 | |
1271 | if (transform == 0) { |
1272 | /* if there were no fragments to predict from, use last |
1273 | * DC saved */ |
1274 | predicted_dc = last_dc[current_frame_type]; |
1275 | } else { |
1276 | /* apply the appropriate predictor transform */ |
1277 | predicted_dc = |
1278 | (predictor_transform[transform][0] * vul) + |
1279 | (predictor_transform[transform][1] * vu) + |
1280 | (predictor_transform[transform][2] * vur) + |
1281 | (predictor_transform[transform][3] * vl); |
1282 | |
1283 | predicted_dc /= 128; |
1284 | |
1285 | /* check for outranging on the [ul u l] and |
1286 | * [ul u ur l] predictors */ |
1287 | if ((transform == 15) || (transform == 13)) { |
1288 | if (FFABS(predicted_dc - vu) > 128) |
1289 | predicted_dc = vu; |
1290 | else if (FFABS(predicted_dc - vl) > 128) |
1291 | predicted_dc = vl; |
1292 | else if (FFABS(predicted_dc - vul) > 128) |
1293 | predicted_dc = vul; |
1294 | } |
1295 | } |
1296 | |
1297 | /* at long last, apply the predictor */ |
1298 | DC_COEFF(i) += predicted_dc; |
1299 | /* save the DC */ |
1300 | last_dc[current_frame_type] = DC_COEFF(i); |
1301 | } |
1302 | } |
1303 | } |
1304 | } |
1305 | |
1306 | static void apply_loop_filter(Vp3DecodeContext *s, int plane, |
1307 | int ystart, int yend) |
1308 | { |
1309 | int x, y; |
1310 | int *bounding_values = s->bounding_values_array + 127; |
1311 | |
1312 | int width = s->fragment_width[!!plane]; |
1313 | int height = s->fragment_height[!!plane]; |
1314 | int fragment = s->fragment_start[plane] + ystart * width; |
1315 | ptrdiff_t stride = s->current_frame.f->linesize[plane]; |
1316 | uint8_t *plane_data = s->current_frame.f->data[plane]; |
1317 | if (!s->flipped_image) |
1318 | stride = -stride; |
1319 | plane_data += s->data_offset[plane] + 8 * ystart * stride; |
1320 | |
1321 | for (y = ystart; y < yend; y++) { |
1322 | for (x = 0; x < width; x++) { |
1323 | /* This code basically just deblocks on the edges of coded blocks. |
1324 | * However, it has to be much more complicated because of the |
1325 | * brain damaged deblock ordering used in VP3/Theora. Order matters |
1326 | * because some pixels get filtered twice. */ |
1327 | if (s->all_fragments[fragment].coding_method != MODE_COPY) { |
1328 | /* do not perform left edge filter for left columns frags */ |
1329 | if (x > 0) { |
1330 | s->vp3dsp.h_loop_filter( |
1331 | plane_data + 8 * x, |
1332 | stride, bounding_values); |
1333 | } |
1334 | |
1335 | /* do not perform top edge filter for top row fragments */ |
1336 | if (y > 0) { |
1337 | s->vp3dsp.v_loop_filter( |
1338 | plane_data + 8 * x, |
1339 | stride, bounding_values); |
1340 | } |
1341 | |
1342 | /* do not perform right edge filter for right column |
1343 | * fragments or if right fragment neighbor is also coded |
1344 | * in this frame (it will be filtered in next iteration) */ |
1345 | if ((x < width - 1) && |
1346 | (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) { |
1347 | s->vp3dsp.h_loop_filter( |
1348 | plane_data + 8 * x + 8, |
1349 | stride, bounding_values); |
1350 | } |
1351 | |
1352 | /* do not perform bottom edge filter for bottom row |
1353 | * fragments or if bottom fragment neighbor is also coded |
1354 | * in this frame (it will be filtered in the next row) */ |
1355 | if ((y < height - 1) && |
1356 | (s->all_fragments[fragment + width].coding_method == MODE_COPY)) { |
1357 | s->vp3dsp.v_loop_filter( |
1358 | plane_data + 8 * x + 8 * stride, |
1359 | stride, bounding_values); |
1360 | } |
1361 | } |
1362 | |
1363 | fragment++; |
1364 | } |
1365 | plane_data += 8 * stride; |
1366 | } |
1367 | } |
1368 | |
1369 | /** |
1370 | * Pull DCT tokens from the 64 levels to decode and dequant the coefficients |
1371 | * for the next block in coding order |
1372 | */ |
1373 | static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag, |
1374 | int plane, int inter, int16_t block[64]) |
1375 | { |
1376 | int16_t *dequantizer = s->qmat[frag->qpi][inter][plane]; |
1377 | uint8_t *perm = s->idct_scantable; |
1378 | int i = 0; |
1379 | |
1380 | do { |
1381 | int token = *s->dct_tokens[plane][i]; |
1382 | switch (token & 3) { |
1383 | case 0: // EOB |
1384 | if (--token < 4) // 0-3 are token types so the EOB run must now be 0 |
1385 | s->dct_tokens[plane][i]++; |
1386 | else |
1387 | *s->dct_tokens[plane][i] = token & ~3; |
1388 | goto end; |
1389 | case 1: // zero run |
1390 | s->dct_tokens[plane][i]++; |
1391 | i += (token >> 2) & 0x7f; |
1392 | if (i > 63) { |
1393 | av_log(s->avctx, AV_LOG_ERROR, "Coefficient index overflow\n"); |
1394 | return i; |
1395 | } |
1396 | block[perm[i]] = (token >> 9) * dequantizer[perm[i]]; |
1397 | i++; |
1398 | break; |
1399 | case 2: // coeff |
1400 | block[perm[i]] = (token >> 2) * dequantizer[perm[i]]; |
1401 | s->dct_tokens[plane][i++]++; |
1402 | break; |
1403 | default: // shouldn't happen |
1404 | return i; |
1405 | } |
1406 | } while (i < 64); |
1407 | // return value is expected to be a valid level |
1408 | i--; |
1409 | end: |
1410 | // the actual DC+prediction is in the fragment structure |
1411 | block[0] = frag->dc * s->qmat[0][inter][plane][0]; |
1412 | return i; |
1413 | } |
1414 | |
1415 | /** |
1416 | * called when all pixels up to row y are complete |
1417 | */ |
1418 | static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y) |
1419 | { |
1420 | int h, cy, i; |
1421 | int offset[AV_NUM_DATA_POINTERS]; |
1422 | |
1423 | if (HAVE_THREADS && s->avctx->active_thread_type & FF_THREAD_FRAME) { |
1424 | int y_flipped = s->flipped_image ? s->height - y : y; |
1425 | |
1426 | /* At the end of the frame, report INT_MAX instead of the height of |
1427 | * the frame. This makes the other threads' ff_thread_await_progress() |
1428 | * calls cheaper, because they don't have to clip their values. */ |
1429 | ff_thread_report_progress(&s->current_frame, |
1430 | y_flipped == s->height ? INT_MAX |
1431 | : y_flipped - 1, |
1432 | 0); |
1433 | } |
1434 | |
1435 | if (!s->avctx->draw_horiz_band) |
1436 | return; |
1437 | |
1438 | h = y - s->last_slice_end; |
1439 | s->last_slice_end = y; |
1440 | y -= h; |
1441 | |
1442 | if (!s->flipped_image) |
1443 | y = s->height - y - h; |
1444 | |
1445 | cy = y >> s->chroma_y_shift; |
1446 | offset[0] = s->current_frame.f->linesize[0] * y; |
1447 | offset[1] = s->current_frame.f->linesize[1] * cy; |
1448 | offset[2] = s->current_frame.f->linesize[2] * cy; |
1449 | for (i = 3; i < AV_NUM_DATA_POINTERS; i++) |
1450 | offset[i] = 0; |
1451 | |
1452 | emms_c(); |
1453 | s->avctx->draw_horiz_band(s->avctx, s->current_frame.f, offset, y, 3, h); |
1454 | } |
1455 | |
1456 | /** |
1457 | * Wait for the reference frame of the current fragment. |
1458 | * The progress value is in luma pixel rows. |
1459 | */ |
1460 | static void await_reference_row(Vp3DecodeContext *s, Vp3Fragment *fragment, |
1461 | int motion_y, int y) |
1462 | { |
1463 | ThreadFrame *ref_frame; |
1464 | int ref_row; |
1465 | int border = motion_y & 1; |
1466 | |
1467 | if (fragment->coding_method == MODE_USING_GOLDEN || |
1468 | fragment->coding_method == MODE_GOLDEN_MV) |
1469 | ref_frame = &s->golden_frame; |
1470 | else |
1471 | ref_frame = &s->last_frame; |
1472 | |
1473 | ref_row = y + (motion_y >> 1); |
1474 | ref_row = FFMAX(FFABS(ref_row), ref_row + 8 + border); |
1475 | |
1476 | ff_thread_await_progress(ref_frame, ref_row, 0); |
1477 | } |
1478 | |
1479 | /* |
1480 | * Perform the final rendering for a particular slice of data. |
1481 | * The slice number ranges from 0..(c_superblock_height - 1). |
1482 | */ |
1483 | static void render_slice(Vp3DecodeContext *s, int slice) |
1484 | { |
1485 | int x, y, i, j, fragment; |
1486 | int16_t *block = s->block; |
1487 | int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef; |
1488 | int motion_halfpel_index; |
1489 | uint8_t *motion_source; |
1490 | int plane, first_pixel; |
1491 | |
1492 | if (slice >= s->c_superblock_height) |
1493 | return; |
1494 | |
1495 | for (plane = 0; plane < 3; plane++) { |
1496 | uint8_t *output_plane = s->current_frame.f->data[plane] + |
1497 | s->data_offset[plane]; |
1498 | uint8_t *last_plane = s->last_frame.f->data[plane] + |
1499 | s->data_offset[plane]; |
1500 | uint8_t *golden_plane = s->golden_frame.f->data[plane] + |
1501 | s->data_offset[plane]; |
1502 | ptrdiff_t stride = s->current_frame.f->linesize[plane]; |
1503 | int plane_width = s->width >> (plane && s->chroma_x_shift); |
1504 | int plane_height = s->height >> (plane && s->chroma_y_shift); |
1505 | int8_t(*motion_val)[2] = s->motion_val[!!plane]; |
1506 | |
1507 | int sb_x, sb_y = slice << (!plane && s->chroma_y_shift); |
1508 | int slice_height = sb_y + 1 + (!plane && s->chroma_y_shift); |
1509 | int slice_width = plane ? s->c_superblock_width |
1510 | : s->y_superblock_width; |
1511 | |
1512 | int fragment_width = s->fragment_width[!!plane]; |
1513 | int fragment_height = s->fragment_height[!!plane]; |
1514 | int fragment_start = s->fragment_start[plane]; |
1515 | |
1516 | int do_await = !plane && HAVE_THREADS && |
1517 | (s->avctx->active_thread_type & FF_THREAD_FRAME); |
1518 | |
1519 | if (!s->flipped_image) |
1520 | stride = -stride; |
1521 | if (CONFIG_GRAY && plane && (s->avctx->flags & AV_CODEC_FLAG_GRAY)) |
1522 | continue; |
1523 | |
1524 | /* for each superblock row in the slice (both of them)... */ |
1525 | for (; sb_y < slice_height; sb_y++) { |
1526 | /* for each superblock in a row... */ |
1527 | for (sb_x = 0; sb_x < slice_width; sb_x++) { |
1528 | /* for each block in a superblock... */ |
1529 | for (j = 0; j < 16; j++) { |
1530 | x = 4 * sb_x + hilbert_offset[j][0]; |
1531 | y = 4 * sb_y + hilbert_offset[j][1]; |
1532 | fragment = y * fragment_width + x; |
1533 | |
1534 | i = fragment_start + fragment; |
1535 | |
1536 | // bounds check |
1537 | if (x >= fragment_width || y >= fragment_height) |
1538 | continue; |
1539 | |
1540 | first_pixel = 8 * y * stride + 8 * x; |
1541 | |
1542 | if (do_await && |
1543 | s->all_fragments[i].coding_method != MODE_INTRA) |
1544 | await_reference_row(s, &s->all_fragments[i], |
1545 | motion_val[fragment][1], |
1546 | (16 * y) >> s->chroma_y_shift); |
1547 | |
1548 | /* transform if this block was coded */ |
1549 | if (s->all_fragments[i].coding_method != MODE_COPY) { |
1550 | if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) || |
1551 | (s->all_fragments[i].coding_method == MODE_GOLDEN_MV)) |
1552 | motion_source = golden_plane; |
1553 | else |
1554 | motion_source = last_plane; |
1555 | |
1556 | motion_source += first_pixel; |
1557 | motion_halfpel_index = 0; |
1558 | |
1559 | /* sort out the motion vector if this fragment is coded |
1560 | * using a motion vector method */ |
1561 | if ((s->all_fragments[i].coding_method > MODE_INTRA) && |
1562 | (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) { |
1563 | int src_x, src_y; |
1564 | motion_x = motion_val[fragment][0]; |
1565 | motion_y = motion_val[fragment][1]; |
1566 | |
1567 | src_x = (motion_x >> 1) + 8 * x; |
1568 | src_y = (motion_y >> 1) + 8 * y; |
1569 | |
1570 | motion_halfpel_index = motion_x & 0x01; |
1571 | motion_source += (motion_x >> 1); |
1572 | |
1573 | motion_halfpel_index |= (motion_y & 0x01) << 1; |
1574 | motion_source += ((motion_y >> 1) * stride); |
1575 | |
1576 | if (src_x < 0 || src_y < 0 || |
1577 | src_x + 9 >= plane_width || |
1578 | src_y + 9 >= plane_height) { |
1579 | uint8_t *temp = s->edge_emu_buffer; |
1580 | if (stride < 0) |
1581 | temp -= 8 * stride; |
1582 | |
1583 | s->vdsp.emulated_edge_mc(temp, motion_source, |
1584 | stride, stride, |
1585 | 9, 9, src_x, src_y, |
1586 | plane_width, |
1587 | plane_height); |
1588 | motion_source = temp; |
1589 | } |
1590 | } |
1591 | |
1592 | /* first, take care of copying a block from either the |
1593 | * previous or the golden frame */ |
1594 | if (s->all_fragments[i].coding_method != MODE_INTRA) { |
1595 | /* Note, it is possible to implement all MC cases |
1596 | * with put_no_rnd_pixels_l2 which would look more |
1597 | * like the VP3 source but this would be slower as |
1598 | * put_no_rnd_pixels_tab is better optimized */ |
1599 | if (motion_halfpel_index != 3) { |
1600 | s->hdsp.put_no_rnd_pixels_tab[1][motion_halfpel_index]( |
1601 | output_plane + first_pixel, |
1602 | motion_source, stride, 8); |
1603 | } else { |
1604 | /* d is 0 if motion_x and _y have the same sign, |
1605 | * else -1 */ |
1606 | int d = (motion_x ^ motion_y) >> 31; |
1607 | s->vp3dsp.put_no_rnd_pixels_l2(output_plane + first_pixel, |
1608 | motion_source - d, |
1609 | motion_source + stride + 1 + d, |
1610 | stride, 8); |
1611 | } |
1612 | } |
1613 | |
1614 | /* invert DCT and place (or add) in final output */ |
1615 | |
1616 | if (s->all_fragments[i].coding_method == MODE_INTRA) { |
1617 | vp3_dequant(s, s->all_fragments + i, |
1618 | plane, 0, block); |
1619 | s->vp3dsp.idct_put(output_plane + first_pixel, |
1620 | stride, |
1621 | block); |
1622 | } else { |
1623 | if (vp3_dequant(s, s->all_fragments + i, |
1624 | plane, 1, block)) { |
1625 | s->vp3dsp.idct_add(output_plane + first_pixel, |
1626 | stride, |
1627 | block); |
1628 | } else { |
1629 | s->vp3dsp.idct_dc_add(output_plane + first_pixel, |
1630 | stride, block); |
1631 | } |
1632 | } |
1633 | } else { |
1634 | /* copy directly from the previous frame */ |
1635 | s->hdsp.put_pixels_tab[1][0]( |
1636 | output_plane + first_pixel, |
1637 | last_plane + first_pixel, |
1638 | stride, 8); |
1639 | } |
1640 | } |
1641 | } |
1642 | |
1643 | // Filter up to the last row in the superblock row |
1644 | if (!s->skip_loop_filter) |
1645 | apply_loop_filter(s, plane, 4 * sb_y - !!sb_y, |
1646 | FFMIN(4 * sb_y + 3, fragment_height - 1)); |
1647 | } |
1648 | } |
1649 | |
1650 | /* this looks like a good place for slice dispatch... */ |
1651 | /* algorithm: |
1652 | * if (slice == s->macroblock_height - 1) |
1653 | * dispatch (both last slice & 2nd-to-last slice); |
1654 | * else if (slice > 0) |
1655 | * dispatch (slice - 1); |
1656 | */ |
1657 | |
1658 | vp3_draw_horiz_band(s, FFMIN((32 << s->chroma_y_shift) * (slice + 1) - 16, |
1659 | s->height - 16)); |
1660 | } |
1661 | |
1662 | /// Allocate tables for per-frame data in Vp3DecodeContext |
1663 | static av_cold int allocate_tables(AVCodecContext *avctx) |
1664 | { |
1665 | Vp3DecodeContext *s = avctx->priv_data; |
1666 | int y_fragment_count, c_fragment_count; |
1667 | |
1668 | free_tables(avctx); |
1669 | |
1670 | y_fragment_count = s->fragment_width[0] * s->fragment_height[0]; |
1671 | c_fragment_count = s->fragment_width[1] * s->fragment_height[1]; |
1672 | |
1673 | s->superblock_coding = av_mallocz(s->superblock_count); |
1674 | s->all_fragments = av_mallocz_array(s->fragment_count, sizeof(Vp3Fragment)); |
1675 | |
1676 | s->coded_fragment_list[0] = av_mallocz_array(s->fragment_count, sizeof(int)); |
1677 | |
1678 | s->dct_tokens_base = av_mallocz_array(s->fragment_count, |
1679 | 64 * sizeof(*s->dct_tokens_base)); |
1680 | s->motion_val[0] = av_mallocz_array(y_fragment_count, sizeof(*s->motion_val[0])); |
1681 | s->motion_val[1] = av_mallocz_array(c_fragment_count, sizeof(*s->motion_val[1])); |
1682 | |
1683 | /* work out the block mapping tables */ |
1684 | s->superblock_fragments = av_mallocz_array(s->superblock_count, 16 * sizeof(int)); |
1685 | s->macroblock_coding = av_mallocz(s->macroblock_count + 1); |
1686 | |
1687 | if (!s->superblock_coding || !s->all_fragments || |
1688 | !s->dct_tokens_base || !s->coded_fragment_list[0] || |
1689 | !s->superblock_fragments || !s->macroblock_coding || |
1690 | !s->motion_val[0] || !s->motion_val[1]) { |
1691 | vp3_decode_end(avctx); |
1692 | return -1; |
1693 | } |
1694 | |
1695 | init_block_mapping(s); |
1696 | |
1697 | return 0; |
1698 | } |
1699 | |
1700 | static av_cold int init_frames(Vp3DecodeContext *s) |
1701 | { |
1702 | s->current_frame.f = av_frame_alloc(); |
1703 | s->last_frame.f = av_frame_alloc(); |
1704 | s->golden_frame.f = av_frame_alloc(); |
1705 | |
1706 | if (!s->current_frame.f || !s->last_frame.f || !s->golden_frame.f) { |
1707 | av_frame_free(&s->current_frame.f); |
1708 | av_frame_free(&s->last_frame.f); |
1709 | av_frame_free(&s->golden_frame.f); |
1710 | return AVERROR(ENOMEM); |
1711 | } |
1712 | |
1713 | return 0; |
1714 | } |
1715 | |
1716 | static av_cold int vp3_decode_init(AVCodecContext *avctx) |
1717 | { |
1718 | Vp3DecodeContext *s = avctx->priv_data; |
1719 | int i, inter, plane, ret; |
1720 | int c_width; |
1721 | int c_height; |
1722 | int y_fragment_count, c_fragment_count; |
1723 | |
1724 | ret = init_frames(s); |
1725 | if (ret < 0) |
1726 | return ret; |
1727 | |
1728 | avctx->internal->allocate_progress = 1; |
1729 | |
1730 | if (avctx->codec_tag == MKTAG('V', 'P', '3', '0')) |
1731 | s->version = 0; |
1732 | else |
1733 | s->version = 1; |
1734 | |
1735 | s->avctx = avctx; |
1736 | s->width = FFALIGN(avctx->coded_width, 16); |
1737 | s->height = FFALIGN(avctx->coded_height, 16); |
1738 | if (avctx->codec_id != AV_CODEC_ID_THEORA) |
1739 | avctx->pix_fmt = AV_PIX_FMT_YUV420P; |
1740 | avctx->chroma_sample_location = AVCHROMA_LOC_CENTER; |
1741 | ff_hpeldsp_init(&s->hdsp, avctx->flags | AV_CODEC_FLAG_BITEXACT); |
1742 | ff_videodsp_init(&s->vdsp, 8); |
1743 | ff_vp3dsp_init(&s->vp3dsp, avctx->flags); |
1744 | |
1745 | for (i = 0; i < 64; i++) { |
1746 | #define TRANSPOSE(x) (((x) >> 3) | (((x) & 7) << 3)) |
1747 | s->idct_permutation[i] = TRANSPOSE(i); |
1748 | s->idct_scantable[i] = TRANSPOSE(ff_zigzag_direct[i]); |
1749 | #undef TRANSPOSE |
1750 | } |
1751 | |
1752 | /* initialize to an impossible value which will force a recalculation |
1753 | * in the first frame decode */ |
1754 | for (i = 0; i < 3; i++) |
1755 | s->qps[i] = -1; |
1756 | |
1757 | avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift); |
1758 | |
1759 | s->y_superblock_width = (s->width + 31) / 32; |
1760 | s->y_superblock_height = (s->height + 31) / 32; |
1761 | s->y_superblock_count = s->y_superblock_width * s->y_superblock_height; |
1762 | |
1763 | /* work out the dimensions for the C planes */ |
1764 | c_width = s->width >> s->chroma_x_shift; |
1765 | c_height = s->height >> s->chroma_y_shift; |
1766 | s->c_superblock_width = (c_width + 31) / 32; |
1767 | s->c_superblock_height = (c_height + 31) / 32; |
1768 | s->c_superblock_count = s->c_superblock_width * s->c_superblock_height; |
1769 | |
1770 | s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2); |
1771 | s->u_superblock_start = s->y_superblock_count; |
1772 | s->v_superblock_start = s->u_superblock_start + s->c_superblock_count; |
1773 | |
1774 | s->macroblock_width = (s->width + 15) / 16; |
1775 | s->macroblock_height = (s->height + 15) / 16; |
1776 | s->macroblock_count = s->macroblock_width * s->macroblock_height; |
1777 | |
1778 | s->fragment_width[0] = s->width / FRAGMENT_PIXELS; |
1779 | s->fragment_height[0] = s->height / FRAGMENT_PIXELS; |
1780 | s->fragment_width[1] = s->fragment_width[0] >> s->chroma_x_shift; |
1781 | s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift; |
1782 | |
1783 | /* fragment count covers all 8x8 blocks for all 3 planes */ |
1784 | y_fragment_count = s->fragment_width[0] * s->fragment_height[0]; |
1785 | c_fragment_count = s->fragment_width[1] * s->fragment_height[1]; |
1786 | s->fragment_count = y_fragment_count + 2 * c_fragment_count; |
1787 | s->fragment_start[1] = y_fragment_count; |
1788 | s->fragment_start[2] = y_fragment_count + c_fragment_count; |
1789 | |
1790 | if (!s->theora_tables) { |
1791 | for (i = 0; i < 64; i++) { |
1792 | s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i]; |
1793 | s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i]; |
1794 | s->base_matrix[0][i] = vp31_intra_y_dequant[i]; |
1795 | s->base_matrix[1][i] = vp31_intra_c_dequant[i]; |
1796 | s->base_matrix[2][i] = vp31_inter_dequant[i]; |
1797 | s->filter_limit_values[i] = vp31_filter_limit_values[i]; |
1798 | } |
1799 | |
1800 | for (inter = 0; inter < 2; inter++) { |
1801 | for (plane = 0; plane < 3; plane++) { |
1802 | s->qr_count[inter][plane] = 1; |
1803 | s->qr_size[inter][plane][0] = 63; |
1804 | s->qr_base[inter][plane][0] = |
1805 | s->qr_base[inter][plane][1] = 2 * inter + (!!plane) * !inter; |
1806 | } |
1807 | } |
1808 | |
1809 | /* init VLC tables */ |
1810 | for (i = 0; i < 16; i++) { |
1811 | /* DC histograms */ |
1812 | init_vlc(&s->dc_vlc[i], 11, 32, |
1813 | &dc_bias[i][0][1], 4, 2, |
1814 | &dc_bias[i][0][0], 4, 2, 0); |
1815 | |
1816 | /* group 1 AC histograms */ |
1817 | init_vlc(&s->ac_vlc_1[i], 11, 32, |
1818 | &ac_bias_0[i][0][1], 4, 2, |
1819 | &ac_bias_0[i][0][0], 4, 2, 0); |
1820 | |
1821 | /* group 2 AC histograms */ |
1822 | init_vlc(&s->ac_vlc_2[i], 11, 32, |
1823 | &ac_bias_1[i][0][1], 4, 2, |
1824 | &ac_bias_1[i][0][0], 4, 2, 0); |
1825 | |
1826 | /* group 3 AC histograms */ |
1827 | init_vlc(&s->ac_vlc_3[i], 11, 32, |
1828 | &ac_bias_2[i][0][1], 4, 2, |
1829 | &ac_bias_2[i][0][0], 4, 2, 0); |
1830 | |
1831 | /* group 4 AC histograms */ |
1832 | init_vlc(&s->ac_vlc_4[i], 11, 32, |
1833 | &ac_bias_3[i][0][1], 4, 2, |
1834 | &ac_bias_3[i][0][0], 4, 2, 0); |
1835 | } |
1836 | } else { |
1837 | for (i = 0; i < 16; i++) { |
1838 | /* DC histograms */ |
1839 | if (init_vlc(&s->dc_vlc[i], 11, 32, |
1840 | &s->huffman_table[i][0][1], 8, 4, |
1841 | &s->huffman_table[i][0][0], 8, 4, 0) < 0) |
1842 | goto vlc_fail; |
1843 | |
1844 | /* group 1 AC histograms */ |
1845 | if (init_vlc(&s->ac_vlc_1[i], 11, 32, |
1846 | &s->huffman_table[i + 16][0][1], 8, 4, |
1847 | &s->huffman_table[i + 16][0][0], 8, 4, 0) < 0) |
1848 | goto vlc_fail; |
1849 | |
1850 | /* group 2 AC histograms */ |
1851 | if (init_vlc(&s->ac_vlc_2[i], 11, 32, |
1852 | &s->huffman_table[i + 16 * 2][0][1], 8, 4, |
1853 | &s->huffman_table[i + 16 * 2][0][0], 8, 4, 0) < 0) |
1854 | goto vlc_fail; |
1855 | |
1856 | /* group 3 AC histograms */ |
1857 | if (init_vlc(&s->ac_vlc_3[i], 11, 32, |
1858 | &s->huffman_table[i + 16 * 3][0][1], 8, 4, |
1859 | &s->huffman_table[i + 16 * 3][0][0], 8, 4, 0) < 0) |
1860 | goto vlc_fail; |
1861 | |
1862 | /* group 4 AC histograms */ |
1863 | if (init_vlc(&s->ac_vlc_4[i], 11, 32, |
1864 | &s->huffman_table[i + 16 * 4][0][1], 8, 4, |
1865 | &s->huffman_table[i + 16 * 4][0][0], 8, 4, 0) < 0) |
1866 | goto vlc_fail; |
1867 | } |
1868 | } |
1869 | |
1870 | init_vlc(&s->superblock_run_length_vlc, 6, 34, |
1871 | &superblock_run_length_vlc_table[0][1], 4, 2, |
1872 | &superblock_run_length_vlc_table[0][0], 4, 2, 0); |
1873 | |
1874 | init_vlc(&s->fragment_run_length_vlc, 5, 30, |
1875 | &fragment_run_length_vlc_table[0][1], 4, 2, |
1876 | &fragment_run_length_vlc_table[0][0], 4, 2, 0); |
1877 | |
1878 | init_vlc(&s->mode_code_vlc, 3, 8, |
1879 | &mode_code_vlc_table[0][1], 2, 1, |
1880 | &mode_code_vlc_table[0][0], 2, 1, 0); |
1881 | |
1882 | init_vlc(&s->motion_vector_vlc, 6, 63, |
1883 | &motion_vector_vlc_table[0][1], 2, 1, |
1884 | &motion_vector_vlc_table[0][0], 2, 1, 0); |
1885 | |
1886 | return allocate_tables(avctx); |
1887 | |
1888 | vlc_fail: |
1889 | av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n"); |
1890 | return -1; |
1891 | } |
1892 | |
1893 | /// Release and shuffle frames after decode finishes |
1894 | static int update_frames(AVCodecContext *avctx) |
1895 | { |
1896 | Vp3DecodeContext *s = avctx->priv_data; |
1897 | int ret = 0; |
1898 | |
1899 | /* shuffle frames (last = current) */ |
1900 | ff_thread_release_buffer(avctx, &s->last_frame); |
1901 | ret = ff_thread_ref_frame(&s->last_frame, &s->current_frame); |
1902 | if (ret < 0) |
1903 | goto fail; |
1904 | |
1905 | if (s->keyframe) { |
1906 | ff_thread_release_buffer(avctx, &s->golden_frame); |
1907 | ret = ff_thread_ref_frame(&s->golden_frame, &s->current_frame); |
1908 | } |
1909 | |
1910 | fail: |
1911 | ff_thread_release_buffer(avctx, &s->current_frame); |
1912 | return ret; |
1913 | } |
1914 | |
1915 | static int ref_frame(Vp3DecodeContext *s, ThreadFrame *dst, ThreadFrame *src) |
1916 | { |
1917 | ff_thread_release_buffer(s->avctx, dst); |
1918 | if (src->f->data[0]) |
1919 | return ff_thread_ref_frame(dst, src); |
1920 | return 0; |
1921 | } |
1922 | |
1923 | static int ref_frames(Vp3DecodeContext *dst, Vp3DecodeContext *src) |
1924 | { |
1925 | int ret; |
1926 | if ((ret = ref_frame(dst, &dst->current_frame, &src->current_frame)) < 0 || |
1927 | (ret = ref_frame(dst, &dst->golden_frame, &src->golden_frame)) < 0 || |
1928 | (ret = ref_frame(dst, &dst->last_frame, &src->last_frame)) < 0) |
1929 | return ret; |
1930 | return 0; |
1931 | } |
1932 | |
1933 | #if HAVE_THREADS |
1934 | static int vp3_update_thread_context(AVCodecContext *dst, const AVCodecContext *src) |
1935 | { |
1936 | Vp3DecodeContext *s = dst->priv_data, *s1 = src->priv_data; |
1937 | int qps_changed = 0, i, err; |
1938 | |
1939 | #define copy_fields(to, from, start_field, end_field) \ |
1940 | memcpy(&to->start_field, &from->start_field, \ |
1941 | (char *) &to->end_field - (char *) &to->start_field) |
1942 | |
1943 | if (!s1->current_frame.f->data[0] || |
1944 | s->width != s1->width || s->height != s1->height) { |
1945 | if (s != s1) |
1946 | ref_frames(s, s1); |
1947 | return -1; |
1948 | } |
1949 | |
1950 | if (s != s1) { |
1951 | if (!s->current_frame.f) |
1952 | return AVERROR(ENOMEM); |
1953 | // init tables if the first frame hasn't been decoded |
1954 | if (!s->current_frame.f->data[0]) { |
1955 | int y_fragment_count, c_fragment_count; |
1956 | s->avctx = dst; |
1957 | err = allocate_tables(dst); |
1958 | if (err) |
1959 | return err; |
1960 | y_fragment_count = s->fragment_width[0] * s->fragment_height[0]; |
1961 | c_fragment_count = s->fragment_width[1] * s->fragment_height[1]; |
1962 | memcpy(s->motion_val[0], s1->motion_val[0], |
1963 | y_fragment_count * sizeof(*s->motion_val[0])); |
1964 | memcpy(s->motion_val[1], s1->motion_val[1], |
1965 | c_fragment_count * sizeof(*s->motion_val[1])); |
1966 | } |
1967 | |
1968 | // copy previous frame data |
1969 | if ((err = ref_frames(s, s1)) < 0) |
1970 | return err; |
1971 | |
1972 | s->keyframe = s1->keyframe; |
1973 | |
1974 | // copy qscale data if necessary |
1975 | for (i = 0; i < 3; i++) { |
1976 | if (s->qps[i] != s1->qps[1]) { |
1977 | qps_changed = 1; |
1978 | memcpy(&s->qmat[i], &s1->qmat[i], sizeof(s->qmat[i])); |
1979 | } |
1980 | } |
1981 | |
1982 | if (s->qps[0] != s1->qps[0]) |
1983 | memcpy(&s->bounding_values_array, &s1->bounding_values_array, |
1984 | sizeof(s->bounding_values_array)); |
1985 | |
1986 | if (qps_changed) |
1987 | copy_fields(s, s1, qps, superblock_count); |
1988 | #undef copy_fields |
1989 | } |
1990 | |
1991 | return update_frames(dst); |
1992 | } |
1993 | #endif |
1994 | |
1995 | static int vp3_decode_frame(AVCodecContext *avctx, |
1996 | void *data, int *got_frame, |
1997 | AVPacket *avpkt) |
1998 | { |
1999 | const uint8_t *buf = avpkt->data; |
2000 | int buf_size = avpkt->size; |
2001 | Vp3DecodeContext *s = avctx->priv_data; |
2002 | GetBitContext gb; |
2003 | int i, ret; |
2004 | |
2005 | if ((ret = init_get_bits8(&gb, buf, buf_size)) < 0) |
2006 | return ret; |
2007 | |
2008 | #if CONFIG_THEORA_DECODER |
2009 | if (s->theora && get_bits1(&gb)) { |
2010 | int type = get_bits(&gb, 7); |
2011 | skip_bits_long(&gb, 6*8); /* "theora" */ |
2012 | |
2013 | if (s->avctx->active_thread_type&FF_THREAD_FRAME) { |
2014 | av_log(avctx, AV_LOG_ERROR, "midstream reconfiguration with multithreading is unsupported, try -threads 1\n"); |
2015 | return AVERROR_PATCHWELCOME; |
2016 | } |
2017 | if (type == 0) { |
2018 | vp3_decode_end(avctx); |
2019 | ret = theora_decode_header(avctx, &gb); |
2020 | |
2021 | if (ret >= 0) |
2022 | ret = vp3_decode_init(avctx); |
2023 | if (ret < 0) { |
2024 | vp3_decode_end(avctx); |
2025 | return ret; |
2026 | } |
2027 | return buf_size; |
2028 | } else if (type == 2) { |
2029 | vp3_decode_end(avctx); |
2030 | ret = theora_decode_tables(avctx, &gb); |
2031 | if (ret >= 0) |
2032 | ret = vp3_decode_init(avctx); |
2033 | if (ret < 0) { |
2034 | vp3_decode_end(avctx); |
2035 | return ret; |
2036 | } |
2037 | return buf_size; |
2038 | } |
2039 | |
2040 | av_log(avctx, AV_LOG_ERROR, |
2041 | "Header packet passed to frame decoder, skipping\n"); |
2042 | return -1; |
2043 | } |
2044 | #endif |
2045 | |
2046 | s->keyframe = !get_bits1(&gb); |
2047 | if (!s->all_fragments) { |
2048 | av_log(avctx, AV_LOG_ERROR, "Data packet without prior valid headers\n"); |
2049 | return -1; |
2050 | } |
2051 | if (!s->theora) |
2052 | skip_bits(&gb, 1); |
2053 | for (i = 0; i < 3; i++) |
2054 | s->last_qps[i] = s->qps[i]; |
2055 | |
2056 | s->nqps = 0; |
2057 | do { |
2058 | s->qps[s->nqps++] = get_bits(&gb, 6); |
2059 | } while (s->theora >= 0x030200 && s->nqps < 3 && get_bits1(&gb)); |
2060 | for (i = s->nqps; i < 3; i++) |
2061 | s->qps[i] = -1; |
2062 | |
2063 | if (s->avctx->debug & FF_DEBUG_PICT_INFO) |
2064 | av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n", |
2065 | s->keyframe ? "key" : "", avctx->frame_number + 1, s->qps[0]); |
2066 | |
2067 | s->skip_loop_filter = !s->filter_limit_values[s->qps[0]] || |
2068 | avctx->skip_loop_filter >= (s->keyframe ? AVDISCARD_ALL |
2069 | : AVDISCARD_NONKEY); |
2070 | |
2071 | if (s->qps[0] != s->last_qps[0]) |
2072 | init_loop_filter(s); |
2073 | |
2074 | for (i = 0; i < s->nqps; i++) |
2075 | // reinit all dequantizers if the first one changed, because |
2076 | // the DC of the first quantizer must be used for all matrices |
2077 | if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0]) |
2078 | init_dequantizer(s, i); |
2079 | |
2080 | if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe) |
2081 | return buf_size; |
2082 | |
2083 | s->current_frame.f->pict_type = s->keyframe ? AV_PICTURE_TYPE_I |
2084 | : AV_PICTURE_TYPE_P; |
2085 | s->current_frame.f->key_frame = s->keyframe; |
2086 | if (ff_thread_get_buffer(avctx, &s->current_frame, AV_GET_BUFFER_FLAG_REF) < 0) |
2087 | goto error; |
2088 | |
2089 | if (!s->edge_emu_buffer) |
2090 | s->edge_emu_buffer = av_malloc(9 * FFABS(s->current_frame.f->linesize[0])); |
2091 | |
2092 | if (s->keyframe) { |
2093 | if (!s->theora) { |
2094 | skip_bits(&gb, 4); /* width code */ |
2095 | skip_bits(&gb, 4); /* height code */ |
2096 | if (s->version) { |
2097 | s->version = get_bits(&gb, 5); |
2098 | if (avctx->frame_number == 0) |
2099 | av_log(s->avctx, AV_LOG_DEBUG, |
2100 | "VP version: %d\n", s->version); |
2101 | } |
2102 | } |
2103 | if (s->version || s->theora) { |
2104 | if (get_bits1(&gb)) |
2105 | av_log(s->avctx, AV_LOG_ERROR, |
2106 | "Warning, unsupported keyframe coding type?!\n"); |
2107 | skip_bits(&gb, 2); /* reserved? */ |
2108 | } |
2109 | } else { |
2110 | if (!s->golden_frame.f->data[0]) { |
2111 | av_log(s->avctx, AV_LOG_WARNING, |
2112 | "vp3: first frame not a keyframe\n"); |
2113 | |
2114 | s->golden_frame.f->pict_type = AV_PICTURE_TYPE_I; |
2115 | if (ff_thread_get_buffer(avctx, &s->golden_frame, |
2116 | AV_GET_BUFFER_FLAG_REF) < 0) |
2117 | goto error; |
2118 | ff_thread_release_buffer(avctx, &s->last_frame); |
2119 | if ((ret = ff_thread_ref_frame(&s->last_frame, |
2120 | &s->golden_frame)) < 0) |
2121 | goto error; |
2122 | ff_thread_report_progress(&s->last_frame, INT_MAX, 0); |
2123 | } |
2124 | } |
2125 | |
2126 | memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment)); |
2127 | ff_thread_finish_setup(avctx); |
2128 | |
2129 | if (unpack_superblocks(s, &gb)) { |
2130 | av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n"); |
2131 | goto error; |
2132 | } |
2133 | if (unpack_modes(s, &gb)) { |
2134 | av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n"); |
2135 | goto error; |
2136 | } |
2137 | if (unpack_vectors(s, &gb)) { |
2138 | av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n"); |
2139 | goto error; |
2140 | } |
2141 | if (unpack_block_qpis(s, &gb)) { |
2142 | av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n"); |
2143 | goto error; |
2144 | } |
2145 | if (unpack_dct_coeffs(s, &gb)) { |
2146 | av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n"); |
2147 | goto error; |
2148 | } |
2149 | |
2150 | for (i = 0; i < 3; i++) { |
2151 | int height = s->height >> (i && s->chroma_y_shift); |
2152 | if (s->flipped_image) |
2153 | s->data_offset[i] = 0; |
2154 | else |
2155 | s->data_offset[i] = (height - 1) * s->current_frame.f->linesize[i]; |
2156 | } |
2157 | |
2158 | s->last_slice_end = 0; |
2159 | for (i = 0; i < s->c_superblock_height; i++) |
2160 | render_slice(s, i); |
2161 | |
2162 | // filter the last row |
2163 | for (i = 0; i < 3; i++) { |
2164 | int row = (s->height >> (3 + (i && s->chroma_y_shift))) - 1; |
2165 | apply_loop_filter(s, i, row, row + 1); |
2166 | } |
2167 | vp3_draw_horiz_band(s, s->height); |
2168 | |
2169 | /* output frame, offset as needed */ |
2170 | if ((ret = av_frame_ref(data, s->current_frame.f)) < 0) |
2171 | return ret; |
2172 | for (i = 0; i < 3; i++) { |
2173 | AVFrame *dst = data; |
2174 | int off = (s->offset_x >> (i && s->chroma_y_shift)) + |
2175 | (s->offset_y >> (i && s->chroma_y_shift)) * dst->linesize[i]; |
2176 | dst->data[i] += off; |
2177 | } |
2178 | *got_frame = 1; |
2179 | |
2180 | if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_FRAME)) { |
2181 | ret = update_frames(avctx); |
2182 | if (ret < 0) |
2183 | return ret; |
2184 | } |
2185 | |
2186 | return buf_size; |
2187 | |
2188 | error: |
2189 | ff_thread_report_progress(&s->current_frame, INT_MAX, 0); |
2190 | |
2191 | if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_FRAME)) |
2192 | av_frame_unref(s->current_frame.f); |
2193 | |
2194 | return -1; |
2195 | } |
2196 | |
2197 | static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb) |
2198 | { |
2199 | Vp3DecodeContext *s = avctx->priv_data; |
2200 | |
2201 | if (get_bits1(gb)) { |
2202 | int token; |
2203 | if (s->entries >= 32) { /* overflow */ |
2204 | av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n"); |
2205 | return -1; |
2206 | } |
2207 | token = get_bits(gb, 5); |
2208 | ff_dlog(avctx, "hti %d hbits %x token %d entry : %d size %d\n", |
2209 | s->hti, s->hbits, token, s->entries, s->huff_code_size); |
2210 | s->huffman_table[s->hti][token][0] = s->hbits; |
2211 | s->huffman_table[s->hti][token][1] = s->huff_code_size; |
2212 | s->entries++; |
2213 | } else { |
2214 | if (s->huff_code_size >= 32) { /* overflow */ |
2215 | av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n"); |
2216 | return -1; |
2217 | } |
2218 | s->huff_code_size++; |
2219 | s->hbits <<= 1; |
2220 | if (read_huffman_tree(avctx, gb)) |
2221 | return -1; |
2222 | s->hbits |= 1; |
2223 | if (read_huffman_tree(avctx, gb)) |
2224 | return -1; |
2225 | s->hbits >>= 1; |
2226 | s->huff_code_size--; |
2227 | } |
2228 | return 0; |
2229 | } |
2230 | |
2231 | #if HAVE_THREADS |
2232 | static int vp3_init_thread_copy(AVCodecContext *avctx) |
2233 | { |
2234 | Vp3DecodeContext *s = avctx->priv_data; |
2235 | |
2236 | s->superblock_coding = NULL; |
2237 | s->all_fragments = NULL; |
2238 | s->coded_fragment_list[0] = NULL; |
2239 | s->dct_tokens_base = NULL; |
2240 | s->superblock_fragments = NULL; |
2241 | s->macroblock_coding = NULL; |
2242 | s->motion_val[0] = NULL; |
2243 | s->motion_val[1] = NULL; |
2244 | s->edge_emu_buffer = NULL; |
2245 | |
2246 | return init_frames(s); |
2247 | } |
2248 | #endif |
2249 | |
2250 | #if CONFIG_THEORA_DECODER |
2251 | static const enum AVPixelFormat theora_pix_fmts[4] = { |
2252 | AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P |
2253 | }; |
2254 | |
2255 | static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb) |
2256 | { |
2257 | Vp3DecodeContext *s = avctx->priv_data; |
2258 | int visible_width, visible_height, colorspace; |
2259 | uint8_t offset_x = 0, offset_y = 0; |
2260 | int ret; |
2261 | AVRational fps, aspect; |
2262 | |
2263 | s->theora_header = 0; |
2264 | s->theora = get_bits_long(gb, 24); |
2265 | av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora); |
2266 | |
2267 | /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 |
2268 | * but previous versions have the image flipped relative to vp3 */ |
2269 | if (s->theora < 0x030200) { |
2270 | s->flipped_image = 1; |
2271 | av_log(avctx, AV_LOG_DEBUG, |
2272 | "Old (<alpha3) Theora bitstream, flipped image\n"); |
2273 | } |
2274 | |
2275 | visible_width = |
2276 | s->width = get_bits(gb, 16) << 4; |
2277 | visible_height = |
2278 | s->height = get_bits(gb, 16) << 4; |
2279 | |
2280 | if (s->theora >= 0x030200) { |
2281 | visible_width = get_bits_long(gb, 24); |
2282 | visible_height = get_bits_long(gb, 24); |
2283 | |
2284 | offset_x = get_bits(gb, 8); /* offset x */ |
2285 | offset_y = get_bits(gb, 8); /* offset y, from bottom */ |
2286 | } |
2287 | |
2288 | /* sanity check */ |
2289 | if (av_image_check_size(visible_width, visible_height, 0, avctx) < 0 || |
2290 | visible_width + offset_x > s->width || |
2291 | visible_height + offset_y > s->height) { |
2292 | av_log(avctx, AV_LOG_ERROR, |
2293 | "Invalid frame dimensions - w:%d h:%d x:%d y:%d (%dx%d).\n", |
2294 | visible_width, visible_height, offset_x, offset_y, |
2295 | s->width, s->height); |
2296 | return AVERROR_INVALIDDATA; |
2297 | } |
2298 | |
2299 | fps.num = get_bits_long(gb, 32); |
2300 | fps.den = get_bits_long(gb, 32); |
2301 | if (fps.num && fps.den) { |
2302 | if (fps.num < 0 || fps.den < 0) { |
2303 | av_log(avctx, AV_LOG_ERROR, "Invalid framerate\n"); |
2304 | return AVERROR_INVALIDDATA; |
2305 | } |
2306 | av_reduce(&avctx->framerate.den, &avctx->framerate.num, |
2307 | fps.den, fps.num, 1 << 30); |
2308 | } |
2309 | |
2310 | aspect.num = get_bits_long(gb, 24); |
2311 | aspect.den = get_bits_long(gb, 24); |
2312 | if (aspect.num && aspect.den) { |
2313 | av_reduce(&avctx->sample_aspect_ratio.num, |
2314 | &avctx->sample_aspect_ratio.den, |
2315 | aspect.num, aspect.den, 1 << 30); |
2316 | ff_set_sar(avctx, avctx->sample_aspect_ratio); |
2317 | } |
2318 | |
2319 | if (s->theora < 0x030200) |
2320 | skip_bits(gb, 5); /* keyframe frequency force */ |
2321 | colorspace = get_bits(gb, 8); |
2322 | skip_bits(gb, 24); /* bitrate */ |
2323 | |
2324 | skip_bits(gb, 6); /* quality hint */ |
2325 | |
2326 | if (s->theora >= 0x030200) { |
2327 | skip_bits(gb, 5); /* keyframe frequency force */ |
2328 | avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)]; |
2329 | if (avctx->pix_fmt == AV_PIX_FMT_NONE) { |
2330 | av_log(avctx, AV_LOG_ERROR, "Invalid pixel format\n"); |
2331 | return AVERROR_INVALIDDATA; |
2332 | } |
2333 | skip_bits(gb, 3); /* reserved */ |
2334 | } else |
2335 | avctx->pix_fmt = AV_PIX_FMT_YUV420P; |
2336 | |
2337 | ret = ff_set_dimensions(avctx, s->width, s->height); |
2338 | if (ret < 0) |
2339 | return ret; |
2340 | if (!(avctx->flags2 & AV_CODEC_FLAG2_IGNORE_CROP)) { |
2341 | avctx->width = visible_width; |
2342 | avctx->height = visible_height; |
2343 | // translate offsets from theora axis ([0,0] lower left) |
2344 | // to normal axis ([0,0] upper left) |
2345 | s->offset_x = offset_x; |
2346 | s->offset_y = s->height - visible_height - offset_y; |
2347 | |
2348 | if ((s->offset_x & 0x1F) && !(avctx->flags & AV_CODEC_FLAG_UNALIGNED)) { |
2349 | s->offset_x &= ~0x1F; |
2350 | if (!s->offset_x_warned) { |
2351 | s->offset_x_warned = 1; |
2352 | av_log(avctx, AV_LOG_WARNING, "Reducing offset_x from %d to %d" |
2353 | "chroma samples to preserve alignment.\n", |
2354 | offset_x, s->offset_x); |
2355 | } |
2356 | } |
2357 | } |
2358 | |
2359 | if (colorspace == 1) |
2360 | avctx->color_primaries = AVCOL_PRI_BT470M; |
2361 | else if (colorspace == 2) |
2362 | avctx->color_primaries = AVCOL_PRI_BT470BG; |
2363 | |
2364 | if (colorspace == 1 || colorspace == 2) { |
2365 | avctx->colorspace = AVCOL_SPC_BT470BG; |
2366 | avctx->color_trc = AVCOL_TRC_BT709; |
2367 | } |
2368 | |
2369 | s->theora_header = 1; |
2370 | return 0; |
2371 | } |
2372 | |
2373 | static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb) |
2374 | { |
2375 | Vp3DecodeContext *s = avctx->priv_data; |
2376 | int i, n, matrices, inter, plane; |
2377 | |
2378 | if (!s->theora_header) |
2379 | return AVERROR_INVALIDDATA; |
2380 | |
2381 | if (s->theora >= 0x030200) { |
2382 | n = get_bits(gb, 3); |
2383 | /* loop filter limit values table */ |
2384 | if (n) |
2385 | for (i = 0; i < 64; i++) |
2386 | s->filter_limit_values[i] = get_bits(gb, n); |
2387 | } |
2388 | |
2389 | if (s->theora >= 0x030200) |
2390 | n = get_bits(gb, 4) + 1; |
2391 | else |
2392 | n = 16; |
2393 | /* quality threshold table */ |
2394 | for (i = 0; i < 64; i++) |
2395 | s->coded_ac_scale_factor[i] = get_bits(gb, n); |
2396 | |
2397 | if (s->theora >= 0x030200) |
2398 | n = get_bits(gb, 4) + 1; |
2399 | else |
2400 | n = 16; |
2401 | /* dc scale factor table */ |
2402 | for (i = 0; i < 64; i++) |
2403 | s->coded_dc_scale_factor[i] = get_bits(gb, n); |
2404 | |
2405 | if (s->theora >= 0x030200) |
2406 | matrices = get_bits(gb, 9) + 1; |
2407 | else |
2408 | matrices = 3; |
2409 | |
2410 | if (matrices > 384) { |
2411 | av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n"); |
2412 | return -1; |
2413 | } |
2414 | |
2415 | for (n = 0; n < matrices; n++) |
2416 | for (i = 0; i < 64; i++) |
2417 | s->base_matrix[n][i] = get_bits(gb, 8); |
2418 | |
2419 | for (inter = 0; inter <= 1; inter++) { |
2420 | for (plane = 0; plane <= 2; plane++) { |
2421 | int newqr = 1; |
2422 | if (inter || plane > 0) |
2423 | newqr = get_bits1(gb); |
2424 | if (!newqr) { |
2425 | int qtj, plj; |
2426 | if (inter && get_bits1(gb)) { |
2427 | qtj = 0; |
2428 | plj = plane; |
2429 | } else { |
2430 | qtj = (3 * inter + plane - 1) / 3; |
2431 | plj = (plane + 2) % 3; |
2432 | } |
2433 | s->qr_count[inter][plane] = s->qr_count[qtj][plj]; |
2434 | memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], |
2435 | sizeof(s->qr_size[0][0])); |
2436 | memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], |
2437 | sizeof(s->qr_base[0][0])); |
2438 | } else { |
2439 | int qri = 0; |
2440 | int qi = 0; |
2441 | |
2442 | for (;;) { |
2443 | i = get_bits(gb, av_log2(matrices - 1) + 1); |
2444 | if (i >= matrices) { |
2445 | av_log(avctx, AV_LOG_ERROR, |
2446 | "invalid base matrix index\n"); |
2447 | return -1; |
2448 | } |
2449 | s->qr_base[inter][plane][qri] = i; |
2450 | if (qi >= 63) |
2451 | break; |
2452 | i = get_bits(gb, av_log2(63 - qi) + 1) + 1; |
2453 | s->qr_size[inter][plane][qri++] = i; |
2454 | qi += i; |
2455 | } |
2456 | |
2457 | if (qi > 63) { |
2458 | av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi); |
2459 | return -1; |
2460 | } |
2461 | s->qr_count[inter][plane] = qri; |
2462 | } |
2463 | } |
2464 | } |
2465 | |
2466 | /* Huffman tables */ |
2467 | for (s->hti = 0; s->hti < 80; s->hti++) { |
2468 | s->entries = 0; |
2469 | s->huff_code_size = 1; |
2470 | if (!get_bits1(gb)) { |
2471 | s->hbits = 0; |
2472 | if (read_huffman_tree(avctx, gb)) |
2473 | return -1; |
2474 | s->hbits = 1; |
2475 | if (read_huffman_tree(avctx, gb)) |
2476 | return -1; |
2477 | } |
2478 | } |
2479 | |
2480 | s->theora_tables = 1; |
2481 | |
2482 | return 0; |
2483 | } |
2484 | |
2485 | static av_cold int theora_decode_init(AVCodecContext *avctx) |
2486 | { |
2487 | Vp3DecodeContext *s = avctx->priv_data; |
2488 | GetBitContext gb; |
2489 | int ptype; |
2490 | const uint8_t *header_start[3]; |
2491 | int header_len[3]; |
2492 | int i; |
2493 | int ret; |
2494 | |
2495 | avctx->pix_fmt = AV_PIX_FMT_YUV420P; |
2496 | |
2497 | s->theora = 1; |
2498 | |
2499 | if (!avctx->extradata_size) { |
2500 | av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n"); |
2501 | return -1; |
2502 | } |
2503 | |
2504 | if (avpriv_split_xiph_headers(avctx->extradata, avctx->extradata_size, |
2505 | 42, header_start, header_len) < 0) { |
2506 | av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n"); |
2507 | return -1; |
2508 | } |
2509 | |
2510 | for (i = 0; i < 3; i++) { |
2511 | if (header_len[i] <= 0) |
2512 | continue; |
2513 | ret = init_get_bits8(&gb, header_start[i], header_len[i]); |
2514 | if (ret < 0) |
2515 | return ret; |
2516 | |
2517 | ptype = get_bits(&gb, 8); |
2518 | |
2519 | if (!(ptype & 0x80)) { |
2520 | av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n"); |
2521 | // return -1; |
2522 | } |
2523 | |
2524 | // FIXME: Check for this as well. |
2525 | skip_bits_long(&gb, 6 * 8); /* "theora" */ |
2526 | |
2527 | switch (ptype) { |
2528 | case 0x80: |
2529 | if (theora_decode_header(avctx, &gb) < 0) |
2530 | return -1; |
2531 | break; |
2532 | case 0x81: |
2533 | // FIXME: is this needed? it breaks sometimes |
2534 | // theora_decode_comments(avctx, gb); |
2535 | break; |
2536 | case 0x82: |
2537 | if (theora_decode_tables(avctx, &gb)) |
2538 | return -1; |
2539 | break; |
2540 | default: |
2541 | av_log(avctx, AV_LOG_ERROR, |
2542 | "Unknown Theora config packet: %d\n", ptype & ~0x80); |
2543 | break; |
2544 | } |
2545 | if (ptype != 0x81 && 8 * header_len[i] != get_bits_count(&gb)) |
2546 | av_log(avctx, AV_LOG_WARNING, |
2547 | "%d bits left in packet %X\n", |
2548 | 8 * header_len[i] - get_bits_count(&gb), ptype); |
2549 | if (s->theora < 0x030200) |
2550 | break; |
2551 | } |
2552 | |
2553 | return vp3_decode_init(avctx); |
2554 | } |
2555 | |
2556 | AVCodec ff_theora_decoder = { |
2557 | .name = "theora", |
2558 | .long_name = NULL_IF_CONFIG_SMALL("Theora"), |
2559 | .type = AVMEDIA_TYPE_VIDEO, |
2560 | .id = AV_CODEC_ID_THEORA, |
2561 | .priv_data_size = sizeof(Vp3DecodeContext), |
2562 | .init = theora_decode_init, |
2563 | .close = vp3_decode_end, |
2564 | .decode = vp3_decode_frame, |
2565 | .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DRAW_HORIZ_BAND | |
2566 | AV_CODEC_CAP_FRAME_THREADS, |
2567 | .flush = vp3_decode_flush, |
2568 | .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy), |
2569 | .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context) |
2570 | }; |
2571 | #endif |
2572 | |
2573 | AVCodec ff_vp3_decoder = { |
2574 | .name = "vp3", |
2575 | .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"), |
2576 | .type = AVMEDIA_TYPE_VIDEO, |
2577 | .id = AV_CODEC_ID_VP3, |
2578 | .priv_data_size = sizeof(Vp3DecodeContext), |
2579 | .init = vp3_decode_init, |
2580 | .close = vp3_decode_end, |
2581 | .decode = vp3_decode_frame, |
2582 | .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DRAW_HORIZ_BAND | |
2583 | AV_CODEC_CAP_FRAME_THREADS, |
2584 | .flush = vp3_decode_flush, |
2585 | .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy), |
2586 | .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context), |
2587 | }; |
2588 |