2 * Copyright (C) 2003-2004 the ffmpeg project
4 * This file is part of FFmpeg.
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.
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.
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
22 * @file libavcodec/vp3.c
23 * On2 VP3 Video Decoder
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
29 * Theora decoder by Alex Beregszaszi
43 #define FRAGMENT_PIXELS 8
45 static av_cold int vp3_decode_end(AVCodecContext *avctx);
47 typedef struct Coeff {
53 //FIXME split things out into their own arrays
54 typedef struct Vp3Fragment {
56 /* address of first pixel taking into account which plane the fragment
57 * lives on as well as the plane stride */
59 /* this is the macroblock that the fragment belongs to */
61 uint8_t coding_method;
67 #define SB_NOT_CODED 0
68 #define SB_PARTIALLY_CODED 1
69 #define SB_FULLY_CODED 2
71 #define MODE_INTER_NO_MV 0
73 #define MODE_INTER_PLUS_MV 2
74 #define MODE_INTER_LAST_MV 3
75 #define MODE_INTER_PRIOR_LAST 4
76 #define MODE_USING_GOLDEN 5
77 #define MODE_GOLDEN_MV 6
78 #define MODE_INTER_FOURMV 7
79 #define CODING_MODE_COUNT 8
81 /* special internal mode */
84 /* There are 6 preset schemes, plus a free-form scheme */
85 static const int ModeAlphabet[6][CODING_MODE_COUNT] =
87 /* scheme 1: Last motion vector dominates */
88 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
89 MODE_INTER_PLUS_MV, MODE_INTER_NO_MV,
90 MODE_INTRA, MODE_USING_GOLDEN,
91 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
94 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
95 MODE_INTER_NO_MV, MODE_INTER_PLUS_MV,
96 MODE_INTRA, MODE_USING_GOLDEN,
97 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
100 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
101 MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
102 MODE_INTRA, MODE_USING_GOLDEN,
103 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
106 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
107 MODE_INTER_NO_MV, MODE_INTER_PRIOR_LAST,
108 MODE_INTRA, MODE_USING_GOLDEN,
109 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
111 /* scheme 5: No motion vector dominates */
112 { MODE_INTER_NO_MV, MODE_INTER_LAST_MV,
113 MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
114 MODE_INTRA, MODE_USING_GOLDEN,
115 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
118 { MODE_INTER_NO_MV, MODE_USING_GOLDEN,
119 MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
120 MODE_INTER_PLUS_MV, MODE_INTRA,
121 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
125 #define MIN_DEQUANT_VAL 2
127 typedef struct Vp3DecodeContext {
128 AVCodecContext *avctx;
129 int theora, theora_tables;
132 AVFrame golden_frame;
134 AVFrame current_frame;
143 int superblock_count;
144 int y_superblock_width;
145 int y_superblock_height;
146 int c_superblock_width;
147 int c_superblock_height;
148 int u_superblock_start;
149 int v_superblock_start;
150 unsigned char *superblock_coding;
152 int macroblock_count;
153 int macroblock_width;
154 int macroblock_height;
160 Vp3Fragment *all_fragments;
161 uint8_t *coeff_counts;
164 int fragment_start[3];
169 uint16_t coded_dc_scale_factor[64];
170 uint32_t coded_ac_scale_factor[64];
171 uint8_t base_matrix[384][64];
172 uint8_t qr_count[2][3];
173 uint8_t qr_size [2][3][64];
174 uint16_t qr_base[2][3][64];
176 /* this is a list of indexes into the all_fragments array indicating
177 * which of the fragments are coded */
178 int *coded_fragment_list;
179 int coded_fragment_list_index;
180 int pixel_addresses_initialized;
182 /* track which fragments have already been decoded; called 'fast'
183 * because this data structure avoids having to iterate through every
184 * fragment in coded_fragment_list; once a fragment has been fully
185 * decoded, it is removed from this list */
186 int *fast_fragment_list;
187 int fragment_list_y_head;
188 int fragment_list_c_head;
196 VLC superblock_run_length_vlc;
197 VLC fragment_run_length_vlc;
199 VLC motion_vector_vlc;
201 /* these arrays need to be on 16-byte boundaries since SSE2 operations
203 DECLARE_ALIGNED_16(int16_t, qmat[3][2][3][64]); //<qmat[qpi][is_inter][plane]
205 /* This table contains superblock_count * 16 entries. Each set of 16
206 * numbers corresponds to the fragment indexes 0..15 of the superblock.
207 * An entry will be -1 to indicate that no entry corresponds to that
209 int *superblock_fragments;
211 /* This table contains superblock_count * 4 entries. Each set of 4
212 * numbers corresponds to the macroblock indexes 0..3 of the superblock.
213 * An entry will be -1 to indicate that no entry corresponds to that
215 int *superblock_macroblocks;
217 /* This table contains macroblock_count * 6 entries. Each set of 6
218 * numbers corresponds to the fragment indexes 0..5 which comprise
219 * the macroblock (4 Y fragments and 2 C fragments). */
220 int *macroblock_fragments;
221 /* This is an array that indicates how a particular macroblock
223 unsigned char *macroblock_coding;
225 int first_coded_y_fragment;
226 int first_coded_c_fragment;
227 int last_coded_y_fragment;
228 int last_coded_c_fragment;
230 uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
231 int8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
238 uint16_t huffman_table[80][32][2];
240 uint8_t filter_limit_values[64];
241 DECLARE_ALIGNED_8(int, bounding_values_array[256+2]);
244 /************************************************************************
245 * VP3 specific functions
246 ************************************************************************/
249 * This function sets up all of the various blocks mappings:
250 * superblocks <-> fragments, macroblocks <-> fragments,
251 * superblocks <-> macroblocks
253 * Returns 0 is successful; returns 1 if *anything* went wrong.
255 static int init_block_mapping(Vp3DecodeContext *s)
258 signed int hilbert_walk_mb[4];
260 int current_fragment = 0;
261 int current_width = 0;
262 int current_height = 0;
265 int superblock_row_inc = 0;
266 int mapping_index = 0;
268 int current_macroblock;
271 static const signed char travel_width[16] = {
278 static const signed char travel_height[16] = {
285 static const signed char travel_width_mb[4] = {
289 static const signed char travel_height_mb[4] = {
293 hilbert_walk_mb[0] = 1;
294 hilbert_walk_mb[1] = s->macroblock_width;
295 hilbert_walk_mb[2] = 1;
296 hilbert_walk_mb[3] = -s->macroblock_width;
298 /* iterate through each superblock (all planes) and map the fragments */
299 for (i = 0; i < s->superblock_count; i++) {
300 /* time to re-assign the limits? */
303 /* start of Y superblocks */
304 right_edge = s->fragment_width;
305 bottom_edge = s->fragment_height;
308 superblock_row_inc = 3 * s->fragment_width -
309 (s->y_superblock_width * 4 - s->fragment_width);
311 /* the first operation for this variable is to advance by 1 */
312 current_fragment = -1;
314 } else if (i == s->u_superblock_start) {
316 /* start of U superblocks */
317 right_edge = s->fragment_width / 2;
318 bottom_edge = s->fragment_height / 2;
321 superblock_row_inc = 3 * (s->fragment_width / 2) -
322 (s->c_superblock_width * 4 - s->fragment_width / 2);
324 /* the first operation for this variable is to advance by 1 */
325 current_fragment = s->fragment_start[1] - 1;
327 } else if (i == s->v_superblock_start) {
329 /* start of V superblocks */
330 right_edge = s->fragment_width / 2;
331 bottom_edge = s->fragment_height / 2;
334 superblock_row_inc = 3 * (s->fragment_width / 2) -
335 (s->c_superblock_width * 4 - s->fragment_width / 2);
337 /* the first operation for this variable is to advance by 1 */
338 current_fragment = s->fragment_start[2] - 1;
342 if (current_width >= right_edge - 1) {
343 /* reset width and move to next superblock row */
347 /* fragment is now at the start of a new superblock row */
348 current_fragment += superblock_row_inc;
351 /* iterate through all 16 fragments in a superblock */
352 for (j = 0; j < 16; j++) {
353 current_fragment += travel_width[j] + right_edge * travel_height[j];
354 current_width += travel_width[j];
355 current_height += travel_height[j];
357 /* check if the fragment is in bounds */
358 if ((current_width < right_edge) &&
359 (current_height < bottom_edge)) {
360 s->superblock_fragments[mapping_index] = current_fragment;
362 s->superblock_fragments[mapping_index] = -1;
369 /* initialize the superblock <-> macroblock mapping; iterate through
370 * all of the Y plane superblocks to build this mapping */
371 right_edge = s->macroblock_width;
372 bottom_edge = s->macroblock_height;
375 superblock_row_inc = s->macroblock_width -
376 (s->y_superblock_width * 2 - s->macroblock_width);
378 current_macroblock = -1;
379 for (i = 0; i < s->u_superblock_start; i++) {
381 if (current_width >= right_edge - 1) {
382 /* reset width and move to next superblock row */
386 /* macroblock is now at the start of a new superblock row */
387 current_macroblock += superblock_row_inc;
390 /* iterate through each potential macroblock in the superblock */
391 for (j = 0; j < 4; j++) {
392 current_macroblock += hilbert_walk_mb[j];
393 current_width += travel_width_mb[j];
394 current_height += travel_height_mb[j];
396 /* check if the macroblock is in bounds */
397 if ((current_width < right_edge) &&
398 (current_height < bottom_edge)) {
399 s->superblock_macroblocks[mapping_index] = current_macroblock;
401 s->superblock_macroblocks[mapping_index] = -1;
408 /* initialize the macroblock <-> fragment mapping */
409 current_fragment = 0;
410 current_macroblock = 0;
412 for (i = 0; i < s->fragment_height; i += 2) {
414 for (j = 0; j < s->fragment_width; j += 2) {
416 s->all_fragments[current_fragment].macroblock = current_macroblock;
417 s->macroblock_fragments[mapping_index++] = current_fragment;
419 if (j + 1 < s->fragment_width) {
420 s->all_fragments[current_fragment + 1].macroblock = current_macroblock;
421 s->macroblock_fragments[mapping_index++] = current_fragment + 1;
423 s->macroblock_fragments[mapping_index++] = -1;
425 if (i + 1 < s->fragment_height) {
426 s->all_fragments[current_fragment + s->fragment_width].macroblock =
428 s->macroblock_fragments[mapping_index++] =
429 current_fragment + s->fragment_width;
431 s->macroblock_fragments[mapping_index++] = -1;
433 if ((j + 1 < s->fragment_width) && (i + 1 < s->fragment_height)) {
434 s->all_fragments[current_fragment + s->fragment_width + 1].macroblock =
436 s->macroblock_fragments[mapping_index++] =
437 current_fragment + s->fragment_width + 1;
439 s->macroblock_fragments[mapping_index++] = -1;
442 c_fragment = s->fragment_start[1] +
443 (i * s->fragment_width / 4) + (j / 2);
444 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
445 s->macroblock_fragments[mapping_index++] = c_fragment;
447 c_fragment = s->fragment_start[2] +
448 (i * s->fragment_width / 4) + (j / 2);
449 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
450 s->macroblock_fragments[mapping_index++] = c_fragment;
452 if (j + 2 <= s->fragment_width)
453 current_fragment += 2;
456 current_macroblock++;
459 current_fragment += s->fragment_width;
462 return 0; /* successful path out */
466 * This function wipes out all of the fragment data.
468 static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
472 /* zero out all of the fragment information */
473 s->coded_fragment_list_index = 0;
474 for (i = 0; i < s->fragment_count; i++) {
475 s->coeff_counts[i] = 0;
476 s->all_fragments[i].motion_x = 127;
477 s->all_fragments[i].motion_y = 127;
478 s->all_fragments[i].next_coeff= NULL;
479 s->all_fragments[i].qpi = 0;
481 s->coeffs[i].coeff=0;
482 s->coeffs[i].next= NULL;
487 * This function sets up the dequantization tables used for a particular
490 static void init_dequantizer(Vp3DecodeContext *s, int qpi)
492 int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
493 int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
494 int i, plane, inter, qri, bmi, bmj, qistart;
496 for(inter=0; inter<2; inter++){
497 for(plane=0; plane<3; plane++){
499 for(qri=0; qri<s->qr_count[inter][plane]; qri++){
500 sum+= s->qr_size[inter][plane][qri];
501 if(s->qps[qpi] <= sum)
504 qistart= sum - s->qr_size[inter][plane][qri];
505 bmi= s->qr_base[inter][plane][qri ];
506 bmj= s->qr_base[inter][plane][qri+1];
508 int coeff= ( 2*(sum -s->qps[qpi])*s->base_matrix[bmi][i]
509 - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
510 + s->qr_size[inter][plane][qri])
511 / (2*s->qr_size[inter][plane][qri]);
513 int qmin= 8<<(inter + !i);
514 int qscale= i ? ac_scale_factor : dc_scale_factor;
516 s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
518 // all DC coefficients use the same quant so as not to interfere with DC prediction
519 s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
523 memset(s->qscale_table, (FFMAX(s->qmat[0][0][0][1], s->qmat[0][0][1][1])+8)/16, 512); //FIXME finetune
527 * This function initializes the loop filter boundary limits if the frame's
528 * quality index is different from the previous frame's.
530 * The filter_limit_values may not be larger than 127.
532 static void init_loop_filter(Vp3DecodeContext *s)
534 int *bounding_values= s->bounding_values_array+127;
539 filter_limit = s->filter_limit_values[s->qps[0]];
541 /* set up the bounding values */
542 memset(s->bounding_values_array, 0, 256 * sizeof(int));
543 for (x = 0; x < filter_limit; x++) {
544 bounding_values[-x] = -x;
545 bounding_values[x] = x;
547 for (x = value = filter_limit; x < 128 && value; x++, value--) {
548 bounding_values[ x] = value;
549 bounding_values[-x] = -value;
552 bounding_values[128] = value;
553 bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
557 * This function unpacks all of the superblock/macroblock/fragment coding
558 * information from the bitstream.
560 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
563 int current_superblock = 0;
565 int decode_fully_flags = 0;
566 int decode_partial_blocks = 0;
567 int first_c_fragment_seen;
570 int current_fragment;
573 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
577 /* unpack the list of partially-coded superblocks */
579 /* toggle the bit because as soon as the first run length is
580 * fetched the bit will be toggled again */
582 while (current_superblock < s->superblock_count) {
583 if (current_run-- == 0) {
585 current_run = get_vlc2(gb,
586 s->superblock_run_length_vlc.table, 6, 2);
587 if (current_run == 33)
588 current_run += get_bits(gb, 12);
590 /* if any of the superblocks are not partially coded, flag
591 * a boolean to decode the list of fully-coded superblocks */
593 decode_fully_flags = 1;
596 /* make a note of the fact that there are partially coded
598 decode_partial_blocks = 1;
601 s->superblock_coding[current_superblock++] = bit;
604 /* unpack the list of fully coded superblocks if any of the blocks were
605 * not marked as partially coded in the previous step */
606 if (decode_fully_flags) {
608 current_superblock = 0;
611 /* toggle the bit because as soon as the first run length is
612 * fetched the bit will be toggled again */
614 while (current_superblock < s->superblock_count) {
616 /* skip any superblocks already marked as partially coded */
617 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
619 if (current_run-- == 0) {
621 current_run = get_vlc2(gb,
622 s->superblock_run_length_vlc.table, 6, 2);
623 if (current_run == 33)
624 current_run += get_bits(gb, 12);
626 s->superblock_coding[current_superblock] = 2*bit;
628 current_superblock++;
632 /* if there were partial blocks, initialize bitstream for
633 * unpacking fragment codings */
634 if (decode_partial_blocks) {
638 /* toggle the bit because as soon as the first run length is
639 * fetched the bit will be toggled again */
644 /* figure out which fragments are coded; iterate through each
645 * superblock (all planes) */
646 s->coded_fragment_list_index = 0;
647 s->next_coeff= s->coeffs + s->fragment_count;
648 s->first_coded_y_fragment = s->first_coded_c_fragment = 0;
649 s->last_coded_y_fragment = s->last_coded_c_fragment = -1;
650 first_c_fragment_seen = 0;
651 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
652 for (i = 0; i < s->superblock_count; i++) {
654 /* iterate through all 16 fragments in a superblock */
655 for (j = 0; j < 16; j++) {
657 /* if the fragment is in bounds, check its coding status */
658 current_fragment = s->superblock_fragments[i * 16 + j];
659 if (current_fragment >= s->fragment_count) {
660 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
661 current_fragment, s->fragment_count);
664 if (current_fragment != -1) {
665 if (s->superblock_coding[i] == SB_NOT_CODED) {
667 /* copy all the fragments from the prior frame */
668 s->all_fragments[current_fragment].coding_method =
671 } else if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
673 /* fragment may or may not be coded; this is the case
674 * that cares about the fragment coding runs */
675 if (current_run-- == 0) {
677 current_run = get_vlc2(gb,
678 s->fragment_run_length_vlc.table, 5, 2);
682 /* default mode; actual mode will be decoded in
684 s->all_fragments[current_fragment].coding_method =
686 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
687 s->coded_fragment_list[s->coded_fragment_list_index] =
689 if ((current_fragment >= s->fragment_start[1]) &&
690 (s->last_coded_y_fragment == -1) &&
691 (!first_c_fragment_seen)) {
692 s->first_coded_c_fragment = s->coded_fragment_list_index;
693 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
694 first_c_fragment_seen = 1;
696 s->coded_fragment_list_index++;
697 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
699 /* not coded; copy this fragment from the prior frame */
700 s->all_fragments[current_fragment].coding_method =
706 /* fragments are fully coded in this superblock; actual
707 * coding will be determined in next step */
708 s->all_fragments[current_fragment].coding_method =
710 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
711 s->coded_fragment_list[s->coded_fragment_list_index] =
713 if ((current_fragment >= s->fragment_start[1]) &&
714 (s->last_coded_y_fragment == -1) &&
715 (!first_c_fragment_seen)) {
716 s->first_coded_c_fragment = s->coded_fragment_list_index;
717 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
718 first_c_fragment_seen = 1;
720 s->coded_fragment_list_index++;
721 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
727 if (!first_c_fragment_seen)
728 /* only Y fragments coded in this frame */
729 s->last_coded_y_fragment = s->coded_fragment_list_index - 1;
731 /* end the list of coded C fragments */
732 s->last_coded_c_fragment = s->coded_fragment_list_index - 1;
734 for (i = 0; i < s->fragment_count - 1; i++) {
735 s->fast_fragment_list[i] = i + 1;
737 s->fast_fragment_list[s->fragment_count - 1] = -1;
739 if (s->last_coded_y_fragment == -1)
740 s->fragment_list_y_head = -1;
742 s->fragment_list_y_head = s->first_coded_y_fragment;
743 s->fast_fragment_list[s->last_coded_y_fragment] = -1;
746 if (s->last_coded_c_fragment == -1)
747 s->fragment_list_c_head = -1;
749 s->fragment_list_c_head = s->first_coded_c_fragment;
750 s->fast_fragment_list[s->last_coded_c_fragment] = -1;
757 * This function unpacks all the coding mode data for individual macroblocks
758 * from the bitstream.
760 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
764 int current_macroblock;
765 int current_fragment;
767 int custom_mode_alphabet[CODING_MODE_COUNT];
770 for (i = 0; i < s->fragment_count; i++)
771 s->all_fragments[i].coding_method = MODE_INTRA;
775 /* fetch the mode coding scheme for this frame */
776 scheme = get_bits(gb, 3);
778 /* is it a custom coding scheme? */
780 for (i = 0; i < 8; i++)
781 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
782 for (i = 0; i < 8; i++)
783 custom_mode_alphabet[get_bits(gb, 3)] = i;
786 /* iterate through all of the macroblocks that contain 1 or more
788 for (i = 0; i < s->u_superblock_start; i++) {
790 for (j = 0; j < 4; j++) {
791 current_macroblock = s->superblock_macroblocks[i * 4 + j];
792 if ((current_macroblock == -1) ||
793 (s->macroblock_coding[current_macroblock] == MODE_COPY))
795 if (current_macroblock >= s->macroblock_count) {
796 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad macroblock number (%d >= %d)\n",
797 current_macroblock, s->macroblock_count);
801 /* mode 7 means get 3 bits for each coding mode */
803 coding_mode = get_bits(gb, 3);
805 coding_mode = custom_mode_alphabet
806 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
808 coding_mode = ModeAlphabet[scheme-1]
809 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
811 s->macroblock_coding[current_macroblock] = coding_mode;
812 for (k = 0; k < 6; k++) {
814 s->macroblock_fragments[current_macroblock * 6 + k];
815 if (current_fragment == -1)
817 if (current_fragment >= s->fragment_count) {
818 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad fragment number (%d >= %d)\n",
819 current_fragment, s->fragment_count);
822 if (s->all_fragments[current_fragment].coding_method !=
824 s->all_fragments[current_fragment].coding_method =
835 * This function unpacks all the motion vectors for the individual
836 * macroblocks from the bitstream.
838 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
844 int last_motion_x = 0;
845 int last_motion_y = 0;
846 int prior_last_motion_x = 0;
847 int prior_last_motion_y = 0;
848 int current_macroblock;
849 int current_fragment;
854 memset(motion_x, 0, 6 * sizeof(int));
855 memset(motion_y, 0, 6 * sizeof(int));
857 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
858 coding_mode = get_bits1(gb);
860 /* iterate through all of the macroblocks that contain 1 or more
862 for (i = 0; i < s->u_superblock_start; i++) {
864 for (j = 0; j < 4; j++) {
865 current_macroblock = s->superblock_macroblocks[i * 4 + j];
866 if ((current_macroblock == -1) ||
867 (s->macroblock_coding[current_macroblock] == MODE_COPY))
869 if (current_macroblock >= s->macroblock_count) {
870 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad macroblock number (%d >= %d)\n",
871 current_macroblock, s->macroblock_count);
875 current_fragment = s->macroblock_fragments[current_macroblock * 6];
876 if (current_fragment >= s->fragment_count) {
877 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d\n",
878 current_fragment, s->fragment_count);
881 switch (s->macroblock_coding[current_macroblock]) {
883 case MODE_INTER_PLUS_MV:
885 /* all 6 fragments use the same motion vector */
886 if (coding_mode == 0) {
887 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
888 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
890 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
891 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
894 for (k = 1; k < 6; k++) {
895 motion_x[k] = motion_x[0];
896 motion_y[k] = motion_y[0];
899 /* vector maintenance, only on MODE_INTER_PLUS_MV */
900 if (s->macroblock_coding[current_macroblock] ==
901 MODE_INTER_PLUS_MV) {
902 prior_last_motion_x = last_motion_x;
903 prior_last_motion_y = last_motion_y;
904 last_motion_x = motion_x[0];
905 last_motion_y = motion_y[0];
909 case MODE_INTER_FOURMV:
910 /* vector maintenance */
911 prior_last_motion_x = last_motion_x;
912 prior_last_motion_y = last_motion_y;
914 /* fetch 4 vectors from the bitstream, one for each
915 * Y fragment, then average for the C fragment vectors */
916 motion_x[4] = motion_y[4] = 0;
917 for (k = 0; k < 4; k++) {
918 for (l = 0; l < s->coded_fragment_list_index; l++)
919 if (s->coded_fragment_list[l] == s->macroblock_fragments[6*current_macroblock + k])
921 if (l < s->coded_fragment_list_index) {
922 if (coding_mode == 0) {
923 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
924 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
926 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
927 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
929 last_motion_x = motion_x[k];
930 last_motion_y = motion_y[k];
935 motion_x[4] += motion_x[k];
936 motion_y[4] += motion_y[k];
940 motion_x[4]= RSHIFT(motion_x[4], 2);
942 motion_y[4]= RSHIFT(motion_y[4], 2);
945 case MODE_INTER_LAST_MV:
946 /* all 6 fragments use the last motion vector */
947 motion_x[0] = last_motion_x;
948 motion_y[0] = last_motion_y;
949 for (k = 1; k < 6; k++) {
950 motion_x[k] = motion_x[0];
951 motion_y[k] = motion_y[0];
954 /* no vector maintenance (last vector remains the
958 case MODE_INTER_PRIOR_LAST:
959 /* all 6 fragments use the motion vector prior to the
960 * last motion vector */
961 motion_x[0] = prior_last_motion_x;
962 motion_y[0] = prior_last_motion_y;
963 for (k = 1; k < 6; k++) {
964 motion_x[k] = motion_x[0];
965 motion_y[k] = motion_y[0];
968 /* vector maintenance */
969 prior_last_motion_x = last_motion_x;
970 prior_last_motion_y = last_motion_y;
971 last_motion_x = motion_x[0];
972 last_motion_y = motion_y[0];
976 /* covers intra, inter without MV, golden without MV */
977 memset(motion_x, 0, 6 * sizeof(int));
978 memset(motion_y, 0, 6 * sizeof(int));
980 /* no vector maintenance */
984 /* assign the motion vectors to the correct fragments */
985 for (k = 0; k < 6; k++) {
987 s->macroblock_fragments[current_macroblock * 6 + k];
988 if (current_fragment == -1)
990 if (current_fragment >= s->fragment_count) {
991 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d)\n",
992 current_fragment, s->fragment_count);
995 s->all_fragments[current_fragment].motion_x = motion_x[k];
996 s->all_fragments[current_fragment].motion_y = motion_y[k];
1004 static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
1006 int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
1007 int num_blocks = s->coded_fragment_list_index;
1009 for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
1010 i = blocks_decoded = num_blocks_at_qpi = 0;
1012 bit = get_bits1(gb);
1015 run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
1016 if (run_length == 34)
1017 run_length += get_bits(gb, 12);
1018 blocks_decoded += run_length;
1021 num_blocks_at_qpi += run_length;
1023 for (j = 0; j < run_length; i++) {
1024 if (i >= s->coded_fragment_list_index)
1027 if (s->all_fragments[s->coded_fragment_list[i]].qpi == qpi) {
1028 s->all_fragments[s->coded_fragment_list[i]].qpi += bit;
1033 if (run_length == 4129)
1034 bit = get_bits1(gb);
1037 } while (blocks_decoded < num_blocks);
1039 num_blocks -= num_blocks_at_qpi;
1046 * This function is called by unpack_dct_coeffs() to extract the VLCs from
1047 * the bitstream. The VLCs encode tokens which are used to unpack DCT
1048 * data. This function unpacks all the VLCs for either the Y plane or both
1049 * C planes, and is called for DC coefficients or different AC coefficient
1050 * levels (since different coefficient types require different VLC tables.
1052 * This function returns a residual eob run. E.g, if a particular token gave
1053 * instructions to EOB the next 5 fragments and there were only 2 fragments
1054 * left in the current fragment range, 3 would be returned so that it could
1055 * be passed into the next call to this same function.
1057 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
1058 VLC *table, int coeff_index,
1066 Vp3Fragment *fragment;
1069 int previous_fragment;
1073 /* local references to structure members to avoid repeated deferences */
1074 uint8_t *perm= s->scantable.permutated;
1075 int *coded_fragment_list = s->coded_fragment_list;
1076 Vp3Fragment *all_fragments = s->all_fragments;
1077 uint8_t *coeff_counts = s->coeff_counts;
1078 VLC_TYPE (*vlc_table)[2] = table->table;
1079 int *fast_fragment_list = s->fast_fragment_list;
1082 next_fragment = s->fragment_list_y_head;
1083 list_head = &s->fragment_list_y_head;
1085 next_fragment = s->fragment_list_c_head;
1086 list_head = &s->fragment_list_c_head;
1090 previous_fragment = -1; /* this indicates that the previous fragment is actually the list head */
1092 fragment_num = coded_fragment_list[i];
1094 if (coeff_counts[fragment_num] > coeff_index) {
1095 previous_fragment = i;
1096 i = fast_fragment_list[i];
1099 fragment = &all_fragments[fragment_num];
1102 /* decode a VLC into a token */
1103 token = get_vlc2(gb, vlc_table, 5, 3);
1104 /* use the token to get a zero run, a coefficient, and an eob run */
1106 eob_run = eob_run_base[token];
1107 if (eob_run_get_bits[token])
1108 eob_run += get_bits(gb, eob_run_get_bits[token]);
1109 coeff = zero_run = 0;
1111 bits_to_get = coeff_get_bits[token];
1113 bits_to_get = get_bits(gb, bits_to_get);
1114 coeff = coeff_tables[token][bits_to_get];
1116 zero_run = zero_run_base[token];
1117 if (zero_run_get_bits[token])
1118 zero_run += get_bits(gb, zero_run_get_bits[token]);
1123 coeff_counts[fragment_num] += zero_run;
1124 if (coeff_counts[fragment_num] < 64){
1125 fragment->next_coeff->coeff= coeff;
1126 fragment->next_coeff->index= perm[coeff_counts[fragment_num]++]; //FIXME perm here already?
1127 fragment->next_coeff->next= s->next_coeff;
1128 s->next_coeff->next=NULL;
1129 fragment->next_coeff= s->next_coeff++;
1131 /* previous fragment is now this fragment */
1132 previous_fragment = i;
1134 coeff_counts[fragment_num] |= 128;
1136 /* remove this fragment from the list */
1137 if (previous_fragment != -1)
1138 fast_fragment_list[previous_fragment] = fast_fragment_list[i];
1140 *list_head = fast_fragment_list[i];
1141 /* previous fragment remains unchanged */
1144 i = fast_fragment_list[i];
1150 static void reverse_dc_prediction(Vp3DecodeContext *s,
1153 int fragment_height);
1155 * This function unpacks all of the DCT coefficient data from the
1158 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1165 int residual_eob_run = 0;
1169 /* fetch the DC table indexes */
1170 dc_y_table = get_bits(gb, 4);
1171 dc_c_table = get_bits(gb, 4);
1173 /* unpack the Y plane DC coefficients */
1174 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1175 1, residual_eob_run);
1177 /* reverse prediction of the Y-plane DC coefficients */
1178 reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
1180 /* unpack the C plane DC coefficients */
1181 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1182 0, residual_eob_run);
1184 /* reverse prediction of the C-plane DC coefficients */
1185 if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1187 reverse_dc_prediction(s, s->fragment_start[1],
1188 s->fragment_width / 2, s->fragment_height / 2);
1189 reverse_dc_prediction(s, s->fragment_start[2],
1190 s->fragment_width / 2, s->fragment_height / 2);
1193 /* fetch the AC table indexes */
1194 ac_y_table = get_bits(gb, 4);
1195 ac_c_table = get_bits(gb, 4);
1197 /* build tables of AC VLC tables */
1198 for (i = 1; i <= 5; i++) {
1199 y_tables[i] = &s->ac_vlc_1[ac_y_table];
1200 c_tables[i] = &s->ac_vlc_1[ac_c_table];
1202 for (i = 6; i <= 14; i++) {
1203 y_tables[i] = &s->ac_vlc_2[ac_y_table];
1204 c_tables[i] = &s->ac_vlc_2[ac_c_table];
1206 for (i = 15; i <= 27; i++) {
1207 y_tables[i] = &s->ac_vlc_3[ac_y_table];
1208 c_tables[i] = &s->ac_vlc_3[ac_c_table];
1210 for (i = 28; i <= 63; i++) {
1211 y_tables[i] = &s->ac_vlc_4[ac_y_table];
1212 c_tables[i] = &s->ac_vlc_4[ac_c_table];
1215 /* decode all AC coefficents */
1216 for (i = 1; i <= 63; i++) {
1217 if (s->fragment_list_y_head != -1)
1218 residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1219 1, residual_eob_run);
1221 if (s->fragment_list_c_head != -1)
1222 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1223 0, residual_eob_run);
1230 * This function reverses the DC prediction for each coded fragment in
1231 * the frame. Much of this function is adapted directly from the original
1234 #define COMPATIBLE_FRAME(x) \
1235 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1236 #define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
1238 static void reverse_dc_prediction(Vp3DecodeContext *s,
1241 int fragment_height)
1250 int i = first_fragment;
1254 /* DC values for the left, up-left, up, and up-right fragments */
1255 int vl, vul, vu, vur;
1257 /* indexes for the left, up-left, up, and up-right fragments */
1261 * The 6 fields mean:
1262 * 0: up-left multiplier
1264 * 2: up-right multiplier
1265 * 3: left multiplier
1267 static const int predictor_transform[16][4] = {
1269 { 0, 0, 0,128}, // PL
1270 { 0, 0,128, 0}, // PUR
1271 { 0, 0, 53, 75}, // PUR|PL
1272 { 0,128, 0, 0}, // PU
1273 { 0, 64, 0, 64}, // PU|PL
1274 { 0,128, 0, 0}, // PU|PUR
1275 { 0, 0, 53, 75}, // PU|PUR|PL
1276 {128, 0, 0, 0}, // PUL
1277 { 0, 0, 0,128}, // PUL|PL
1278 { 64, 0, 64, 0}, // PUL|PUR
1279 { 0, 0, 53, 75}, // PUL|PUR|PL
1280 { 0,128, 0, 0}, // PUL|PU
1281 {-104,116, 0,116}, // PUL|PU|PL
1282 { 24, 80, 24, 0}, // PUL|PU|PUR
1283 {-104,116, 0,116} // PUL|PU|PUR|PL
1286 /* This table shows which types of blocks can use other blocks for
1287 * prediction. For example, INTRA is the only mode in this table to
1288 * have a frame number of 0. That means INTRA blocks can only predict
1289 * from other INTRA blocks. There are 2 golden frame coding types;
1290 * blocks encoding in these modes can only predict from other blocks
1291 * that were encoded with these 1 of these 2 modes. */
1292 static const unsigned char compatible_frame[9] = {
1293 1, /* MODE_INTER_NO_MV */
1295 1, /* MODE_INTER_PLUS_MV */
1296 1, /* MODE_INTER_LAST_MV */
1297 1, /* MODE_INTER_PRIOR_MV */
1298 2, /* MODE_USING_GOLDEN */
1299 2, /* MODE_GOLDEN_MV */
1300 1, /* MODE_INTER_FOUR_MV */
1303 int current_frame_type;
1305 /* there is a last DC predictor for each of the 3 frame types */
1310 vul = vu = vur = vl = 0;
1311 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1313 /* for each fragment row... */
1314 for (y = 0; y < fragment_height; y++) {
1316 /* for each fragment in a row... */
1317 for (x = 0; x < fragment_width; x++, i++) {
1319 /* reverse prediction if this block was coded */
1320 if (s->all_fragments[i].coding_method != MODE_COPY) {
1322 current_frame_type =
1323 compatible_frame[s->all_fragments[i].coding_method];
1329 if(COMPATIBLE_FRAME(l))
1333 u= i-fragment_width;
1335 if(COMPATIBLE_FRAME(u))
1338 ul= i-fragment_width-1;
1340 if(COMPATIBLE_FRAME(ul))
1343 if(x + 1 < fragment_width){
1344 ur= i-fragment_width+1;
1346 if(COMPATIBLE_FRAME(ur))
1351 if (transform == 0) {
1353 /* if there were no fragments to predict from, use last
1355 predicted_dc = last_dc[current_frame_type];
1358 /* apply the appropriate predictor transform */
1360 (predictor_transform[transform][0] * vul) +
1361 (predictor_transform[transform][1] * vu) +
1362 (predictor_transform[transform][2] * vur) +
1363 (predictor_transform[transform][3] * vl);
1365 predicted_dc /= 128;
1367 /* check for outranging on the [ul u l] and
1368 * [ul u ur l] predictors */
1369 if ((transform == 15) || (transform == 13)) {
1370 if (FFABS(predicted_dc - vu) > 128)
1372 else if (FFABS(predicted_dc - vl) > 128)
1374 else if (FFABS(predicted_dc - vul) > 128)
1379 /* at long last, apply the predictor */
1380 if(s->coeffs[i].index){
1381 *s->next_coeff= s->coeffs[i];
1382 s->coeffs[i].index=0;
1383 s->coeffs[i].coeff=0;
1384 s->coeffs[i].next= s->next_coeff++;
1386 s->coeffs[i].coeff += predicted_dc;
1388 last_dc[current_frame_type] = DC_COEFF(i);
1389 if(DC_COEFF(i) && !(s->coeff_counts[i]&127)){
1390 s->coeff_counts[i]= 129;
1391 // s->all_fragments[i].next_coeff= s->next_coeff;
1392 s->coeffs[i].next= s->next_coeff;
1393 (s->next_coeff++)->next=NULL;
1401 * Perform the final rendering for a particular slice of data.
1402 * The slice number ranges from 0..(macroblock_height - 1).
1404 static void render_slice(Vp3DecodeContext *s, int slice)
1407 int16_t *dequantizer;
1408 DECLARE_ALIGNED_16(DCTELEM, block[64]);
1409 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1410 int motion_halfpel_index;
1411 uint8_t *motion_source;
1413 int current_macroblock_entry = slice * s->macroblock_width * 6;
1415 if (slice >= s->macroblock_height)
1418 for (plane = 0; plane < 3; plane++) {
1419 uint8_t *output_plane = s->current_frame.data [plane];
1420 uint8_t * last_plane = s-> last_frame.data [plane];
1421 uint8_t *golden_plane = s-> golden_frame.data [plane];
1422 int stride = s->current_frame.linesize[plane];
1423 int plane_width = s->width >> !!plane;
1424 int plane_height = s->height >> !!plane;
1425 int y = slice * FRAGMENT_PIXELS << !plane ;
1426 int slice_height = y + (FRAGMENT_PIXELS << !plane);
1427 int i = s->macroblock_fragments[current_macroblock_entry + plane + 3*!!plane];
1429 if (!s->flipped_image) stride = -stride;
1432 if(FFABS(stride) > 2048)
1433 return; //various tables are fixed size
1435 /* for each fragment row in the slice (both of them)... */
1436 for (; y < slice_height; y += 8) {
1438 /* for each fragment in a row... */
1439 for (x = 0; x < plane_width; x += 8, i++) {
1441 if ((i < 0) || (i >= s->fragment_count)) {
1442 av_log(s->avctx, AV_LOG_ERROR, " vp3:render_slice(): bad fragment number (%d)\n", i);
1446 /* transform if this block was coded */
1447 if ((s->all_fragments[i].coding_method != MODE_COPY) &&
1448 !((s->avctx->flags & CODEC_FLAG_GRAY) && plane)) {
1450 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1451 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1452 motion_source= golden_plane;
1454 motion_source= last_plane;
1456 motion_source += s->all_fragments[i].first_pixel;
1457 motion_halfpel_index = 0;
1459 /* sort out the motion vector if this fragment is coded
1460 * using a motion vector method */
1461 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1462 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1464 motion_x = s->all_fragments[i].motion_x;
1465 motion_y = s->all_fragments[i].motion_y;
1467 motion_x= (motion_x>>1) | (motion_x&1);
1468 motion_y= (motion_y>>1) | (motion_y&1);
1471 src_x= (motion_x>>1) + x;
1472 src_y= (motion_y>>1) + y;
1473 if ((motion_x == 127) || (motion_y == 127))
1474 av_log(s->avctx, AV_LOG_ERROR, " help! got invalid motion vector! (%X, %X)\n", motion_x, motion_y);
1476 motion_halfpel_index = motion_x & 0x01;
1477 motion_source += (motion_x >> 1);
1479 motion_halfpel_index |= (motion_y & 0x01) << 1;
1480 motion_source += ((motion_y >> 1) * stride);
1482 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1483 uint8_t *temp= s->edge_emu_buffer;
1484 if(stride<0) temp -= 9*stride;
1485 else temp += 9*stride;
1487 ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1488 motion_source= temp;
1493 /* first, take care of copying a block from either the
1494 * previous or the golden frame */
1495 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1496 /* Note, it is possible to implement all MC cases with
1497 put_no_rnd_pixels_l2 which would look more like the
1498 VP3 source but this would be slower as
1499 put_no_rnd_pixels_tab is better optimzed */
1500 if(motion_halfpel_index != 3){
1501 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1502 output_plane + s->all_fragments[i].first_pixel,
1503 motion_source, stride, 8);
1505 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1506 s->dsp.put_no_rnd_pixels_l2[1](
1507 output_plane + s->all_fragments[i].first_pixel,
1509 motion_source + stride + 1 + d,
1512 dequantizer = s->qmat[s->all_fragments[i].qpi][1][plane];
1514 dequantizer = s->qmat[s->all_fragments[i].qpi][0][plane];
1517 /* dequantize the DCT coefficients */
1518 if(s->avctx->idct_algo==FF_IDCT_VP3){
1519 Coeff *coeff= s->coeffs + i;
1520 s->dsp.clear_block(block);
1522 block[coeff->index]= coeff->coeff * dequantizer[coeff->index];
1526 Coeff *coeff= s->coeffs + i;
1527 s->dsp.clear_block(block);
1529 block[coeff->index]= (coeff->coeff * dequantizer[coeff->index] + 2)>>2;
1534 /* invert DCT and place (or add) in final output */
1536 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1537 if(s->avctx->idct_algo!=FF_IDCT_VP3)
1540 output_plane + s->all_fragments[i].first_pixel,
1545 output_plane + s->all_fragments[i].first_pixel,
1551 /* copy directly from the previous frame */
1552 s->dsp.put_pixels_tab[1][0](
1553 output_plane + s->all_fragments[i].first_pixel,
1554 last_plane + s->all_fragments[i].first_pixel,
1559 /* perform the left edge filter if:
1560 * - the fragment is not on the left column
1561 * - the fragment is coded in this frame
1562 * - the fragment is not coded in this frame but the left
1563 * fragment is coded in this frame (this is done instead
1564 * of a right edge filter when rendering the left fragment
1565 * since this fragment is not available yet) */
1567 ((s->all_fragments[i].coding_method != MODE_COPY) ||
1568 ((s->all_fragments[i].coding_method == MODE_COPY) &&
1569 (s->all_fragments[i - 1].coding_method != MODE_COPY)) )) {
1571 output_plane + s->all_fragments[i].first_pixel + 7*stride,
1572 -stride, s->bounding_values_array + 127);
1575 /* perform the top edge filter if:
1576 * - the fragment is not on the top row
1577 * - the fragment is coded in this frame
1578 * - the fragment is not coded in this frame but the above
1579 * fragment is coded in this frame (this is done instead
1580 * of a bottom edge filter when rendering the above
1581 * fragment since this fragment is not available yet) */
1583 ((s->all_fragments[i].coding_method != MODE_COPY) ||
1584 ((s->all_fragments[i].coding_method == MODE_COPY) &&
1585 (s->all_fragments[i - fragment_width].coding_method != MODE_COPY)) )) {
1587 output_plane + s->all_fragments[i].first_pixel - stride,
1588 -stride, s->bounding_values_array + 127);
1595 /* this looks like a good place for slice dispatch... */
1597 * if (slice == s->macroblock_height - 1)
1598 * dispatch (both last slice & 2nd-to-last slice);
1599 * else if (slice > 0)
1600 * dispatch (slice - 1);
1606 static void apply_loop_filter(Vp3DecodeContext *s)
1610 int *bounding_values= s->bounding_values_array+127;
1613 int bounding_values_array[256];
1616 /* find the right loop limit value */
1617 for (x = 63; x >= 0; x--) {
1618 if (vp31_ac_scale_factor[x] >= s->quality_index)
1621 filter_limit = vp31_filter_limit_values[s->quality_index];
1623 /* set up the bounding values */
1624 memset(bounding_values_array, 0, 256 * sizeof(int));
1625 for (x = 0; x < filter_limit; x++) {
1626 bounding_values[-x - filter_limit] = -filter_limit + x;
1627 bounding_values[-x] = -x;
1628 bounding_values[x] = x;
1629 bounding_values[x + filter_limit] = filter_limit - x;
1633 for (plane = 0; plane < 3; plane++) {
1634 int width = s->fragment_width >> !!plane;
1635 int height = s->fragment_height >> !!plane;
1636 int fragment = s->fragment_start [plane];
1637 int stride = s->current_frame.linesize[plane];
1638 uint8_t *plane_data = s->current_frame.data [plane];
1639 if (!s->flipped_image) stride = -stride;
1641 for (y = 0; y < height; y++) {
1643 for (x = 0; x < width; x++) {
1644 /* This code basically just deblocks on the edges of coded blocks.
1645 * However, it has to be much more complicated because of the
1646 * braindamaged deblock ordering used in VP3/Theora. Order matters
1647 * because some pixels get filtered twice. */
1648 if( s->all_fragments[fragment].coding_method != MODE_COPY )
1650 /* do not perform left edge filter for left columns frags */
1652 s->dsp.vp3_h_loop_filter(
1653 plane_data + s->all_fragments[fragment].first_pixel,
1654 stride, bounding_values);
1657 /* do not perform top edge filter for top row fragments */
1659 s->dsp.vp3_v_loop_filter(
1660 plane_data + s->all_fragments[fragment].first_pixel,
1661 stride, bounding_values);
1664 /* do not perform right edge filter for right column
1665 * fragments or if right fragment neighbor is also coded
1666 * in this frame (it will be filtered in next iteration) */
1667 if ((x < width - 1) &&
1668 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1669 s->dsp.vp3_h_loop_filter(
1670 plane_data + s->all_fragments[fragment + 1].first_pixel,
1671 stride, bounding_values);
1674 /* do not perform bottom edge filter for bottom row
1675 * fragments or if bottom fragment neighbor is also coded
1676 * in this frame (it will be filtered in the next row) */
1677 if ((y < height - 1) &&
1678 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1679 s->dsp.vp3_v_loop_filter(
1680 plane_data + s->all_fragments[fragment + width].first_pixel,
1681 stride, bounding_values);
1692 * This function computes the first pixel addresses for each fragment.
1693 * This function needs to be invoked after the first frame is allocated
1694 * so that it has access to the plane strides.
1696 static void vp3_calculate_pixel_addresses(Vp3DecodeContext *s)
1698 #define Y_INITIAL(chroma_shift) s->flipped_image ? 1 : s->fragment_height >> chroma_shift
1699 #define Y_FINISHED(chroma_shift) s->flipped_image ? y <= s->fragment_height >> chroma_shift : y > 0
1702 const int y_inc = s->flipped_image ? 1 : -1;
1704 /* figure out the first pixel addresses for each of the fragments */
1707 for (y = Y_INITIAL(0); Y_FINISHED(0); y += y_inc) {
1708 for (x = 0; x < s->fragment_width; x++) {
1709 s->all_fragments[i++].first_pixel =
1710 s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
1711 s->golden_frame.linesize[0] +
1712 x * FRAGMENT_PIXELS;
1717 i = s->fragment_start[1];
1718 for (y = Y_INITIAL(1); Y_FINISHED(1); y += y_inc) {
1719 for (x = 0; x < s->fragment_width / 2; x++) {
1720 s->all_fragments[i++].first_pixel =
1721 s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
1722 s->golden_frame.linesize[1] +
1723 x * FRAGMENT_PIXELS;
1728 i = s->fragment_start[2];
1729 for (y = Y_INITIAL(1); Y_FINISHED(1); y += y_inc) {
1730 for (x = 0; x < s->fragment_width / 2; x++) {
1731 s->all_fragments[i++].first_pixel =
1732 s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
1733 s->golden_frame.linesize[2] +
1734 x * FRAGMENT_PIXELS;
1740 * This is the ffmpeg/libavcodec API init function.
1742 static av_cold int vp3_decode_init(AVCodecContext *avctx)
1744 Vp3DecodeContext *s = avctx->priv_data;
1745 int i, inter, plane;
1748 int y_superblock_count;
1749 int c_superblock_count;
1751 if (avctx->codec_tag == MKTAG('V','P','3','0'))
1757 s->width = FFALIGN(avctx->width, 16);
1758 s->height = FFALIGN(avctx->height, 16);
1759 avctx->pix_fmt = PIX_FMT_YUV420P;
1760 avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1761 if(avctx->idct_algo==FF_IDCT_AUTO)
1762 avctx->idct_algo=FF_IDCT_VP3;
1763 dsputil_init(&s->dsp, avctx);
1765 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1767 /* initialize to an impossible value which will force a recalculation
1768 * in the first frame decode */
1769 for (i = 0; i < 3; i++)
1772 s->y_superblock_width = (s->width + 31) / 32;
1773 s->y_superblock_height = (s->height + 31) / 32;
1774 y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1776 /* work out the dimensions for the C planes */
1777 c_width = s->width / 2;
1778 c_height = s->height / 2;
1779 s->c_superblock_width = (c_width + 31) / 32;
1780 s->c_superblock_height = (c_height + 31) / 32;
1781 c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1783 s->superblock_count = y_superblock_count + (c_superblock_count * 2);
1784 s->u_superblock_start = y_superblock_count;
1785 s->v_superblock_start = s->u_superblock_start + c_superblock_count;
1786 s->superblock_coding = av_malloc(s->superblock_count);
1788 s->macroblock_width = (s->width + 15) / 16;
1789 s->macroblock_height = (s->height + 15) / 16;
1790 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1792 s->fragment_width = s->width / FRAGMENT_PIXELS;
1793 s->fragment_height = s->height / FRAGMENT_PIXELS;
1795 /* fragment count covers all 8x8 blocks for all 3 planes */
1796 s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
1797 s->fragment_start[1] = s->fragment_width * s->fragment_height;
1798 s->fragment_start[2] = s->fragment_width * s->fragment_height * 5 / 4;
1800 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1801 s->coeff_counts = av_malloc(s->fragment_count * sizeof(*s->coeff_counts));
1802 s->coeffs = av_malloc(s->fragment_count * sizeof(Coeff) * 65);
1803 s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int));
1804 s->fast_fragment_list = av_malloc(s->fragment_count * sizeof(int));
1805 s->pixel_addresses_initialized = 0;
1806 if (!s->superblock_coding || !s->all_fragments || !s->coeff_counts ||
1807 !s->coeffs || !s->coded_fragment_list || !s->fast_fragment_list) {
1808 vp3_decode_end(avctx);
1812 if (!s->theora_tables)
1814 for (i = 0; i < 64; i++) {
1815 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1816 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1817 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1818 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1819 s->base_matrix[2][i] = vp31_inter_dequant[i];
1820 s->filter_limit_values[i] = vp31_filter_limit_values[i];
1823 for(inter=0; inter<2; inter++){
1824 for(plane=0; plane<3; plane++){
1825 s->qr_count[inter][plane]= 1;
1826 s->qr_size [inter][plane][0]= 63;
1827 s->qr_base [inter][plane][0]=
1828 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1832 /* init VLC tables */
1833 for (i = 0; i < 16; i++) {
1836 init_vlc(&s->dc_vlc[i], 5, 32,
1837 &dc_bias[i][0][1], 4, 2,
1838 &dc_bias[i][0][0], 4, 2, 0);
1840 /* group 1 AC histograms */
1841 init_vlc(&s->ac_vlc_1[i], 5, 32,
1842 &ac_bias_0[i][0][1], 4, 2,
1843 &ac_bias_0[i][0][0], 4, 2, 0);
1845 /* group 2 AC histograms */
1846 init_vlc(&s->ac_vlc_2[i], 5, 32,
1847 &ac_bias_1[i][0][1], 4, 2,
1848 &ac_bias_1[i][0][0], 4, 2, 0);
1850 /* group 3 AC histograms */
1851 init_vlc(&s->ac_vlc_3[i], 5, 32,
1852 &ac_bias_2[i][0][1], 4, 2,
1853 &ac_bias_2[i][0][0], 4, 2, 0);
1855 /* group 4 AC histograms */
1856 init_vlc(&s->ac_vlc_4[i], 5, 32,
1857 &ac_bias_3[i][0][1], 4, 2,
1858 &ac_bias_3[i][0][0], 4, 2, 0);
1861 for (i = 0; i < 16; i++) {
1864 if (init_vlc(&s->dc_vlc[i], 5, 32,
1865 &s->huffman_table[i][0][1], 4, 2,
1866 &s->huffman_table[i][0][0], 4, 2, 0) < 0)
1869 /* group 1 AC histograms */
1870 if (init_vlc(&s->ac_vlc_1[i], 5, 32,
1871 &s->huffman_table[i+16][0][1], 4, 2,
1872 &s->huffman_table[i+16][0][0], 4, 2, 0) < 0)
1875 /* group 2 AC histograms */
1876 if (init_vlc(&s->ac_vlc_2[i], 5, 32,
1877 &s->huffman_table[i+16*2][0][1], 4, 2,
1878 &s->huffman_table[i+16*2][0][0], 4, 2, 0) < 0)
1881 /* group 3 AC histograms */
1882 if (init_vlc(&s->ac_vlc_3[i], 5, 32,
1883 &s->huffman_table[i+16*3][0][1], 4, 2,
1884 &s->huffman_table[i+16*3][0][0], 4, 2, 0) < 0)
1887 /* group 4 AC histograms */
1888 if (init_vlc(&s->ac_vlc_4[i], 5, 32,
1889 &s->huffman_table[i+16*4][0][1], 4, 2,
1890 &s->huffman_table[i+16*4][0][0], 4, 2, 0) < 0)
1895 init_vlc(&s->superblock_run_length_vlc, 6, 34,
1896 &superblock_run_length_vlc_table[0][1], 4, 2,
1897 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1899 init_vlc(&s->fragment_run_length_vlc, 5, 30,
1900 &fragment_run_length_vlc_table[0][1], 4, 2,
1901 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1903 init_vlc(&s->mode_code_vlc, 3, 8,
1904 &mode_code_vlc_table[0][1], 2, 1,
1905 &mode_code_vlc_table[0][0], 2, 1, 0);
1907 init_vlc(&s->motion_vector_vlc, 6, 63,
1908 &motion_vector_vlc_table[0][1], 2, 1,
1909 &motion_vector_vlc_table[0][0], 2, 1, 0);
1911 /* work out the block mapping tables */
1912 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1913 s->superblock_macroblocks = av_malloc(s->superblock_count * 4 * sizeof(int));
1914 s->macroblock_fragments = av_malloc(s->macroblock_count * 6 * sizeof(int));
1915 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1916 if (!s->superblock_fragments || !s->superblock_macroblocks ||
1917 !s->macroblock_fragments || !s->macroblock_coding) {
1918 vp3_decode_end(avctx);
1921 init_block_mapping(s);
1923 for (i = 0; i < 3; i++) {
1924 s->current_frame.data[i] = NULL;
1925 s->last_frame.data[i] = NULL;
1926 s->golden_frame.data[i] = NULL;
1932 av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1937 * This is the ffmpeg/libavcodec API frame decode function.
1939 static int vp3_decode_frame(AVCodecContext *avctx,
1940 void *data, int *data_size,
1943 const uint8_t *buf = avpkt->data;
1944 int buf_size = avpkt->size;
1945 Vp3DecodeContext *s = avctx->priv_data;
1947 static int counter = 0;
1950 init_get_bits(&gb, buf, buf_size * 8);
1952 if (s->theora && get_bits1(&gb))
1954 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1958 s->keyframe = !get_bits1(&gb);
1961 for (i = 0; i < 3; i++)
1962 s->last_qps[i] = s->qps[i];
1966 s->qps[s->nqps++]= get_bits(&gb, 6);
1967 } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1968 for (i = s->nqps; i < 3; i++)
1971 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1972 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1973 s->keyframe?"key":"", counter, s->qps[0]);
1976 if (s->qps[0] != s->last_qps[0])
1977 init_loop_filter(s);
1979 for (i = 0; i < s->nqps; i++)
1980 // reinit all dequantizers if the first one changed, because
1981 // the DC of the first quantizer must be used for all matrices
1982 if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1983 init_dequantizer(s, i);
1985 if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1991 skip_bits(&gb, 4); /* width code */
1992 skip_bits(&gb, 4); /* height code */
1995 s->version = get_bits(&gb, 5);
1997 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
2000 if (s->version || s->theora)
2003 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
2004 skip_bits(&gb, 2); /* reserved? */
2007 if (s->last_frame.data[0] == s->golden_frame.data[0]) {
2008 if (s->golden_frame.data[0])
2009 avctx->release_buffer(avctx, &s->golden_frame);
2010 s->last_frame= s->golden_frame; /* ensure that we catch any access to this released frame */
2012 if (s->golden_frame.data[0])
2013 avctx->release_buffer(avctx, &s->golden_frame);
2014 if (s->last_frame.data[0])
2015 avctx->release_buffer(avctx, &s->last_frame);
2018 s->golden_frame.reference = 3;
2019 if(avctx->get_buffer(avctx, &s->golden_frame) < 0) {
2020 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
2024 /* golden frame is also the current frame */
2025 s->current_frame= s->golden_frame;
2027 /* time to figure out pixel addresses? */
2028 if (!s->pixel_addresses_initialized)
2030 vp3_calculate_pixel_addresses(s);
2031 s->pixel_addresses_initialized = 1;
2034 /* allocate a new current frame */
2035 s->current_frame.reference = 3;
2036 if (!s->pixel_addresses_initialized) {
2037 av_log(s->avctx, AV_LOG_ERROR, "vp3: first frame not a keyframe\n");
2040 if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
2041 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
2046 s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
2047 s->current_frame.qstride= 0;
2051 if (unpack_superblocks(s, &gb)){
2052 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
2055 if (unpack_modes(s, &gb)){
2056 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
2059 if (unpack_vectors(s, &gb)){
2060 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2063 if (unpack_block_qpis(s, &gb)){
2064 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
2067 if (unpack_dct_coeffs(s, &gb)){
2068 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
2072 for (i = 0; i < s->macroblock_height; i++)
2075 apply_loop_filter(s);
2077 *data_size=sizeof(AVFrame);
2078 *(AVFrame*)data= s->current_frame;
2080 /* release the last frame, if it is allocated and if it is not the
2082 if ((s->last_frame.data[0]) &&
2083 (s->last_frame.data[0] != s->golden_frame.data[0]))
2084 avctx->release_buffer(avctx, &s->last_frame);
2086 /* shuffle frames (last = current) */
2087 s->last_frame= s->current_frame;
2088 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
2094 * This is the ffmpeg/libavcodec API module cleanup function.
2096 static av_cold int vp3_decode_end(AVCodecContext *avctx)
2098 Vp3DecodeContext *s = avctx->priv_data;
2101 av_free(s->superblock_coding);
2102 av_free(s->all_fragments);
2103 av_free(s->coeff_counts);
2105 av_free(s->coded_fragment_list);
2106 av_free(s->fast_fragment_list);
2107 av_free(s->superblock_fragments);
2108 av_free(s->superblock_macroblocks);
2109 av_free(s->macroblock_fragments);
2110 av_free(s->macroblock_coding);
2112 for (i = 0; i < 16; i++) {
2113 free_vlc(&s->dc_vlc[i]);
2114 free_vlc(&s->ac_vlc_1[i]);
2115 free_vlc(&s->ac_vlc_2[i]);
2116 free_vlc(&s->ac_vlc_3[i]);
2117 free_vlc(&s->ac_vlc_4[i]);
2120 free_vlc(&s->superblock_run_length_vlc);
2121 free_vlc(&s->fragment_run_length_vlc);
2122 free_vlc(&s->mode_code_vlc);
2123 free_vlc(&s->motion_vector_vlc);
2125 /* release all frames */
2126 if (s->golden_frame.data[0] && s->golden_frame.data[0] != s->last_frame.data[0])
2127 avctx->release_buffer(avctx, &s->golden_frame);
2128 if (s->last_frame.data[0])
2129 avctx->release_buffer(avctx, &s->last_frame);
2130 /* no need to release the current_frame since it will always be pointing
2131 * to the same frame as either the golden or last frame */
2136 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2138 Vp3DecodeContext *s = avctx->priv_data;
2140 if (get_bits1(gb)) {
2142 if (s->entries >= 32) { /* overflow */
2143 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2146 token = get_bits(gb, 5);
2147 //av_log(avctx, AV_LOG_DEBUG, "hti %d hbits %x token %d entry : %d size %d\n", s->hti, s->hbits, token, s->entries, s->huff_code_size);
2148 s->huffman_table[s->hti][token][0] = s->hbits;
2149 s->huffman_table[s->hti][token][1] = s->huff_code_size;
2153 if (s->huff_code_size >= 32) {/* overflow */
2154 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2157 s->huff_code_size++;
2159 if (read_huffman_tree(avctx, gb))
2162 if (read_huffman_tree(avctx, gb))
2165 s->huff_code_size--;
2170 #if CONFIG_THEORA_DECODER
2171 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2173 Vp3DecodeContext *s = avctx->priv_data;
2174 int visible_width, visible_height;
2176 s->theora = get_bits_long(gb, 24);
2177 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2179 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2180 /* but previous versions have the image flipped relative to vp3 */
2181 if (s->theora < 0x030200)
2183 s->flipped_image = 1;
2184 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2187 visible_width = s->width = get_bits(gb, 16) << 4;
2188 visible_height = s->height = get_bits(gb, 16) << 4;
2190 if(avcodec_check_dimensions(avctx, s->width, s->height)){
2191 av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2192 s->width= s->height= 0;
2196 if (s->theora >= 0x030400)
2198 skip_bits(gb, 32); /* total number of superblocks in a frame */
2199 // fixme, the next field is 36bits long
2200 skip_bits(gb, 32); /* total number of blocks in a frame */
2201 skip_bits(gb, 4); /* total number of blocks in a frame */
2202 skip_bits(gb, 32); /* total number of macroblocks in a frame */
2205 if (s->theora >= 0x030200) {
2206 visible_width = get_bits_long(gb, 24);
2207 visible_height = get_bits_long(gb, 24);
2209 skip_bits(gb, 8); /* offset x */
2210 skip_bits(gb, 8); /* offset y */
2213 skip_bits(gb, 32); /* fps numerator */
2214 skip_bits(gb, 32); /* fps denumerator */
2215 skip_bits(gb, 24); /* aspect numerator */
2216 skip_bits(gb, 24); /* aspect denumerator */
2218 if (s->theora < 0x030200)
2219 skip_bits(gb, 5); /* keyframe frequency force */
2220 skip_bits(gb, 8); /* colorspace */
2221 if (s->theora >= 0x030400)
2222 skip_bits(gb, 2); /* pixel format: 420,res,422,444 */
2223 skip_bits(gb, 24); /* bitrate */
2225 skip_bits(gb, 6); /* quality hint */
2227 if (s->theora >= 0x030200)
2229 skip_bits(gb, 5); /* keyframe frequency force */
2231 if (s->theora < 0x030400)
2232 skip_bits(gb, 5); /* spare bits */
2235 // align_get_bits(gb);
2237 if ( visible_width <= s->width && visible_width > s->width-16
2238 && visible_height <= s->height && visible_height > s->height-16)
2239 avcodec_set_dimensions(avctx, visible_width, visible_height);
2241 avcodec_set_dimensions(avctx, s->width, s->height);
2246 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2248 Vp3DecodeContext *s = avctx->priv_data;
2249 int i, n, matrices, inter, plane;
2251 if (s->theora >= 0x030200) {
2252 n = get_bits(gb, 3);
2253 /* loop filter limit values table */
2254 for (i = 0; i < 64; i++) {
2255 s->filter_limit_values[i] = get_bits(gb, n);
2256 if (s->filter_limit_values[i] > 127) {
2257 av_log(avctx, AV_LOG_ERROR, "filter limit value too large (%i > 127), clamping\n", s->filter_limit_values[i]);
2258 s->filter_limit_values[i] = 127;
2263 if (s->theora >= 0x030200)
2264 n = get_bits(gb, 4) + 1;
2267 /* quality threshold table */
2268 for (i = 0; i < 64; i++)
2269 s->coded_ac_scale_factor[i] = get_bits(gb, n);
2271 if (s->theora >= 0x030200)
2272 n = get_bits(gb, 4) + 1;
2275 /* dc scale factor table */
2276 for (i = 0; i < 64; i++)
2277 s->coded_dc_scale_factor[i] = get_bits(gb, n);
2279 if (s->theora >= 0x030200)
2280 matrices = get_bits(gb, 9) + 1;
2285 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2289 for(n=0; n<matrices; n++){
2290 for (i = 0; i < 64; i++)
2291 s->base_matrix[n][i]= get_bits(gb, 8);
2294 for (inter = 0; inter <= 1; inter++) {
2295 for (plane = 0; plane <= 2; plane++) {
2297 if (inter || plane > 0)
2298 newqr = get_bits1(gb);
2301 if(inter && get_bits1(gb)){
2305 qtj= (3*inter + plane - 1) / 3;
2306 plj= (plane + 2) % 3;
2308 s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2309 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2310 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2316 i= get_bits(gb, av_log2(matrices-1)+1);
2318 av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2321 s->qr_base[inter][plane][qri]= i;
2324 i = get_bits(gb, av_log2(63-qi)+1) + 1;
2325 s->qr_size[inter][plane][qri++]= i;
2330 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2333 s->qr_count[inter][plane]= qri;
2338 /* Huffman tables */
2339 for (s->hti = 0; s->hti < 80; s->hti++) {
2341 s->huff_code_size = 1;
2342 if (!get_bits1(gb)) {
2344 if(read_huffman_tree(avctx, gb))
2347 if(read_huffman_tree(avctx, gb))
2352 s->theora_tables = 1;
2357 static av_cold int theora_decode_init(AVCodecContext *avctx)
2359 Vp3DecodeContext *s = avctx->priv_data;
2362 uint8_t *header_start[3];
2368 if (!avctx->extradata_size)
2370 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2374 if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2375 42, header_start, header_len) < 0) {
2376 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2381 init_get_bits(&gb, header_start[i], header_len[i] * 8);
2383 ptype = get_bits(&gb, 8);
2385 if (!(ptype & 0x80))
2387 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2391 // FIXME: Check for this as well.
2392 skip_bits_long(&gb, 6*8); /* "theora" */
2397 theora_decode_header(avctx, &gb);
2400 // FIXME: is this needed? it breaks sometimes
2401 // theora_decode_comments(avctx, gb);
2404 if (theora_decode_tables(avctx, &gb))
2408 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2411 if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2412 av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2413 if (s->theora < 0x030200)
2417 return vp3_decode_init(avctx);
2420 AVCodec theora_decoder = {
2424 sizeof(Vp3DecodeContext),
2431 .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2435 AVCodec vp3_decoder = {
2439 sizeof(Vp3DecodeContext),
2446 .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),