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 uint8_t coding_method;
62 #define SB_NOT_CODED 0
63 #define SB_PARTIALLY_CODED 1
64 #define SB_FULLY_CODED 2
66 #define MODE_INTER_NO_MV 0
68 #define MODE_INTER_PLUS_MV 2
69 #define MODE_INTER_LAST_MV 3
70 #define MODE_INTER_PRIOR_LAST 4
71 #define MODE_USING_GOLDEN 5
72 #define MODE_GOLDEN_MV 6
73 #define MODE_INTER_FOURMV 7
74 #define CODING_MODE_COUNT 8
76 /* special internal mode */
79 /* There are 6 preset schemes, plus a free-form scheme */
80 static const int ModeAlphabet[6][CODING_MODE_COUNT] =
82 /* scheme 1: Last motion vector dominates */
83 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
84 MODE_INTER_PLUS_MV, MODE_INTER_NO_MV,
85 MODE_INTRA, MODE_USING_GOLDEN,
86 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
89 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
90 MODE_INTER_NO_MV, MODE_INTER_PLUS_MV,
91 MODE_INTRA, MODE_USING_GOLDEN,
92 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
95 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
96 MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
97 MODE_INTRA, MODE_USING_GOLDEN,
98 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
101 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
102 MODE_INTER_NO_MV, MODE_INTER_PRIOR_LAST,
103 MODE_INTRA, MODE_USING_GOLDEN,
104 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
106 /* scheme 5: No motion vector dominates */
107 { MODE_INTER_NO_MV, MODE_INTER_LAST_MV,
108 MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
109 MODE_INTRA, MODE_USING_GOLDEN,
110 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
113 { MODE_INTER_NO_MV, MODE_USING_GOLDEN,
114 MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
115 MODE_INTER_PLUS_MV, MODE_INTRA,
116 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
120 #define MIN_DEQUANT_VAL 2
122 typedef struct Vp3DecodeContext {
123 AVCodecContext *avctx;
124 int theora, theora_tables;
127 AVFrame golden_frame;
129 AVFrame current_frame;
139 int superblock_count;
140 int y_superblock_width;
141 int y_superblock_height;
142 int c_superblock_width;
143 int c_superblock_height;
144 int u_superblock_start;
145 int v_superblock_start;
146 unsigned char *superblock_coding;
148 int macroblock_count;
149 int macroblock_width;
150 int macroblock_height;
156 Vp3Fragment *all_fragments;
157 uint8_t *coeff_counts;
160 int fragment_start[3];
166 uint16_t coded_dc_scale_factor[64];
167 uint32_t coded_ac_scale_factor[64];
168 uint8_t base_matrix[384][64];
169 uint8_t qr_count[2][3];
170 uint8_t qr_size [2][3][64];
171 uint16_t qr_base[2][3][64];
173 /* this is a list of indexes into the all_fragments array indicating
174 * which of the fragments are coded */
175 int *coded_fragment_list;
176 int coded_fragment_list_index;
178 /* track which fragments have already been decoded; called 'fast'
179 * because this data structure avoids having to iterate through every
180 * fragment in coded_fragment_list; once a fragment has been fully
181 * decoded, it is removed from this list */
182 int *fast_fragment_list;
183 int fragment_list_y_head;
184 int fragment_list_c_head;
192 VLC superblock_run_length_vlc;
193 VLC fragment_run_length_vlc;
195 VLC motion_vector_vlc;
197 /* these arrays need to be on 16-byte boundaries since SSE2 operations
199 DECLARE_ALIGNED_16(int16_t, qmat)[3][2][3][64]; //<qmat[qpi][is_inter][plane]
201 /* This table contains superblock_count * 16 entries. Each set of 16
202 * numbers corresponds to the fragment indexes 0..15 of the superblock.
203 * An entry will be -1 to indicate that no entry corresponds to that
205 int *superblock_fragments;
207 /* This is an array that indicates how a particular macroblock
209 unsigned char *macroblock_coding;
211 int first_coded_y_fragment;
212 int first_coded_c_fragment;
213 int last_coded_y_fragment;
214 int last_coded_c_fragment;
216 uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
217 int8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
224 uint16_t huffman_table[80][32][2];
226 uint8_t filter_limit_values[64];
227 DECLARE_ALIGNED_8(int, bounding_values_array)[256+2];
230 /************************************************************************
231 * VP3 specific functions
232 ************************************************************************/
235 * This function sets up all of the various blocks mappings:
236 * superblocks <-> fragments, macroblocks <-> fragments,
237 * superblocks <-> macroblocks
239 * Returns 0 is successful; returns 1 if *anything* went wrong.
241 static int init_block_mapping(Vp3DecodeContext *s)
244 signed int hilbert_walk_mb[4];
246 int current_fragment = 0;
247 int current_width = 0;
248 int current_height = 0;
251 int superblock_row_inc = 0;
252 int mapping_index = 0;
254 int current_macroblock;
257 static const signed char travel_width[16] = {
264 static const signed char travel_height[16] = {
271 hilbert_walk_mb[0] = 1;
272 hilbert_walk_mb[1] = s->macroblock_width;
273 hilbert_walk_mb[2] = 1;
274 hilbert_walk_mb[3] = -s->macroblock_width;
276 /* iterate through each superblock (all planes) and map the fragments */
277 for (i = 0; i < s->superblock_count; i++) {
278 /* time to re-assign the limits? */
281 /* start of Y superblocks */
282 right_edge = s->fragment_width;
283 bottom_edge = s->fragment_height;
286 superblock_row_inc = 3 * s->fragment_width -
287 (s->y_superblock_width * 4 - s->fragment_width);
289 /* the first operation for this variable is to advance by 1 */
290 current_fragment = -1;
292 } else if (i == s->u_superblock_start) {
294 /* start of U superblocks */
295 right_edge = s->fragment_width / 2;
296 bottom_edge = s->fragment_height / 2;
299 superblock_row_inc = 3 * (s->fragment_width / 2) -
300 (s->c_superblock_width * 4 - s->fragment_width / 2);
302 /* the first operation for this variable is to advance by 1 */
303 current_fragment = s->fragment_start[1] - 1;
305 } else if (i == s->v_superblock_start) {
307 /* start of V superblocks */
308 right_edge = s->fragment_width / 2;
309 bottom_edge = s->fragment_height / 2;
312 superblock_row_inc = 3 * (s->fragment_width / 2) -
313 (s->c_superblock_width * 4 - s->fragment_width / 2);
315 /* the first operation for this variable is to advance by 1 */
316 current_fragment = s->fragment_start[2] - 1;
320 if (current_width >= right_edge - 1) {
321 /* reset width and move to next superblock row */
325 /* fragment is now at the start of a new superblock row */
326 current_fragment += superblock_row_inc;
329 /* iterate through all 16 fragments in a superblock */
330 for (j = 0; j < 16; j++) {
331 current_fragment += travel_width[j] + right_edge * travel_height[j];
332 current_width += travel_width[j];
333 current_height += travel_height[j];
335 /* check if the fragment is in bounds */
336 if ((current_width < right_edge) &&
337 (current_height < bottom_edge)) {
338 s->superblock_fragments[mapping_index] = current_fragment;
340 s->superblock_fragments[mapping_index] = -1;
347 return 0; /* successful path out */
351 * This function wipes out all of the fragment data.
353 static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
357 /* zero out all of the fragment information */
358 s->coded_fragment_list_index = 0;
359 for (i = 0; i < s->fragment_count; i++) {
360 s->coeff_counts[i] = 0;
361 s->all_fragments[i].motion_x = 127;
362 s->all_fragments[i].motion_y = 127;
363 s->all_fragments[i].next_coeff= NULL;
364 s->all_fragments[i].qpi = 0;
366 s->coeffs[i].coeff=0;
367 s->coeffs[i].next= NULL;
372 * This function sets up the dequantization tables used for a particular
375 static void init_dequantizer(Vp3DecodeContext *s, int qpi)
377 int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
378 int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
379 int i, plane, inter, qri, bmi, bmj, qistart;
381 for(inter=0; inter<2; inter++){
382 for(plane=0; plane<3; plane++){
384 for(qri=0; qri<s->qr_count[inter][plane]; qri++){
385 sum+= s->qr_size[inter][plane][qri];
386 if(s->qps[qpi] <= sum)
389 qistart= sum - s->qr_size[inter][plane][qri];
390 bmi= s->qr_base[inter][plane][qri ];
391 bmj= s->qr_base[inter][plane][qri+1];
393 int coeff= ( 2*(sum -s->qps[qpi])*s->base_matrix[bmi][i]
394 - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
395 + s->qr_size[inter][plane][qri])
396 / (2*s->qr_size[inter][plane][qri]);
398 int qmin= 8<<(inter + !i);
399 int qscale= i ? ac_scale_factor : dc_scale_factor;
401 s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
403 // all DC coefficients use the same quant so as not to interfere with DC prediction
404 s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
408 memset(s->qscale_table, (FFMAX(s->qmat[0][0][0][1], s->qmat[0][0][1][1])+8)/16, 512); //FIXME finetune
412 * This function initializes the loop filter boundary limits if the frame's
413 * quality index is different from the previous frame's.
415 * The filter_limit_values may not be larger than 127.
417 static void init_loop_filter(Vp3DecodeContext *s)
419 int *bounding_values= s->bounding_values_array+127;
424 filter_limit = s->filter_limit_values[s->qps[0]];
426 /* set up the bounding values */
427 memset(s->bounding_values_array, 0, 256 * sizeof(int));
428 for (x = 0; x < filter_limit; x++) {
429 bounding_values[-x] = -x;
430 bounding_values[x] = x;
432 for (x = value = filter_limit; x < 128 && value; x++, value--) {
433 bounding_values[ x] = value;
434 bounding_values[-x] = -value;
437 bounding_values[128] = value;
438 bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
442 * This function unpacks all of the superblock/macroblock/fragment coding
443 * information from the bitstream.
445 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
448 int current_superblock = 0;
450 int decode_fully_flags = 0;
451 int decode_partial_blocks = 0;
452 int first_c_fragment_seen;
455 int current_fragment;
458 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
462 /* unpack the list of partially-coded superblocks */
464 /* toggle the bit because as soon as the first run length is
465 * fetched the bit will be toggled again */
467 while (current_superblock < s->superblock_count) {
468 if (current_run-- == 0) {
470 current_run = get_vlc2(gb,
471 s->superblock_run_length_vlc.table, 6, 2);
472 if (current_run == 33)
473 current_run += get_bits(gb, 12);
475 /* if any of the superblocks are not partially coded, flag
476 * a boolean to decode the list of fully-coded superblocks */
478 decode_fully_flags = 1;
481 /* make a note of the fact that there are partially coded
483 decode_partial_blocks = 1;
486 s->superblock_coding[current_superblock++] = bit;
489 /* unpack the list of fully coded superblocks if any of the blocks were
490 * not marked as partially coded in the previous step */
491 if (decode_fully_flags) {
493 current_superblock = 0;
496 /* toggle the bit because as soon as the first run length is
497 * fetched the bit will be toggled again */
499 while (current_superblock < s->superblock_count) {
501 /* skip any superblocks already marked as partially coded */
502 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
504 if (current_run-- == 0) {
506 current_run = get_vlc2(gb,
507 s->superblock_run_length_vlc.table, 6, 2);
508 if (current_run == 33)
509 current_run += get_bits(gb, 12);
511 s->superblock_coding[current_superblock] = 2*bit;
513 current_superblock++;
517 /* if there were partial blocks, initialize bitstream for
518 * unpacking fragment codings */
519 if (decode_partial_blocks) {
523 /* toggle the bit because as soon as the first run length is
524 * fetched the bit will be toggled again */
529 /* figure out which fragments are coded; iterate through each
530 * superblock (all planes) */
531 s->coded_fragment_list_index = 0;
532 s->next_coeff= s->coeffs + s->fragment_count;
533 s->first_coded_y_fragment = s->first_coded_c_fragment = 0;
534 s->last_coded_y_fragment = s->last_coded_c_fragment = -1;
535 first_c_fragment_seen = 0;
536 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
537 for (i = 0; i < s->superblock_count; i++) {
539 /* iterate through all 16 fragments in a superblock */
540 for (j = 0; j < 16; j++) {
542 /* if the fragment is in bounds, check its coding status */
543 current_fragment = s->superblock_fragments[i * 16 + j];
544 if (current_fragment >= s->fragment_count) {
545 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
546 current_fragment, s->fragment_count);
549 if (current_fragment != -1) {
550 if (s->superblock_coding[i] == SB_NOT_CODED) {
552 /* copy all the fragments from the prior frame */
553 s->all_fragments[current_fragment].coding_method =
556 } else if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
558 /* fragment may or may not be coded; this is the case
559 * that cares about the fragment coding runs */
560 if (current_run-- == 0) {
562 current_run = get_vlc2(gb,
563 s->fragment_run_length_vlc.table, 5, 2);
567 /* default mode; actual mode will be decoded in
569 s->all_fragments[current_fragment].coding_method =
571 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
572 s->coded_fragment_list[s->coded_fragment_list_index] =
574 if ((current_fragment >= s->fragment_start[1]) &&
575 (s->last_coded_y_fragment == -1) &&
576 (!first_c_fragment_seen)) {
577 s->first_coded_c_fragment = s->coded_fragment_list_index;
578 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
579 first_c_fragment_seen = 1;
581 s->coded_fragment_list_index++;
583 /* not coded; copy this fragment from the prior frame */
584 s->all_fragments[current_fragment].coding_method =
590 /* fragments are fully coded in this superblock; actual
591 * coding will be determined in next step */
592 s->all_fragments[current_fragment].coding_method =
594 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
595 s->coded_fragment_list[s->coded_fragment_list_index] =
597 if ((current_fragment >= s->fragment_start[1]) &&
598 (s->last_coded_y_fragment == -1) &&
599 (!first_c_fragment_seen)) {
600 s->first_coded_c_fragment = s->coded_fragment_list_index;
601 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
602 first_c_fragment_seen = 1;
604 s->coded_fragment_list_index++;
610 if (!first_c_fragment_seen)
611 /* only Y fragments coded in this frame */
612 s->last_coded_y_fragment = s->coded_fragment_list_index - 1;
614 /* end the list of coded C fragments */
615 s->last_coded_c_fragment = s->coded_fragment_list_index - 1;
617 for (i = 0; i < s->fragment_count - 1; i++) {
618 s->fast_fragment_list[i] = i + 1;
620 s->fast_fragment_list[s->fragment_count - 1] = -1;
622 if (s->last_coded_y_fragment == -1)
623 s->fragment_list_y_head = -1;
625 s->fragment_list_y_head = s->first_coded_y_fragment;
626 s->fast_fragment_list[s->last_coded_y_fragment] = -1;
629 if (s->last_coded_c_fragment == -1)
630 s->fragment_list_c_head = -1;
632 s->fragment_list_c_head = s->first_coded_c_fragment;
633 s->fast_fragment_list[s->last_coded_c_fragment] = -1;
640 * This function unpacks all the coding mode data for individual macroblocks
641 * from the bitstream.
643 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
645 int i, j, k, sb_x, sb_y;
647 int current_macroblock;
648 int current_fragment;
650 int custom_mode_alphabet[CODING_MODE_COUNT];
654 for (i = 0; i < s->fragment_count; i++)
655 s->all_fragments[i].coding_method = MODE_INTRA;
659 /* fetch the mode coding scheme for this frame */
660 scheme = get_bits(gb, 3);
662 /* is it a custom coding scheme? */
664 for (i = 0; i < 8; i++)
665 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
666 for (i = 0; i < 8; i++)
667 custom_mode_alphabet[get_bits(gb, 3)] = i;
668 alphabet = custom_mode_alphabet;
670 alphabet = ModeAlphabet[scheme-1];
672 /* iterate through all of the macroblocks that contain 1 or more
674 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
675 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
677 for (j = 0; j < 4; j++) {
678 int mb_x = 2*sb_x + (j>>1);
679 int mb_y = 2*sb_y + (((j>>1)+j)&1);
681 current_macroblock = mb_y * s->macroblock_width + mb_x;
683 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height)
686 #define BLOCK_X (2*mb_x + (k&1))
687 #define BLOCK_Y (2*mb_y + (k>>1))
688 /* coding modes are only stored if the macroblock has at least one
689 * luma block coded, otherwise it must be INTER_NO_MV */
690 for (k = 0; k < 4; k++) {
691 current_fragment = BLOCK_Y*s->fragment_width + BLOCK_X;
692 if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
696 s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
700 /* mode 7 means get 3 bits for each coding mode */
702 coding_mode = get_bits(gb, 3);
704 coding_mode = alphabet
705 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
707 s->macroblock_coding[current_macroblock] = coding_mode;
708 for (k = 0; k < 4; k++) {
710 BLOCK_Y*s->fragment_width + BLOCK_X;
711 if (s->all_fragments[current_fragment].coding_method !=
713 s->all_fragments[current_fragment].coding_method =
716 for (k = 0; k < 2; k++) {
717 current_fragment = s->fragment_start[k+1] +
718 mb_y*(s->fragment_width>>1) + mb_x;
719 if (s->all_fragments[current_fragment].coding_method !=
721 s->all_fragments[current_fragment].coding_method =
733 * This function unpacks all the motion vectors for the individual
734 * macroblocks from the bitstream.
736 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
738 int j, k, sb_x, sb_y;
742 int last_motion_x = 0;
743 int last_motion_y = 0;
744 int prior_last_motion_x = 0;
745 int prior_last_motion_y = 0;
746 int current_macroblock;
747 int current_fragment;
752 memset(motion_x, 0, 6 * sizeof(int));
753 memset(motion_y, 0, 6 * sizeof(int));
755 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
756 coding_mode = get_bits1(gb);
758 /* iterate through all of the macroblocks that contain 1 or more
760 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
761 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
763 for (j = 0; j < 4; j++) {
764 int mb_x = 2*sb_x + (j>>1);
765 int mb_y = 2*sb_y + (((j>>1)+j)&1);
766 current_macroblock = mb_y * s->macroblock_width + mb_x;
768 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
769 (s->macroblock_coding[current_macroblock] == MODE_COPY))
772 switch (s->macroblock_coding[current_macroblock]) {
774 case MODE_INTER_PLUS_MV:
776 /* all 6 fragments use the same motion vector */
777 if (coding_mode == 0) {
778 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
779 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
781 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
782 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
785 /* vector maintenance, only on MODE_INTER_PLUS_MV */
786 if (s->macroblock_coding[current_macroblock] ==
787 MODE_INTER_PLUS_MV) {
788 prior_last_motion_x = last_motion_x;
789 prior_last_motion_y = last_motion_y;
790 last_motion_x = motion_x[0];
791 last_motion_y = motion_y[0];
795 case MODE_INTER_FOURMV:
796 /* vector maintenance */
797 prior_last_motion_x = last_motion_x;
798 prior_last_motion_y = last_motion_y;
800 /* fetch 4 vectors from the bitstream, one for each
801 * Y fragment, then average for the C fragment vectors */
802 motion_x[4] = motion_y[4] = 0;
803 for (k = 0; k < 4; k++) {
804 current_fragment = BLOCK_Y*s->fragment_width + BLOCK_X;
805 if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
806 if (coding_mode == 0) {
807 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
808 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
810 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
811 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
813 last_motion_x = motion_x[k];
814 last_motion_y = motion_y[k];
819 motion_x[4] += motion_x[k];
820 motion_y[4] += motion_y[k];
824 motion_x[4]= RSHIFT(motion_x[4], 2);
826 motion_y[4]= RSHIFT(motion_y[4], 2);
829 case MODE_INTER_LAST_MV:
830 /* all 6 fragments use the last motion vector */
831 motion_x[0] = last_motion_x;
832 motion_y[0] = last_motion_y;
834 /* no vector maintenance (last vector remains the
838 case MODE_INTER_PRIOR_LAST:
839 /* all 6 fragments use the motion vector prior to the
840 * last motion vector */
841 motion_x[0] = prior_last_motion_x;
842 motion_y[0] = prior_last_motion_y;
844 /* vector maintenance */
845 prior_last_motion_x = last_motion_x;
846 prior_last_motion_y = last_motion_y;
847 last_motion_x = motion_x[0];
848 last_motion_y = motion_y[0];
852 /* covers intra, inter without MV, golden without MV */
856 /* no vector maintenance */
860 /* assign the motion vectors to the correct fragments */
861 for (k = 0; k < 4; k++) {
863 BLOCK_Y*s->fragment_width + BLOCK_X;
864 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
865 s->all_fragments[current_fragment].motion_x = motion_x[k];
866 s->all_fragments[current_fragment].motion_y = motion_y[k];
868 s->all_fragments[current_fragment].motion_x = motion_x[0];
869 s->all_fragments[current_fragment].motion_y = motion_y[0];
872 for (k = 0; k < 2; k++) {
873 current_fragment = s->fragment_start[k+1] +
874 mb_y*(s->fragment_width>>1) + mb_x;
875 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
876 s->all_fragments[current_fragment].motion_x = motion_x[k+4];
877 s->all_fragments[current_fragment].motion_y = motion_y[k+4];
879 s->all_fragments[current_fragment].motion_x = motion_x[0];
880 s->all_fragments[current_fragment].motion_y = motion_y[0];
890 static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
892 int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
893 int num_blocks = s->coded_fragment_list_index;
895 for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
896 i = blocks_decoded = num_blocks_at_qpi = 0;
901 run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
902 if (run_length == 34)
903 run_length += get_bits(gb, 12);
904 blocks_decoded += run_length;
907 num_blocks_at_qpi += run_length;
909 for (j = 0; j < run_length; i++) {
910 if (i >= s->coded_fragment_list_index)
913 if (s->all_fragments[s->coded_fragment_list[i]].qpi == qpi) {
914 s->all_fragments[s->coded_fragment_list[i]].qpi += bit;
919 if (run_length == 4129)
923 } while (blocks_decoded < num_blocks);
925 num_blocks -= num_blocks_at_qpi;
932 * This function is called by unpack_dct_coeffs() to extract the VLCs from
933 * the bitstream. The VLCs encode tokens which are used to unpack DCT
934 * data. This function unpacks all the VLCs for either the Y plane or both
935 * C planes, and is called for DC coefficients or different AC coefficient
936 * levels (since different coefficient types require different VLC tables.
938 * This function returns a residual eob run. E.g, if a particular token gave
939 * instructions to EOB the next 5 fragments and there were only 2 fragments
940 * left in the current fragment range, 3 would be returned so that it could
941 * be passed into the next call to this same function.
943 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
944 VLC *table, int coeff_index,
952 Vp3Fragment *fragment;
955 int previous_fragment;
959 /* local references to structure members to avoid repeated deferences */
960 uint8_t *perm= s->scantable.permutated;
961 int *coded_fragment_list = s->coded_fragment_list;
962 Vp3Fragment *all_fragments = s->all_fragments;
963 uint8_t *coeff_counts = s->coeff_counts;
964 VLC_TYPE (*vlc_table)[2] = table->table;
965 int *fast_fragment_list = s->fast_fragment_list;
968 next_fragment = s->fragment_list_y_head;
969 list_head = &s->fragment_list_y_head;
971 next_fragment = s->fragment_list_c_head;
972 list_head = &s->fragment_list_c_head;
976 previous_fragment = -1; /* this indicates that the previous fragment is actually the list head */
978 fragment_num = coded_fragment_list[i];
980 if (coeff_counts[fragment_num] > coeff_index) {
981 previous_fragment = i;
982 i = fast_fragment_list[i];
985 fragment = &all_fragments[fragment_num];
988 /* decode a VLC into a token */
989 token = get_vlc2(gb, vlc_table, 5, 3);
990 /* use the token to get a zero run, a coefficient, and an eob run */
992 eob_run = eob_run_base[token];
993 if (eob_run_get_bits[token])
994 eob_run += get_bits(gb, eob_run_get_bits[token]);
995 coeff = zero_run = 0;
997 bits_to_get = coeff_get_bits[token];
999 bits_to_get = get_bits(gb, bits_to_get);
1000 coeff = coeff_tables[token][bits_to_get];
1002 zero_run = zero_run_base[token];
1003 if (zero_run_get_bits[token])
1004 zero_run += get_bits(gb, zero_run_get_bits[token]);
1009 coeff_counts[fragment_num] += zero_run;
1010 if (coeff_counts[fragment_num] < 64){
1011 fragment->next_coeff->coeff= coeff;
1012 fragment->next_coeff->index= perm[coeff_counts[fragment_num]++]; //FIXME perm here already?
1013 fragment->next_coeff->next= s->next_coeff;
1014 s->next_coeff->next=NULL;
1015 fragment->next_coeff= s->next_coeff++;
1017 /* previous fragment is now this fragment */
1018 previous_fragment = i;
1020 coeff_counts[fragment_num] |= 128;
1022 /* remove this fragment from the list */
1023 if (previous_fragment != -1)
1024 fast_fragment_list[previous_fragment] = fast_fragment_list[i];
1026 *list_head = fast_fragment_list[i];
1027 /* previous fragment remains unchanged */
1030 i = fast_fragment_list[i];
1036 static void reverse_dc_prediction(Vp3DecodeContext *s,
1039 int fragment_height);
1041 * This function unpacks all of the DCT coefficient data from the
1044 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1051 int residual_eob_run = 0;
1055 /* fetch the DC table indexes */
1056 dc_y_table = get_bits(gb, 4);
1057 dc_c_table = get_bits(gb, 4);
1059 /* unpack the Y plane DC coefficients */
1060 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1061 1, residual_eob_run);
1063 /* reverse prediction of the Y-plane DC coefficients */
1064 reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
1066 /* unpack the C plane DC coefficients */
1067 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1068 0, residual_eob_run);
1070 /* reverse prediction of the C-plane DC coefficients */
1071 if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1073 reverse_dc_prediction(s, s->fragment_start[1],
1074 s->fragment_width / 2, s->fragment_height / 2);
1075 reverse_dc_prediction(s, s->fragment_start[2],
1076 s->fragment_width / 2, s->fragment_height / 2);
1079 /* fetch the AC table indexes */
1080 ac_y_table = get_bits(gb, 4);
1081 ac_c_table = get_bits(gb, 4);
1083 /* build tables of AC VLC tables */
1084 for (i = 1; i <= 5; i++) {
1085 y_tables[i] = &s->ac_vlc_1[ac_y_table];
1086 c_tables[i] = &s->ac_vlc_1[ac_c_table];
1088 for (i = 6; i <= 14; i++) {
1089 y_tables[i] = &s->ac_vlc_2[ac_y_table];
1090 c_tables[i] = &s->ac_vlc_2[ac_c_table];
1092 for (i = 15; i <= 27; i++) {
1093 y_tables[i] = &s->ac_vlc_3[ac_y_table];
1094 c_tables[i] = &s->ac_vlc_3[ac_c_table];
1096 for (i = 28; i <= 63; i++) {
1097 y_tables[i] = &s->ac_vlc_4[ac_y_table];
1098 c_tables[i] = &s->ac_vlc_4[ac_c_table];
1101 /* decode all AC coefficents */
1102 for (i = 1; i <= 63; i++) {
1103 if (s->fragment_list_y_head != -1)
1104 residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1105 1, residual_eob_run);
1107 if (s->fragment_list_c_head != -1)
1108 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1109 0, residual_eob_run);
1116 * This function reverses the DC prediction for each coded fragment in
1117 * the frame. Much of this function is adapted directly from the original
1120 #define COMPATIBLE_FRAME(x) \
1121 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1122 #define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
1124 static void reverse_dc_prediction(Vp3DecodeContext *s,
1127 int fragment_height)
1136 int i = first_fragment;
1140 /* DC values for the left, up-left, up, and up-right fragments */
1141 int vl, vul, vu, vur;
1143 /* indexes for the left, up-left, up, and up-right fragments */
1147 * The 6 fields mean:
1148 * 0: up-left multiplier
1150 * 2: up-right multiplier
1151 * 3: left multiplier
1153 static const int predictor_transform[16][4] = {
1155 { 0, 0, 0,128}, // PL
1156 { 0, 0,128, 0}, // PUR
1157 { 0, 0, 53, 75}, // PUR|PL
1158 { 0,128, 0, 0}, // PU
1159 { 0, 64, 0, 64}, // PU|PL
1160 { 0,128, 0, 0}, // PU|PUR
1161 { 0, 0, 53, 75}, // PU|PUR|PL
1162 {128, 0, 0, 0}, // PUL
1163 { 0, 0, 0,128}, // PUL|PL
1164 { 64, 0, 64, 0}, // PUL|PUR
1165 { 0, 0, 53, 75}, // PUL|PUR|PL
1166 { 0,128, 0, 0}, // PUL|PU
1167 {-104,116, 0,116}, // PUL|PU|PL
1168 { 24, 80, 24, 0}, // PUL|PU|PUR
1169 {-104,116, 0,116} // PUL|PU|PUR|PL
1172 /* This table shows which types of blocks can use other blocks for
1173 * prediction. For example, INTRA is the only mode in this table to
1174 * have a frame number of 0. That means INTRA blocks can only predict
1175 * from other INTRA blocks. There are 2 golden frame coding types;
1176 * blocks encoding in these modes can only predict from other blocks
1177 * that were encoded with these 1 of these 2 modes. */
1178 static const unsigned char compatible_frame[9] = {
1179 1, /* MODE_INTER_NO_MV */
1181 1, /* MODE_INTER_PLUS_MV */
1182 1, /* MODE_INTER_LAST_MV */
1183 1, /* MODE_INTER_PRIOR_MV */
1184 2, /* MODE_USING_GOLDEN */
1185 2, /* MODE_GOLDEN_MV */
1186 1, /* MODE_INTER_FOUR_MV */
1189 int current_frame_type;
1191 /* there is a last DC predictor for each of the 3 frame types */
1196 vul = vu = vur = vl = 0;
1197 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1199 /* for each fragment row... */
1200 for (y = 0; y < fragment_height; y++) {
1202 /* for each fragment in a row... */
1203 for (x = 0; x < fragment_width; x++, i++) {
1205 /* reverse prediction if this block was coded */
1206 if (s->all_fragments[i].coding_method != MODE_COPY) {
1208 current_frame_type =
1209 compatible_frame[s->all_fragments[i].coding_method];
1215 if(COMPATIBLE_FRAME(l))
1219 u= i-fragment_width;
1221 if(COMPATIBLE_FRAME(u))
1224 ul= i-fragment_width-1;
1226 if(COMPATIBLE_FRAME(ul))
1229 if(x + 1 < fragment_width){
1230 ur= i-fragment_width+1;
1232 if(COMPATIBLE_FRAME(ur))
1237 if (transform == 0) {
1239 /* if there were no fragments to predict from, use last
1241 predicted_dc = last_dc[current_frame_type];
1244 /* apply the appropriate predictor transform */
1246 (predictor_transform[transform][0] * vul) +
1247 (predictor_transform[transform][1] * vu) +
1248 (predictor_transform[transform][2] * vur) +
1249 (predictor_transform[transform][3] * vl);
1251 predicted_dc /= 128;
1253 /* check for outranging on the [ul u l] and
1254 * [ul u ur l] predictors */
1255 if ((transform == 15) || (transform == 13)) {
1256 if (FFABS(predicted_dc - vu) > 128)
1258 else if (FFABS(predicted_dc - vl) > 128)
1260 else if (FFABS(predicted_dc - vul) > 128)
1265 /* at long last, apply the predictor */
1266 if(s->coeffs[i].index){
1267 *s->next_coeff= s->coeffs[i];
1268 s->coeffs[i].index=0;
1269 s->coeffs[i].coeff=0;
1270 s->coeffs[i].next= s->next_coeff++;
1272 s->coeffs[i].coeff += predicted_dc;
1274 last_dc[current_frame_type] = DC_COEFF(i);
1275 if(DC_COEFF(i) && !(s->coeff_counts[i]&127)){
1276 s->coeff_counts[i]= 129;
1277 // s->all_fragments[i].next_coeff= s->next_coeff;
1278 s->coeffs[i].next= s->next_coeff;
1279 (s->next_coeff++)->next=NULL;
1286 static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
1289 int *bounding_values= s->bounding_values_array+127;
1291 int width = s->fragment_width >> !!plane;
1292 int height = s->fragment_height >> !!plane;
1293 int fragment = s->fragment_start [plane] + ystart * width;
1294 int stride = s->current_frame.linesize[plane];
1295 uint8_t *plane_data = s->current_frame.data [plane];
1296 if (!s->flipped_image) stride = -stride;
1297 plane_data += s->data_offset[plane] + 8*ystart*stride;
1299 for (y = ystart; y < yend; y++) {
1301 for (x = 0; x < width; x++) {
1302 /* This code basically just deblocks on the edges of coded blocks.
1303 * However, it has to be much more complicated because of the
1304 * braindamaged deblock ordering used in VP3/Theora. Order matters
1305 * because some pixels get filtered twice. */
1306 if( s->all_fragments[fragment].coding_method != MODE_COPY )
1308 /* do not perform left edge filter for left columns frags */
1310 s->dsp.vp3_h_loop_filter(
1312 stride, bounding_values);
1315 /* do not perform top edge filter for top row fragments */
1317 s->dsp.vp3_v_loop_filter(
1319 stride, bounding_values);
1322 /* do not perform right edge filter for right column
1323 * fragments or if right fragment neighbor is also coded
1324 * in this frame (it will be filtered in next iteration) */
1325 if ((x < width - 1) &&
1326 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1327 s->dsp.vp3_h_loop_filter(
1328 plane_data + 8*x + 8,
1329 stride, bounding_values);
1332 /* do not perform bottom edge filter for bottom row
1333 * fragments or if bottom fragment neighbor is also coded
1334 * in this frame (it will be filtered in the next row) */
1335 if ((y < height - 1) &&
1336 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1337 s->dsp.vp3_v_loop_filter(
1338 plane_data + 8*x + 8*stride,
1339 stride, bounding_values);
1345 plane_data += 8*stride;
1350 * called when all pixels up to row y are complete
1352 static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1357 if(s->avctx->draw_horiz_band==NULL)
1360 h= y - s->last_slice_end;
1363 if (!s->flipped_image) {
1365 h -= s->height - s->avctx->height; // account for non-mod16
1366 y = s->height - y - h;
1370 offset[0] = s->current_frame.linesize[0]*y;
1371 offset[1] = s->current_frame.linesize[1]*cy;
1372 offset[2] = s->current_frame.linesize[2]*cy;
1376 s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
1377 s->last_slice_end= y + h;
1381 * Perform the final rendering for a particular slice of data.
1382 * The slice number ranges from 0..(macroblock_height - 1).
1384 static void render_slice(Vp3DecodeContext *s, int slice)
1387 int16_t *dequantizer;
1388 DECLARE_ALIGNED_16(DCTELEM, block)[64];
1389 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1390 int motion_halfpel_index;
1391 uint8_t *motion_source;
1394 if (slice >= s->macroblock_height)
1397 for (plane = 0; plane < 3; plane++) {
1398 uint8_t *output_plane = s->current_frame.data [plane] + s->data_offset[plane];
1399 uint8_t * last_plane = s-> last_frame.data [plane] + s->data_offset[plane];
1400 uint8_t *golden_plane = s-> golden_frame.data [plane] + s->data_offset[plane];
1401 int stride = s->current_frame.linesize[plane];
1402 int plane_width = s->width >> !!plane;
1403 int plane_height = s->height >> !!plane;
1404 int y = slice * FRAGMENT_PIXELS << !plane ;
1405 int slice_height = y + (FRAGMENT_PIXELS << !plane);
1406 int i = s->fragment_start[plane] + (y>>3)*(s->fragment_width>>!!plane);
1408 if (!s->flipped_image) stride = -stride;
1411 if(FFABS(stride) > 2048)
1412 return; //various tables are fixed size
1414 /* for each fragment row in the slice (both of them)... */
1415 for (; y < slice_height; y += 8) {
1417 /* for each fragment in a row... */
1418 for (x = 0; x < plane_width; x += 8, i++) {
1419 int first_pixel = y*stride + x;
1421 if ((i < 0) || (i >= s->fragment_count)) {
1422 av_log(s->avctx, AV_LOG_ERROR, " vp3:render_slice(): bad fragment number (%d)\n", i);
1426 /* transform if this block was coded */
1427 if ((s->all_fragments[i].coding_method != MODE_COPY) &&
1428 !((s->avctx->flags & CODEC_FLAG_GRAY) && plane)) {
1430 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1431 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1432 motion_source= golden_plane;
1434 motion_source= last_plane;
1436 motion_source += first_pixel;
1437 motion_halfpel_index = 0;
1439 /* sort out the motion vector if this fragment is coded
1440 * using a motion vector method */
1441 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1442 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1444 motion_x = s->all_fragments[i].motion_x;
1445 motion_y = s->all_fragments[i].motion_y;
1447 motion_x= (motion_x>>1) | (motion_x&1);
1448 motion_y= (motion_y>>1) | (motion_y&1);
1451 src_x= (motion_x>>1) + x;
1452 src_y= (motion_y>>1) + y;
1453 if ((motion_x == 127) || (motion_y == 127))
1454 av_log(s->avctx, AV_LOG_ERROR, " help! got invalid motion vector! (%X, %X)\n", motion_x, motion_y);
1456 motion_halfpel_index = motion_x & 0x01;
1457 motion_source += (motion_x >> 1);
1459 motion_halfpel_index |= (motion_y & 0x01) << 1;
1460 motion_source += ((motion_y >> 1) * stride);
1462 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1463 uint8_t *temp= s->edge_emu_buffer;
1464 if(stride<0) temp -= 9*stride;
1465 else temp += 9*stride;
1467 ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1468 motion_source= temp;
1473 /* first, take care of copying a block from either the
1474 * previous or the golden frame */
1475 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1476 /* Note, it is possible to implement all MC cases with
1477 put_no_rnd_pixels_l2 which would look more like the
1478 VP3 source but this would be slower as
1479 put_no_rnd_pixels_tab is better optimzed */
1480 if(motion_halfpel_index != 3){
1481 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1482 output_plane + first_pixel,
1483 motion_source, stride, 8);
1485 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1486 s->dsp.put_no_rnd_pixels_l2[1](
1487 output_plane + first_pixel,
1489 motion_source + stride + 1 + d,
1492 dequantizer = s->qmat[s->all_fragments[i].qpi][1][plane];
1494 dequantizer = s->qmat[s->all_fragments[i].qpi][0][plane];
1497 /* dequantize the DCT coefficients */
1498 if(s->avctx->idct_algo==FF_IDCT_VP3){
1499 Coeff *coeff= s->coeffs + i;
1500 s->dsp.clear_block(block);
1502 block[coeff->index]= coeff->coeff * dequantizer[coeff->index];
1506 Coeff *coeff= s->coeffs + i;
1507 s->dsp.clear_block(block);
1509 block[coeff->index]= (coeff->coeff * dequantizer[coeff->index] + 2)>>2;
1514 /* invert DCT and place (or add) in final output */
1516 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1517 if(s->avctx->idct_algo!=FF_IDCT_VP3)
1520 output_plane + first_pixel,
1525 output_plane + first_pixel,
1531 /* copy directly from the previous frame */
1532 s->dsp.put_pixels_tab[1][0](
1533 output_plane + first_pixel,
1534 last_plane + first_pixel,
1539 // Filter the previous block row. We can't filter the current row yet
1540 // since it needs pixels from the next row
1542 apply_loop_filter(s, plane, (y>>3)-1, (y>>3));
1546 /* this looks like a good place for slice dispatch... */
1548 * if (slice == s->macroblock_height - 1)
1549 * dispatch (both last slice & 2nd-to-last slice);
1550 * else if (slice > 0)
1551 * dispatch (slice - 1);
1554 // now that we've filtered the last rows, they're safe to display
1556 vp3_draw_horiz_band(s, 16*slice);
1560 * This is the ffmpeg/libavcodec API init function.
1562 static av_cold int vp3_decode_init(AVCodecContext *avctx)
1564 Vp3DecodeContext *s = avctx->priv_data;
1565 int i, inter, plane;
1568 int y_superblock_count;
1569 int c_superblock_count;
1571 if (avctx->codec_tag == MKTAG('V','P','3','0'))
1577 s->width = FFALIGN(avctx->width, 16);
1578 s->height = FFALIGN(avctx->height, 16);
1579 avctx->pix_fmt = PIX_FMT_YUV420P;
1580 avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1581 if(avctx->idct_algo==FF_IDCT_AUTO)
1582 avctx->idct_algo=FF_IDCT_VP3;
1583 dsputil_init(&s->dsp, avctx);
1585 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1587 /* initialize to an impossible value which will force a recalculation
1588 * in the first frame decode */
1589 for (i = 0; i < 3; i++)
1592 s->y_superblock_width = (s->width + 31) / 32;
1593 s->y_superblock_height = (s->height + 31) / 32;
1594 y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1596 /* work out the dimensions for the C planes */
1597 c_width = s->width / 2;
1598 c_height = s->height / 2;
1599 s->c_superblock_width = (c_width + 31) / 32;
1600 s->c_superblock_height = (c_height + 31) / 32;
1601 c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1603 s->superblock_count = y_superblock_count + (c_superblock_count * 2);
1604 s->u_superblock_start = y_superblock_count;
1605 s->v_superblock_start = s->u_superblock_start + c_superblock_count;
1606 s->superblock_coding = av_malloc(s->superblock_count);
1608 s->macroblock_width = (s->width + 15) / 16;
1609 s->macroblock_height = (s->height + 15) / 16;
1610 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1612 s->fragment_width = s->width / FRAGMENT_PIXELS;
1613 s->fragment_height = s->height / FRAGMENT_PIXELS;
1615 /* fragment count covers all 8x8 blocks for all 3 planes */
1616 s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
1617 s->fragment_start[1] = s->fragment_width * s->fragment_height;
1618 s->fragment_start[2] = s->fragment_width * s->fragment_height * 5 / 4;
1620 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1621 s->coeff_counts = av_malloc(s->fragment_count * sizeof(*s->coeff_counts));
1622 s->coeffs = av_malloc(s->fragment_count * sizeof(Coeff) * 65);
1623 s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int));
1624 s->fast_fragment_list = av_malloc(s->fragment_count * sizeof(int));
1625 if (!s->superblock_coding || !s->all_fragments || !s->coeff_counts ||
1626 !s->coeffs || !s->coded_fragment_list || !s->fast_fragment_list) {
1627 vp3_decode_end(avctx);
1631 if (!s->theora_tables)
1633 for (i = 0; i < 64; i++) {
1634 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1635 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1636 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1637 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1638 s->base_matrix[2][i] = vp31_inter_dequant[i];
1639 s->filter_limit_values[i] = vp31_filter_limit_values[i];
1642 for(inter=0; inter<2; inter++){
1643 for(plane=0; plane<3; plane++){
1644 s->qr_count[inter][plane]= 1;
1645 s->qr_size [inter][plane][0]= 63;
1646 s->qr_base [inter][plane][0]=
1647 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1651 /* init VLC tables */
1652 for (i = 0; i < 16; i++) {
1655 init_vlc(&s->dc_vlc[i], 5, 32,
1656 &dc_bias[i][0][1], 4, 2,
1657 &dc_bias[i][0][0], 4, 2, 0);
1659 /* group 1 AC histograms */
1660 init_vlc(&s->ac_vlc_1[i], 5, 32,
1661 &ac_bias_0[i][0][1], 4, 2,
1662 &ac_bias_0[i][0][0], 4, 2, 0);
1664 /* group 2 AC histograms */
1665 init_vlc(&s->ac_vlc_2[i], 5, 32,
1666 &ac_bias_1[i][0][1], 4, 2,
1667 &ac_bias_1[i][0][0], 4, 2, 0);
1669 /* group 3 AC histograms */
1670 init_vlc(&s->ac_vlc_3[i], 5, 32,
1671 &ac_bias_2[i][0][1], 4, 2,
1672 &ac_bias_2[i][0][0], 4, 2, 0);
1674 /* group 4 AC histograms */
1675 init_vlc(&s->ac_vlc_4[i], 5, 32,
1676 &ac_bias_3[i][0][1], 4, 2,
1677 &ac_bias_3[i][0][0], 4, 2, 0);
1680 for (i = 0; i < 16; i++) {
1683 if (init_vlc(&s->dc_vlc[i], 5, 32,
1684 &s->huffman_table[i][0][1], 4, 2,
1685 &s->huffman_table[i][0][0], 4, 2, 0) < 0)
1688 /* group 1 AC histograms */
1689 if (init_vlc(&s->ac_vlc_1[i], 5, 32,
1690 &s->huffman_table[i+16][0][1], 4, 2,
1691 &s->huffman_table[i+16][0][0], 4, 2, 0) < 0)
1694 /* group 2 AC histograms */
1695 if (init_vlc(&s->ac_vlc_2[i], 5, 32,
1696 &s->huffman_table[i+16*2][0][1], 4, 2,
1697 &s->huffman_table[i+16*2][0][0], 4, 2, 0) < 0)
1700 /* group 3 AC histograms */
1701 if (init_vlc(&s->ac_vlc_3[i], 5, 32,
1702 &s->huffman_table[i+16*3][0][1], 4, 2,
1703 &s->huffman_table[i+16*3][0][0], 4, 2, 0) < 0)
1706 /* group 4 AC histograms */
1707 if (init_vlc(&s->ac_vlc_4[i], 5, 32,
1708 &s->huffman_table[i+16*4][0][1], 4, 2,
1709 &s->huffman_table[i+16*4][0][0], 4, 2, 0) < 0)
1714 init_vlc(&s->superblock_run_length_vlc, 6, 34,
1715 &superblock_run_length_vlc_table[0][1], 4, 2,
1716 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1718 init_vlc(&s->fragment_run_length_vlc, 5, 30,
1719 &fragment_run_length_vlc_table[0][1], 4, 2,
1720 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1722 init_vlc(&s->mode_code_vlc, 3, 8,
1723 &mode_code_vlc_table[0][1], 2, 1,
1724 &mode_code_vlc_table[0][0], 2, 1, 0);
1726 init_vlc(&s->motion_vector_vlc, 6, 63,
1727 &motion_vector_vlc_table[0][1], 2, 1,
1728 &motion_vector_vlc_table[0][0], 2, 1, 0);
1730 /* work out the block mapping tables */
1731 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1732 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1733 if (!s->superblock_fragments || !s->macroblock_coding) {
1734 vp3_decode_end(avctx);
1737 init_block_mapping(s);
1739 for (i = 0; i < 3; i++) {
1740 s->current_frame.data[i] = NULL;
1741 s->last_frame.data[i] = NULL;
1742 s->golden_frame.data[i] = NULL;
1748 av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1753 * This is the ffmpeg/libavcodec API frame decode function.
1755 static int vp3_decode_frame(AVCodecContext *avctx,
1756 void *data, int *data_size,
1759 const uint8_t *buf = avpkt->data;
1760 int buf_size = avpkt->size;
1761 Vp3DecodeContext *s = avctx->priv_data;
1763 static int counter = 0;
1766 init_get_bits(&gb, buf, buf_size * 8);
1768 if (s->theora && get_bits1(&gb))
1770 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1774 s->keyframe = !get_bits1(&gb);
1777 for (i = 0; i < 3; i++)
1778 s->last_qps[i] = s->qps[i];
1782 s->qps[s->nqps++]= get_bits(&gb, 6);
1783 } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1784 for (i = s->nqps; i < 3; i++)
1787 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1788 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1789 s->keyframe?"key":"", counter, s->qps[0]);
1792 if (s->qps[0] != s->last_qps[0])
1793 init_loop_filter(s);
1795 for (i = 0; i < s->nqps; i++)
1796 // reinit all dequantizers if the first one changed, because
1797 // the DC of the first quantizer must be used for all matrices
1798 if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1799 init_dequantizer(s, i);
1801 if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1807 skip_bits(&gb, 4); /* width code */
1808 skip_bits(&gb, 4); /* height code */
1811 s->version = get_bits(&gb, 5);
1813 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1816 if (s->version || s->theora)
1819 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1820 skip_bits(&gb, 2); /* reserved? */
1823 if (s->last_frame.data[0] == s->golden_frame.data[0]) {
1824 if (s->golden_frame.data[0])
1825 avctx->release_buffer(avctx, &s->golden_frame);
1826 s->last_frame= s->golden_frame; /* ensure that we catch any access to this released frame */
1828 if (s->golden_frame.data[0])
1829 avctx->release_buffer(avctx, &s->golden_frame);
1830 if (s->last_frame.data[0])
1831 avctx->release_buffer(avctx, &s->last_frame);
1834 s->golden_frame.reference = 3;
1835 if(avctx->get_buffer(avctx, &s->golden_frame) < 0) {
1836 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
1840 /* golden frame is also the current frame */
1841 s->current_frame= s->golden_frame;
1843 /* allocate a new current frame */
1844 s->current_frame.reference = 3;
1845 if (!s->golden_frame.data[0]) {
1846 av_log(s->avctx, AV_LOG_ERROR, "vp3: first frame not a keyframe\n");
1849 if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
1850 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
1855 s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
1856 s->current_frame.qstride= 0;
1860 if (unpack_superblocks(s, &gb)){
1861 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
1864 if (unpack_modes(s, &gb)){
1865 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
1868 if (unpack_vectors(s, &gb)){
1869 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
1872 if (unpack_block_qpis(s, &gb)){
1873 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
1876 if (unpack_dct_coeffs(s, &gb)){
1877 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
1881 for (i = 0; i < 3; i++) {
1882 if (s->flipped_image)
1883 s->data_offset[i] = 0;
1885 s->data_offset[i] = ((s->height>>!!i)-1) * s->current_frame.linesize[i];
1888 s->last_slice_end = 0;
1889 for (i = 0; i < s->macroblock_height; i++)
1892 // filter the last row
1893 for (i = 0; i < 3; i++) {
1894 int row = (s->height >> (3+!!i)) - 1;
1895 apply_loop_filter(s, i, row, row+1);
1897 vp3_draw_horiz_band(s, s->height);
1899 *data_size=sizeof(AVFrame);
1900 *(AVFrame*)data= s->current_frame;
1902 /* release the last frame, if it is allocated and if it is not the
1904 if ((s->last_frame.data[0]) &&
1905 (s->last_frame.data[0] != s->golden_frame.data[0]))
1906 avctx->release_buffer(avctx, &s->last_frame);
1908 /* shuffle frames (last = current) */
1909 s->last_frame= s->current_frame;
1910 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
1916 * This is the ffmpeg/libavcodec API module cleanup function.
1918 static av_cold int vp3_decode_end(AVCodecContext *avctx)
1920 Vp3DecodeContext *s = avctx->priv_data;
1923 av_free(s->superblock_coding);
1924 av_free(s->all_fragments);
1925 av_free(s->coeff_counts);
1927 av_free(s->coded_fragment_list);
1928 av_free(s->fast_fragment_list);
1929 av_free(s->superblock_fragments);
1930 av_free(s->macroblock_coding);
1932 for (i = 0; i < 16; i++) {
1933 free_vlc(&s->dc_vlc[i]);
1934 free_vlc(&s->ac_vlc_1[i]);
1935 free_vlc(&s->ac_vlc_2[i]);
1936 free_vlc(&s->ac_vlc_3[i]);
1937 free_vlc(&s->ac_vlc_4[i]);
1940 free_vlc(&s->superblock_run_length_vlc);
1941 free_vlc(&s->fragment_run_length_vlc);
1942 free_vlc(&s->mode_code_vlc);
1943 free_vlc(&s->motion_vector_vlc);
1945 /* release all frames */
1946 if (s->golden_frame.data[0] && s->golden_frame.data[0] != s->last_frame.data[0])
1947 avctx->release_buffer(avctx, &s->golden_frame);
1948 if (s->last_frame.data[0])
1949 avctx->release_buffer(avctx, &s->last_frame);
1950 /* no need to release the current_frame since it will always be pointing
1951 * to the same frame as either the golden or last frame */
1956 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
1958 Vp3DecodeContext *s = avctx->priv_data;
1960 if (get_bits1(gb)) {
1962 if (s->entries >= 32) { /* overflow */
1963 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1966 token = get_bits(gb, 5);
1967 //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);
1968 s->huffman_table[s->hti][token][0] = s->hbits;
1969 s->huffman_table[s->hti][token][1] = s->huff_code_size;
1973 if (s->huff_code_size >= 32) {/* overflow */
1974 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1977 s->huff_code_size++;
1979 if (read_huffman_tree(avctx, gb))
1982 if (read_huffman_tree(avctx, gb))
1985 s->huff_code_size--;
1990 #if CONFIG_THEORA_DECODER
1991 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
1993 Vp3DecodeContext *s = avctx->priv_data;
1994 int visible_width, visible_height, colorspace;
1996 s->theora = get_bits_long(gb, 24);
1997 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
1999 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2000 /* but previous versions have the image flipped relative to vp3 */
2001 if (s->theora < 0x030200)
2003 s->flipped_image = 1;
2004 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2007 visible_width = s->width = get_bits(gb, 16) << 4;
2008 visible_height = s->height = get_bits(gb, 16) << 4;
2010 if(avcodec_check_dimensions(avctx, s->width, s->height)){
2011 av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2012 s->width= s->height= 0;
2016 if (s->theora >= 0x030200) {
2017 visible_width = get_bits_long(gb, 24);
2018 visible_height = get_bits_long(gb, 24);
2020 skip_bits(gb, 8); /* offset x */
2021 skip_bits(gb, 8); /* offset y */
2024 skip_bits(gb, 32); /* fps numerator */
2025 skip_bits(gb, 32); /* fps denumerator */
2026 skip_bits(gb, 24); /* aspect numerator */
2027 skip_bits(gb, 24); /* aspect denumerator */
2029 if (s->theora < 0x030200)
2030 skip_bits(gb, 5); /* keyframe frequency force */
2031 colorspace = get_bits(gb, 8);
2032 skip_bits(gb, 24); /* bitrate */
2034 skip_bits(gb, 6); /* quality hint */
2036 if (s->theora >= 0x030200)
2038 skip_bits(gb, 5); /* keyframe frequency force */
2039 skip_bits(gb, 2); /* pixel format: 420,res,422,444 */
2040 skip_bits(gb, 3); /* reserved */
2043 // align_get_bits(gb);
2045 if ( visible_width <= s->width && visible_width > s->width-16
2046 && visible_height <= s->height && visible_height > s->height-16)
2047 avcodec_set_dimensions(avctx, visible_width, visible_height);
2049 avcodec_set_dimensions(avctx, s->width, s->height);
2051 if (colorspace == 1) {
2052 avctx->color_primaries = AVCOL_PRI_BT470M;
2053 } else if (colorspace == 2) {
2054 avctx->color_primaries = AVCOL_PRI_BT470BG;
2056 if (colorspace == 1 || colorspace == 2) {
2057 avctx->colorspace = AVCOL_SPC_BT470BG;
2058 avctx->color_trc = AVCOL_TRC_BT709;
2064 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2066 Vp3DecodeContext *s = avctx->priv_data;
2067 int i, n, matrices, inter, plane;
2069 if (s->theora >= 0x030200) {
2070 n = get_bits(gb, 3);
2071 /* loop filter limit values table */
2072 for (i = 0; i < 64; i++) {
2073 s->filter_limit_values[i] = get_bits(gb, n);
2074 if (s->filter_limit_values[i] > 127) {
2075 av_log(avctx, AV_LOG_ERROR, "filter limit value too large (%i > 127), clamping\n", s->filter_limit_values[i]);
2076 s->filter_limit_values[i] = 127;
2081 if (s->theora >= 0x030200)
2082 n = get_bits(gb, 4) + 1;
2085 /* quality threshold table */
2086 for (i = 0; i < 64; i++)
2087 s->coded_ac_scale_factor[i] = get_bits(gb, n);
2089 if (s->theora >= 0x030200)
2090 n = get_bits(gb, 4) + 1;
2093 /* dc scale factor table */
2094 for (i = 0; i < 64; i++)
2095 s->coded_dc_scale_factor[i] = get_bits(gb, n);
2097 if (s->theora >= 0x030200)
2098 matrices = get_bits(gb, 9) + 1;
2103 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2107 for(n=0; n<matrices; n++){
2108 for (i = 0; i < 64; i++)
2109 s->base_matrix[n][i]= get_bits(gb, 8);
2112 for (inter = 0; inter <= 1; inter++) {
2113 for (plane = 0; plane <= 2; plane++) {
2115 if (inter || plane > 0)
2116 newqr = get_bits1(gb);
2119 if(inter && get_bits1(gb)){
2123 qtj= (3*inter + plane - 1) / 3;
2124 plj= (plane + 2) % 3;
2126 s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2127 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2128 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2134 i= get_bits(gb, av_log2(matrices-1)+1);
2136 av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2139 s->qr_base[inter][plane][qri]= i;
2142 i = get_bits(gb, av_log2(63-qi)+1) + 1;
2143 s->qr_size[inter][plane][qri++]= i;
2148 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2151 s->qr_count[inter][plane]= qri;
2156 /* Huffman tables */
2157 for (s->hti = 0; s->hti < 80; s->hti++) {
2159 s->huff_code_size = 1;
2160 if (!get_bits1(gb)) {
2162 if(read_huffman_tree(avctx, gb))
2165 if(read_huffman_tree(avctx, gb))
2170 s->theora_tables = 1;
2175 static av_cold int theora_decode_init(AVCodecContext *avctx)
2177 Vp3DecodeContext *s = avctx->priv_data;
2180 uint8_t *header_start[3];
2186 if (!avctx->extradata_size)
2188 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2192 if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2193 42, header_start, header_len) < 0) {
2194 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2199 init_get_bits(&gb, header_start[i], header_len[i] * 8);
2201 ptype = get_bits(&gb, 8);
2203 if (!(ptype & 0x80))
2205 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2209 // FIXME: Check for this as well.
2210 skip_bits_long(&gb, 6*8); /* "theora" */
2215 theora_decode_header(avctx, &gb);
2218 // FIXME: is this needed? it breaks sometimes
2219 // theora_decode_comments(avctx, gb);
2222 if (theora_decode_tables(avctx, &gb))
2226 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2229 if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2230 av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2231 if (s->theora < 0x030200)
2235 return vp3_decode_init(avctx);
2238 AVCodec theora_decoder = {
2242 sizeof(Vp3DecodeContext),
2247 CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2249 .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2253 AVCodec vp3_decoder = {
2257 sizeof(Vp3DecodeContext),
2262 CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2264 .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),