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
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
36 #include "libavutil/imgutils.h"
45 #define FRAGMENT_PIXELS 8
47 static av_cold int vp3_decode_end(AVCodecContext *avctx);
49 //FIXME split things out into their own arrays
50 typedef struct Vp3Fragment {
52 uint8_t coding_method;
56 #define SB_NOT_CODED 0
57 #define SB_PARTIALLY_CODED 1
58 #define SB_FULLY_CODED 2
60 // This is the maximum length of a single long bit run that can be encoded
61 // for superblock coding or block qps. Theora special-cases this to read a
62 // bit instead of flipping the current bit to allow for runs longer than 4129.
63 #define MAXIMUM_LONG_BIT_RUN 4129
65 #define MODE_INTER_NO_MV 0
67 #define MODE_INTER_PLUS_MV 2
68 #define MODE_INTER_LAST_MV 3
69 #define MODE_INTER_PRIOR_LAST 4
70 #define MODE_USING_GOLDEN 5
71 #define MODE_GOLDEN_MV 6
72 #define MODE_INTER_FOURMV 7
73 #define CODING_MODE_COUNT 8
75 /* special internal mode */
78 /* There are 6 preset schemes, plus a free-form scheme */
79 static const int ModeAlphabet[6][CODING_MODE_COUNT] =
81 /* scheme 1: Last motion vector dominates */
82 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
83 MODE_INTER_PLUS_MV, MODE_INTER_NO_MV,
84 MODE_INTRA, MODE_USING_GOLDEN,
85 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
88 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
89 MODE_INTER_NO_MV, MODE_INTER_PLUS_MV,
90 MODE_INTRA, MODE_USING_GOLDEN,
91 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
94 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
95 MODE_INTER_PRIOR_LAST, MODE_INTER_NO_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_NO_MV, MODE_INTER_PRIOR_LAST,
102 MODE_INTRA, MODE_USING_GOLDEN,
103 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
105 /* scheme 5: No motion vector dominates */
106 { MODE_INTER_NO_MV, MODE_INTER_LAST_MV,
107 MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
108 MODE_INTRA, MODE_USING_GOLDEN,
109 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
112 { MODE_INTER_NO_MV, MODE_USING_GOLDEN,
113 MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
114 MODE_INTER_PLUS_MV, MODE_INTRA,
115 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
119 static const uint8_t hilbert_offset[16][2] = {
120 {0,0}, {1,0}, {1,1}, {0,1},
121 {0,2}, {0,3}, {1,3}, {1,2},
122 {2,2}, {2,3}, {3,3}, {3,2},
123 {3,1}, {2,1}, {2,0}, {3,0}
126 #define MIN_DEQUANT_VAL 2
128 typedef struct Vp3DecodeContext {
129 AVCodecContext *avctx;
130 int theora, theora_tables;
133 int chroma_x_shift, chroma_y_shift;
134 AVFrame golden_frame;
136 AVFrame current_frame;
141 int skip_loop_filter;
147 int superblock_count;
148 int y_superblock_width;
149 int y_superblock_height;
150 int y_superblock_count;
151 int c_superblock_width;
152 int c_superblock_height;
153 int c_superblock_count;
154 int u_superblock_start;
155 int v_superblock_start;
156 unsigned char *superblock_coding;
158 int macroblock_count;
159 int macroblock_width;
160 int macroblock_height;
163 int fragment_width[2];
164 int fragment_height[2];
166 Vp3Fragment *all_fragments;
167 int fragment_start[3];
170 int8_t (*motion_val[2])[2];
175 uint16_t coded_dc_scale_factor[64];
176 uint32_t coded_ac_scale_factor[64];
177 uint8_t base_matrix[384][64];
178 uint8_t qr_count[2][3];
179 uint8_t qr_size [2][3][64];
180 uint16_t qr_base[2][3][64];
183 * This is a list of all tokens in bitstream order. Reordering takes place
184 * by pulling from each level during IDCT. As a consequence, IDCT must be
185 * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32
186 * otherwise. The 32 different tokens with up to 12 bits of extradata are
187 * collapsed into 3 types, packed as follows:
188 * (from the low to high bits)
190 * 2 bits: type (0,1,2)
191 * 0: EOB run, 14 bits for run length (12 needed)
192 * 1: zero run, 7 bits for run length
193 * 7 bits for the next coefficient (3 needed)
194 * 2: coefficient, 14 bits (11 needed)
196 * Coefficients are signed, so are packed in the highest bits for automatic
199 int16_t *dct_tokens[3][64];
200 int16_t *dct_tokens_base;
201 #define TOKEN_EOB(eob_run) ((eob_run) << 2)
202 #define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) << 9) + ((zero_run) << 2) + 1)
203 #define TOKEN_COEFF(coeff) (((coeff) << 2) + 2)
206 * number of blocks that contain DCT coefficients at the given level or higher
208 int num_coded_frags[3][64];
209 int total_num_coded_frags;
211 /* this is a list of indexes into the all_fragments array indicating
212 * which of the fragments are coded */
213 int *coded_fragment_list[3];
221 VLC superblock_run_length_vlc;
222 VLC fragment_run_length_vlc;
224 VLC motion_vector_vlc;
226 /* these arrays need to be on 16-byte boundaries since SSE2 operations
228 DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64]; ///< qmat[qpi][is_inter][plane]
230 /* This table contains superblock_count * 16 entries. Each set of 16
231 * numbers corresponds to the fragment indexes 0..15 of the superblock.
232 * An entry will be -1 to indicate that no entry corresponds to that
234 int *superblock_fragments;
236 /* This is an array that indicates how a particular macroblock
238 unsigned char *macroblock_coding;
240 uint8_t *edge_emu_buffer;
247 uint32_t huffman_table[80][32][2];
249 uint8_t filter_limit_values[64];
250 DECLARE_ALIGNED(8, int, bounding_values_array)[256+2];
253 /************************************************************************
254 * VP3 specific functions
255 ************************************************************************/
258 * This function sets up all of the various blocks mappings:
259 * superblocks <-> fragments, macroblocks <-> fragments,
260 * superblocks <-> macroblocks
262 * @return 0 is successful; returns 1 if *anything* went wrong.
264 static int init_block_mapping(Vp3DecodeContext *s)
266 int sb_x, sb_y, plane;
269 for (plane = 0; plane < 3; plane++) {
270 int sb_width = plane ? s->c_superblock_width : s->y_superblock_width;
271 int sb_height = plane ? s->c_superblock_height : s->y_superblock_height;
272 int frag_width = s->fragment_width[!!plane];
273 int frag_height = s->fragment_height[!!plane];
275 for (sb_y = 0; sb_y < sb_height; sb_y++)
276 for (sb_x = 0; sb_x < sb_width; sb_x++)
277 for (i = 0; i < 16; i++) {
278 x = 4*sb_x + hilbert_offset[i][0];
279 y = 4*sb_y + hilbert_offset[i][1];
281 if (x < frag_width && y < frag_height)
282 s->superblock_fragments[j++] = s->fragment_start[plane] + y*frag_width + x;
284 s->superblock_fragments[j++] = -1;
288 return 0; /* successful path out */
292 * This function sets up the dequantization tables used for a particular
295 static void init_dequantizer(Vp3DecodeContext *s, int qpi)
297 int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
298 int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
299 int i, plane, inter, qri, bmi, bmj, qistart;
301 for(inter=0; inter<2; inter++){
302 for(plane=0; plane<3; plane++){
304 for(qri=0; qri<s->qr_count[inter][plane]; qri++){
305 sum+= s->qr_size[inter][plane][qri];
306 if(s->qps[qpi] <= sum)
309 qistart= sum - s->qr_size[inter][plane][qri];
310 bmi= s->qr_base[inter][plane][qri ];
311 bmj= s->qr_base[inter][plane][qri+1];
313 int coeff= ( 2*(sum -s->qps[qpi])*s->base_matrix[bmi][i]
314 - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
315 + s->qr_size[inter][plane][qri])
316 / (2*s->qr_size[inter][plane][qri]);
318 int qmin= 8<<(inter + !i);
319 int qscale= i ? ac_scale_factor : dc_scale_factor;
321 s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
323 // all DC coefficients use the same quant so as not to interfere with DC prediction
324 s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
330 * This function initializes the loop filter boundary limits if the frame's
331 * quality index is different from the previous frame's.
333 * The filter_limit_values may not be larger than 127.
335 static void init_loop_filter(Vp3DecodeContext *s)
337 int *bounding_values= s->bounding_values_array+127;
342 filter_limit = s->filter_limit_values[s->qps[0]];
344 /* set up the bounding values */
345 memset(s->bounding_values_array, 0, 256 * sizeof(int));
346 for (x = 0; x < filter_limit; x++) {
347 bounding_values[-x] = -x;
348 bounding_values[x] = x;
350 for (x = value = filter_limit; x < 128 && value; x++, value--) {
351 bounding_values[ x] = value;
352 bounding_values[-x] = -value;
355 bounding_values[128] = value;
356 bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
360 * This function unpacks all of the superblock/macroblock/fragment coding
361 * information from the bitstream.
363 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
365 int superblock_starts[3] = { 0, s->u_superblock_start, s->v_superblock_start };
367 int current_superblock = 0;
369 int num_partial_superblocks = 0;
372 int current_fragment;
376 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
380 /* unpack the list of partially-coded superblocks */
381 bit = get_bits1(gb) ^ 1;
384 while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) {
385 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
390 current_run = get_vlc2(gb,
391 s->superblock_run_length_vlc.table, 6, 2) + 1;
392 if (current_run == 34)
393 current_run += get_bits(gb, 12);
395 if (current_superblock + current_run > s->superblock_count) {
396 av_log(s->avctx, AV_LOG_ERROR, "Invalid partially coded superblock run length\n");
400 memset(s->superblock_coding + current_superblock, bit, current_run);
402 current_superblock += current_run;
404 num_partial_superblocks += current_run;
407 /* unpack the list of fully coded superblocks if any of the blocks were
408 * not marked as partially coded in the previous step */
409 if (num_partial_superblocks < s->superblock_count) {
410 int superblocks_decoded = 0;
412 current_superblock = 0;
413 bit = get_bits1(gb) ^ 1;
416 while (superblocks_decoded < s->superblock_count - num_partial_superblocks
417 && get_bits_left(gb) > 0) {
419 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
424 current_run = get_vlc2(gb,
425 s->superblock_run_length_vlc.table, 6, 2) + 1;
426 if (current_run == 34)
427 current_run += get_bits(gb, 12);
429 for (j = 0; j < current_run; current_superblock++) {
430 if (current_superblock >= s->superblock_count) {
431 av_log(s->avctx, AV_LOG_ERROR, "Invalid fully coded superblock run length\n");
435 /* skip any superblocks already marked as partially coded */
436 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
437 s->superblock_coding[current_superblock] = 2*bit;
441 superblocks_decoded += current_run;
445 /* if there were partial blocks, initialize bitstream for
446 * unpacking fragment codings */
447 if (num_partial_superblocks) {
451 /* toggle the bit because as soon as the first run length is
452 * fetched the bit will be toggled again */
457 /* figure out which fragments are coded; iterate through each
458 * superblock (all planes) */
459 s->total_num_coded_frags = 0;
460 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
462 for (plane = 0; plane < 3; plane++) {
463 int sb_start = superblock_starts[plane];
464 int sb_end = sb_start + (plane ? s->c_superblock_count : s->y_superblock_count);
465 int num_coded_frags = 0;
467 for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) {
469 /* iterate through all 16 fragments in a superblock */
470 for (j = 0; j < 16; j++) {
472 /* if the fragment is in bounds, check its coding status */
473 current_fragment = s->superblock_fragments[i * 16 + j];
474 if (current_fragment != -1) {
475 int coded = s->superblock_coding[i];
477 if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
479 /* fragment may or may not be coded; this is the case
480 * that cares about the fragment coding runs */
481 if (current_run-- == 0) {
483 current_run = get_vlc2(gb,
484 s->fragment_run_length_vlc.table, 5, 2);
490 /* default mode; actual mode will be decoded in
492 s->all_fragments[current_fragment].coding_method =
494 s->coded_fragment_list[plane][num_coded_frags++] =
497 /* not coded; copy this fragment from the prior frame */
498 s->all_fragments[current_fragment].coding_method =
504 s->total_num_coded_frags += num_coded_frags;
505 for (i = 0; i < 64; i++)
506 s->num_coded_frags[plane][i] = num_coded_frags;
508 s->coded_fragment_list[plane+1] = s->coded_fragment_list[plane] + num_coded_frags;
514 * This function unpacks all the coding mode data for individual macroblocks
515 * from the bitstream.
517 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
519 int i, j, k, sb_x, sb_y;
521 int current_macroblock;
522 int current_fragment;
524 int custom_mode_alphabet[CODING_MODE_COUNT];
529 for (i = 0; i < s->fragment_count; i++)
530 s->all_fragments[i].coding_method = MODE_INTRA;
534 /* fetch the mode coding scheme for this frame */
535 scheme = get_bits(gb, 3);
537 /* is it a custom coding scheme? */
539 for (i = 0; i < 8; i++)
540 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
541 for (i = 0; i < 8; i++)
542 custom_mode_alphabet[get_bits(gb, 3)] = i;
543 alphabet = custom_mode_alphabet;
545 alphabet = ModeAlphabet[scheme-1];
547 /* iterate through all of the macroblocks that contain 1 or more
549 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
550 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
551 if (get_bits_left(gb) <= 0)
554 for (j = 0; j < 4; j++) {
555 int mb_x = 2*sb_x + (j>>1);
556 int mb_y = 2*sb_y + (((j>>1)+j)&1);
557 current_macroblock = mb_y * s->macroblock_width + mb_x;
559 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height)
562 #define BLOCK_X (2*mb_x + (k&1))
563 #define BLOCK_Y (2*mb_y + (k>>1))
564 /* coding modes are only stored if the macroblock has at least one
565 * luma block coded, otherwise it must be INTER_NO_MV */
566 for (k = 0; k < 4; k++) {
567 current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
568 if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
572 s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
576 /* mode 7 means get 3 bits for each coding mode */
578 coding_mode = get_bits(gb, 3);
580 coding_mode = alphabet
581 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
583 s->macroblock_coding[current_macroblock] = coding_mode;
584 for (k = 0; k < 4; k++) {
585 frag = s->all_fragments + BLOCK_Y*s->fragment_width[0] + BLOCK_X;
586 if (frag->coding_method != MODE_COPY)
587 frag->coding_method = coding_mode;
590 #define SET_CHROMA_MODES \
591 if (frag[s->fragment_start[1]].coding_method != MODE_COPY) \
592 frag[s->fragment_start[1]].coding_method = coding_mode;\
593 if (frag[s->fragment_start[2]].coding_method != MODE_COPY) \
594 frag[s->fragment_start[2]].coding_method = coding_mode;
596 if (s->chroma_y_shift) {
597 frag = s->all_fragments + mb_y*s->fragment_width[1] + mb_x;
599 } else if (s->chroma_x_shift) {
600 frag = s->all_fragments + 2*mb_y*s->fragment_width[1] + mb_x;
601 for (k = 0; k < 2; k++) {
603 frag += s->fragment_width[1];
606 for (k = 0; k < 4; k++) {
607 frag = s->all_fragments + BLOCK_Y*s->fragment_width[1] + BLOCK_X;
620 * This function unpacks all the motion vectors for the individual
621 * macroblocks from the bitstream.
623 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
625 int j, k, sb_x, sb_y;
629 int last_motion_x = 0;
630 int last_motion_y = 0;
631 int prior_last_motion_x = 0;
632 int prior_last_motion_y = 0;
633 int current_macroblock;
634 int current_fragment;
640 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
641 coding_mode = get_bits1(gb);
643 /* iterate through all of the macroblocks that contain 1 or more
645 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
646 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
647 if (get_bits_left(gb) <= 0)
650 for (j = 0; j < 4; j++) {
651 int mb_x = 2*sb_x + (j>>1);
652 int mb_y = 2*sb_y + (((j>>1)+j)&1);
653 current_macroblock = mb_y * s->macroblock_width + mb_x;
655 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
656 (s->macroblock_coding[current_macroblock] == MODE_COPY))
659 switch (s->macroblock_coding[current_macroblock]) {
661 case MODE_INTER_PLUS_MV:
663 /* all 6 fragments use the same motion vector */
664 if (coding_mode == 0) {
665 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
666 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
668 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
669 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
672 /* vector maintenance, only on MODE_INTER_PLUS_MV */
673 if (s->macroblock_coding[current_macroblock] ==
674 MODE_INTER_PLUS_MV) {
675 prior_last_motion_x = last_motion_x;
676 prior_last_motion_y = last_motion_y;
677 last_motion_x = motion_x[0];
678 last_motion_y = motion_y[0];
682 case MODE_INTER_FOURMV:
683 /* vector maintenance */
684 prior_last_motion_x = last_motion_x;
685 prior_last_motion_y = last_motion_y;
687 /* fetch 4 vectors from the bitstream, one for each
688 * Y fragment, then average for the C fragment vectors */
689 for (k = 0; k < 4; k++) {
690 current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
691 if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
692 if (coding_mode == 0) {
693 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
694 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
696 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
697 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
699 last_motion_x = motion_x[k];
700 last_motion_y = motion_y[k];
708 case MODE_INTER_LAST_MV:
709 /* all 6 fragments use the last motion vector */
710 motion_x[0] = last_motion_x;
711 motion_y[0] = last_motion_y;
713 /* no vector maintenance (last vector remains the
717 case MODE_INTER_PRIOR_LAST:
718 /* all 6 fragments use the motion vector prior to the
719 * last motion vector */
720 motion_x[0] = prior_last_motion_x;
721 motion_y[0] = prior_last_motion_y;
723 /* vector maintenance */
724 prior_last_motion_x = last_motion_x;
725 prior_last_motion_y = last_motion_y;
726 last_motion_x = motion_x[0];
727 last_motion_y = motion_y[0];
731 /* covers intra, inter without MV, golden without MV */
735 /* no vector maintenance */
739 /* assign the motion vectors to the correct fragments */
740 for (k = 0; k < 4; k++) {
742 BLOCK_Y*s->fragment_width[0] + BLOCK_X;
743 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
744 s->motion_val[0][current_fragment][0] = motion_x[k];
745 s->motion_val[0][current_fragment][1] = motion_y[k];
747 s->motion_val[0][current_fragment][0] = motion_x[0];
748 s->motion_val[0][current_fragment][1] = motion_y[0];
752 if (s->chroma_y_shift) {
753 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
754 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] + motion_x[2] + motion_x[3], 2);
755 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] + motion_y[2] + motion_y[3], 2);
757 motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
758 motion_y[0] = (motion_y[0]>>1) | (motion_y[0]&1);
759 frag = mb_y*s->fragment_width[1] + mb_x;
760 s->motion_val[1][frag][0] = motion_x[0];
761 s->motion_val[1][frag][1] = motion_y[0];
762 } else if (s->chroma_x_shift) {
763 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
764 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1);
765 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1);
766 motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1);
767 motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1);
769 motion_x[1] = motion_x[0];
770 motion_y[1] = motion_y[0];
772 motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
773 motion_x[1] = (motion_x[1]>>1) | (motion_x[1]&1);
775 frag = 2*mb_y*s->fragment_width[1] + mb_x;
776 for (k = 0; k < 2; k++) {
777 s->motion_val[1][frag][0] = motion_x[k];
778 s->motion_val[1][frag][1] = motion_y[k];
779 frag += s->fragment_width[1];
782 for (k = 0; k < 4; k++) {
783 frag = BLOCK_Y*s->fragment_width[1] + BLOCK_X;
784 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
785 s->motion_val[1][frag][0] = motion_x[k];
786 s->motion_val[1][frag][1] = motion_y[k];
788 s->motion_val[1][frag][0] = motion_x[0];
789 s->motion_val[1][frag][1] = motion_y[0];
800 static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
802 int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
803 int num_blocks = s->total_num_coded_frags;
805 for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
806 i = blocks_decoded = num_blocks_at_qpi = 0;
808 bit = get_bits1(gb) ^ 1;
812 if (run_length == MAXIMUM_LONG_BIT_RUN)
817 run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
818 if (run_length == 34)
819 run_length += get_bits(gb, 12);
820 blocks_decoded += run_length;
823 num_blocks_at_qpi += run_length;
825 for (j = 0; j < run_length; i++) {
826 if (i >= s->total_num_coded_frags)
829 if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
830 s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
834 } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0);
836 num_blocks -= num_blocks_at_qpi;
843 * This function is called by unpack_dct_coeffs() to extract the VLCs from
844 * the bitstream. The VLCs encode tokens which are used to unpack DCT
845 * data. This function unpacks all the VLCs for either the Y plane or both
846 * C planes, and is called for DC coefficients or different AC coefficient
847 * levels (since different coefficient types require different VLC tables.
849 * This function returns a residual eob run. E.g, if a particular token gave
850 * instructions to EOB the next 5 fragments and there were only 2 fragments
851 * left in the current fragment range, 3 would be returned so that it could
852 * be passed into the next call to this same function.
854 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
855 VLC *table, int coeff_index,
866 int num_coeffs = s->num_coded_frags[plane][coeff_index];
867 int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
869 /* local references to structure members to avoid repeated deferences */
870 int *coded_fragment_list = s->coded_fragment_list[plane];
871 Vp3Fragment *all_fragments = s->all_fragments;
872 VLC_TYPE (*vlc_table)[2] = table->table;
875 av_log(s->avctx, AV_LOG_ERROR, "Invalid number of coefficents at level %d\n", coeff_index);
877 if (eob_run > num_coeffs) {
878 coeff_i = blocks_ended = num_coeffs;
879 eob_run -= num_coeffs;
881 coeff_i = blocks_ended = eob_run;
885 // insert fake EOB token to cover the split between planes or zzi
887 dct_tokens[j++] = blocks_ended << 2;
889 while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
890 /* decode a VLC into a token */
891 token = get_vlc2(gb, vlc_table, 11, 3);
892 /* use the token to get a zero run, a coefficient, and an eob run */
894 eob_run = eob_run_base[token];
895 if (eob_run_get_bits[token])
896 eob_run += get_bits(gb, eob_run_get_bits[token]);
898 // record only the number of blocks ended in this plane,
899 // any spill will be recorded in the next plane.
900 if (eob_run > num_coeffs - coeff_i) {
901 dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
902 blocks_ended += num_coeffs - coeff_i;
903 eob_run -= num_coeffs - coeff_i;
904 coeff_i = num_coeffs;
906 dct_tokens[j++] = TOKEN_EOB(eob_run);
907 blocks_ended += eob_run;
912 bits_to_get = coeff_get_bits[token];
914 bits_to_get = get_bits(gb, bits_to_get);
915 coeff = coeff_tables[token][bits_to_get];
917 zero_run = zero_run_base[token];
918 if (zero_run_get_bits[token])
919 zero_run += get_bits(gb, zero_run_get_bits[token]);
922 dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
924 // Save DC into the fragment structure. DC prediction is
925 // done in raster order, so the actual DC can't be in with
926 // other tokens. We still need the token in dct_tokens[]
927 // however, or else the structure collapses on itself.
929 all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
931 dct_tokens[j++] = TOKEN_COEFF(coeff);
934 if (coeff_index + zero_run > 64) {
935 av_log(s->avctx, AV_LOG_DEBUG, "Invalid zero run of %d with"
936 " %d coeffs left\n", zero_run, 64-coeff_index);
937 zero_run = 64 - coeff_index;
940 // zero runs code multiple coefficients,
941 // so don't try to decode coeffs for those higher levels
942 for (i = coeff_index+1; i <= coeff_index+zero_run; i++)
943 s->num_coded_frags[plane][i]--;
948 if (blocks_ended > s->num_coded_frags[plane][coeff_index])
949 av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
951 // decrement the number of blocks that have higher coeffecients for each
952 // EOB run at this level
954 for (i = coeff_index+1; i < 64; i++)
955 s->num_coded_frags[plane][i] -= blocks_ended;
957 // setup the next buffer
959 s->dct_tokens[plane+1][coeff_index] = dct_tokens + j;
960 else if (coeff_index < 63)
961 s->dct_tokens[0][coeff_index+1] = dct_tokens + j;
966 static void reverse_dc_prediction(Vp3DecodeContext *s,
969 int fragment_height);
971 * This function unpacks all of the DCT coefficient data from the
974 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
981 int residual_eob_run = 0;
985 s->dct_tokens[0][0] = s->dct_tokens_base;
987 /* fetch the DC table indexes */
988 dc_y_table = get_bits(gb, 4);
989 dc_c_table = get_bits(gb, 4);
991 /* unpack the Y plane DC coefficients */
992 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
993 0, residual_eob_run);
995 /* reverse prediction of the Y-plane DC coefficients */
996 reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]);
998 /* unpack the C plane DC coefficients */
999 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1000 1, residual_eob_run);
1001 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1002 2, residual_eob_run);
1004 /* reverse prediction of the C-plane DC coefficients */
1005 if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1007 reverse_dc_prediction(s, s->fragment_start[1],
1008 s->fragment_width[1], s->fragment_height[1]);
1009 reverse_dc_prediction(s, s->fragment_start[2],
1010 s->fragment_width[1], s->fragment_height[1]);
1013 /* fetch the AC table indexes */
1014 ac_y_table = get_bits(gb, 4);
1015 ac_c_table = get_bits(gb, 4);
1017 /* build tables of AC VLC tables */
1018 for (i = 1; i <= 5; i++) {
1019 y_tables[i] = &s->ac_vlc_1[ac_y_table];
1020 c_tables[i] = &s->ac_vlc_1[ac_c_table];
1022 for (i = 6; i <= 14; i++) {
1023 y_tables[i] = &s->ac_vlc_2[ac_y_table];
1024 c_tables[i] = &s->ac_vlc_2[ac_c_table];
1026 for (i = 15; i <= 27; i++) {
1027 y_tables[i] = &s->ac_vlc_3[ac_y_table];
1028 c_tables[i] = &s->ac_vlc_3[ac_c_table];
1030 for (i = 28; i <= 63; i++) {
1031 y_tables[i] = &s->ac_vlc_4[ac_y_table];
1032 c_tables[i] = &s->ac_vlc_4[ac_c_table];
1035 /* decode all AC coefficents */
1036 for (i = 1; i <= 63; i++) {
1037 residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1038 0, residual_eob_run);
1040 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1041 1, residual_eob_run);
1042 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1043 2, residual_eob_run);
1050 * This function reverses the DC prediction for each coded fragment in
1051 * the frame. Much of this function is adapted directly from the original
1054 #define COMPATIBLE_FRAME(x) \
1055 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1056 #define DC_COEFF(u) s->all_fragments[u].dc
1058 static void reverse_dc_prediction(Vp3DecodeContext *s,
1061 int fragment_height)
1070 int i = first_fragment;
1074 /* DC values for the left, up-left, up, and up-right fragments */
1075 int vl, vul, vu, vur;
1077 /* indexes for the left, up-left, up, and up-right fragments */
1081 * The 6 fields mean:
1082 * 0: up-left multiplier
1084 * 2: up-right multiplier
1085 * 3: left multiplier
1087 static const int predictor_transform[16][4] = {
1089 { 0, 0, 0,128}, // PL
1090 { 0, 0,128, 0}, // PUR
1091 { 0, 0, 53, 75}, // PUR|PL
1092 { 0,128, 0, 0}, // PU
1093 { 0, 64, 0, 64}, // PU|PL
1094 { 0,128, 0, 0}, // PU|PUR
1095 { 0, 0, 53, 75}, // PU|PUR|PL
1096 {128, 0, 0, 0}, // PUL
1097 { 0, 0, 0,128}, // PUL|PL
1098 { 64, 0, 64, 0}, // PUL|PUR
1099 { 0, 0, 53, 75}, // PUL|PUR|PL
1100 { 0,128, 0, 0}, // PUL|PU
1101 {-104,116, 0,116}, // PUL|PU|PL
1102 { 24, 80, 24, 0}, // PUL|PU|PUR
1103 {-104,116, 0,116} // PUL|PU|PUR|PL
1106 /* This table shows which types of blocks can use other blocks for
1107 * prediction. For example, INTRA is the only mode in this table to
1108 * have a frame number of 0. That means INTRA blocks can only predict
1109 * from other INTRA blocks. There are 2 golden frame coding types;
1110 * blocks encoding in these modes can only predict from other blocks
1111 * that were encoded with these 1 of these 2 modes. */
1112 static const unsigned char compatible_frame[9] = {
1113 1, /* MODE_INTER_NO_MV */
1115 1, /* MODE_INTER_PLUS_MV */
1116 1, /* MODE_INTER_LAST_MV */
1117 1, /* MODE_INTER_PRIOR_MV */
1118 2, /* MODE_USING_GOLDEN */
1119 2, /* MODE_GOLDEN_MV */
1120 1, /* MODE_INTER_FOUR_MV */
1123 int current_frame_type;
1125 /* there is a last DC predictor for each of the 3 frame types */
1130 vul = vu = vur = vl = 0;
1131 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1133 /* for each fragment row... */
1134 for (y = 0; y < fragment_height; y++) {
1136 /* for each fragment in a row... */
1137 for (x = 0; x < fragment_width; x++, i++) {
1139 /* reverse prediction if this block was coded */
1140 if (s->all_fragments[i].coding_method != MODE_COPY) {
1142 current_frame_type =
1143 compatible_frame[s->all_fragments[i].coding_method];
1149 if(COMPATIBLE_FRAME(l))
1153 u= i-fragment_width;
1155 if(COMPATIBLE_FRAME(u))
1158 ul= i-fragment_width-1;
1160 if(COMPATIBLE_FRAME(ul))
1163 if(x + 1 < fragment_width){
1164 ur= i-fragment_width+1;
1166 if(COMPATIBLE_FRAME(ur))
1171 if (transform == 0) {
1173 /* if there were no fragments to predict from, use last
1175 predicted_dc = last_dc[current_frame_type];
1178 /* apply the appropriate predictor transform */
1180 (predictor_transform[transform][0] * vul) +
1181 (predictor_transform[transform][1] * vu) +
1182 (predictor_transform[transform][2] * vur) +
1183 (predictor_transform[transform][3] * vl);
1185 predicted_dc /= 128;
1187 /* check for outranging on the [ul u l] and
1188 * [ul u ur l] predictors */
1189 if ((transform == 15) || (transform == 13)) {
1190 if (FFABS(predicted_dc - vu) > 128)
1192 else if (FFABS(predicted_dc - vl) > 128)
1194 else if (FFABS(predicted_dc - vul) > 128)
1199 /* at long last, apply the predictor */
1200 DC_COEFF(i) += predicted_dc;
1202 last_dc[current_frame_type] = DC_COEFF(i);
1208 static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
1211 int *bounding_values= s->bounding_values_array+127;
1213 int width = s->fragment_width[!!plane];
1214 int height = s->fragment_height[!!plane];
1215 int fragment = s->fragment_start [plane] + ystart * width;
1216 int stride = s->current_frame.linesize[plane];
1217 uint8_t *plane_data = s->current_frame.data [plane];
1218 if (!s->flipped_image) stride = -stride;
1219 plane_data += s->data_offset[plane] + 8*ystart*stride;
1221 for (y = ystart; y < yend; y++) {
1223 for (x = 0; x < width; x++) {
1224 /* This code basically just deblocks on the edges of coded blocks.
1225 * However, it has to be much more complicated because of the
1226 * braindamaged deblock ordering used in VP3/Theora. Order matters
1227 * because some pixels get filtered twice. */
1228 if( s->all_fragments[fragment].coding_method != MODE_COPY )
1230 /* do not perform left edge filter for left columns frags */
1232 s->dsp.vp3_h_loop_filter(
1234 stride, bounding_values);
1237 /* do not perform top edge filter for top row fragments */
1239 s->dsp.vp3_v_loop_filter(
1241 stride, bounding_values);
1244 /* do not perform right edge filter for right column
1245 * fragments or if right fragment neighbor is also coded
1246 * in this frame (it will be filtered in next iteration) */
1247 if ((x < width - 1) &&
1248 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1249 s->dsp.vp3_h_loop_filter(
1250 plane_data + 8*x + 8,
1251 stride, bounding_values);
1254 /* do not perform bottom edge filter for bottom row
1255 * fragments or if bottom fragment neighbor is also coded
1256 * in this frame (it will be filtered in the next row) */
1257 if ((y < height - 1) &&
1258 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1259 s->dsp.vp3_v_loop_filter(
1260 plane_data + 8*x + 8*stride,
1261 stride, bounding_values);
1267 plane_data += 8*stride;
1272 * Pull DCT tokens from the 64 levels to decode and dequant the coefficients
1273 * for the next block in coding order
1275 static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1276 int plane, int inter, DCTELEM block[64])
1278 int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1279 uint8_t *perm = s->scantable.permutated;
1283 int token = *s->dct_tokens[plane][i];
1284 switch (token & 3) {
1286 if (--token < 4) // 0-3 are token types, so the EOB run must now be 0
1287 s->dct_tokens[plane][i]++;
1289 *s->dct_tokens[plane][i] = token & ~3;
1292 s->dct_tokens[plane][i]++;
1293 i += (token >> 2) & 0x7f;
1294 block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1298 block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1299 s->dct_tokens[plane][i++]++;
1301 default: // shouldn't happen
1306 // the actual DC+prediction is in the fragment structure
1307 block[0] = frag->dc * s->qmat[0][inter][plane][0];
1312 * called when all pixels up to row y are complete
1314 static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1319 if (HAVE_PTHREADS && s->avctx->active_thread_type&FF_THREAD_FRAME) {
1320 int y_flipped = s->flipped_image ? s->avctx->height-y : y;
1322 // At the end of the frame, report INT_MAX instead of the height of the frame.
1323 // This makes the other threads' ff_thread_await_progress() calls cheaper, because
1324 // they don't have to clip their values.
1325 ff_thread_report_progress(&s->current_frame, y_flipped==s->avctx->height ? INT_MAX : y_flipped-1, 0);
1328 if(s->avctx->draw_horiz_band==NULL)
1331 h= y - s->last_slice_end;
1332 s->last_slice_end= y;
1335 if (!s->flipped_image) {
1336 y = s->avctx->height - y - h;
1339 cy = y >> s->chroma_y_shift;
1340 offset[0] = s->current_frame.linesize[0]*y;
1341 offset[1] = s->current_frame.linesize[1]*cy;
1342 offset[2] = s->current_frame.linesize[2]*cy;
1346 s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
1350 * Wait for the reference frame of the current fragment.
1351 * The progress value is in luma pixel rows.
1353 static void await_reference_row(Vp3DecodeContext *s, Vp3Fragment *fragment, int motion_y, int y)
1357 int border = motion_y&1;
1359 if (fragment->coding_method == MODE_USING_GOLDEN ||
1360 fragment->coding_method == MODE_GOLDEN_MV)
1361 ref_frame = &s->golden_frame;
1363 ref_frame = &s->last_frame;
1365 ref_row = y + (motion_y>>1);
1366 ref_row = FFMAX(FFABS(ref_row), ref_row + 8 + border);
1368 ff_thread_await_progress(ref_frame, ref_row, 0);
1372 * Perform the final rendering for a particular slice of data.
1373 * The slice number ranges from 0..(c_superblock_height - 1).
1375 static void render_slice(Vp3DecodeContext *s, int slice)
1377 int x, y, i, j, fragment;
1378 LOCAL_ALIGNED_16(DCTELEM, block, [64]);
1379 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1380 int motion_halfpel_index;
1381 uint8_t *motion_source;
1382 int plane, first_pixel;
1384 if (slice >= s->c_superblock_height)
1387 for (plane = 0; plane < 3; plane++) {
1388 uint8_t *output_plane = s->current_frame.data [plane] + s->data_offset[plane];
1389 uint8_t * last_plane = s-> last_frame.data [plane] + s->data_offset[plane];
1390 uint8_t *golden_plane = s-> golden_frame.data [plane] + s->data_offset[plane];
1391 int stride = s->current_frame.linesize[plane];
1392 int plane_width = s->width >> (plane && s->chroma_x_shift);
1393 int plane_height = s->height >> (plane && s->chroma_y_shift);
1394 int8_t (*motion_val)[2] = s->motion_val[!!plane];
1396 int sb_x, sb_y = slice << (!plane && s->chroma_y_shift);
1397 int slice_height = sb_y + 1 + (!plane && s->chroma_y_shift);
1398 int slice_width = plane ? s->c_superblock_width : s->y_superblock_width;
1400 int fragment_width = s->fragment_width[!!plane];
1401 int fragment_height = s->fragment_height[!!plane];
1402 int fragment_start = s->fragment_start[plane];
1403 int do_await = !plane && HAVE_PTHREADS && (s->avctx->active_thread_type&FF_THREAD_FRAME);
1405 if (!s->flipped_image) stride = -stride;
1406 if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
1409 /* for each superblock row in the slice (both of them)... */
1410 for (; sb_y < slice_height; sb_y++) {
1412 /* for each superblock in a row... */
1413 for (sb_x = 0; sb_x < slice_width; sb_x++) {
1415 /* for each block in a superblock... */
1416 for (j = 0; j < 16; j++) {
1417 x = 4*sb_x + hilbert_offset[j][0];
1418 y = 4*sb_y + hilbert_offset[j][1];
1419 fragment = y*fragment_width + x;
1421 i = fragment_start + fragment;
1424 if (x >= fragment_width || y >= fragment_height)
1427 first_pixel = 8*y*stride + 8*x;
1429 if (do_await && s->all_fragments[i].coding_method != MODE_INTRA)
1430 await_reference_row(s, &s->all_fragments[i], motion_val[fragment][1], (16*y) >> s->chroma_y_shift);
1432 /* transform if this block was coded */
1433 if (s->all_fragments[i].coding_method != MODE_COPY) {
1434 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1435 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1436 motion_source= golden_plane;
1438 motion_source= last_plane;
1440 motion_source += first_pixel;
1441 motion_halfpel_index = 0;
1443 /* sort out the motion vector if this fragment is coded
1444 * using a motion vector method */
1445 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1446 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1448 motion_x = motion_val[fragment][0];
1449 motion_y = motion_val[fragment][1];
1451 src_x= (motion_x>>1) + 8*x;
1452 src_y= (motion_y>>1) + 8*y;
1454 motion_halfpel_index = motion_x & 0x01;
1455 motion_source += (motion_x >> 1);
1457 motion_halfpel_index |= (motion_y & 0x01) << 1;
1458 motion_source += ((motion_y >> 1) * stride);
1460 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1461 uint8_t *temp= s->edge_emu_buffer;
1462 if(stride<0) temp -= 8*stride;
1464 s->dsp.emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1465 motion_source= temp;
1470 /* first, take care of copying a block from either the
1471 * previous or the golden frame */
1472 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1473 /* Note, it is possible to implement all MC cases with
1474 put_no_rnd_pixels_l2 which would look more like the
1475 VP3 source but this would be slower as
1476 put_no_rnd_pixels_tab is better optimzed */
1477 if(motion_halfpel_index != 3){
1478 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1479 output_plane + first_pixel,
1480 motion_source, stride, 8);
1482 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1483 s->dsp.put_no_rnd_pixels_l2[1](
1484 output_plane + first_pixel,
1486 motion_source + stride + 1 + d,
1491 s->dsp.clear_block(block);
1493 /* invert DCT and place (or add) in final output */
1495 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1496 vp3_dequant(s, s->all_fragments + i, plane, 0, block);
1497 if(s->avctx->idct_algo!=FF_IDCT_VP3)
1500 output_plane + first_pixel,
1504 if (vp3_dequant(s, s->all_fragments + i, plane, 1, block)) {
1506 output_plane + first_pixel,
1510 s->dsp.vp3_idct_dc_add(output_plane + first_pixel, stride, block);
1515 /* copy directly from the previous frame */
1516 s->dsp.put_pixels_tab[1][0](
1517 output_plane + first_pixel,
1518 last_plane + first_pixel,
1525 // Filter up to the last row in the superblock row
1526 if (!s->skip_loop_filter)
1527 apply_loop_filter(s, plane, 4*sb_y - !!sb_y, FFMIN(4*sb_y+3, fragment_height-1));
1531 /* this looks like a good place for slice dispatch... */
1533 * if (slice == s->macroblock_height - 1)
1534 * dispatch (both last slice & 2nd-to-last slice);
1535 * else if (slice > 0)
1536 * dispatch (slice - 1);
1539 vp3_draw_horiz_band(s, FFMIN((32 << s->chroma_y_shift) * (slice + 1) -16, s->height-16));
1542 /// Allocate tables for per-frame data in Vp3DecodeContext
1543 static av_cold int allocate_tables(AVCodecContext *avctx)
1545 Vp3DecodeContext *s = avctx->priv_data;
1546 int y_fragment_count, c_fragment_count;
1548 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1549 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1551 s->superblock_coding = av_malloc(s->superblock_count);
1552 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1553 s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int));
1554 s->dct_tokens_base = av_malloc(64*s->fragment_count * sizeof(*s->dct_tokens_base));
1555 s->motion_val[0] = av_malloc(y_fragment_count * sizeof(*s->motion_val[0]));
1556 s->motion_val[1] = av_malloc(c_fragment_count * sizeof(*s->motion_val[1]));
1558 /* work out the block mapping tables */
1559 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1560 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1562 if (!s->superblock_coding || !s->all_fragments || !s->dct_tokens_base ||
1563 !s->coded_fragment_list[0] || !s->superblock_fragments || !s->macroblock_coding ||
1564 !s->motion_val[0] || !s->motion_val[1]) {
1565 vp3_decode_end(avctx);
1569 init_block_mapping(s);
1575 * This is the ffmpeg/libavcodec API init function.
1577 static av_cold int vp3_decode_init(AVCodecContext *avctx)
1579 Vp3DecodeContext *s = avctx->priv_data;
1580 int i, inter, plane;
1583 int y_fragment_count, c_fragment_count;
1585 if (avctx->codec_tag == MKTAG('V','P','3','0'))
1591 s->width = FFALIGN(avctx->width, 16);
1592 s->height = FFALIGN(avctx->height, 16);
1593 if (avctx->pix_fmt == PIX_FMT_NONE)
1594 avctx->pix_fmt = PIX_FMT_YUV420P;
1595 avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1596 if(avctx->idct_algo==FF_IDCT_AUTO)
1597 avctx->idct_algo=FF_IDCT_VP3;
1598 dsputil_init(&s->dsp, avctx);
1600 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1602 /* initialize to an impossible value which will force a recalculation
1603 * in the first frame decode */
1604 for (i = 0; i < 3; i++)
1607 avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift);
1609 s->y_superblock_width = (s->width + 31) / 32;
1610 s->y_superblock_height = (s->height + 31) / 32;
1611 s->y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1613 /* work out the dimensions for the C planes */
1614 c_width = s->width >> s->chroma_x_shift;
1615 c_height = s->height >> s->chroma_y_shift;
1616 s->c_superblock_width = (c_width + 31) / 32;
1617 s->c_superblock_height = (c_height + 31) / 32;
1618 s->c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1620 s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
1621 s->u_superblock_start = s->y_superblock_count;
1622 s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
1624 s->macroblock_width = (s->width + 15) / 16;
1625 s->macroblock_height = (s->height + 15) / 16;
1626 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1628 s->fragment_width[0] = s->width / FRAGMENT_PIXELS;
1629 s->fragment_height[0] = s->height / FRAGMENT_PIXELS;
1630 s->fragment_width[1] = s->fragment_width[0] >> s->chroma_x_shift;
1631 s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift;
1633 /* fragment count covers all 8x8 blocks for all 3 planes */
1634 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1635 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1636 s->fragment_count = y_fragment_count + 2*c_fragment_count;
1637 s->fragment_start[1] = y_fragment_count;
1638 s->fragment_start[2] = y_fragment_count + c_fragment_count;
1640 if (!s->theora_tables)
1642 for (i = 0; i < 64; i++) {
1643 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1644 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1645 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1646 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1647 s->base_matrix[2][i] = vp31_inter_dequant[i];
1648 s->filter_limit_values[i] = vp31_filter_limit_values[i];
1651 for(inter=0; inter<2; inter++){
1652 for(plane=0; plane<3; plane++){
1653 s->qr_count[inter][plane]= 1;
1654 s->qr_size [inter][plane][0]= 63;
1655 s->qr_base [inter][plane][0]=
1656 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1660 /* init VLC tables */
1661 for (i = 0; i < 16; i++) {
1664 init_vlc(&s->dc_vlc[i], 11, 32,
1665 &dc_bias[i][0][1], 4, 2,
1666 &dc_bias[i][0][0], 4, 2, 0);
1668 /* group 1 AC histograms */
1669 init_vlc(&s->ac_vlc_1[i], 11, 32,
1670 &ac_bias_0[i][0][1], 4, 2,
1671 &ac_bias_0[i][0][0], 4, 2, 0);
1673 /* group 2 AC histograms */
1674 init_vlc(&s->ac_vlc_2[i], 11, 32,
1675 &ac_bias_1[i][0][1], 4, 2,
1676 &ac_bias_1[i][0][0], 4, 2, 0);
1678 /* group 3 AC histograms */
1679 init_vlc(&s->ac_vlc_3[i], 11, 32,
1680 &ac_bias_2[i][0][1], 4, 2,
1681 &ac_bias_2[i][0][0], 4, 2, 0);
1683 /* group 4 AC histograms */
1684 init_vlc(&s->ac_vlc_4[i], 11, 32,
1685 &ac_bias_3[i][0][1], 4, 2,
1686 &ac_bias_3[i][0][0], 4, 2, 0);
1690 for (i = 0; i < 16; i++) {
1692 if (init_vlc(&s->dc_vlc[i], 11, 32,
1693 &s->huffman_table[i][0][1], 8, 4,
1694 &s->huffman_table[i][0][0], 8, 4, 0) < 0)
1697 /* group 1 AC histograms */
1698 if (init_vlc(&s->ac_vlc_1[i], 11, 32,
1699 &s->huffman_table[i+16][0][1], 8, 4,
1700 &s->huffman_table[i+16][0][0], 8, 4, 0) < 0)
1703 /* group 2 AC histograms */
1704 if (init_vlc(&s->ac_vlc_2[i], 11, 32,
1705 &s->huffman_table[i+16*2][0][1], 8, 4,
1706 &s->huffman_table[i+16*2][0][0], 8, 4, 0) < 0)
1709 /* group 3 AC histograms */
1710 if (init_vlc(&s->ac_vlc_3[i], 11, 32,
1711 &s->huffman_table[i+16*3][0][1], 8, 4,
1712 &s->huffman_table[i+16*3][0][0], 8, 4, 0) < 0)
1715 /* group 4 AC histograms */
1716 if (init_vlc(&s->ac_vlc_4[i], 11, 32,
1717 &s->huffman_table[i+16*4][0][1], 8, 4,
1718 &s->huffman_table[i+16*4][0][0], 8, 4, 0) < 0)
1723 init_vlc(&s->superblock_run_length_vlc, 6, 34,
1724 &superblock_run_length_vlc_table[0][1], 4, 2,
1725 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1727 init_vlc(&s->fragment_run_length_vlc, 5, 30,
1728 &fragment_run_length_vlc_table[0][1], 4, 2,
1729 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1731 init_vlc(&s->mode_code_vlc, 3, 8,
1732 &mode_code_vlc_table[0][1], 2, 1,
1733 &mode_code_vlc_table[0][0], 2, 1, 0);
1735 init_vlc(&s->motion_vector_vlc, 6, 63,
1736 &motion_vector_vlc_table[0][1], 2, 1,
1737 &motion_vector_vlc_table[0][0], 2, 1, 0);
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;
1745 return allocate_tables(avctx);
1748 av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1752 /// Release and shuffle frames after decode finishes
1753 static void update_frames(AVCodecContext *avctx)
1755 Vp3DecodeContext *s = avctx->priv_data;
1757 /* release the last frame, if it is allocated and if it is not the
1759 if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
1760 ff_thread_release_buffer(avctx, &s->last_frame);
1762 /* shuffle frames (last = current) */
1763 s->last_frame= s->current_frame;
1766 if (s->golden_frame.data[0])
1767 ff_thread_release_buffer(avctx, &s->golden_frame);
1768 s->golden_frame = s->current_frame;
1769 s->last_frame.type = FF_BUFFER_TYPE_COPY;
1772 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
1775 static int vp3_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
1777 Vp3DecodeContext *s = dst->priv_data, *s1 = src->priv_data;
1778 int qps_changed = 0, i, err;
1780 if (!s1->current_frame.data[0]
1781 ||s->width != s1->width
1782 ||s->height!= s1->height)
1786 // init tables if the first frame hasn't been decoded
1787 if (!s->current_frame.data[0]) {
1788 int y_fragment_count, c_fragment_count;
1790 err = allocate_tables(dst);
1793 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1794 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1795 memcpy(s->motion_val[0], s1->motion_val[0], y_fragment_count * sizeof(*s->motion_val[0]));
1796 memcpy(s->motion_val[1], s1->motion_val[1], c_fragment_count * sizeof(*s->motion_val[1]));
1799 #define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
1801 // copy previous frame data
1802 copy_fields(s, s1, golden_frame, dsp);
1804 // copy qscale data if necessary
1805 for (i = 0; i < 3; i++) {
1806 if (s->qps[i] != s1->qps[1]) {
1808 memcpy(&s->qmat[i], &s1->qmat[i], sizeof(s->qmat[i]));
1812 if (s->qps[0] != s1->qps[0])
1813 memcpy(&s->bounding_values_array, &s1->bounding_values_array, sizeof(s->bounding_values_array));
1816 copy_fields(s, s1, qps, superblock_count);
1826 * This is the ffmpeg/libavcodec API frame decode function.
1828 static int vp3_decode_frame(AVCodecContext *avctx,
1829 void *data, int *data_size,
1832 const uint8_t *buf = avpkt->data;
1833 int buf_size = avpkt->size;
1834 Vp3DecodeContext *s = avctx->priv_data;
1838 init_get_bits(&gb, buf, buf_size * 8);
1840 if (s->theora && get_bits1(&gb))
1842 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1846 s->keyframe = !get_bits1(&gb);
1849 for (i = 0; i < 3; i++)
1850 s->last_qps[i] = s->qps[i];
1854 s->qps[s->nqps++]= get_bits(&gb, 6);
1855 } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1856 for (i = s->nqps; i < 3; i++)
1859 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1860 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1861 s->keyframe?"key":"", avctx->frame_number+1, s->qps[0]);
1863 s->skip_loop_filter = !s->filter_limit_values[s->qps[0]] ||
1864 avctx->skip_loop_filter >= (s->keyframe ? AVDISCARD_ALL : AVDISCARD_NONKEY);
1866 if (s->qps[0] != s->last_qps[0])
1867 init_loop_filter(s);
1869 for (i = 0; i < s->nqps; i++)
1870 // reinit all dequantizers if the first one changed, because
1871 // the DC of the first quantizer must be used for all matrices
1872 if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1873 init_dequantizer(s, i);
1875 if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1878 s->current_frame.reference = 3;
1879 s->current_frame.pict_type = s->keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P;
1880 s->current_frame.key_frame = s->keyframe;
1881 if (ff_thread_get_buffer(avctx, &s->current_frame) < 0) {
1882 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1886 if (!s->edge_emu_buffer)
1887 s->edge_emu_buffer = av_malloc(9*FFABS(s->current_frame.linesize[0]));
1892 skip_bits(&gb, 4); /* width code */
1893 skip_bits(&gb, 4); /* height code */
1896 s->version = get_bits(&gb, 5);
1897 if (avctx->frame_number == 0)
1898 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1901 if (s->version || s->theora)
1904 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1905 skip_bits(&gb, 2); /* reserved? */
1908 if (!s->golden_frame.data[0]) {
1909 av_log(s->avctx, AV_LOG_WARNING, "vp3: first frame not a keyframe\n");
1911 s->golden_frame.reference = 3;
1912 s->golden_frame.pict_type = AV_PICTURE_TYPE_I;
1913 if (ff_thread_get_buffer(avctx, &s->golden_frame) < 0) {
1914 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1917 s->last_frame = s->golden_frame;
1918 s->last_frame.type = FF_BUFFER_TYPE_COPY;
1919 ff_thread_report_progress(&s->last_frame, INT_MAX, 0);
1923 memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
1924 ff_thread_finish_setup(avctx);
1926 if (unpack_superblocks(s, &gb)){
1927 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
1930 if (unpack_modes(s, &gb)){
1931 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
1934 if (unpack_vectors(s, &gb)){
1935 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
1938 if (unpack_block_qpis(s, &gb)){
1939 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
1942 if (unpack_dct_coeffs(s, &gb)){
1943 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
1947 for (i = 0; i < 3; i++) {
1948 int height = s->height >> (i && s->chroma_y_shift);
1949 if (s->flipped_image)
1950 s->data_offset[i] = 0;
1952 s->data_offset[i] = (height-1) * s->current_frame.linesize[i];
1955 s->last_slice_end = 0;
1956 for (i = 0; i < s->c_superblock_height; i++)
1959 // filter the last row
1960 for (i = 0; i < 3; i++) {
1961 int row = (s->height >> (3+(i && s->chroma_y_shift))) - 1;
1962 apply_loop_filter(s, i, row, row+1);
1964 vp3_draw_horiz_band(s, s->avctx->height);
1966 *data_size=sizeof(AVFrame);
1967 *(AVFrame*)data= s->current_frame;
1969 if (!HAVE_PTHREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME))
1970 update_frames(avctx);
1975 ff_thread_report_progress(&s->current_frame, INT_MAX, 0);
1977 if (!HAVE_PTHREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME))
1978 avctx->release_buffer(avctx, &s->current_frame);
1984 * This is the ffmpeg/libavcodec API module cleanup function.
1986 static av_cold int vp3_decode_end(AVCodecContext *avctx)
1988 Vp3DecodeContext *s = avctx->priv_data;
1991 if (avctx->is_copy && !s->current_frame.data[0])
1994 av_free(s->superblock_coding);
1995 av_free(s->all_fragments);
1996 av_free(s->coded_fragment_list[0]);
1997 av_free(s->dct_tokens_base);
1998 av_free(s->superblock_fragments);
1999 av_free(s->macroblock_coding);
2000 av_free(s->motion_val[0]);
2001 av_free(s->motion_val[1]);
2002 av_free(s->edge_emu_buffer);
2004 if (avctx->is_copy) return 0;
2006 for (i = 0; i < 16; i++) {
2007 free_vlc(&s->dc_vlc[i]);
2008 free_vlc(&s->ac_vlc_1[i]);
2009 free_vlc(&s->ac_vlc_2[i]);
2010 free_vlc(&s->ac_vlc_3[i]);
2011 free_vlc(&s->ac_vlc_4[i]);
2014 free_vlc(&s->superblock_run_length_vlc);
2015 free_vlc(&s->fragment_run_length_vlc);
2016 free_vlc(&s->mode_code_vlc);
2017 free_vlc(&s->motion_vector_vlc);
2019 /* release all frames */
2020 if (s->golden_frame.data[0])
2021 ff_thread_release_buffer(avctx, &s->golden_frame);
2022 if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
2023 ff_thread_release_buffer(avctx, &s->last_frame);
2024 /* no need to release the current_frame since it will always be pointing
2025 * to the same frame as either the golden or last frame */
2030 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2032 Vp3DecodeContext *s = avctx->priv_data;
2034 if (get_bits1(gb)) {
2036 if (s->entries >= 32) { /* overflow */
2037 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2040 token = get_bits(gb, 5);
2041 //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);
2042 s->huffman_table[s->hti][token][0] = s->hbits;
2043 s->huffman_table[s->hti][token][1] = s->huff_code_size;
2047 if (s->huff_code_size >= 32) {/* overflow */
2048 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2051 s->huff_code_size++;
2053 if (read_huffman_tree(avctx, gb))
2056 if (read_huffman_tree(avctx, gb))
2059 s->huff_code_size--;
2064 #if CONFIG_THEORA_DECODER
2065 static const enum PixelFormat theora_pix_fmts[4] = {
2066 PIX_FMT_YUV420P, PIX_FMT_NONE, PIX_FMT_YUV422P, PIX_FMT_YUV444P
2069 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2071 Vp3DecodeContext *s = avctx->priv_data;
2072 int visible_width, visible_height, colorspace;
2073 int offset_x = 0, offset_y = 0;
2074 AVRational fps, aspect;
2076 s->theora = get_bits_long(gb, 24);
2077 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2079 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2080 /* but previous versions have the image flipped relative to vp3 */
2081 if (s->theora < 0x030200)
2083 s->flipped_image = 1;
2084 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2087 visible_width = s->width = get_bits(gb, 16) << 4;
2088 visible_height = s->height = get_bits(gb, 16) << 4;
2090 if(av_image_check_size(s->width, s->height, 0, avctx)){
2091 av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2092 s->width= s->height= 0;
2096 if (s->theora >= 0x030200) {
2097 visible_width = get_bits_long(gb, 24);
2098 visible_height = get_bits_long(gb, 24);
2100 offset_x = get_bits(gb, 8); /* offset x */
2101 offset_y = get_bits(gb, 8); /* offset y, from bottom */
2104 fps.num = get_bits_long(gb, 32);
2105 fps.den = get_bits_long(gb, 32);
2106 if (fps.num && fps.den) {
2107 av_reduce(&avctx->time_base.num, &avctx->time_base.den,
2108 fps.den, fps.num, 1<<30);
2111 aspect.num = get_bits_long(gb, 24);
2112 aspect.den = get_bits_long(gb, 24);
2113 if (aspect.num && aspect.den) {
2114 av_reduce(&avctx->sample_aspect_ratio.num,
2115 &avctx->sample_aspect_ratio.den,
2116 aspect.num, aspect.den, 1<<30);
2119 if (s->theora < 0x030200)
2120 skip_bits(gb, 5); /* keyframe frequency force */
2121 colorspace = get_bits(gb, 8);
2122 skip_bits(gb, 24); /* bitrate */
2124 skip_bits(gb, 6); /* quality hint */
2126 if (s->theora >= 0x030200)
2128 skip_bits(gb, 5); /* keyframe frequency force */
2129 avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)];
2130 skip_bits(gb, 3); /* reserved */
2133 // align_get_bits(gb);
2135 if ( visible_width <= s->width && visible_width > s->width-16
2136 && visible_height <= s->height && visible_height > s->height-16
2137 && !offset_x && (offset_y == s->height - visible_height))
2138 avcodec_set_dimensions(avctx, visible_width, visible_height);
2140 avcodec_set_dimensions(avctx, s->width, s->height);
2142 if (colorspace == 1) {
2143 avctx->color_primaries = AVCOL_PRI_BT470M;
2144 } else if (colorspace == 2) {
2145 avctx->color_primaries = AVCOL_PRI_BT470BG;
2147 if (colorspace == 1 || colorspace == 2) {
2148 avctx->colorspace = AVCOL_SPC_BT470BG;
2149 avctx->color_trc = AVCOL_TRC_BT709;
2155 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2157 Vp3DecodeContext *s = avctx->priv_data;
2158 int i, n, matrices, inter, plane;
2160 if (s->theora >= 0x030200) {
2161 n = get_bits(gb, 3);
2162 /* loop filter limit values table */
2164 for (i = 0; i < 64; i++)
2165 s->filter_limit_values[i] = get_bits(gb, n);
2168 if (s->theora >= 0x030200)
2169 n = get_bits(gb, 4) + 1;
2172 /* quality threshold table */
2173 for (i = 0; i < 64; i++)
2174 s->coded_ac_scale_factor[i] = get_bits(gb, n);
2176 if (s->theora >= 0x030200)
2177 n = get_bits(gb, 4) + 1;
2180 /* dc scale factor table */
2181 for (i = 0; i < 64; i++)
2182 s->coded_dc_scale_factor[i] = get_bits(gb, n);
2184 if (s->theora >= 0x030200)
2185 matrices = get_bits(gb, 9) + 1;
2190 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2194 for(n=0; n<matrices; n++){
2195 for (i = 0; i < 64; i++)
2196 s->base_matrix[n][i]= get_bits(gb, 8);
2199 for (inter = 0; inter <= 1; inter++) {
2200 for (plane = 0; plane <= 2; plane++) {
2202 if (inter || plane > 0)
2203 newqr = get_bits1(gb);
2206 if(inter && get_bits1(gb)){
2210 qtj= (3*inter + plane - 1) / 3;
2211 plj= (plane + 2) % 3;
2213 s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2214 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2215 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2221 i= get_bits(gb, av_log2(matrices-1)+1);
2223 av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2226 s->qr_base[inter][plane][qri]= i;
2229 i = get_bits(gb, av_log2(63-qi)+1) + 1;
2230 s->qr_size[inter][plane][qri++]= i;
2235 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2238 s->qr_count[inter][plane]= qri;
2243 /* Huffman tables */
2244 for (s->hti = 0; s->hti < 80; s->hti++) {
2246 s->huff_code_size = 1;
2247 if (!get_bits1(gb)) {
2249 if(read_huffman_tree(avctx, gb))
2252 if(read_huffman_tree(avctx, gb))
2257 s->theora_tables = 1;
2262 static av_cold int theora_decode_init(AVCodecContext *avctx)
2264 Vp3DecodeContext *s = avctx->priv_data;
2267 uint8_t *header_start[3];
2273 if (!avctx->extradata_size)
2275 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2279 if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2280 42, header_start, header_len) < 0) {
2281 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2286 init_get_bits(&gb, header_start[i], header_len[i] * 8);
2288 ptype = get_bits(&gb, 8);
2290 if (!(ptype & 0x80))
2292 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2296 // FIXME: Check for this as well.
2297 skip_bits_long(&gb, 6*8); /* "theora" */
2302 theora_decode_header(avctx, &gb);
2305 // FIXME: is this needed? it breaks sometimes
2306 // theora_decode_comments(avctx, gb);
2309 if (theora_decode_tables(avctx, &gb))
2313 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2316 if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2317 av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2318 if (s->theora < 0x030200)
2322 return vp3_decode_init(avctx);
2325 static void vp3_decode_flush(AVCodecContext *avctx)
2327 Vp3DecodeContext *s = avctx->priv_data;
2329 if (s->golden_frame.data[0]) {
2330 if (s->golden_frame.data[0] == s->last_frame.data[0])
2331 memset(&s->last_frame, 0, sizeof(AVFrame));
2332 if (s->current_frame.data[0] == s->golden_frame.data[0])
2333 memset(&s->current_frame, 0, sizeof(AVFrame));
2334 ff_thread_release_buffer(avctx, &s->golden_frame);
2336 if (s->last_frame.data[0]) {
2337 if (s->current_frame.data[0] == s->last_frame.data[0])
2338 memset(&s->current_frame, 0, sizeof(AVFrame));
2339 ff_thread_release_buffer(avctx, &s->last_frame);
2341 if (s->current_frame.data[0])
2342 ff_thread_release_buffer(avctx, &s->current_frame);
2345 AVCodec ff_theora_decoder = {
2347 .type = AVMEDIA_TYPE_VIDEO,
2348 .id = CODEC_ID_THEORA,
2349 .priv_data_size = sizeof(Vp3DecodeContext),
2350 .init = theora_decode_init,
2351 .close = vp3_decode_end,
2352 .decode = vp3_decode_frame,
2353 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND | CODEC_CAP_FRAME_THREADS,
2354 .flush = vp3_decode_flush,
2355 .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2356 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context)
2360 AVCodec ff_vp3_decoder = {
2362 .type = AVMEDIA_TYPE_VIDEO,
2364 .priv_data_size = sizeof(Vp3DecodeContext),
2365 .init = vp3_decode_init,
2366 .close = vp3_decode_end,
2367 .decode = vp3_decode_frame,
2368 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND | CODEC_CAP_FRAME_THREADS,
2369 .flush = vp3_decode_flush,
2370 .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2371 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context)