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"
46 #define FRAGMENT_PIXELS 8
48 static av_cold int vp3_decode_end(AVCodecContext *avctx);
49 static void vp3_decode_flush(AVCodecContext *avctx);
51 //FIXME split things out into their own arrays
52 typedef struct Vp3Fragment {
54 uint8_t coding_method;
58 #define SB_NOT_CODED 0
59 #define SB_PARTIALLY_CODED 1
60 #define SB_FULLY_CODED 2
62 // This is the maximum length of a single long bit run that can be encoded
63 // for superblock coding or block qps. Theora special-cases this to read a
64 // bit instead of flipping the current bit to allow for runs longer than 4129.
65 #define MAXIMUM_LONG_BIT_RUN 4129
67 #define MODE_INTER_NO_MV 0
69 #define MODE_INTER_PLUS_MV 2
70 #define MODE_INTER_LAST_MV 3
71 #define MODE_INTER_PRIOR_LAST 4
72 #define MODE_USING_GOLDEN 5
73 #define MODE_GOLDEN_MV 6
74 #define MODE_INTER_FOURMV 7
75 #define CODING_MODE_COUNT 8
77 /* special internal mode */
80 /* There are 6 preset schemes, plus a free-form scheme */
81 static const int ModeAlphabet[6][CODING_MODE_COUNT] =
83 /* scheme 1: Last motion vector dominates */
84 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
85 MODE_INTER_PLUS_MV, MODE_INTER_NO_MV,
86 MODE_INTRA, MODE_USING_GOLDEN,
87 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
90 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
91 MODE_INTER_NO_MV, MODE_INTER_PLUS_MV,
92 MODE_INTRA, MODE_USING_GOLDEN,
93 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
96 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
97 MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
98 MODE_INTRA, MODE_USING_GOLDEN,
99 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
102 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
103 MODE_INTER_NO_MV, MODE_INTER_PRIOR_LAST,
104 MODE_INTRA, MODE_USING_GOLDEN,
105 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
107 /* scheme 5: No motion vector dominates */
108 { MODE_INTER_NO_MV, MODE_INTER_LAST_MV,
109 MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
110 MODE_INTRA, MODE_USING_GOLDEN,
111 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
114 { MODE_INTER_NO_MV, MODE_USING_GOLDEN,
115 MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
116 MODE_INTER_PLUS_MV, MODE_INTRA,
117 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
121 static const uint8_t hilbert_offset[16][2] = {
122 {0,0}, {1,0}, {1,1}, {0,1},
123 {0,2}, {0,3}, {1,3}, {1,2},
124 {2,2}, {2,3}, {3,3}, {3,2},
125 {3,1}, {2,1}, {2,0}, {3,0}
128 #define MIN_DEQUANT_VAL 2
130 typedef struct Vp3DecodeContext {
131 AVCodecContext *avctx;
132 int theora, theora_tables;
135 int chroma_x_shift, chroma_y_shift;
136 AVFrame golden_frame;
138 AVFrame current_frame;
143 int skip_loop_filter;
149 int superblock_count;
150 int y_superblock_width;
151 int y_superblock_height;
152 int y_superblock_count;
153 int c_superblock_width;
154 int c_superblock_height;
155 int c_superblock_count;
156 int u_superblock_start;
157 int v_superblock_start;
158 unsigned char *superblock_coding;
160 int macroblock_count;
161 int macroblock_width;
162 int macroblock_height;
165 int fragment_width[2];
166 int fragment_height[2];
168 Vp3Fragment *all_fragments;
169 int fragment_start[3];
172 int8_t (*motion_val[2])[2];
177 uint16_t coded_dc_scale_factor[64];
178 uint32_t coded_ac_scale_factor[64];
179 uint8_t base_matrix[384][64];
180 uint8_t qr_count[2][3];
181 uint8_t qr_size [2][3][64];
182 uint16_t qr_base[2][3][64];
185 * This is a list of all tokens in bitstream order. Reordering takes place
186 * by pulling from each level during IDCT. As a consequence, IDCT must be
187 * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32
188 * otherwise. The 32 different tokens with up to 12 bits of extradata are
189 * collapsed into 3 types, packed as follows:
190 * (from the low to high bits)
192 * 2 bits: type (0,1,2)
193 * 0: EOB run, 14 bits for run length (12 needed)
194 * 1: zero run, 7 bits for run length
195 * 7 bits for the next coefficient (3 needed)
196 * 2: coefficient, 14 bits (11 needed)
198 * Coefficients are signed, so are packed in the highest bits for automatic
201 int16_t *dct_tokens[3][64];
202 int16_t *dct_tokens_base;
203 #define TOKEN_EOB(eob_run) ((eob_run) << 2)
204 #define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) << 9) + ((zero_run) << 2) + 1)
205 #define TOKEN_COEFF(coeff) (((coeff) << 2) + 2)
208 * number of blocks that contain DCT coefficients at the given level or higher
210 int num_coded_frags[3][64];
211 int total_num_coded_frags;
213 /* this is a list of indexes into the all_fragments array indicating
214 * which of the fragments are coded */
215 int *coded_fragment_list[3];
223 VLC superblock_run_length_vlc;
224 VLC fragment_run_length_vlc;
226 VLC motion_vector_vlc;
228 /* these arrays need to be on 16-byte boundaries since SSE2 operations
230 DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64]; ///< qmat[qpi][is_inter][plane]
232 /* This table contains superblock_count * 16 entries. Each set of 16
233 * numbers corresponds to the fragment indexes 0..15 of the superblock.
234 * An entry will be -1 to indicate that no entry corresponds to that
236 int *superblock_fragments;
238 /* This is an array that indicates how a particular macroblock
240 unsigned char *macroblock_coding;
242 uint8_t *edge_emu_buffer;
249 uint32_t huffman_table[80][32][2];
251 uint8_t filter_limit_values[64];
252 DECLARE_ALIGNED(8, int, bounding_values_array)[256+2];
255 /************************************************************************
256 * VP3 specific functions
257 ************************************************************************/
260 * This function sets up all of the various blocks mappings:
261 * superblocks <-> fragments, macroblocks <-> fragments,
262 * superblocks <-> macroblocks
264 * @return 0 is successful; returns 1 if *anything* went wrong.
266 static int init_block_mapping(Vp3DecodeContext *s)
268 int sb_x, sb_y, plane;
271 for (plane = 0; plane < 3; plane++) {
272 int sb_width = plane ? s->c_superblock_width : s->y_superblock_width;
273 int sb_height = plane ? s->c_superblock_height : s->y_superblock_height;
274 int frag_width = s->fragment_width[!!plane];
275 int frag_height = s->fragment_height[!!plane];
277 for (sb_y = 0; sb_y < sb_height; sb_y++)
278 for (sb_x = 0; sb_x < sb_width; sb_x++)
279 for (i = 0; i < 16; i++) {
280 x = 4*sb_x + hilbert_offset[i][0];
281 y = 4*sb_y + hilbert_offset[i][1];
283 if (x < frag_width && y < frag_height)
284 s->superblock_fragments[j++] = s->fragment_start[plane] + y*frag_width + x;
286 s->superblock_fragments[j++] = -1;
290 return 0; /* successful path out */
294 * This function sets up the dequantization tables used for a particular
297 static void init_dequantizer(Vp3DecodeContext *s, int qpi)
299 int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
300 int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
301 int i, plane, inter, qri, bmi, bmj, qistart;
303 for(inter=0; inter<2; inter++){
304 for(plane=0; plane<3; plane++){
306 for(qri=0; qri<s->qr_count[inter][plane]; qri++){
307 sum+= s->qr_size[inter][plane][qri];
308 if(s->qps[qpi] <= sum)
311 qistart= sum - s->qr_size[inter][plane][qri];
312 bmi= s->qr_base[inter][plane][qri ];
313 bmj= s->qr_base[inter][plane][qri+1];
315 int coeff= ( 2*(sum -s->qps[qpi])*s->base_matrix[bmi][i]
316 - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
317 + s->qr_size[inter][plane][qri])
318 / (2*s->qr_size[inter][plane][qri]);
320 int qmin= 8<<(inter + !i);
321 int qscale= i ? ac_scale_factor : dc_scale_factor;
323 s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
325 // all DC coefficients use the same quant so as not to interfere with DC prediction
326 s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
332 * This function initializes the loop filter boundary limits if the frame's
333 * quality index is different from the previous frame's.
335 * The filter_limit_values may not be larger than 127.
337 static void init_loop_filter(Vp3DecodeContext *s)
339 int *bounding_values= s->bounding_values_array+127;
344 filter_limit = s->filter_limit_values[s->qps[0]];
346 /* set up the bounding values */
347 memset(s->bounding_values_array, 0, 256 * sizeof(int));
348 for (x = 0; x < filter_limit; x++) {
349 bounding_values[-x] = -x;
350 bounding_values[x] = x;
352 for (x = value = filter_limit; x < 128 && value; x++, value--) {
353 bounding_values[ x] = value;
354 bounding_values[-x] = -value;
357 bounding_values[128] = value;
358 bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
362 * This function unpacks all of the superblock/macroblock/fragment coding
363 * information from the bitstream.
365 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
367 int superblock_starts[3] = { 0, s->u_superblock_start, s->v_superblock_start };
369 int current_superblock = 0;
371 int num_partial_superblocks = 0;
374 int current_fragment;
378 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
382 /* unpack the list of partially-coded superblocks */
383 bit = get_bits1(gb) ^ 1;
386 while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) {
387 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
392 current_run = get_vlc2(gb,
393 s->superblock_run_length_vlc.table, 6, 2) + 1;
394 if (current_run == 34)
395 current_run += get_bits(gb, 12);
397 if (current_superblock + current_run > s->superblock_count) {
398 av_log(s->avctx, AV_LOG_ERROR, "Invalid partially coded superblock run length\n");
402 memset(s->superblock_coding + current_superblock, bit, current_run);
404 current_superblock += current_run;
406 num_partial_superblocks += current_run;
409 /* unpack the list of fully coded superblocks if any of the blocks were
410 * not marked as partially coded in the previous step */
411 if (num_partial_superblocks < s->superblock_count) {
412 int superblocks_decoded = 0;
414 current_superblock = 0;
415 bit = get_bits1(gb) ^ 1;
418 while (superblocks_decoded < s->superblock_count - num_partial_superblocks
419 && get_bits_left(gb) > 0) {
421 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
426 current_run = get_vlc2(gb,
427 s->superblock_run_length_vlc.table, 6, 2) + 1;
428 if (current_run == 34)
429 current_run += get_bits(gb, 12);
431 for (j = 0; j < current_run; current_superblock++) {
432 if (current_superblock >= s->superblock_count) {
433 av_log(s->avctx, AV_LOG_ERROR, "Invalid fully coded superblock run length\n");
437 /* skip any superblocks already marked as partially coded */
438 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
439 s->superblock_coding[current_superblock] = 2*bit;
443 superblocks_decoded += current_run;
447 /* if there were partial blocks, initialize bitstream for
448 * unpacking fragment codings */
449 if (num_partial_superblocks) {
453 /* toggle the bit because as soon as the first run length is
454 * fetched the bit will be toggled again */
459 /* figure out which fragments are coded; iterate through each
460 * superblock (all planes) */
461 s->total_num_coded_frags = 0;
462 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
464 for (plane = 0; plane < 3; plane++) {
465 int sb_start = superblock_starts[plane];
466 int sb_end = sb_start + (plane ? s->c_superblock_count : s->y_superblock_count);
467 int num_coded_frags = 0;
469 for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) {
471 /* iterate through all 16 fragments in a superblock */
472 for (j = 0; j < 16; j++) {
474 /* if the fragment is in bounds, check its coding status */
475 current_fragment = s->superblock_fragments[i * 16 + j];
476 if (current_fragment != -1) {
477 int coded = s->superblock_coding[i];
479 if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
481 /* fragment may or may not be coded; this is the case
482 * that cares about the fragment coding runs */
483 if (current_run-- == 0) {
485 current_run = get_vlc2(gb,
486 s->fragment_run_length_vlc.table, 5, 2);
492 /* default mode; actual mode will be decoded in
494 s->all_fragments[current_fragment].coding_method =
496 s->coded_fragment_list[plane][num_coded_frags++] =
499 /* not coded; copy this fragment from the prior frame */
500 s->all_fragments[current_fragment].coding_method =
506 s->total_num_coded_frags += num_coded_frags;
507 for (i = 0; i < 64; i++)
508 s->num_coded_frags[plane][i] = num_coded_frags;
510 s->coded_fragment_list[plane+1] = s->coded_fragment_list[plane] + num_coded_frags;
516 * This function unpacks all the coding mode data for individual macroblocks
517 * from the bitstream.
519 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
521 int i, j, k, sb_x, sb_y;
523 int current_macroblock;
524 int current_fragment;
526 int custom_mode_alphabet[CODING_MODE_COUNT];
531 for (i = 0; i < s->fragment_count; i++)
532 s->all_fragments[i].coding_method = MODE_INTRA;
536 /* fetch the mode coding scheme for this frame */
537 scheme = get_bits(gb, 3);
539 /* is it a custom coding scheme? */
541 for (i = 0; i < 8; i++)
542 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
543 for (i = 0; i < 8; i++)
544 custom_mode_alphabet[get_bits(gb, 3)] = i;
545 alphabet = custom_mode_alphabet;
547 alphabet = ModeAlphabet[scheme-1];
549 /* iterate through all of the macroblocks that contain 1 or more
551 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
552 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
553 if (get_bits_left(gb) <= 0)
556 for (j = 0; j < 4; j++) {
557 int mb_x = 2*sb_x + (j>>1);
558 int mb_y = 2*sb_y + (((j>>1)+j)&1);
559 current_macroblock = mb_y * s->macroblock_width + mb_x;
561 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height)
564 #define BLOCK_X (2*mb_x + (k&1))
565 #define BLOCK_Y (2*mb_y + (k>>1))
566 /* coding modes are only stored if the macroblock has at least one
567 * luma block coded, otherwise it must be INTER_NO_MV */
568 for (k = 0; k < 4; k++) {
569 current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
570 if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
574 s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
578 /* mode 7 means get 3 bits for each coding mode */
580 coding_mode = get_bits(gb, 3);
582 coding_mode = alphabet
583 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
585 s->macroblock_coding[current_macroblock] = coding_mode;
586 for (k = 0; k < 4; k++) {
587 frag = s->all_fragments + BLOCK_Y*s->fragment_width[0] + BLOCK_X;
588 if (frag->coding_method != MODE_COPY)
589 frag->coding_method = coding_mode;
592 #define SET_CHROMA_MODES \
593 if (frag[s->fragment_start[1]].coding_method != MODE_COPY) \
594 frag[s->fragment_start[1]].coding_method = coding_mode;\
595 if (frag[s->fragment_start[2]].coding_method != MODE_COPY) \
596 frag[s->fragment_start[2]].coding_method = coding_mode;
598 if (s->chroma_y_shift) {
599 frag = s->all_fragments + mb_y*s->fragment_width[1] + mb_x;
601 } else if (s->chroma_x_shift) {
602 frag = s->all_fragments + 2*mb_y*s->fragment_width[1] + mb_x;
603 for (k = 0; k < 2; k++) {
605 frag += s->fragment_width[1];
608 for (k = 0; k < 4; k++) {
609 frag = s->all_fragments + BLOCK_Y*s->fragment_width[1] + BLOCK_X;
622 * This function unpacks all the motion vectors for the individual
623 * macroblocks from the bitstream.
625 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
627 int j, k, sb_x, sb_y;
631 int last_motion_x = 0;
632 int last_motion_y = 0;
633 int prior_last_motion_x = 0;
634 int prior_last_motion_y = 0;
635 int current_macroblock;
636 int current_fragment;
642 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
643 coding_mode = get_bits1(gb);
645 /* iterate through all of the macroblocks that contain 1 or more
647 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
648 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
649 if (get_bits_left(gb) <= 0)
652 for (j = 0; j < 4; j++) {
653 int mb_x = 2*sb_x + (j>>1);
654 int mb_y = 2*sb_y + (((j>>1)+j)&1);
655 current_macroblock = mb_y * s->macroblock_width + mb_x;
657 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
658 (s->macroblock_coding[current_macroblock] == MODE_COPY))
661 switch (s->macroblock_coding[current_macroblock]) {
663 case MODE_INTER_PLUS_MV:
665 /* all 6 fragments use the same motion vector */
666 if (coding_mode == 0) {
667 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
668 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
670 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
671 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
674 /* vector maintenance, only on MODE_INTER_PLUS_MV */
675 if (s->macroblock_coding[current_macroblock] ==
676 MODE_INTER_PLUS_MV) {
677 prior_last_motion_x = last_motion_x;
678 prior_last_motion_y = last_motion_y;
679 last_motion_x = motion_x[0];
680 last_motion_y = motion_y[0];
684 case MODE_INTER_FOURMV:
685 /* vector maintenance */
686 prior_last_motion_x = last_motion_x;
687 prior_last_motion_y = last_motion_y;
689 /* fetch 4 vectors from the bitstream, one for each
690 * Y fragment, then average for the C fragment vectors */
691 for (k = 0; k < 4; k++) {
692 current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
693 if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
694 if (coding_mode == 0) {
695 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
696 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
698 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
699 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
701 last_motion_x = motion_x[k];
702 last_motion_y = motion_y[k];
710 case MODE_INTER_LAST_MV:
711 /* all 6 fragments use the last motion vector */
712 motion_x[0] = last_motion_x;
713 motion_y[0] = last_motion_y;
715 /* no vector maintenance (last vector remains the
719 case MODE_INTER_PRIOR_LAST:
720 /* all 6 fragments use the motion vector prior to the
721 * last motion vector */
722 motion_x[0] = prior_last_motion_x;
723 motion_y[0] = prior_last_motion_y;
725 /* vector maintenance */
726 prior_last_motion_x = last_motion_x;
727 prior_last_motion_y = last_motion_y;
728 last_motion_x = motion_x[0];
729 last_motion_y = motion_y[0];
733 /* covers intra, inter without MV, golden without MV */
737 /* no vector maintenance */
741 /* assign the motion vectors to the correct fragments */
742 for (k = 0; k < 4; k++) {
744 BLOCK_Y*s->fragment_width[0] + BLOCK_X;
745 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
746 s->motion_val[0][current_fragment][0] = motion_x[k];
747 s->motion_val[0][current_fragment][1] = motion_y[k];
749 s->motion_val[0][current_fragment][0] = motion_x[0];
750 s->motion_val[0][current_fragment][1] = motion_y[0];
754 if (s->chroma_y_shift) {
755 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
756 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] + motion_x[2] + motion_x[3], 2);
757 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] + motion_y[2] + motion_y[3], 2);
759 motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
760 motion_y[0] = (motion_y[0]>>1) | (motion_y[0]&1);
761 frag = mb_y*s->fragment_width[1] + mb_x;
762 s->motion_val[1][frag][0] = motion_x[0];
763 s->motion_val[1][frag][1] = motion_y[0];
764 } else if (s->chroma_x_shift) {
765 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
766 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1);
767 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1);
768 motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1);
769 motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1);
771 motion_x[1] = motion_x[0];
772 motion_y[1] = motion_y[0];
774 motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
775 motion_x[1] = (motion_x[1]>>1) | (motion_x[1]&1);
777 frag = 2*mb_y*s->fragment_width[1] + mb_x;
778 for (k = 0; k < 2; k++) {
779 s->motion_val[1][frag][0] = motion_x[k];
780 s->motion_val[1][frag][1] = motion_y[k];
781 frag += s->fragment_width[1];
784 for (k = 0; k < 4; k++) {
785 frag = BLOCK_Y*s->fragment_width[1] + BLOCK_X;
786 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
787 s->motion_val[1][frag][0] = motion_x[k];
788 s->motion_val[1][frag][1] = motion_y[k];
790 s->motion_val[1][frag][0] = motion_x[0];
791 s->motion_val[1][frag][1] = motion_y[0];
802 static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
804 int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
805 int num_blocks = s->total_num_coded_frags;
807 for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
808 i = blocks_decoded = num_blocks_at_qpi = 0;
810 bit = get_bits1(gb) ^ 1;
814 if (run_length == MAXIMUM_LONG_BIT_RUN)
819 run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
820 if (run_length == 34)
821 run_length += get_bits(gb, 12);
822 blocks_decoded += run_length;
825 num_blocks_at_qpi += run_length;
827 for (j = 0; j < run_length; i++) {
828 if (i >= s->total_num_coded_frags)
831 if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
832 s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
836 } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0);
838 num_blocks -= num_blocks_at_qpi;
845 * This function is called by unpack_dct_coeffs() to extract the VLCs from
846 * the bitstream. The VLCs encode tokens which are used to unpack DCT
847 * data. This function unpacks all the VLCs for either the Y plane or both
848 * C planes, and is called for DC coefficients or different AC coefficient
849 * levels (since different coefficient types require different VLC tables.
851 * This function returns a residual eob run. E.g, if a particular token gave
852 * instructions to EOB the next 5 fragments and there were only 2 fragments
853 * left in the current fragment range, 3 would be returned so that it could
854 * be passed into the next call to this same function.
856 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
857 VLC *table, int coeff_index,
868 int num_coeffs = s->num_coded_frags[plane][coeff_index];
869 int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
871 /* local references to structure members to avoid repeated deferences */
872 int *coded_fragment_list = s->coded_fragment_list[plane];
873 Vp3Fragment *all_fragments = s->all_fragments;
874 VLC_TYPE (*vlc_table)[2] = table->table;
877 av_log(s->avctx, AV_LOG_ERROR, "Invalid number of coefficents at level %d\n", coeff_index);
879 if (eob_run > num_coeffs) {
880 coeff_i = blocks_ended = num_coeffs;
881 eob_run -= num_coeffs;
883 coeff_i = blocks_ended = eob_run;
887 // insert fake EOB token to cover the split between planes or zzi
889 dct_tokens[j++] = blocks_ended << 2;
891 while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
892 /* decode a VLC into a token */
893 token = get_vlc2(gb, vlc_table, 11, 3);
894 /* use the token to get a zero run, a coefficient, and an eob run */
895 if ((unsigned) token <= 6U) {
896 eob_run = eob_run_base[token];
897 if (eob_run_get_bits[token])
898 eob_run += get_bits(gb, eob_run_get_bits[token]);
900 // record only the number of blocks ended in this plane,
901 // any spill will be recorded in the next plane.
902 if (eob_run > num_coeffs - coeff_i) {
903 dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
904 blocks_ended += num_coeffs - coeff_i;
905 eob_run -= num_coeffs - coeff_i;
906 coeff_i = num_coeffs;
908 dct_tokens[j++] = TOKEN_EOB(eob_run);
909 blocks_ended += eob_run;
913 } else if (token >= 0) {
914 bits_to_get = coeff_get_bits[token];
916 bits_to_get = get_bits(gb, bits_to_get);
917 coeff = coeff_tables[token][bits_to_get];
919 zero_run = zero_run_base[token];
920 if (zero_run_get_bits[token])
921 zero_run += get_bits(gb, zero_run_get_bits[token]);
924 dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
926 // Save DC into the fragment structure. DC prediction is
927 // done in raster order, so the actual DC can't be in with
928 // other tokens. We still need the token in dct_tokens[]
929 // however, or else the structure collapses on itself.
931 all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
933 dct_tokens[j++] = TOKEN_COEFF(coeff);
936 if (coeff_index + zero_run > 64) {
937 av_log(s->avctx, AV_LOG_DEBUG, "Invalid zero run of %d with"
938 " %d coeffs left\n", zero_run, 64-coeff_index);
939 zero_run = 64 - coeff_index;
942 // zero runs code multiple coefficients,
943 // so don't try to decode coeffs for those higher levels
944 for (i = coeff_index+1; i <= coeff_index+zero_run; i++)
945 s->num_coded_frags[plane][i]--;
948 av_log(s->avctx, AV_LOG_ERROR,
949 "Invalid token %d\n", token);
954 if (blocks_ended > s->num_coded_frags[plane][coeff_index])
955 av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
957 // decrement the number of blocks that have higher coeffecients for each
958 // EOB run at this level
960 for (i = coeff_index+1; i < 64; i++)
961 s->num_coded_frags[plane][i] -= blocks_ended;
963 // setup the next buffer
965 s->dct_tokens[plane+1][coeff_index] = dct_tokens + j;
966 else if (coeff_index < 63)
967 s->dct_tokens[0][coeff_index+1] = dct_tokens + j;
972 static void reverse_dc_prediction(Vp3DecodeContext *s,
975 int fragment_height);
977 * This function unpacks all of the DCT coefficient data from the
980 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
987 int residual_eob_run = 0;
991 s->dct_tokens[0][0] = s->dct_tokens_base;
993 /* fetch the DC table indexes */
994 dc_y_table = get_bits(gb, 4);
995 dc_c_table = get_bits(gb, 4);
997 /* unpack the Y plane DC coefficients */
998 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
999 0, residual_eob_run);
1000 if (residual_eob_run < 0)
1001 return residual_eob_run;
1003 /* reverse prediction of the Y-plane DC coefficients */
1004 reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]);
1006 /* unpack the C plane DC coefficients */
1007 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1008 1, residual_eob_run);
1009 if (residual_eob_run < 0)
1010 return residual_eob_run;
1011 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1012 2, residual_eob_run);
1013 if (residual_eob_run < 0)
1014 return residual_eob_run;
1016 /* reverse prediction of the C-plane DC coefficients */
1017 if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1019 reverse_dc_prediction(s, s->fragment_start[1],
1020 s->fragment_width[1], s->fragment_height[1]);
1021 reverse_dc_prediction(s, s->fragment_start[2],
1022 s->fragment_width[1], s->fragment_height[1]);
1025 /* fetch the AC table indexes */
1026 ac_y_table = get_bits(gb, 4);
1027 ac_c_table = get_bits(gb, 4);
1029 /* build tables of AC VLC tables */
1030 for (i = 1; i <= 5; i++) {
1031 y_tables[i] = &s->ac_vlc_1[ac_y_table];
1032 c_tables[i] = &s->ac_vlc_1[ac_c_table];
1034 for (i = 6; i <= 14; i++) {
1035 y_tables[i] = &s->ac_vlc_2[ac_y_table];
1036 c_tables[i] = &s->ac_vlc_2[ac_c_table];
1038 for (i = 15; i <= 27; i++) {
1039 y_tables[i] = &s->ac_vlc_3[ac_y_table];
1040 c_tables[i] = &s->ac_vlc_3[ac_c_table];
1042 for (i = 28; i <= 63; i++) {
1043 y_tables[i] = &s->ac_vlc_4[ac_y_table];
1044 c_tables[i] = &s->ac_vlc_4[ac_c_table];
1047 /* decode all AC coefficents */
1048 for (i = 1; i <= 63; i++) {
1049 residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1050 0, residual_eob_run);
1051 if (residual_eob_run < 0)
1052 return residual_eob_run;
1054 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1055 1, residual_eob_run);
1056 if (residual_eob_run < 0)
1057 return residual_eob_run;
1058 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1059 2, residual_eob_run);
1060 if (residual_eob_run < 0)
1061 return residual_eob_run;
1068 * This function reverses the DC prediction for each coded fragment in
1069 * the frame. Much of this function is adapted directly from the original
1072 #define COMPATIBLE_FRAME(x) \
1073 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1074 #define DC_COEFF(u) s->all_fragments[u].dc
1076 static void reverse_dc_prediction(Vp3DecodeContext *s,
1079 int fragment_height)
1088 int i = first_fragment;
1092 /* DC values for the left, up-left, up, and up-right fragments */
1093 int vl, vul, vu, vur;
1095 /* indexes for the left, up-left, up, and up-right fragments */
1099 * The 6 fields mean:
1100 * 0: up-left multiplier
1102 * 2: up-right multiplier
1103 * 3: left multiplier
1105 static const int predictor_transform[16][4] = {
1107 { 0, 0, 0,128}, // PL
1108 { 0, 0,128, 0}, // PUR
1109 { 0, 0, 53, 75}, // PUR|PL
1110 { 0,128, 0, 0}, // PU
1111 { 0, 64, 0, 64}, // PU|PL
1112 { 0,128, 0, 0}, // PU|PUR
1113 { 0, 0, 53, 75}, // PU|PUR|PL
1114 {128, 0, 0, 0}, // PUL
1115 { 0, 0, 0,128}, // PUL|PL
1116 { 64, 0, 64, 0}, // PUL|PUR
1117 { 0, 0, 53, 75}, // PUL|PUR|PL
1118 { 0,128, 0, 0}, // PUL|PU
1119 {-104,116, 0,116}, // PUL|PU|PL
1120 { 24, 80, 24, 0}, // PUL|PU|PUR
1121 {-104,116, 0,116} // PUL|PU|PUR|PL
1124 /* This table shows which types of blocks can use other blocks for
1125 * prediction. For example, INTRA is the only mode in this table to
1126 * have a frame number of 0. That means INTRA blocks can only predict
1127 * from other INTRA blocks. There are 2 golden frame coding types;
1128 * blocks encoding in these modes can only predict from other blocks
1129 * that were encoded with these 1 of these 2 modes. */
1130 static const unsigned char compatible_frame[9] = {
1131 1, /* MODE_INTER_NO_MV */
1133 1, /* MODE_INTER_PLUS_MV */
1134 1, /* MODE_INTER_LAST_MV */
1135 1, /* MODE_INTER_PRIOR_MV */
1136 2, /* MODE_USING_GOLDEN */
1137 2, /* MODE_GOLDEN_MV */
1138 1, /* MODE_INTER_FOUR_MV */
1141 int current_frame_type;
1143 /* there is a last DC predictor for each of the 3 frame types */
1148 vul = vu = vur = vl = 0;
1149 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1151 /* for each fragment row... */
1152 for (y = 0; y < fragment_height; y++) {
1154 /* for each fragment in a row... */
1155 for (x = 0; x < fragment_width; x++, i++) {
1157 /* reverse prediction if this block was coded */
1158 if (s->all_fragments[i].coding_method != MODE_COPY) {
1160 current_frame_type =
1161 compatible_frame[s->all_fragments[i].coding_method];
1167 if(COMPATIBLE_FRAME(l))
1171 u= i-fragment_width;
1173 if(COMPATIBLE_FRAME(u))
1176 ul= i-fragment_width-1;
1178 if(COMPATIBLE_FRAME(ul))
1181 if(x + 1 < fragment_width){
1182 ur= i-fragment_width+1;
1184 if(COMPATIBLE_FRAME(ur))
1189 if (transform == 0) {
1191 /* if there were no fragments to predict from, use last
1193 predicted_dc = last_dc[current_frame_type];
1196 /* apply the appropriate predictor transform */
1198 (predictor_transform[transform][0] * vul) +
1199 (predictor_transform[transform][1] * vu) +
1200 (predictor_transform[transform][2] * vur) +
1201 (predictor_transform[transform][3] * vl);
1203 predicted_dc /= 128;
1205 /* check for outranging on the [ul u l] and
1206 * [ul u ur l] predictors */
1207 if ((transform == 15) || (transform == 13)) {
1208 if (FFABS(predicted_dc - vu) > 128)
1210 else if (FFABS(predicted_dc - vl) > 128)
1212 else if (FFABS(predicted_dc - vul) > 128)
1217 /* at long last, apply the predictor */
1218 DC_COEFF(i) += predicted_dc;
1220 last_dc[current_frame_type] = DC_COEFF(i);
1226 static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
1229 int *bounding_values= s->bounding_values_array+127;
1231 int width = s->fragment_width[!!plane];
1232 int height = s->fragment_height[!!plane];
1233 int fragment = s->fragment_start [plane] + ystart * width;
1234 int stride = s->current_frame.linesize[plane];
1235 uint8_t *plane_data = s->current_frame.data [plane];
1236 if (!s->flipped_image) stride = -stride;
1237 plane_data += s->data_offset[plane] + 8*ystart*stride;
1239 for (y = ystart; y < yend; y++) {
1241 for (x = 0; x < width; x++) {
1242 /* This code basically just deblocks on the edges of coded blocks.
1243 * However, it has to be much more complicated because of the
1244 * braindamaged deblock ordering used in VP3/Theora. Order matters
1245 * because some pixels get filtered twice. */
1246 if( s->all_fragments[fragment].coding_method != MODE_COPY )
1248 /* do not perform left edge filter for left columns frags */
1250 s->dsp.vp3_h_loop_filter(
1252 stride, bounding_values);
1255 /* do not perform top edge filter for top row fragments */
1257 s->dsp.vp3_v_loop_filter(
1259 stride, bounding_values);
1262 /* do not perform right edge filter for right column
1263 * fragments or if right fragment neighbor is also coded
1264 * in this frame (it will be filtered in next iteration) */
1265 if ((x < width - 1) &&
1266 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1267 s->dsp.vp3_h_loop_filter(
1268 plane_data + 8*x + 8,
1269 stride, bounding_values);
1272 /* do not perform bottom edge filter for bottom row
1273 * fragments or if bottom fragment neighbor is also coded
1274 * in this frame (it will be filtered in the next row) */
1275 if ((y < height - 1) &&
1276 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1277 s->dsp.vp3_v_loop_filter(
1278 plane_data + 8*x + 8*stride,
1279 stride, bounding_values);
1285 plane_data += 8*stride;
1290 * Pull DCT tokens from the 64 levels to decode and dequant the coefficients
1291 * for the next block in coding order
1293 static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1294 int plane, int inter, DCTELEM block[64])
1296 int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1297 uint8_t *perm = s->scantable.permutated;
1301 int token = *s->dct_tokens[plane][i];
1302 switch (token & 3) {
1304 if (--token < 4) // 0-3 are token types, so the EOB run must now be 0
1305 s->dct_tokens[plane][i]++;
1307 *s->dct_tokens[plane][i] = token & ~3;
1310 s->dct_tokens[plane][i]++;
1311 i += (token >> 2) & 0x7f;
1313 av_log(s->avctx, AV_LOG_ERROR, "Coefficient index overflow\n");
1316 block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1320 block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1321 s->dct_tokens[plane][i++]++;
1323 default: // shouldn't happen
1328 // the actual DC+prediction is in the fragment structure
1329 block[0] = frag->dc * s->qmat[0][inter][plane][0];
1334 * called when all pixels up to row y are complete
1336 static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1339 int offset[AV_NUM_DATA_POINTERS];
1341 if (HAVE_THREADS && s->avctx->active_thread_type&FF_THREAD_FRAME) {
1342 int y_flipped = s->flipped_image ? s->avctx->height-y : y;
1344 // At the end of the frame, report INT_MAX instead of the height of the frame.
1345 // This makes the other threads' ff_thread_await_progress() calls cheaper, because
1346 // they don't have to clip their values.
1347 ff_thread_report_progress(&s->current_frame, y_flipped==s->avctx->height ? INT_MAX : y_flipped-1, 0);
1350 if(s->avctx->draw_horiz_band==NULL)
1353 h= y - s->last_slice_end;
1354 s->last_slice_end= y;
1357 if (!s->flipped_image) {
1358 y = s->avctx->height - y - h;
1361 cy = y >> s->chroma_y_shift;
1362 offset[0] = s->current_frame.linesize[0]*y;
1363 offset[1] = s->current_frame.linesize[1]*cy;
1364 offset[2] = s->current_frame.linesize[2]*cy;
1365 for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
1369 s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
1373 * Wait for the reference frame of the current fragment.
1374 * The progress value is in luma pixel rows.
1376 static void await_reference_row(Vp3DecodeContext *s, Vp3Fragment *fragment, int motion_y, int y)
1380 int border = motion_y&1;
1382 if (fragment->coding_method == MODE_USING_GOLDEN ||
1383 fragment->coding_method == MODE_GOLDEN_MV)
1384 ref_frame = &s->golden_frame;
1386 ref_frame = &s->last_frame;
1388 ref_row = y + (motion_y>>1);
1389 ref_row = FFMAX(FFABS(ref_row), ref_row + 8 + border);
1391 ff_thread_await_progress(ref_frame, ref_row, 0);
1395 * Perform the final rendering for a particular slice of data.
1396 * The slice number ranges from 0..(c_superblock_height - 1).
1398 static void render_slice(Vp3DecodeContext *s, int slice)
1400 int x, y, i, j, fragment;
1401 LOCAL_ALIGNED_16(DCTELEM, block, [64]);
1402 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1403 int motion_halfpel_index;
1404 uint8_t *motion_source;
1405 int plane, first_pixel;
1407 if (slice >= s->c_superblock_height)
1410 for (plane = 0; plane < 3; plane++) {
1411 uint8_t *output_plane = s->current_frame.data [plane] + s->data_offset[plane];
1412 uint8_t * last_plane = s-> last_frame.data [plane] + s->data_offset[plane];
1413 uint8_t *golden_plane = s-> golden_frame.data [plane] + s->data_offset[plane];
1414 int stride = s->current_frame.linesize[plane];
1415 int plane_width = s->width >> (plane && s->chroma_x_shift);
1416 int plane_height = s->height >> (plane && s->chroma_y_shift);
1417 int8_t (*motion_val)[2] = s->motion_val[!!plane];
1419 int sb_x, sb_y = slice << (!plane && s->chroma_y_shift);
1420 int slice_height = sb_y + 1 + (!plane && s->chroma_y_shift);
1421 int slice_width = plane ? s->c_superblock_width : s->y_superblock_width;
1423 int fragment_width = s->fragment_width[!!plane];
1424 int fragment_height = s->fragment_height[!!plane];
1425 int fragment_start = s->fragment_start[plane];
1426 int do_await = !plane && HAVE_THREADS && (s->avctx->active_thread_type&FF_THREAD_FRAME);
1428 if (!s->flipped_image) stride = -stride;
1429 if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
1432 /* for each superblock row in the slice (both of them)... */
1433 for (; sb_y < slice_height; sb_y++) {
1435 /* for each superblock in a row... */
1436 for (sb_x = 0; sb_x < slice_width; sb_x++) {
1438 /* for each block in a superblock... */
1439 for (j = 0; j < 16; j++) {
1440 x = 4*sb_x + hilbert_offset[j][0];
1441 y = 4*sb_y + hilbert_offset[j][1];
1442 fragment = y*fragment_width + x;
1444 i = fragment_start + fragment;
1447 if (x >= fragment_width || y >= fragment_height)
1450 first_pixel = 8*y*stride + 8*x;
1452 if (do_await && s->all_fragments[i].coding_method != MODE_INTRA)
1453 await_reference_row(s, &s->all_fragments[i], motion_val[fragment][1], (16*y) >> s->chroma_y_shift);
1455 /* transform if this block was coded */
1456 if (s->all_fragments[i].coding_method != MODE_COPY) {
1457 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1458 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1459 motion_source= golden_plane;
1461 motion_source= last_plane;
1463 motion_source += first_pixel;
1464 motion_halfpel_index = 0;
1466 /* sort out the motion vector if this fragment is coded
1467 * using a motion vector method */
1468 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1469 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1471 motion_x = motion_val[fragment][0];
1472 motion_y = motion_val[fragment][1];
1474 src_x= (motion_x>>1) + 8*x;
1475 src_y= (motion_y>>1) + 8*y;
1477 motion_halfpel_index = motion_x & 0x01;
1478 motion_source += (motion_x >> 1);
1480 motion_halfpel_index |= (motion_y & 0x01) << 1;
1481 motion_source += ((motion_y >> 1) * stride);
1483 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1484 uint8_t *temp= s->edge_emu_buffer;
1485 if(stride<0) temp -= 8*stride;
1487 s->dsp.emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1488 motion_source= temp;
1493 /* first, take care of copying a block from either the
1494 * previous or the golden frame */
1495 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1496 /* Note, it is possible to implement all MC cases with
1497 put_no_rnd_pixels_l2 which would look more like the
1498 VP3 source but this would be slower as
1499 put_no_rnd_pixels_tab is better optimzed */
1500 if(motion_halfpel_index != 3){
1501 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1502 output_plane + first_pixel,
1503 motion_source, stride, 8);
1505 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1506 s->dsp.put_no_rnd_pixels_l2[1](
1507 output_plane + first_pixel,
1509 motion_source + stride + 1 + d,
1514 s->dsp.clear_block(block);
1516 /* invert DCT and place (or add) in final output */
1518 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1519 vp3_dequant(s, s->all_fragments + i, plane, 0, block);
1520 if(s->avctx->idct_algo!=FF_IDCT_VP3)
1523 output_plane + first_pixel,
1527 if (vp3_dequant(s, s->all_fragments + i, plane, 1, block)) {
1529 output_plane + first_pixel,
1533 s->dsp.vp3_idct_dc_add(output_plane + first_pixel, stride, block);
1538 /* copy directly from the previous frame */
1539 s->dsp.put_pixels_tab[1][0](
1540 output_plane + first_pixel,
1541 last_plane + first_pixel,
1548 // Filter up to the last row in the superblock row
1549 if (!s->skip_loop_filter)
1550 apply_loop_filter(s, plane, 4*sb_y - !!sb_y, FFMIN(4*sb_y+3, fragment_height-1));
1554 /* this looks like a good place for slice dispatch... */
1556 * if (slice == s->macroblock_height - 1)
1557 * dispatch (both last slice & 2nd-to-last slice);
1558 * else if (slice > 0)
1559 * dispatch (slice - 1);
1562 vp3_draw_horiz_band(s, FFMIN((32 << s->chroma_y_shift) * (slice + 1) -16, s->height-16));
1565 /// Allocate tables for per-frame data in Vp3DecodeContext
1566 static av_cold int allocate_tables(AVCodecContext *avctx)
1568 Vp3DecodeContext *s = avctx->priv_data;
1569 int y_fragment_count, c_fragment_count;
1571 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1572 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1574 s->superblock_coding = av_malloc(s->superblock_count);
1575 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1576 s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int));
1577 s->dct_tokens_base = av_malloc(64*s->fragment_count * sizeof(*s->dct_tokens_base));
1578 s->motion_val[0] = av_malloc(y_fragment_count * sizeof(*s->motion_val[0]));
1579 s->motion_val[1] = av_malloc(c_fragment_count * sizeof(*s->motion_val[1]));
1581 /* work out the block mapping tables */
1582 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1583 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1585 if (!s->superblock_coding || !s->all_fragments || !s->dct_tokens_base ||
1586 !s->coded_fragment_list[0] || !s->superblock_fragments || !s->macroblock_coding ||
1587 !s->motion_val[0] || !s->motion_val[1]) {
1588 vp3_decode_end(avctx);
1592 init_block_mapping(s);
1597 static av_cold int vp3_decode_init(AVCodecContext *avctx)
1599 Vp3DecodeContext *s = avctx->priv_data;
1600 int i, inter, plane;
1603 int y_fragment_count, c_fragment_count;
1605 if (avctx->codec_tag == MKTAG('V','P','3','0'))
1611 s->width = FFALIGN(avctx->width, 16);
1612 s->height = FFALIGN(avctx->height, 16);
1613 if (avctx->pix_fmt == PIX_FMT_NONE)
1614 avctx->pix_fmt = PIX_FMT_YUV420P;
1615 avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1616 if(avctx->idct_algo==FF_IDCT_AUTO)
1617 avctx->idct_algo=FF_IDCT_VP3;
1618 dsputil_init(&s->dsp, avctx);
1620 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1622 /* initialize to an impossible value which will force a recalculation
1623 * in the first frame decode */
1624 for (i = 0; i < 3; i++)
1627 avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift);
1629 s->y_superblock_width = (s->width + 31) / 32;
1630 s->y_superblock_height = (s->height + 31) / 32;
1631 s->y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1633 /* work out the dimensions for the C planes */
1634 c_width = s->width >> s->chroma_x_shift;
1635 c_height = s->height >> s->chroma_y_shift;
1636 s->c_superblock_width = (c_width + 31) / 32;
1637 s->c_superblock_height = (c_height + 31) / 32;
1638 s->c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1640 s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
1641 s->u_superblock_start = s->y_superblock_count;
1642 s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
1644 s->macroblock_width = (s->width + 15) / 16;
1645 s->macroblock_height = (s->height + 15) / 16;
1646 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1648 s->fragment_width[0] = s->width / FRAGMENT_PIXELS;
1649 s->fragment_height[0] = s->height / FRAGMENT_PIXELS;
1650 s->fragment_width[1] = s->fragment_width[0] >> s->chroma_x_shift;
1651 s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift;
1653 /* fragment count covers all 8x8 blocks for all 3 planes */
1654 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1655 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1656 s->fragment_count = y_fragment_count + 2*c_fragment_count;
1657 s->fragment_start[1] = y_fragment_count;
1658 s->fragment_start[2] = y_fragment_count + c_fragment_count;
1660 if (!s->theora_tables)
1662 for (i = 0; i < 64; i++) {
1663 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1664 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1665 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1666 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1667 s->base_matrix[2][i] = vp31_inter_dequant[i];
1668 s->filter_limit_values[i] = vp31_filter_limit_values[i];
1671 for(inter=0; inter<2; inter++){
1672 for(plane=0; plane<3; plane++){
1673 s->qr_count[inter][plane]= 1;
1674 s->qr_size [inter][plane][0]= 63;
1675 s->qr_base [inter][plane][0]=
1676 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1680 /* init VLC tables */
1681 for (i = 0; i < 16; i++) {
1684 init_vlc(&s->dc_vlc[i], 11, 32,
1685 &dc_bias[i][0][1], 4, 2,
1686 &dc_bias[i][0][0], 4, 2, 0);
1688 /* group 1 AC histograms */
1689 init_vlc(&s->ac_vlc_1[i], 11, 32,
1690 &ac_bias_0[i][0][1], 4, 2,
1691 &ac_bias_0[i][0][0], 4, 2, 0);
1693 /* group 2 AC histograms */
1694 init_vlc(&s->ac_vlc_2[i], 11, 32,
1695 &ac_bias_1[i][0][1], 4, 2,
1696 &ac_bias_1[i][0][0], 4, 2, 0);
1698 /* group 3 AC histograms */
1699 init_vlc(&s->ac_vlc_3[i], 11, 32,
1700 &ac_bias_2[i][0][1], 4, 2,
1701 &ac_bias_2[i][0][0], 4, 2, 0);
1703 /* group 4 AC histograms */
1704 init_vlc(&s->ac_vlc_4[i], 11, 32,
1705 &ac_bias_3[i][0][1], 4, 2,
1706 &ac_bias_3[i][0][0], 4, 2, 0);
1710 for (i = 0; i < 16; i++) {
1712 if (init_vlc(&s->dc_vlc[i], 11, 32,
1713 &s->huffman_table[i][0][1], 8, 4,
1714 &s->huffman_table[i][0][0], 8, 4, 0) < 0)
1717 /* group 1 AC histograms */
1718 if (init_vlc(&s->ac_vlc_1[i], 11, 32,
1719 &s->huffman_table[i+16][0][1], 8, 4,
1720 &s->huffman_table[i+16][0][0], 8, 4, 0) < 0)
1723 /* group 2 AC histograms */
1724 if (init_vlc(&s->ac_vlc_2[i], 11, 32,
1725 &s->huffman_table[i+16*2][0][1], 8, 4,
1726 &s->huffman_table[i+16*2][0][0], 8, 4, 0) < 0)
1729 /* group 3 AC histograms */
1730 if (init_vlc(&s->ac_vlc_3[i], 11, 32,
1731 &s->huffman_table[i+16*3][0][1], 8, 4,
1732 &s->huffman_table[i+16*3][0][0], 8, 4, 0) < 0)
1735 /* group 4 AC histograms */
1736 if (init_vlc(&s->ac_vlc_4[i], 11, 32,
1737 &s->huffman_table[i+16*4][0][1], 8, 4,
1738 &s->huffman_table[i+16*4][0][0], 8, 4, 0) < 0)
1743 init_vlc(&s->superblock_run_length_vlc, 6, 34,
1744 &superblock_run_length_vlc_table[0][1], 4, 2,
1745 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1747 init_vlc(&s->fragment_run_length_vlc, 5, 30,
1748 &fragment_run_length_vlc_table[0][1], 4, 2,
1749 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1751 init_vlc(&s->mode_code_vlc, 3, 8,
1752 &mode_code_vlc_table[0][1], 2, 1,
1753 &mode_code_vlc_table[0][0], 2, 1, 0);
1755 init_vlc(&s->motion_vector_vlc, 6, 63,
1756 &motion_vector_vlc_table[0][1], 2, 1,
1757 &motion_vector_vlc_table[0][0], 2, 1, 0);
1759 for (i = 0; i < 3; i++) {
1760 s->current_frame.data[i] = NULL;
1761 s->last_frame.data[i] = NULL;
1762 s->golden_frame.data[i] = NULL;
1765 return allocate_tables(avctx);
1768 av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1772 /// Release and shuffle frames after decode finishes
1773 static void update_frames(AVCodecContext *avctx)
1775 Vp3DecodeContext *s = avctx->priv_data;
1777 /* release the last frame, if it is allocated and if it is not the
1779 if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
1780 ff_thread_release_buffer(avctx, &s->last_frame);
1782 /* shuffle frames (last = current) */
1783 s->last_frame= s->current_frame;
1786 if (s->golden_frame.data[0])
1787 ff_thread_release_buffer(avctx, &s->golden_frame);
1788 s->golden_frame = s->current_frame;
1789 s->last_frame.type = FF_BUFFER_TYPE_COPY;
1792 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
1795 static int vp3_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
1797 Vp3DecodeContext *s = dst->priv_data, *s1 = src->priv_data;
1798 int qps_changed = 0, i, err;
1800 #define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
1802 if (!s1->current_frame.data[0]
1803 ||s->width != s1->width
1804 ||s->height!= s1->height) {
1806 copy_fields(s, s1, golden_frame, current_frame);
1811 // init tables if the first frame hasn't been decoded
1812 if (!s->current_frame.data[0]) {
1813 int y_fragment_count, c_fragment_count;
1815 err = allocate_tables(dst);
1818 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1819 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1820 memcpy(s->motion_val[0], s1->motion_val[0], y_fragment_count * sizeof(*s->motion_val[0]));
1821 memcpy(s->motion_val[1], s1->motion_val[1], c_fragment_count * sizeof(*s->motion_val[1]));
1824 // copy previous frame data
1825 copy_fields(s, s1, golden_frame, dsp);
1827 // copy qscale data if necessary
1828 for (i = 0; i < 3; i++) {
1829 if (s->qps[i] != s1->qps[1]) {
1831 memcpy(&s->qmat[i], &s1->qmat[i], sizeof(s->qmat[i]));
1835 if (s->qps[0] != s1->qps[0])
1836 memcpy(&s->bounding_values_array, &s1->bounding_values_array, sizeof(s->bounding_values_array));
1839 copy_fields(s, s1, qps, superblock_count);
1848 static int vp3_decode_frame(AVCodecContext *avctx,
1849 void *data, int *data_size,
1852 const uint8_t *buf = avpkt->data;
1853 int buf_size = avpkt->size;
1854 Vp3DecodeContext *s = avctx->priv_data;
1858 init_get_bits(&gb, buf, buf_size * 8);
1860 if (s->theora && get_bits1(&gb))
1862 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1866 s->keyframe = !get_bits1(&gb);
1869 for (i = 0; i < 3; i++)
1870 s->last_qps[i] = s->qps[i];
1874 s->qps[s->nqps++]= get_bits(&gb, 6);
1875 } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1876 for (i = s->nqps; i < 3; i++)
1879 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1880 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1881 s->keyframe?"key":"", avctx->frame_number+1, s->qps[0]);
1883 s->skip_loop_filter = !s->filter_limit_values[s->qps[0]] ||
1884 avctx->skip_loop_filter >= (s->keyframe ? AVDISCARD_ALL : AVDISCARD_NONKEY);
1886 if (s->qps[0] != s->last_qps[0])
1887 init_loop_filter(s);
1889 for (i = 0; i < s->nqps; i++)
1890 // reinit all dequantizers if the first one changed, because
1891 // the DC of the first quantizer must be used for all matrices
1892 if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1893 init_dequantizer(s, i);
1895 if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1898 s->current_frame.reference = 3;
1899 s->current_frame.pict_type = s->keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P;
1900 s->current_frame.key_frame = s->keyframe;
1901 if (ff_thread_get_buffer(avctx, &s->current_frame) < 0) {
1902 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1906 if (!s->edge_emu_buffer)
1907 s->edge_emu_buffer = av_malloc(9*FFABS(s->current_frame.linesize[0]));
1912 skip_bits(&gb, 4); /* width code */
1913 skip_bits(&gb, 4); /* height code */
1916 s->version = get_bits(&gb, 5);
1917 if (avctx->frame_number == 0)
1918 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1921 if (s->version || s->theora)
1924 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1925 skip_bits(&gb, 2); /* reserved? */
1928 if (!s->golden_frame.data[0]) {
1929 av_log(s->avctx, AV_LOG_WARNING, "vp3: first frame not a keyframe\n");
1931 s->golden_frame.reference = 3;
1932 s->golden_frame.pict_type = AV_PICTURE_TYPE_I;
1933 if (ff_thread_get_buffer(avctx, &s->golden_frame) < 0) {
1934 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1937 s->last_frame = s->golden_frame;
1938 s->last_frame.type = FF_BUFFER_TYPE_COPY;
1939 ff_thread_report_progress(&s->last_frame, INT_MAX, 0);
1943 memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
1944 ff_thread_finish_setup(avctx);
1946 if (unpack_superblocks(s, &gb)){
1947 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
1950 if (unpack_modes(s, &gb)){
1951 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
1954 if (unpack_vectors(s, &gb)){
1955 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
1958 if (unpack_block_qpis(s, &gb)){
1959 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
1962 if (unpack_dct_coeffs(s, &gb)){
1963 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
1967 for (i = 0; i < 3; i++) {
1968 int height = s->height >> (i && s->chroma_y_shift);
1969 if (s->flipped_image)
1970 s->data_offset[i] = 0;
1972 s->data_offset[i] = (height-1) * s->current_frame.linesize[i];
1975 s->last_slice_end = 0;
1976 for (i = 0; i < s->c_superblock_height; i++)
1979 // filter the last row
1980 for (i = 0; i < 3; i++) {
1981 int row = (s->height >> (3+(i && s->chroma_y_shift))) - 1;
1982 apply_loop_filter(s, i, row, row+1);
1984 vp3_draw_horiz_band(s, s->avctx->height);
1986 *data_size=sizeof(AVFrame);
1987 *(AVFrame*)data= s->current_frame;
1989 if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME))
1990 update_frames(avctx);
1995 ff_thread_report_progress(&s->current_frame, INT_MAX, 0);
1997 if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME))
1998 avctx->release_buffer(avctx, &s->current_frame);
2003 static void vp3_decode_flush(AVCodecContext *avctx);
2005 static av_cold int vp3_decode_end(AVCodecContext *avctx)
2007 Vp3DecodeContext *s = avctx->priv_data;
2010 av_free(s->superblock_coding);
2011 av_free(s->all_fragments);
2012 av_free(s->coded_fragment_list[0]);
2013 av_free(s->dct_tokens_base);
2014 av_free(s->superblock_fragments);
2015 av_free(s->macroblock_coding);
2016 av_free(s->motion_val[0]);
2017 av_free(s->motion_val[1]);
2018 av_free(s->edge_emu_buffer);
2020 if (avctx->internal->is_copy)
2023 for (i = 0; i < 16; i++) {
2024 free_vlc(&s->dc_vlc[i]);
2025 free_vlc(&s->ac_vlc_1[i]);
2026 free_vlc(&s->ac_vlc_2[i]);
2027 free_vlc(&s->ac_vlc_3[i]);
2028 free_vlc(&s->ac_vlc_4[i]);
2031 free_vlc(&s->superblock_run_length_vlc);
2032 free_vlc(&s->fragment_run_length_vlc);
2033 free_vlc(&s->mode_code_vlc);
2034 free_vlc(&s->motion_vector_vlc);
2036 /* release all frames */
2037 vp3_decode_flush(avctx);
2042 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2044 Vp3DecodeContext *s = avctx->priv_data;
2046 if (get_bits1(gb)) {
2048 if (s->entries >= 32) { /* overflow */
2049 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2052 token = get_bits(gb, 5);
2053 //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);
2054 s->huffman_table[s->hti][token][0] = s->hbits;
2055 s->huffman_table[s->hti][token][1] = s->huff_code_size;
2059 if (s->huff_code_size >= 32) {/* overflow */
2060 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2063 s->huff_code_size++;
2065 if (read_huffman_tree(avctx, gb))
2068 if (read_huffman_tree(avctx, gb))
2071 s->huff_code_size--;
2076 #if CONFIG_THEORA_DECODER
2077 static const enum PixelFormat theora_pix_fmts[4] = {
2078 PIX_FMT_YUV420P, PIX_FMT_NONE, PIX_FMT_YUV422P, PIX_FMT_YUV444P
2081 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2083 Vp3DecodeContext *s = avctx->priv_data;
2084 int visible_width, visible_height, colorspace;
2085 int offset_x = 0, offset_y = 0;
2086 AVRational fps, aspect;
2088 s->theora = get_bits_long(gb, 24);
2089 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2091 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2092 /* but previous versions have the image flipped relative to vp3 */
2093 if (s->theora < 0x030200)
2095 s->flipped_image = 1;
2096 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2099 visible_width = s->width = get_bits(gb, 16) << 4;
2100 visible_height = s->height = get_bits(gb, 16) << 4;
2102 if(av_image_check_size(s->width, s->height, 0, avctx)){
2103 av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2104 s->width= s->height= 0;
2108 if (s->theora >= 0x030200) {
2109 visible_width = get_bits_long(gb, 24);
2110 visible_height = get_bits_long(gb, 24);
2112 offset_x = get_bits(gb, 8); /* offset x */
2113 offset_y = get_bits(gb, 8); /* offset y, from bottom */
2116 fps.num = get_bits_long(gb, 32);
2117 fps.den = get_bits_long(gb, 32);
2118 if (fps.num && fps.den) {
2119 av_reduce(&avctx->time_base.num, &avctx->time_base.den,
2120 fps.den, fps.num, 1<<30);
2123 aspect.num = get_bits_long(gb, 24);
2124 aspect.den = get_bits_long(gb, 24);
2125 if (aspect.num && aspect.den) {
2126 av_reduce(&avctx->sample_aspect_ratio.num,
2127 &avctx->sample_aspect_ratio.den,
2128 aspect.num, aspect.den, 1<<30);
2131 if (s->theora < 0x030200)
2132 skip_bits(gb, 5); /* keyframe frequency force */
2133 colorspace = get_bits(gb, 8);
2134 skip_bits(gb, 24); /* bitrate */
2136 skip_bits(gb, 6); /* quality hint */
2138 if (s->theora >= 0x030200)
2140 skip_bits(gb, 5); /* keyframe frequency force */
2141 avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)];
2142 skip_bits(gb, 3); /* reserved */
2145 // align_get_bits(gb);
2147 if ( visible_width <= s->width && visible_width > s->width-16
2148 && visible_height <= s->height && visible_height > s->height-16
2149 && !offset_x && (offset_y == s->height - visible_height))
2150 avcodec_set_dimensions(avctx, visible_width, visible_height);
2152 avcodec_set_dimensions(avctx, s->width, s->height);
2154 if (colorspace == 1) {
2155 avctx->color_primaries = AVCOL_PRI_BT470M;
2156 } else if (colorspace == 2) {
2157 avctx->color_primaries = AVCOL_PRI_BT470BG;
2159 if (colorspace == 1 || colorspace == 2) {
2160 avctx->colorspace = AVCOL_SPC_BT470BG;
2161 avctx->color_trc = AVCOL_TRC_BT709;
2167 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2169 Vp3DecodeContext *s = avctx->priv_data;
2170 int i, n, matrices, inter, plane;
2172 if (s->theora >= 0x030200) {
2173 n = get_bits(gb, 3);
2174 /* loop filter limit values table */
2176 for (i = 0; i < 64; i++)
2177 s->filter_limit_values[i] = get_bits(gb, n);
2180 if (s->theora >= 0x030200)
2181 n = get_bits(gb, 4) + 1;
2184 /* quality threshold table */
2185 for (i = 0; i < 64; i++)
2186 s->coded_ac_scale_factor[i] = get_bits(gb, n);
2188 if (s->theora >= 0x030200)
2189 n = get_bits(gb, 4) + 1;
2192 /* dc scale factor table */
2193 for (i = 0; i < 64; i++)
2194 s->coded_dc_scale_factor[i] = get_bits(gb, n);
2196 if (s->theora >= 0x030200)
2197 matrices = get_bits(gb, 9) + 1;
2202 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2206 for(n=0; n<matrices; n++){
2207 for (i = 0; i < 64; i++)
2208 s->base_matrix[n][i]= get_bits(gb, 8);
2211 for (inter = 0; inter <= 1; inter++) {
2212 for (plane = 0; plane <= 2; plane++) {
2214 if (inter || plane > 0)
2215 newqr = get_bits1(gb);
2218 if(inter && get_bits1(gb)){
2222 qtj= (3*inter + plane - 1) / 3;
2223 plj= (plane + 2) % 3;
2225 s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2226 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2227 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2233 i= get_bits(gb, av_log2(matrices-1)+1);
2235 av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2238 s->qr_base[inter][plane][qri]= i;
2241 i = get_bits(gb, av_log2(63-qi)+1) + 1;
2242 s->qr_size[inter][plane][qri++]= i;
2247 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2250 s->qr_count[inter][plane]= qri;
2255 /* Huffman tables */
2256 for (s->hti = 0; s->hti < 80; s->hti++) {
2258 s->huff_code_size = 1;
2259 if (!get_bits1(gb)) {
2261 if(read_huffman_tree(avctx, gb))
2264 if(read_huffman_tree(avctx, gb))
2269 s->theora_tables = 1;
2274 static av_cold int theora_decode_init(AVCodecContext *avctx)
2276 Vp3DecodeContext *s = avctx->priv_data;
2279 uint8_t *header_start[3];
2285 if (!avctx->extradata_size)
2287 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2291 if (avpriv_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2292 42, header_start, header_len) < 0) {
2293 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2298 init_get_bits(&gb, header_start[i], header_len[i] * 8);
2300 ptype = get_bits(&gb, 8);
2302 if (!(ptype & 0x80))
2304 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2308 // FIXME: Check for this as well.
2309 skip_bits_long(&gb, 6*8); /* "theora" */
2314 theora_decode_header(avctx, &gb);
2317 // FIXME: is this needed? it breaks sometimes
2318 // theora_decode_comments(avctx, gb);
2321 if (theora_decode_tables(avctx, &gb))
2325 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2328 if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2329 av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2330 if (s->theora < 0x030200)
2334 return vp3_decode_init(avctx);
2337 static void vp3_decode_flush(AVCodecContext *avctx)
2339 Vp3DecodeContext *s = avctx->priv_data;
2341 if (s->golden_frame.data[0]) {
2342 if (s->golden_frame.data[0] == s->last_frame.data[0])
2343 memset(&s->last_frame, 0, sizeof(AVFrame));
2344 if (s->current_frame.data[0] == s->golden_frame.data[0])
2345 memset(&s->current_frame, 0, sizeof(AVFrame));
2346 ff_thread_release_buffer(avctx, &s->golden_frame);
2348 if (s->last_frame.data[0]) {
2349 if (s->current_frame.data[0] == s->last_frame.data[0])
2350 memset(&s->current_frame, 0, sizeof(AVFrame));
2351 ff_thread_release_buffer(avctx, &s->last_frame);
2353 if (s->current_frame.data[0])
2354 ff_thread_release_buffer(avctx, &s->current_frame);
2357 static int vp3_init_thread_copy(AVCodecContext *avctx)
2359 Vp3DecodeContext *s = avctx->priv_data;
2361 s->superblock_coding = NULL;
2362 s->all_fragments = NULL;
2363 s->coded_fragment_list[0] = NULL;
2364 s->dct_tokens_base = NULL;
2365 s->superblock_fragments = NULL;
2366 s->macroblock_coding = NULL;
2367 s->motion_val[0] = NULL;
2368 s->motion_val[1] = NULL;
2369 s->edge_emu_buffer = NULL;
2374 AVCodec ff_theora_decoder = {
2376 .type = AVMEDIA_TYPE_VIDEO,
2377 .id = CODEC_ID_THEORA,
2378 .priv_data_size = sizeof(Vp3DecodeContext),
2379 .init = theora_decode_init,
2380 .close = vp3_decode_end,
2381 .decode = vp3_decode_frame,
2382 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND | CODEC_CAP_FRAME_THREADS,
2383 .flush = vp3_decode_flush,
2384 .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2385 .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
2386 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context)
2390 AVCodec ff_vp3_decoder = {
2392 .type = AVMEDIA_TYPE_VIDEO,
2394 .priv_data_size = sizeof(Vp3DecodeContext),
2395 .init = vp3_decode_init,
2396 .close = vp3_decode_end,
2397 .decode = vp3_decode_frame,
2398 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND | CODEC_CAP_FRAME_THREADS,
2399 .flush = vp3_decode_flush,
2400 .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2401 .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
2402 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context)