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"
47 #define FRAGMENT_PIXELS 8
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;
139 VP3DSPContext vp3dsp;
142 int skip_loop_filter;
148 int superblock_count;
149 int y_superblock_width;
150 int y_superblock_height;
151 int y_superblock_count;
152 int c_superblock_width;
153 int c_superblock_height;
154 int c_superblock_count;
155 int u_superblock_start;
156 int v_superblock_start;
157 unsigned char *superblock_coding;
159 int macroblock_count;
160 int macroblock_width;
161 int macroblock_height;
164 int fragment_width[2];
165 int fragment_height[2];
167 Vp3Fragment *all_fragments;
168 int fragment_start[3];
171 int8_t (*motion_val[2])[2];
176 uint16_t coded_dc_scale_factor[64];
177 uint32_t coded_ac_scale_factor[64];
178 uint8_t base_matrix[384][64];
179 uint8_t qr_count[2][3];
180 uint8_t qr_size [2][3][64];
181 uint16_t qr_base[2][3][64];
184 * This is a list of all tokens in bitstream order. Reordering takes place
185 * by pulling from each level during IDCT. As a consequence, IDCT must be
186 * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32
187 * otherwise. The 32 different tokens with up to 12 bits of extradata are
188 * collapsed into 3 types, packed as follows:
189 * (from the low to high bits)
191 * 2 bits: type (0,1,2)
192 * 0: EOB run, 14 bits for run length (12 needed)
193 * 1: zero run, 7 bits for run length
194 * 7 bits for the next coefficient (3 needed)
195 * 2: coefficient, 14 bits (11 needed)
197 * Coefficients are signed, so are packed in the highest bits for automatic
200 int16_t *dct_tokens[3][64];
201 int16_t *dct_tokens_base;
202 #define TOKEN_EOB(eob_run) ((eob_run) << 2)
203 #define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) << 9) + ((zero_run) << 2) + 1)
204 #define TOKEN_COEFF(coeff) (((coeff) << 2) + 2)
207 * number of blocks that contain DCT coefficients at the given level or higher
209 int num_coded_frags[3][64];
210 int total_num_coded_frags;
212 /* this is a list of indexes into the all_fragments array indicating
213 * which of the fragments are coded */
214 int *coded_fragment_list[3];
222 VLC superblock_run_length_vlc;
223 VLC fragment_run_length_vlc;
225 VLC motion_vector_vlc;
227 /* these arrays need to be on 16-byte boundaries since SSE2 operations
229 DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64]; ///< qmat[qpi][is_inter][plane]
231 /* This table contains superblock_count * 16 entries. Each set of 16
232 * numbers corresponds to the fragment indexes 0..15 of the superblock.
233 * An entry will be -1 to indicate that no entry corresponds to that
235 int *superblock_fragments;
237 /* This is an array that indicates how a particular macroblock
239 unsigned char *macroblock_coding;
241 uint8_t *edge_emu_buffer;
248 uint32_t huffman_table[80][32][2];
250 uint8_t filter_limit_values[64];
251 DECLARE_ALIGNED(8, int, bounding_values_array)[256+2];
254 /************************************************************************
255 * VP3 specific functions
256 ************************************************************************/
258 static void vp3_decode_flush(AVCodecContext *avctx)
260 Vp3DecodeContext *s = avctx->priv_data;
262 if (s->golden_frame.data[0]) {
263 if (s->golden_frame.data[0] == s->last_frame.data[0])
264 memset(&s->last_frame, 0, sizeof(AVFrame));
265 if (s->current_frame.data[0] == s->golden_frame.data[0])
266 memset(&s->current_frame, 0, sizeof(AVFrame));
267 ff_thread_release_buffer(avctx, &s->golden_frame);
269 if (s->last_frame.data[0]) {
270 if (s->current_frame.data[0] == s->last_frame.data[0])
271 memset(&s->current_frame, 0, sizeof(AVFrame));
272 ff_thread_release_buffer(avctx, &s->last_frame);
274 if (s->current_frame.data[0])
275 ff_thread_release_buffer(avctx, &s->current_frame);
278 static av_cold int vp3_decode_end(AVCodecContext *avctx)
280 Vp3DecodeContext *s = avctx->priv_data;
283 av_free(s->superblock_coding);
284 av_free(s->all_fragments);
285 av_free(s->coded_fragment_list[0]);
286 av_free(s->dct_tokens_base);
287 av_free(s->superblock_fragments);
288 av_free(s->macroblock_coding);
289 av_free(s->motion_val[0]);
290 av_free(s->motion_val[1]);
291 av_free(s->edge_emu_buffer);
293 if (avctx->internal->is_copy)
296 for (i = 0; i < 16; i++) {
297 ff_free_vlc(&s->dc_vlc[i]);
298 ff_free_vlc(&s->ac_vlc_1[i]);
299 ff_free_vlc(&s->ac_vlc_2[i]);
300 ff_free_vlc(&s->ac_vlc_3[i]);
301 ff_free_vlc(&s->ac_vlc_4[i]);
304 ff_free_vlc(&s->superblock_run_length_vlc);
305 ff_free_vlc(&s->fragment_run_length_vlc);
306 ff_free_vlc(&s->mode_code_vlc);
307 ff_free_vlc(&s->motion_vector_vlc);
309 /* release all frames */
310 vp3_decode_flush(avctx);
316 * This function sets up all of the various blocks mappings:
317 * superblocks <-> fragments, macroblocks <-> fragments,
318 * superblocks <-> macroblocks
320 * @return 0 is successful; returns 1 if *anything* went wrong.
322 static int init_block_mapping(Vp3DecodeContext *s)
324 int sb_x, sb_y, plane;
327 for (plane = 0; plane < 3; plane++) {
328 int sb_width = plane ? s->c_superblock_width : s->y_superblock_width;
329 int sb_height = plane ? s->c_superblock_height : s->y_superblock_height;
330 int frag_width = s->fragment_width[!!plane];
331 int frag_height = s->fragment_height[!!plane];
333 for (sb_y = 0; sb_y < sb_height; sb_y++)
334 for (sb_x = 0; sb_x < sb_width; sb_x++)
335 for (i = 0; i < 16; i++) {
336 x = 4*sb_x + hilbert_offset[i][0];
337 y = 4*sb_y + hilbert_offset[i][1];
339 if (x < frag_width && y < frag_height)
340 s->superblock_fragments[j++] = s->fragment_start[plane] + y*frag_width + x;
342 s->superblock_fragments[j++] = -1;
346 return 0; /* successful path out */
350 * This function sets up the dequantization tables used for a particular
353 static void init_dequantizer(Vp3DecodeContext *s, int qpi)
355 int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
356 int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
357 int i, plane, inter, qri, bmi, bmj, qistart;
359 for(inter=0; inter<2; inter++){
360 for(plane=0; plane<3; plane++){
362 for(qri=0; qri<s->qr_count[inter][plane]; qri++){
363 sum+= s->qr_size[inter][plane][qri];
364 if(s->qps[qpi] <= sum)
367 qistart= sum - s->qr_size[inter][plane][qri];
368 bmi= s->qr_base[inter][plane][qri ];
369 bmj= s->qr_base[inter][plane][qri+1];
371 int coeff= ( 2*(sum -s->qps[qpi])*s->base_matrix[bmi][i]
372 - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
373 + s->qr_size[inter][plane][qri])
374 / (2*s->qr_size[inter][plane][qri]);
376 int qmin= 8<<(inter + !i);
377 int qscale= i ? ac_scale_factor : dc_scale_factor;
379 s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
381 // all DC coefficients use the same quant so as not to interfere with DC prediction
382 s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
388 * This function initializes the loop filter boundary limits if the frame's
389 * quality index is different from the previous frame's.
391 * The filter_limit_values may not be larger than 127.
393 static void init_loop_filter(Vp3DecodeContext *s)
395 int *bounding_values= s->bounding_values_array+127;
400 filter_limit = s->filter_limit_values[s->qps[0]];
401 av_assert0(filter_limit < 128U);
403 /* set up the bounding values */
404 memset(s->bounding_values_array, 0, 256 * sizeof(int));
405 for (x = 0; x < filter_limit; x++) {
406 bounding_values[-x] = -x;
407 bounding_values[x] = x;
409 for (x = value = filter_limit; x < 128 && value; x++, value--) {
410 bounding_values[ x] = value;
411 bounding_values[-x] = -value;
414 bounding_values[128] = value;
415 bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
419 * This function unpacks all of the superblock/macroblock/fragment coding
420 * information from the bitstream.
422 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
424 int superblock_starts[3] = { 0, s->u_superblock_start, s->v_superblock_start };
426 int current_superblock = 0;
428 int num_partial_superblocks = 0;
431 int current_fragment;
435 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
439 /* unpack the list of partially-coded superblocks */
440 bit = get_bits1(gb) ^ 1;
443 while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) {
444 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
449 current_run = get_vlc2(gb,
450 s->superblock_run_length_vlc.table, 6, 2) + 1;
451 if (current_run == 34)
452 current_run += get_bits(gb, 12);
454 if (current_superblock + current_run > s->superblock_count) {
455 av_log(s->avctx, AV_LOG_ERROR, "Invalid partially coded superblock run length\n");
459 memset(s->superblock_coding + current_superblock, bit, current_run);
461 current_superblock += current_run;
463 num_partial_superblocks += current_run;
466 /* unpack the list of fully coded superblocks if any of the blocks were
467 * not marked as partially coded in the previous step */
468 if (num_partial_superblocks < s->superblock_count) {
469 int superblocks_decoded = 0;
471 current_superblock = 0;
472 bit = get_bits1(gb) ^ 1;
475 while (superblocks_decoded < s->superblock_count - num_partial_superblocks
476 && get_bits_left(gb) > 0) {
478 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
483 current_run = get_vlc2(gb,
484 s->superblock_run_length_vlc.table, 6, 2) + 1;
485 if (current_run == 34)
486 current_run += get_bits(gb, 12);
488 for (j = 0; j < current_run; current_superblock++) {
489 if (current_superblock >= s->superblock_count) {
490 av_log(s->avctx, AV_LOG_ERROR, "Invalid fully coded superblock run length\n");
494 /* skip any superblocks already marked as partially coded */
495 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
496 s->superblock_coding[current_superblock] = 2*bit;
500 superblocks_decoded += current_run;
504 /* if there were partial blocks, initialize bitstream for
505 * unpacking fragment codings */
506 if (num_partial_superblocks) {
510 /* toggle the bit because as soon as the first run length is
511 * fetched the bit will be toggled again */
516 /* figure out which fragments are coded; iterate through each
517 * superblock (all planes) */
518 s->total_num_coded_frags = 0;
519 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
521 for (plane = 0; plane < 3; plane++) {
522 int sb_start = superblock_starts[plane];
523 int sb_end = sb_start + (plane ? s->c_superblock_count : s->y_superblock_count);
524 int num_coded_frags = 0;
526 for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) {
528 /* iterate through all 16 fragments in a superblock */
529 for (j = 0; j < 16; j++) {
531 /* if the fragment is in bounds, check its coding status */
532 current_fragment = s->superblock_fragments[i * 16 + j];
533 if (current_fragment != -1) {
534 int coded = s->superblock_coding[i];
536 if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
538 /* fragment may or may not be coded; this is the case
539 * that cares about the fragment coding runs */
540 if (current_run-- == 0) {
542 current_run = get_vlc2(gb,
543 s->fragment_run_length_vlc.table, 5, 2);
549 /* default mode; actual mode will be decoded in
551 s->all_fragments[current_fragment].coding_method =
553 s->coded_fragment_list[plane][num_coded_frags++] =
556 /* not coded; copy this fragment from the prior frame */
557 s->all_fragments[current_fragment].coding_method =
563 s->total_num_coded_frags += num_coded_frags;
564 for (i = 0; i < 64; i++)
565 s->num_coded_frags[plane][i] = num_coded_frags;
567 s->coded_fragment_list[plane+1] = s->coded_fragment_list[plane] + num_coded_frags;
573 * This function unpacks all the coding mode data for individual macroblocks
574 * from the bitstream.
576 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
578 int i, j, k, sb_x, sb_y;
580 int current_macroblock;
581 int current_fragment;
583 int custom_mode_alphabet[CODING_MODE_COUNT];
588 for (i = 0; i < s->fragment_count; i++)
589 s->all_fragments[i].coding_method = MODE_INTRA;
593 /* fetch the mode coding scheme for this frame */
594 scheme = get_bits(gb, 3);
596 /* is it a custom coding scheme? */
598 for (i = 0; i < 8; i++)
599 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
600 for (i = 0; i < 8; i++)
601 custom_mode_alphabet[get_bits(gb, 3)] = i;
602 alphabet = custom_mode_alphabet;
604 alphabet = ModeAlphabet[scheme-1];
606 /* iterate through all of the macroblocks that contain 1 or more
608 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
609 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
610 if (get_bits_left(gb) <= 0)
613 for (j = 0; j < 4; j++) {
614 int mb_x = 2*sb_x + (j>>1);
615 int mb_y = 2*sb_y + (((j>>1)+j)&1);
616 current_macroblock = mb_y * s->macroblock_width + mb_x;
618 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height)
621 #define BLOCK_X (2*mb_x + (k&1))
622 #define BLOCK_Y (2*mb_y + (k>>1))
623 /* coding modes are only stored if the macroblock has at least one
624 * luma block coded, otherwise it must be INTER_NO_MV */
625 for (k = 0; k < 4; k++) {
626 current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
627 if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
631 s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
635 /* mode 7 means get 3 bits for each coding mode */
637 coding_mode = get_bits(gb, 3);
639 coding_mode = alphabet
640 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
642 s->macroblock_coding[current_macroblock] = coding_mode;
643 for (k = 0; k < 4; k++) {
644 frag = s->all_fragments + BLOCK_Y*s->fragment_width[0] + BLOCK_X;
645 if (frag->coding_method != MODE_COPY)
646 frag->coding_method = coding_mode;
649 #define SET_CHROMA_MODES \
650 if (frag[s->fragment_start[1]].coding_method != MODE_COPY) \
651 frag[s->fragment_start[1]].coding_method = coding_mode;\
652 if (frag[s->fragment_start[2]].coding_method != MODE_COPY) \
653 frag[s->fragment_start[2]].coding_method = coding_mode;
655 if (s->chroma_y_shift) {
656 frag = s->all_fragments + mb_y*s->fragment_width[1] + mb_x;
658 } else if (s->chroma_x_shift) {
659 frag = s->all_fragments + 2*mb_y*s->fragment_width[1] + mb_x;
660 for (k = 0; k < 2; k++) {
662 frag += s->fragment_width[1];
665 for (k = 0; k < 4; k++) {
666 frag = s->all_fragments + BLOCK_Y*s->fragment_width[1] + BLOCK_X;
679 * This function unpacks all the motion vectors for the individual
680 * macroblocks from the bitstream.
682 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
684 int j, k, sb_x, sb_y;
688 int last_motion_x = 0;
689 int last_motion_y = 0;
690 int prior_last_motion_x = 0;
691 int prior_last_motion_y = 0;
692 int current_macroblock;
693 int current_fragment;
699 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
700 coding_mode = get_bits1(gb);
702 /* iterate through all of the macroblocks that contain 1 or more
704 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
705 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
706 if (get_bits_left(gb) <= 0)
709 for (j = 0; j < 4; j++) {
710 int mb_x = 2*sb_x + (j>>1);
711 int mb_y = 2*sb_y + (((j>>1)+j)&1);
712 current_macroblock = mb_y * s->macroblock_width + mb_x;
714 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
715 (s->macroblock_coding[current_macroblock] == MODE_COPY))
718 switch (s->macroblock_coding[current_macroblock]) {
720 case MODE_INTER_PLUS_MV:
722 /* all 6 fragments use the same motion vector */
723 if (coding_mode == 0) {
724 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
725 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
727 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
728 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
731 /* vector maintenance, only on MODE_INTER_PLUS_MV */
732 if (s->macroblock_coding[current_macroblock] ==
733 MODE_INTER_PLUS_MV) {
734 prior_last_motion_x = last_motion_x;
735 prior_last_motion_y = last_motion_y;
736 last_motion_x = motion_x[0];
737 last_motion_y = motion_y[0];
741 case MODE_INTER_FOURMV:
742 /* vector maintenance */
743 prior_last_motion_x = last_motion_x;
744 prior_last_motion_y = last_motion_y;
746 /* fetch 4 vectors from the bitstream, one for each
747 * Y fragment, then average for the C fragment vectors */
748 for (k = 0; k < 4; k++) {
749 current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
750 if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
751 if (coding_mode == 0) {
752 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
753 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
755 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
756 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
758 last_motion_x = motion_x[k];
759 last_motion_y = motion_y[k];
767 case MODE_INTER_LAST_MV:
768 /* all 6 fragments use the last motion vector */
769 motion_x[0] = last_motion_x;
770 motion_y[0] = last_motion_y;
772 /* no vector maintenance (last vector remains the
776 case MODE_INTER_PRIOR_LAST:
777 /* all 6 fragments use the motion vector prior to the
778 * last motion vector */
779 motion_x[0] = prior_last_motion_x;
780 motion_y[0] = prior_last_motion_y;
782 /* vector maintenance */
783 prior_last_motion_x = last_motion_x;
784 prior_last_motion_y = last_motion_y;
785 last_motion_x = motion_x[0];
786 last_motion_y = motion_y[0];
790 /* covers intra, inter without MV, golden without MV */
794 /* no vector maintenance */
798 /* assign the motion vectors to the correct fragments */
799 for (k = 0; k < 4; k++) {
801 BLOCK_Y*s->fragment_width[0] + BLOCK_X;
802 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
803 s->motion_val[0][current_fragment][0] = motion_x[k];
804 s->motion_val[0][current_fragment][1] = motion_y[k];
806 s->motion_val[0][current_fragment][0] = motion_x[0];
807 s->motion_val[0][current_fragment][1] = motion_y[0];
811 if (s->chroma_y_shift) {
812 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
813 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] + motion_x[2] + motion_x[3], 2);
814 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] + motion_y[2] + motion_y[3], 2);
816 motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
817 motion_y[0] = (motion_y[0]>>1) | (motion_y[0]&1);
818 frag = mb_y*s->fragment_width[1] + mb_x;
819 s->motion_val[1][frag][0] = motion_x[0];
820 s->motion_val[1][frag][1] = motion_y[0];
821 } else if (s->chroma_x_shift) {
822 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
823 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1);
824 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1);
825 motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1);
826 motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1);
828 motion_x[1] = motion_x[0];
829 motion_y[1] = motion_y[0];
831 motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
832 motion_x[1] = (motion_x[1]>>1) | (motion_x[1]&1);
834 frag = 2*mb_y*s->fragment_width[1] + mb_x;
835 for (k = 0; k < 2; k++) {
836 s->motion_val[1][frag][0] = motion_x[k];
837 s->motion_val[1][frag][1] = motion_y[k];
838 frag += s->fragment_width[1];
841 for (k = 0; k < 4; k++) {
842 frag = BLOCK_Y*s->fragment_width[1] + BLOCK_X;
843 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
844 s->motion_val[1][frag][0] = motion_x[k];
845 s->motion_val[1][frag][1] = motion_y[k];
847 s->motion_val[1][frag][0] = motion_x[0];
848 s->motion_val[1][frag][1] = motion_y[0];
859 static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
861 int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
862 int num_blocks = s->total_num_coded_frags;
864 for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
865 i = blocks_decoded = num_blocks_at_qpi = 0;
867 bit = get_bits1(gb) ^ 1;
871 if (run_length == MAXIMUM_LONG_BIT_RUN)
876 run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
877 if (run_length == 34)
878 run_length += get_bits(gb, 12);
879 blocks_decoded += run_length;
882 num_blocks_at_qpi += run_length;
884 for (j = 0; j < run_length; i++) {
885 if (i >= s->total_num_coded_frags)
888 if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
889 s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
893 } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0);
895 num_blocks -= num_blocks_at_qpi;
902 * This function is called by unpack_dct_coeffs() to extract the VLCs from
903 * the bitstream. The VLCs encode tokens which are used to unpack DCT
904 * data. This function unpacks all the VLCs for either the Y plane or both
905 * C planes, and is called for DC coefficients or different AC coefficient
906 * levels (since different coefficient types require different VLC tables.
908 * This function returns a residual eob run. E.g, if a particular token gave
909 * instructions to EOB the next 5 fragments and there were only 2 fragments
910 * left in the current fragment range, 3 would be returned so that it could
911 * be passed into the next call to this same function.
913 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
914 VLC *table, int coeff_index,
925 int num_coeffs = s->num_coded_frags[plane][coeff_index];
926 int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
928 /* local references to structure members to avoid repeated deferences */
929 int *coded_fragment_list = s->coded_fragment_list[plane];
930 Vp3Fragment *all_fragments = s->all_fragments;
931 VLC_TYPE (*vlc_table)[2] = table->table;
934 av_log(s->avctx, AV_LOG_ERROR, "Invalid number of coefficents at level %d\n", coeff_index);
936 if (eob_run > num_coeffs) {
937 coeff_i = blocks_ended = num_coeffs;
938 eob_run -= num_coeffs;
940 coeff_i = blocks_ended = eob_run;
944 // insert fake EOB token to cover the split between planes or zzi
946 dct_tokens[j++] = blocks_ended << 2;
948 while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
949 /* decode a VLC into a token */
950 token = get_vlc2(gb, vlc_table, 11, 3);
951 /* use the token to get a zero run, a coefficient, and an eob run */
952 if ((unsigned) token <= 6U) {
953 eob_run = eob_run_base[token];
954 if (eob_run_get_bits[token])
955 eob_run += get_bits(gb, eob_run_get_bits[token]);
957 // record only the number of blocks ended in this plane,
958 // any spill will be recorded in the next plane.
959 if (eob_run > num_coeffs - coeff_i) {
960 dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
961 blocks_ended += num_coeffs - coeff_i;
962 eob_run -= num_coeffs - coeff_i;
963 coeff_i = num_coeffs;
965 dct_tokens[j++] = TOKEN_EOB(eob_run);
966 blocks_ended += eob_run;
970 } else if (token >= 0) {
971 bits_to_get = coeff_get_bits[token];
973 bits_to_get = get_bits(gb, bits_to_get);
974 coeff = coeff_tables[token][bits_to_get];
976 zero_run = zero_run_base[token];
977 if (zero_run_get_bits[token])
978 zero_run += get_bits(gb, zero_run_get_bits[token]);
981 dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
983 // Save DC into the fragment structure. DC prediction is
984 // done in raster order, so the actual DC can't be in with
985 // other tokens. We still need the token in dct_tokens[]
986 // however, or else the structure collapses on itself.
988 all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
990 dct_tokens[j++] = TOKEN_COEFF(coeff);
993 if (coeff_index + zero_run > 64) {
994 av_log(s->avctx, AV_LOG_DEBUG, "Invalid zero run of %d with"
995 " %d coeffs left\n", zero_run, 64-coeff_index);
996 zero_run = 64 - coeff_index;
999 // zero runs code multiple coefficients,
1000 // so don't try to decode coeffs for those higher levels
1001 for (i = coeff_index+1; i <= coeff_index+zero_run; i++)
1002 s->num_coded_frags[plane][i]--;
1005 av_log(s->avctx, AV_LOG_ERROR,
1006 "Invalid token %d\n", token);
1011 if (blocks_ended > s->num_coded_frags[plane][coeff_index])
1012 av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
1014 // decrement the number of blocks that have higher coeffecients for each
1015 // EOB run at this level
1017 for (i = coeff_index+1; i < 64; i++)
1018 s->num_coded_frags[plane][i] -= blocks_ended;
1020 // setup the next buffer
1022 s->dct_tokens[plane+1][coeff_index] = dct_tokens + j;
1023 else if (coeff_index < 63)
1024 s->dct_tokens[0][coeff_index+1] = dct_tokens + j;
1029 static void reverse_dc_prediction(Vp3DecodeContext *s,
1032 int fragment_height);
1034 * This function unpacks all of the DCT coefficient data from the
1037 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1044 int residual_eob_run = 0;
1048 s->dct_tokens[0][0] = s->dct_tokens_base;
1050 /* fetch the DC table indexes */
1051 dc_y_table = get_bits(gb, 4);
1052 dc_c_table = get_bits(gb, 4);
1054 /* unpack the Y plane DC coefficients */
1055 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1056 0, residual_eob_run);
1057 if (residual_eob_run < 0)
1058 return residual_eob_run;
1060 /* reverse prediction of the Y-plane DC coefficients */
1061 reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]);
1063 /* unpack the C plane DC coefficients */
1064 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1065 1, residual_eob_run);
1066 if (residual_eob_run < 0)
1067 return residual_eob_run;
1068 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1069 2, residual_eob_run);
1070 if (residual_eob_run < 0)
1071 return residual_eob_run;
1073 /* reverse prediction of the C-plane DC coefficients */
1074 if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1076 reverse_dc_prediction(s, s->fragment_start[1],
1077 s->fragment_width[1], s->fragment_height[1]);
1078 reverse_dc_prediction(s, s->fragment_start[2],
1079 s->fragment_width[1], s->fragment_height[1]);
1082 /* fetch the AC table indexes */
1083 ac_y_table = get_bits(gb, 4);
1084 ac_c_table = get_bits(gb, 4);
1086 /* build tables of AC VLC tables */
1087 for (i = 1; i <= 5; i++) {
1088 y_tables[i] = &s->ac_vlc_1[ac_y_table];
1089 c_tables[i] = &s->ac_vlc_1[ac_c_table];
1091 for (i = 6; i <= 14; i++) {
1092 y_tables[i] = &s->ac_vlc_2[ac_y_table];
1093 c_tables[i] = &s->ac_vlc_2[ac_c_table];
1095 for (i = 15; i <= 27; i++) {
1096 y_tables[i] = &s->ac_vlc_3[ac_y_table];
1097 c_tables[i] = &s->ac_vlc_3[ac_c_table];
1099 for (i = 28; i <= 63; i++) {
1100 y_tables[i] = &s->ac_vlc_4[ac_y_table];
1101 c_tables[i] = &s->ac_vlc_4[ac_c_table];
1104 /* decode all AC coefficents */
1105 for (i = 1; i <= 63; i++) {
1106 residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1107 0, residual_eob_run);
1108 if (residual_eob_run < 0)
1109 return residual_eob_run;
1111 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1112 1, residual_eob_run);
1113 if (residual_eob_run < 0)
1114 return residual_eob_run;
1115 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1116 2, residual_eob_run);
1117 if (residual_eob_run < 0)
1118 return residual_eob_run;
1125 * This function reverses the DC prediction for each coded fragment in
1126 * the frame. Much of this function is adapted directly from the original
1129 #define COMPATIBLE_FRAME(x) \
1130 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1131 #define DC_COEFF(u) s->all_fragments[u].dc
1133 static void reverse_dc_prediction(Vp3DecodeContext *s,
1136 int fragment_height)
1145 int i = first_fragment;
1149 /* DC values for the left, up-left, up, and up-right fragments */
1150 int vl, vul, vu, vur;
1152 /* indexes for the left, up-left, up, and up-right fragments */
1156 * The 6 fields mean:
1157 * 0: up-left multiplier
1159 * 2: up-right multiplier
1160 * 3: left multiplier
1162 static const int predictor_transform[16][4] = {
1164 { 0, 0, 0,128}, // PL
1165 { 0, 0,128, 0}, // PUR
1166 { 0, 0, 53, 75}, // PUR|PL
1167 { 0,128, 0, 0}, // PU
1168 { 0, 64, 0, 64}, // PU|PL
1169 { 0,128, 0, 0}, // PU|PUR
1170 { 0, 0, 53, 75}, // PU|PUR|PL
1171 {128, 0, 0, 0}, // PUL
1172 { 0, 0, 0,128}, // PUL|PL
1173 { 64, 0, 64, 0}, // PUL|PUR
1174 { 0, 0, 53, 75}, // PUL|PUR|PL
1175 { 0,128, 0, 0}, // PUL|PU
1176 {-104,116, 0,116}, // PUL|PU|PL
1177 { 24, 80, 24, 0}, // PUL|PU|PUR
1178 {-104,116, 0,116} // PUL|PU|PUR|PL
1181 /* This table shows which types of blocks can use other blocks for
1182 * prediction. For example, INTRA is the only mode in this table to
1183 * have a frame number of 0. That means INTRA blocks can only predict
1184 * from other INTRA blocks. There are 2 golden frame coding types;
1185 * blocks encoding in these modes can only predict from other blocks
1186 * that were encoded with these 1 of these 2 modes. */
1187 static const unsigned char compatible_frame[9] = {
1188 1, /* MODE_INTER_NO_MV */
1190 1, /* MODE_INTER_PLUS_MV */
1191 1, /* MODE_INTER_LAST_MV */
1192 1, /* MODE_INTER_PRIOR_MV */
1193 2, /* MODE_USING_GOLDEN */
1194 2, /* MODE_GOLDEN_MV */
1195 1, /* MODE_INTER_FOUR_MV */
1198 int current_frame_type;
1200 /* there is a last DC predictor for each of the 3 frame types */
1205 vul = vu = vur = vl = 0;
1206 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1208 /* for each fragment row... */
1209 for (y = 0; y < fragment_height; y++) {
1211 /* for each fragment in a row... */
1212 for (x = 0; x < fragment_width; x++, i++) {
1214 /* reverse prediction if this block was coded */
1215 if (s->all_fragments[i].coding_method != MODE_COPY) {
1217 current_frame_type =
1218 compatible_frame[s->all_fragments[i].coding_method];
1224 if(COMPATIBLE_FRAME(l))
1228 u= i-fragment_width;
1230 if(COMPATIBLE_FRAME(u))
1233 ul= i-fragment_width-1;
1235 if(COMPATIBLE_FRAME(ul))
1238 if(x + 1 < fragment_width){
1239 ur= i-fragment_width+1;
1241 if(COMPATIBLE_FRAME(ur))
1246 if (transform == 0) {
1248 /* if there were no fragments to predict from, use last
1250 predicted_dc = last_dc[current_frame_type];
1253 /* apply the appropriate predictor transform */
1255 (predictor_transform[transform][0] * vul) +
1256 (predictor_transform[transform][1] * vu) +
1257 (predictor_transform[transform][2] * vur) +
1258 (predictor_transform[transform][3] * vl);
1260 predicted_dc /= 128;
1262 /* check for outranging on the [ul u l] and
1263 * [ul u ur l] predictors */
1264 if ((transform == 15) || (transform == 13)) {
1265 if (FFABS(predicted_dc - vu) > 128)
1267 else if (FFABS(predicted_dc - vl) > 128)
1269 else if (FFABS(predicted_dc - vul) > 128)
1274 /* at long last, apply the predictor */
1275 DC_COEFF(i) += predicted_dc;
1277 last_dc[current_frame_type] = DC_COEFF(i);
1283 static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
1286 int *bounding_values= s->bounding_values_array+127;
1288 int width = s->fragment_width[!!plane];
1289 int height = s->fragment_height[!!plane];
1290 int fragment = s->fragment_start [plane] + ystart * width;
1291 int stride = s->current_frame.linesize[plane];
1292 uint8_t *plane_data = s->current_frame.data [plane];
1293 if (!s->flipped_image) stride = -stride;
1294 plane_data += s->data_offset[plane] + 8*ystart*stride;
1296 for (y = ystart; y < yend; y++) {
1298 for (x = 0; x < width; x++) {
1299 /* This code basically just deblocks on the edges of coded blocks.
1300 * However, it has to be much more complicated because of the
1301 * braindamaged deblock ordering used in VP3/Theora. Order matters
1302 * because some pixels get filtered twice. */
1303 if( s->all_fragments[fragment].coding_method != MODE_COPY )
1305 /* do not perform left edge filter for left columns frags */
1307 s->vp3dsp.h_loop_filter(
1309 stride, bounding_values);
1312 /* do not perform top edge filter for top row fragments */
1314 s->vp3dsp.v_loop_filter(
1316 stride, bounding_values);
1319 /* do not perform right edge filter for right column
1320 * fragments or if right fragment neighbor is also coded
1321 * in this frame (it will be filtered in next iteration) */
1322 if ((x < width - 1) &&
1323 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1324 s->vp3dsp.h_loop_filter(
1325 plane_data + 8*x + 8,
1326 stride, bounding_values);
1329 /* do not perform bottom edge filter for bottom row
1330 * fragments or if bottom fragment neighbor is also coded
1331 * in this frame (it will be filtered in the next row) */
1332 if ((y < height - 1) &&
1333 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1334 s->vp3dsp.v_loop_filter(
1335 plane_data + 8*x + 8*stride,
1336 stride, bounding_values);
1342 plane_data += 8*stride;
1347 * Pull DCT tokens from the 64 levels to decode and dequant the coefficients
1348 * for the next block in coding order
1350 static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1351 int plane, int inter, DCTELEM block[64])
1353 int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1354 uint8_t *perm = s->scantable.permutated;
1358 int token = *s->dct_tokens[plane][i];
1359 switch (token & 3) {
1361 if (--token < 4) // 0-3 are token types, so the EOB run must now be 0
1362 s->dct_tokens[plane][i]++;
1364 *s->dct_tokens[plane][i] = token & ~3;
1367 s->dct_tokens[plane][i]++;
1368 i += (token >> 2) & 0x7f;
1370 av_log(s->avctx, AV_LOG_ERROR, "Coefficient index overflow\n");
1373 block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1377 block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1378 s->dct_tokens[plane][i++]++;
1380 default: // shouldn't happen
1384 // return value is expected to be a valid level
1387 // the actual DC+prediction is in the fragment structure
1388 block[0] = frag->dc * s->qmat[0][inter][plane][0];
1393 * called when all pixels up to row y are complete
1395 static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1398 int offset[AV_NUM_DATA_POINTERS];
1400 if (HAVE_THREADS && s->avctx->active_thread_type&FF_THREAD_FRAME) {
1401 int y_flipped = s->flipped_image ? s->avctx->height-y : y;
1403 // At the end of the frame, report INT_MAX instead of the height of the frame.
1404 // This makes the other threads' ff_thread_await_progress() calls cheaper, because
1405 // they don't have to clip their values.
1406 ff_thread_report_progress(&s->current_frame, y_flipped==s->avctx->height ? INT_MAX : y_flipped-1, 0);
1409 if(s->avctx->draw_horiz_band==NULL)
1412 h= y - s->last_slice_end;
1413 s->last_slice_end= y;
1416 if (!s->flipped_image) {
1417 y = s->avctx->height - y - h;
1420 cy = y >> s->chroma_y_shift;
1421 offset[0] = s->current_frame.linesize[0]*y;
1422 offset[1] = s->current_frame.linesize[1]*cy;
1423 offset[2] = s->current_frame.linesize[2]*cy;
1424 for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
1428 s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
1432 * Wait for the reference frame of the current fragment.
1433 * The progress value is in luma pixel rows.
1435 static void await_reference_row(Vp3DecodeContext *s, Vp3Fragment *fragment, int motion_y, int y)
1439 int border = motion_y&1;
1441 if (fragment->coding_method == MODE_USING_GOLDEN ||
1442 fragment->coding_method == MODE_GOLDEN_MV)
1443 ref_frame = &s->golden_frame;
1445 ref_frame = &s->last_frame;
1447 ref_row = y + (motion_y>>1);
1448 ref_row = FFMAX(FFABS(ref_row), ref_row + 8 + border);
1450 ff_thread_await_progress(ref_frame, ref_row, 0);
1454 * Perform the final rendering for a particular slice of data.
1455 * The slice number ranges from 0..(c_superblock_height - 1).
1457 static void render_slice(Vp3DecodeContext *s, int slice)
1459 int x, y, i, j, fragment;
1460 LOCAL_ALIGNED_16(DCTELEM, block, [64]);
1461 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1462 int motion_halfpel_index;
1463 uint8_t *motion_source;
1464 int plane, first_pixel;
1466 if (slice >= s->c_superblock_height)
1469 for (plane = 0; plane < 3; plane++) {
1470 uint8_t *output_plane = s->current_frame.data [plane] + s->data_offset[plane];
1471 uint8_t * last_plane = s-> last_frame.data [plane] + s->data_offset[plane];
1472 uint8_t *golden_plane = s-> golden_frame.data [plane] + s->data_offset[plane];
1473 int stride = s->current_frame.linesize[plane];
1474 int plane_width = s->width >> (plane && s->chroma_x_shift);
1475 int plane_height = s->height >> (plane && s->chroma_y_shift);
1476 int8_t (*motion_val)[2] = s->motion_val[!!plane];
1478 int sb_x, sb_y = slice << (!plane && s->chroma_y_shift);
1479 int slice_height = sb_y + 1 + (!plane && s->chroma_y_shift);
1480 int slice_width = plane ? s->c_superblock_width : s->y_superblock_width;
1482 int fragment_width = s->fragment_width[!!plane];
1483 int fragment_height = s->fragment_height[!!plane];
1484 int fragment_start = s->fragment_start[plane];
1485 int do_await = !plane && HAVE_THREADS && (s->avctx->active_thread_type&FF_THREAD_FRAME);
1487 if (!s->flipped_image) stride = -stride;
1488 if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
1491 /* for each superblock row in the slice (both of them)... */
1492 for (; sb_y < slice_height; sb_y++) {
1494 /* for each superblock in a row... */
1495 for (sb_x = 0; sb_x < slice_width; sb_x++) {
1497 /* for each block in a superblock... */
1498 for (j = 0; j < 16; j++) {
1499 x = 4*sb_x + hilbert_offset[j][0];
1500 y = 4*sb_y + hilbert_offset[j][1];
1501 fragment = y*fragment_width + x;
1503 i = fragment_start + fragment;
1506 if (x >= fragment_width || y >= fragment_height)
1509 first_pixel = 8*y*stride + 8*x;
1511 if (do_await && s->all_fragments[i].coding_method != MODE_INTRA)
1512 await_reference_row(s, &s->all_fragments[i], motion_val[fragment][1], (16*y) >> s->chroma_y_shift);
1514 /* transform if this block was coded */
1515 if (s->all_fragments[i].coding_method != MODE_COPY) {
1516 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1517 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1518 motion_source= golden_plane;
1520 motion_source= last_plane;
1522 motion_source += first_pixel;
1523 motion_halfpel_index = 0;
1525 /* sort out the motion vector if this fragment is coded
1526 * using a motion vector method */
1527 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1528 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1530 motion_x = motion_val[fragment][0];
1531 motion_y = motion_val[fragment][1];
1533 src_x= (motion_x>>1) + 8*x;
1534 src_y= (motion_y>>1) + 8*y;
1536 motion_halfpel_index = motion_x & 0x01;
1537 motion_source += (motion_x >> 1);
1539 motion_halfpel_index |= (motion_y & 0x01) << 1;
1540 motion_source += ((motion_y >> 1) * stride);
1542 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1543 uint8_t *temp= s->edge_emu_buffer;
1544 if(stride<0) temp -= 8*stride;
1546 s->dsp.emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1547 motion_source= temp;
1552 /* first, take care of copying a block from either the
1553 * previous or the golden frame */
1554 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1555 /* Note, it is possible to implement all MC cases with
1556 put_no_rnd_pixels_l2 which would look more like the
1557 VP3 source but this would be slower as
1558 put_no_rnd_pixels_tab is better optimzed */
1559 if(motion_halfpel_index != 3){
1560 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1561 output_plane + first_pixel,
1562 motion_source, stride, 8);
1564 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1565 s->dsp.put_no_rnd_pixels_l2[1](
1566 output_plane + first_pixel,
1568 motion_source + stride + 1 + d,
1573 s->dsp.clear_block(block);
1575 /* invert DCT and place (or add) in final output */
1577 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1578 vp3_dequant(s, s->all_fragments + i, plane, 0, block);
1580 output_plane + first_pixel,
1584 if (vp3_dequant(s, s->all_fragments + i, plane, 1, block)) {
1586 output_plane + first_pixel,
1590 s->vp3dsp.idct_dc_add(output_plane + first_pixel, stride, block);
1595 /* copy directly from the previous frame */
1596 s->dsp.put_pixels_tab[1][0](
1597 output_plane + first_pixel,
1598 last_plane + first_pixel,
1605 // Filter up to the last row in the superblock row
1606 if (!s->skip_loop_filter)
1607 apply_loop_filter(s, plane, 4*sb_y - !!sb_y, FFMIN(4*sb_y+3, fragment_height-1));
1611 /* this looks like a good place for slice dispatch... */
1613 * if (slice == s->macroblock_height - 1)
1614 * dispatch (both last slice & 2nd-to-last slice);
1615 * else if (slice > 0)
1616 * dispatch (slice - 1);
1619 vp3_draw_horiz_band(s, FFMIN((32 << s->chroma_y_shift) * (slice + 1) -16, s->height-16));
1622 /// Allocate tables for per-frame data in Vp3DecodeContext
1623 static av_cold int allocate_tables(AVCodecContext *avctx)
1625 Vp3DecodeContext *s = avctx->priv_data;
1626 int y_fragment_count, c_fragment_count;
1628 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1629 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1631 s->superblock_coding = av_malloc(s->superblock_count);
1632 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1633 s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int));
1634 s->dct_tokens_base = av_malloc(64*s->fragment_count * sizeof(*s->dct_tokens_base));
1635 s->motion_val[0] = av_malloc(y_fragment_count * sizeof(*s->motion_val[0]));
1636 s->motion_val[1] = av_malloc(c_fragment_count * sizeof(*s->motion_val[1]));
1638 /* work out the block mapping tables */
1639 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1640 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1642 if (!s->superblock_coding || !s->all_fragments || !s->dct_tokens_base ||
1643 !s->coded_fragment_list[0] || !s->superblock_fragments || !s->macroblock_coding ||
1644 !s->motion_val[0] || !s->motion_val[1]) {
1645 vp3_decode_end(avctx);
1649 init_block_mapping(s);
1654 static av_cold int vp3_decode_init(AVCodecContext *avctx)
1656 Vp3DecodeContext *s = avctx->priv_data;
1657 int i, inter, plane;
1660 int y_fragment_count, c_fragment_count;
1662 if (avctx->codec_tag == MKTAG('V','P','3','0'))
1668 s->width = FFALIGN(avctx->width, 16);
1669 s->height = FFALIGN(avctx->height, 16);
1670 if (avctx->codec_id != AV_CODEC_ID_THEORA)
1671 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
1672 avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1673 ff_dsputil_init(&s->dsp, avctx);
1674 ff_vp3dsp_init(&s->vp3dsp, avctx->flags);
1676 ff_init_scantable_permutation(s->dsp.idct_permutation, s->vp3dsp.idct_perm);
1677 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1679 /* initialize to an impossible value which will force a recalculation
1680 * in the first frame decode */
1681 for (i = 0; i < 3; i++)
1684 avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift);
1686 s->y_superblock_width = (s->width + 31) / 32;
1687 s->y_superblock_height = (s->height + 31) / 32;
1688 s->y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1690 /* work out the dimensions for the C planes */
1691 c_width = s->width >> s->chroma_x_shift;
1692 c_height = s->height >> s->chroma_y_shift;
1693 s->c_superblock_width = (c_width + 31) / 32;
1694 s->c_superblock_height = (c_height + 31) / 32;
1695 s->c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1697 s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
1698 s->u_superblock_start = s->y_superblock_count;
1699 s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
1701 s->macroblock_width = (s->width + 15) / 16;
1702 s->macroblock_height = (s->height + 15) / 16;
1703 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1705 s->fragment_width[0] = s->width / FRAGMENT_PIXELS;
1706 s->fragment_height[0] = s->height / FRAGMENT_PIXELS;
1707 s->fragment_width[1] = s->fragment_width[0] >> s->chroma_x_shift;
1708 s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift;
1710 /* fragment count covers all 8x8 blocks for all 3 planes */
1711 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1712 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1713 s->fragment_count = y_fragment_count + 2*c_fragment_count;
1714 s->fragment_start[1] = y_fragment_count;
1715 s->fragment_start[2] = y_fragment_count + c_fragment_count;
1717 if (!s->theora_tables)
1719 for (i = 0; i < 64; i++) {
1720 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1721 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1722 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1723 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1724 s->base_matrix[2][i] = vp31_inter_dequant[i];
1725 s->filter_limit_values[i] = vp31_filter_limit_values[i];
1728 for(inter=0; inter<2; inter++){
1729 for(plane=0; plane<3; plane++){
1730 s->qr_count[inter][plane]= 1;
1731 s->qr_size [inter][plane][0]= 63;
1732 s->qr_base [inter][plane][0]=
1733 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1737 /* init VLC tables */
1738 for (i = 0; i < 16; i++) {
1741 init_vlc(&s->dc_vlc[i], 11, 32,
1742 &dc_bias[i][0][1], 4, 2,
1743 &dc_bias[i][0][0], 4, 2, 0);
1745 /* group 1 AC histograms */
1746 init_vlc(&s->ac_vlc_1[i], 11, 32,
1747 &ac_bias_0[i][0][1], 4, 2,
1748 &ac_bias_0[i][0][0], 4, 2, 0);
1750 /* group 2 AC histograms */
1751 init_vlc(&s->ac_vlc_2[i], 11, 32,
1752 &ac_bias_1[i][0][1], 4, 2,
1753 &ac_bias_1[i][0][0], 4, 2, 0);
1755 /* group 3 AC histograms */
1756 init_vlc(&s->ac_vlc_3[i], 11, 32,
1757 &ac_bias_2[i][0][1], 4, 2,
1758 &ac_bias_2[i][0][0], 4, 2, 0);
1760 /* group 4 AC histograms */
1761 init_vlc(&s->ac_vlc_4[i], 11, 32,
1762 &ac_bias_3[i][0][1], 4, 2,
1763 &ac_bias_3[i][0][0], 4, 2, 0);
1767 for (i = 0; i < 16; i++) {
1769 if (init_vlc(&s->dc_vlc[i], 11, 32,
1770 &s->huffman_table[i][0][1], 8, 4,
1771 &s->huffman_table[i][0][0], 8, 4, 0) < 0)
1774 /* group 1 AC histograms */
1775 if (init_vlc(&s->ac_vlc_1[i], 11, 32,
1776 &s->huffman_table[i+16][0][1], 8, 4,
1777 &s->huffman_table[i+16][0][0], 8, 4, 0) < 0)
1780 /* group 2 AC histograms */
1781 if (init_vlc(&s->ac_vlc_2[i], 11, 32,
1782 &s->huffman_table[i+16*2][0][1], 8, 4,
1783 &s->huffman_table[i+16*2][0][0], 8, 4, 0) < 0)
1786 /* group 3 AC histograms */
1787 if (init_vlc(&s->ac_vlc_3[i], 11, 32,
1788 &s->huffman_table[i+16*3][0][1], 8, 4,
1789 &s->huffman_table[i+16*3][0][0], 8, 4, 0) < 0)
1792 /* group 4 AC histograms */
1793 if (init_vlc(&s->ac_vlc_4[i], 11, 32,
1794 &s->huffman_table[i+16*4][0][1], 8, 4,
1795 &s->huffman_table[i+16*4][0][0], 8, 4, 0) < 0)
1800 init_vlc(&s->superblock_run_length_vlc, 6, 34,
1801 &superblock_run_length_vlc_table[0][1], 4, 2,
1802 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1804 init_vlc(&s->fragment_run_length_vlc, 5, 30,
1805 &fragment_run_length_vlc_table[0][1], 4, 2,
1806 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1808 init_vlc(&s->mode_code_vlc, 3, 8,
1809 &mode_code_vlc_table[0][1], 2, 1,
1810 &mode_code_vlc_table[0][0], 2, 1, 0);
1812 init_vlc(&s->motion_vector_vlc, 6, 63,
1813 &motion_vector_vlc_table[0][1], 2, 1,
1814 &motion_vector_vlc_table[0][0], 2, 1, 0);
1816 for (i = 0; i < 3; i++) {
1817 s->current_frame.data[i] = NULL;
1818 s->last_frame.data[i] = NULL;
1819 s->golden_frame.data[i] = NULL;
1822 return allocate_tables(avctx);
1825 av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1829 /// Release and shuffle frames after decode finishes
1830 static void update_frames(AVCodecContext *avctx)
1832 Vp3DecodeContext *s = avctx->priv_data;
1834 /* release the last frame, if it is allocated and if it is not the
1836 if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
1837 ff_thread_release_buffer(avctx, &s->last_frame);
1839 /* shuffle frames (last = current) */
1840 s->last_frame= s->current_frame;
1843 if (s->golden_frame.data[0])
1844 ff_thread_release_buffer(avctx, &s->golden_frame);
1845 s->golden_frame = s->current_frame;
1846 s->last_frame.type = FF_BUFFER_TYPE_COPY;
1849 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
1852 static int vp3_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
1854 Vp3DecodeContext *s = dst->priv_data, *s1 = src->priv_data;
1855 int qps_changed = 0, i, err;
1857 #define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
1859 if (!s1->current_frame.data[0]
1860 ||s->width != s1->width
1861 ||s->height!= s1->height) {
1863 copy_fields(s, s1, golden_frame, keyframe);
1868 // init tables if the first frame hasn't been decoded
1869 if (!s->current_frame.data[0]) {
1870 int y_fragment_count, c_fragment_count;
1872 err = allocate_tables(dst);
1875 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1876 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1877 memcpy(s->motion_val[0], s1->motion_val[0], y_fragment_count * sizeof(*s->motion_val[0]));
1878 memcpy(s->motion_val[1], s1->motion_val[1], c_fragment_count * sizeof(*s->motion_val[1]));
1881 // copy previous frame data
1882 copy_fields(s, s1, golden_frame, dsp);
1884 // copy qscale data if necessary
1885 for (i = 0; i < 3; i++) {
1886 if (s->qps[i] != s1->qps[1]) {
1888 memcpy(&s->qmat[i], &s1->qmat[i], sizeof(s->qmat[i]));
1892 if (s->qps[0] != s1->qps[0])
1893 memcpy(&s->bounding_values_array, &s1->bounding_values_array, sizeof(s->bounding_values_array));
1896 copy_fields(s, s1, qps, superblock_count);
1905 static int vp3_decode_frame(AVCodecContext *avctx,
1906 void *data, int *got_frame,
1909 const uint8_t *buf = avpkt->data;
1910 int buf_size = avpkt->size;
1911 Vp3DecodeContext *s = avctx->priv_data;
1915 init_get_bits(&gb, buf, buf_size * 8);
1917 if (s->theora && get_bits1(&gb))
1919 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1923 s->keyframe = !get_bits1(&gb);
1926 for (i = 0; i < 3; i++)
1927 s->last_qps[i] = s->qps[i];
1931 s->qps[s->nqps++]= get_bits(&gb, 6);
1932 } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1933 for (i = s->nqps; i < 3; i++)
1936 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1937 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1938 s->keyframe?"key":"", avctx->frame_number+1, s->qps[0]);
1940 s->skip_loop_filter = !s->filter_limit_values[s->qps[0]] ||
1941 avctx->skip_loop_filter >= (s->keyframe ? AVDISCARD_ALL : AVDISCARD_NONKEY);
1943 if (s->qps[0] != s->last_qps[0])
1944 init_loop_filter(s);
1946 for (i = 0; i < s->nqps; i++)
1947 // reinit all dequantizers if the first one changed, because
1948 // the DC of the first quantizer must be used for all matrices
1949 if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1950 init_dequantizer(s, i);
1952 if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1955 s->current_frame.reference = 3;
1956 s->current_frame.pict_type = s->keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P;
1957 s->current_frame.key_frame = s->keyframe;
1958 if (ff_thread_get_buffer(avctx, &s->current_frame) < 0) {
1959 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1963 if (!s->edge_emu_buffer)
1964 s->edge_emu_buffer = av_malloc(9*FFABS(s->current_frame.linesize[0]));
1969 skip_bits(&gb, 4); /* width code */
1970 skip_bits(&gb, 4); /* height code */
1973 s->version = get_bits(&gb, 5);
1974 if (avctx->frame_number == 0)
1975 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1978 if (s->version || s->theora)
1981 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1982 skip_bits(&gb, 2); /* reserved? */
1985 if (!s->golden_frame.data[0]) {
1986 av_log(s->avctx, AV_LOG_WARNING, "vp3: first frame not a keyframe\n");
1988 s->golden_frame.reference = 3;
1989 s->golden_frame.pict_type = AV_PICTURE_TYPE_I;
1990 if (ff_thread_get_buffer(avctx, &s->golden_frame) < 0) {
1991 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1994 s->last_frame = s->golden_frame;
1995 s->last_frame.type = FF_BUFFER_TYPE_COPY;
1996 ff_thread_report_progress(&s->last_frame, INT_MAX, 0);
2000 memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
2001 ff_thread_finish_setup(avctx);
2003 if (unpack_superblocks(s, &gb)){
2004 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
2007 if (unpack_modes(s, &gb)){
2008 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
2011 if (unpack_vectors(s, &gb)){
2012 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2015 if (unpack_block_qpis(s, &gb)){
2016 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
2019 if (unpack_dct_coeffs(s, &gb)){
2020 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
2024 for (i = 0; i < 3; i++) {
2025 int height = s->height >> (i && s->chroma_y_shift);
2026 if (s->flipped_image)
2027 s->data_offset[i] = 0;
2029 s->data_offset[i] = (height-1) * s->current_frame.linesize[i];
2032 s->last_slice_end = 0;
2033 for (i = 0; i < s->c_superblock_height; i++)
2036 // filter the last row
2037 for (i = 0; i < 3; i++) {
2038 int row = (s->height >> (3+(i && s->chroma_y_shift))) - 1;
2039 apply_loop_filter(s, i, row, row+1);
2041 vp3_draw_horiz_band(s, s->avctx->height);
2044 *(AVFrame*)data= s->current_frame;
2046 if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME))
2047 update_frames(avctx);
2052 ff_thread_report_progress(&s->current_frame, INT_MAX, 0);
2054 if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME))
2055 avctx->release_buffer(avctx, &s->current_frame);
2060 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2062 Vp3DecodeContext *s = avctx->priv_data;
2064 if (get_bits1(gb)) {
2066 if (s->entries >= 32) { /* overflow */
2067 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2070 token = get_bits(gb, 5);
2071 av_dlog(avctx, "hti %d hbits %x token %d entry : %d size %d\n",
2072 s->hti, s->hbits, token, s->entries, s->huff_code_size);
2073 s->huffman_table[s->hti][token][0] = s->hbits;
2074 s->huffman_table[s->hti][token][1] = s->huff_code_size;
2078 if (s->huff_code_size >= 32) {/* overflow */
2079 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2082 s->huff_code_size++;
2084 if (read_huffman_tree(avctx, gb))
2087 if (read_huffman_tree(avctx, gb))
2090 s->huff_code_size--;
2095 static int vp3_init_thread_copy(AVCodecContext *avctx)
2097 Vp3DecodeContext *s = avctx->priv_data;
2099 s->superblock_coding = NULL;
2100 s->all_fragments = NULL;
2101 s->coded_fragment_list[0] = NULL;
2102 s->dct_tokens_base = NULL;
2103 s->superblock_fragments = NULL;
2104 s->macroblock_coding = NULL;
2105 s->motion_val[0] = NULL;
2106 s->motion_val[1] = NULL;
2107 s->edge_emu_buffer = NULL;
2112 #if CONFIG_THEORA_DECODER
2113 static const enum AVPixelFormat theora_pix_fmts[4] = {
2114 AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P
2117 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2119 Vp3DecodeContext *s = avctx->priv_data;
2120 int visible_width, visible_height, colorspace;
2121 int offset_x = 0, offset_y = 0;
2122 AVRational fps, aspect;
2124 s->theora = get_bits_long(gb, 24);
2125 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2127 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2128 /* but previous versions have the image flipped relative to vp3 */
2129 if (s->theora < 0x030200)
2131 s->flipped_image = 1;
2132 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2135 visible_width = s->width = get_bits(gb, 16) << 4;
2136 visible_height = s->height = get_bits(gb, 16) << 4;
2138 if(av_image_check_size(s->width, s->height, 0, avctx)){
2139 av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2140 s->width= s->height= 0;
2144 if (s->theora >= 0x030200) {
2145 visible_width = get_bits_long(gb, 24);
2146 visible_height = get_bits_long(gb, 24);
2148 offset_x = get_bits(gb, 8); /* offset x */
2149 offset_y = get_bits(gb, 8); /* offset y, from bottom */
2152 fps.num = get_bits_long(gb, 32);
2153 fps.den = get_bits_long(gb, 32);
2154 if (fps.num && fps.den) {
2155 av_reduce(&avctx->time_base.num, &avctx->time_base.den,
2156 fps.den, fps.num, 1<<30);
2159 aspect.num = get_bits_long(gb, 24);
2160 aspect.den = get_bits_long(gb, 24);
2161 if (aspect.num && aspect.den) {
2162 av_reduce(&avctx->sample_aspect_ratio.num,
2163 &avctx->sample_aspect_ratio.den,
2164 aspect.num, aspect.den, 1<<30);
2167 if (s->theora < 0x030200)
2168 skip_bits(gb, 5); /* keyframe frequency force */
2169 colorspace = get_bits(gb, 8);
2170 skip_bits(gb, 24); /* bitrate */
2172 skip_bits(gb, 6); /* quality hint */
2174 if (s->theora >= 0x030200)
2176 skip_bits(gb, 5); /* keyframe frequency force */
2177 avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)];
2178 if (avctx->pix_fmt == AV_PIX_FMT_NONE) {
2179 av_log(avctx, AV_LOG_ERROR, "Invalid pixel format\n");
2180 return AVERROR_INVALIDDATA;
2182 skip_bits(gb, 3); /* reserved */
2185 // align_get_bits(gb);
2187 if ( visible_width <= s->width && visible_width > s->width-16
2188 && visible_height <= s->height && visible_height > s->height-16
2189 && !offset_x && (offset_y == s->height - visible_height))
2190 avcodec_set_dimensions(avctx, visible_width, visible_height);
2192 avcodec_set_dimensions(avctx, s->width, s->height);
2194 if (colorspace == 1) {
2195 avctx->color_primaries = AVCOL_PRI_BT470M;
2196 } else if (colorspace == 2) {
2197 avctx->color_primaries = AVCOL_PRI_BT470BG;
2199 if (colorspace == 1 || colorspace == 2) {
2200 avctx->colorspace = AVCOL_SPC_BT470BG;
2201 avctx->color_trc = AVCOL_TRC_BT709;
2207 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2209 Vp3DecodeContext *s = avctx->priv_data;
2210 int i, n, matrices, inter, plane;
2212 if (s->theora >= 0x030200) {
2213 n = get_bits(gb, 3);
2214 /* loop filter limit values table */
2216 for (i = 0; i < 64; i++)
2217 s->filter_limit_values[i] = get_bits(gb, n);
2220 if (s->theora >= 0x030200)
2221 n = get_bits(gb, 4) + 1;
2224 /* quality threshold table */
2225 for (i = 0; i < 64; i++)
2226 s->coded_ac_scale_factor[i] = get_bits(gb, n);
2228 if (s->theora >= 0x030200)
2229 n = get_bits(gb, 4) + 1;
2232 /* dc scale factor table */
2233 for (i = 0; i < 64; i++)
2234 s->coded_dc_scale_factor[i] = get_bits(gb, n);
2236 if (s->theora >= 0x030200)
2237 matrices = get_bits(gb, 9) + 1;
2242 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2246 for(n=0; n<matrices; n++){
2247 for (i = 0; i < 64; i++)
2248 s->base_matrix[n][i]= get_bits(gb, 8);
2251 for (inter = 0; inter <= 1; inter++) {
2252 for (plane = 0; plane <= 2; plane++) {
2254 if (inter || plane > 0)
2255 newqr = get_bits1(gb);
2258 if(inter && get_bits1(gb)){
2262 qtj= (3*inter + plane - 1) / 3;
2263 plj= (plane + 2) % 3;
2265 s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2266 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2267 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2273 i= get_bits(gb, av_log2(matrices-1)+1);
2275 av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2278 s->qr_base[inter][plane][qri]= i;
2281 i = get_bits(gb, av_log2(63-qi)+1) + 1;
2282 s->qr_size[inter][plane][qri++]= i;
2287 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2290 s->qr_count[inter][plane]= qri;
2295 /* Huffman tables */
2296 for (s->hti = 0; s->hti < 80; s->hti++) {
2298 s->huff_code_size = 1;
2299 if (!get_bits1(gb)) {
2301 if(read_huffman_tree(avctx, gb))
2304 if(read_huffman_tree(avctx, gb))
2309 s->theora_tables = 1;
2314 static av_cold int theora_decode_init(AVCodecContext *avctx)
2316 Vp3DecodeContext *s = avctx->priv_data;
2319 uint8_t *header_start[3];
2323 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
2327 if (!avctx->extradata_size)
2329 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2333 if (avpriv_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2334 42, header_start, header_len) < 0) {
2335 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2340 init_get_bits(&gb, header_start[i], header_len[i] * 8);
2342 ptype = get_bits(&gb, 8);
2344 if (!(ptype & 0x80))
2346 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2350 // FIXME: Check for this as well.
2351 skip_bits_long(&gb, 6*8); /* "theora" */
2356 if (theora_decode_header(avctx, &gb) < 0)
2360 // FIXME: is this needed? it breaks sometimes
2361 // theora_decode_comments(avctx, gb);
2364 if (theora_decode_tables(avctx, &gb))
2368 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2371 if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2372 av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2373 if (s->theora < 0x030200)
2377 return vp3_decode_init(avctx);
2380 AVCodec ff_theora_decoder = {
2382 .type = AVMEDIA_TYPE_VIDEO,
2383 .id = AV_CODEC_ID_THEORA,
2384 .priv_data_size = sizeof(Vp3DecodeContext),
2385 .init = theora_decode_init,
2386 .close = vp3_decode_end,
2387 .decode = vp3_decode_frame,
2388 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND |
2389 CODEC_CAP_FRAME_THREADS,
2390 .flush = vp3_decode_flush,
2391 .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2392 .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
2393 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context)
2397 AVCodec ff_vp3_decoder = {
2399 .type = AVMEDIA_TYPE_VIDEO,
2400 .id = AV_CODEC_ID_VP3,
2401 .priv_data_size = sizeof(Vp3DecodeContext),
2402 .init = vp3_decode_init,
2403 .close = vp3_decode_end,
2404 .decode = vp3_decode_frame,
2405 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND |
2406 CODEC_CAP_FRAME_THREADS,
2407 .flush = vp3_decode_flush,
2408 .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2409 .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
2410 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context),