2 * Copyright (C) 2003-2004 the ffmpeg project
4 * This file is part of Libav.
6 * Libav 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 * Libav 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 Libav; 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"
49 #define FRAGMENT_PIXELS 8
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] = {
82 /* scheme 1: Last motion vector dominates */
83 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
84 MODE_INTER_PLUS_MV, MODE_INTER_NO_MV,
85 MODE_INTRA, MODE_USING_GOLDEN,
86 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
89 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
90 MODE_INTER_NO_MV, MODE_INTER_PLUS_MV,
91 MODE_INTRA, MODE_USING_GOLDEN,
92 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
95 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
96 MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
97 MODE_INTRA, MODE_USING_GOLDEN,
98 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
101 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
102 MODE_INTER_NO_MV, MODE_INTER_PRIOR_LAST,
103 MODE_INTRA, MODE_USING_GOLDEN,
104 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
106 /* scheme 5: No motion vector dominates */
107 { MODE_INTER_NO_MV, MODE_INTER_LAST_MV,
108 MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
109 MODE_INTRA, MODE_USING_GOLDEN,
110 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
113 { MODE_INTER_NO_MV, MODE_USING_GOLDEN,
114 MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
115 MODE_INTER_PLUS_MV, MODE_INTRA,
116 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
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 ThreadFrame golden_frame;
135 ThreadFrame last_frame;
136 ThreadFrame current_frame;
138 uint8_t idct_permutation[64];
139 uint8_t idct_scantable[64];
141 VideoDSPContext vdsp;
142 VP3DSPContext vp3dsp;
143 DECLARE_ALIGNED(16, int16_t, block)[64];
146 int skip_loop_filter;
152 int superblock_count;
153 int y_superblock_width;
154 int y_superblock_height;
155 int y_superblock_count;
156 int c_superblock_width;
157 int c_superblock_height;
158 int c_superblock_count;
159 int u_superblock_start;
160 int v_superblock_start;
161 unsigned char *superblock_coding;
163 int macroblock_count;
164 int macroblock_width;
165 int macroblock_height;
168 int fragment_width[2];
169 int fragment_height[2];
171 Vp3Fragment *all_fragments;
172 int fragment_start[3];
177 int8_t (*motion_val[2])[2];
180 uint16_t coded_dc_scale_factor[64];
181 uint32_t coded_ac_scale_factor[64];
182 uint8_t base_matrix[384][64];
183 uint8_t qr_count[2][3];
184 uint8_t qr_size[2][3][64];
185 uint16_t qr_base[2][3][64];
188 * This is a list of all tokens in bitstream order. Reordering takes place
189 * by pulling from each level during IDCT. As a consequence, IDCT must be
190 * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32
191 * otherwise. The 32 different tokens with up to 12 bits of extradata are
192 * collapsed into 3 types, packed as follows:
193 * (from the low to high bits)
195 * 2 bits: type (0,1,2)
196 * 0: EOB run, 14 bits for run length (12 needed)
197 * 1: zero run, 7 bits for run length
198 * 7 bits for the next coefficient (3 needed)
199 * 2: coefficient, 14 bits (11 needed)
201 * Coefficients are signed, so are packed in the highest bits for automatic
204 int16_t *dct_tokens[3][64];
205 int16_t *dct_tokens_base;
206 #define TOKEN_EOB(eob_run) ((eob_run) << 2)
207 #define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) << 9) + ((zero_run) << 2) + 1)
208 #define TOKEN_COEFF(coeff) (((coeff) << 2) + 2)
211 * number of blocks that contain DCT coefficients at
212 * the given level or higher
214 int num_coded_frags[3][64];
215 int total_num_coded_frags;
217 /* this is a list of indexes into the all_fragments array indicating
218 * which of the fragments are coded */
219 int *coded_fragment_list[3];
227 VLC superblock_run_length_vlc;
228 VLC fragment_run_length_vlc;
230 VLC motion_vector_vlc;
232 /* these arrays need to be on 16-byte boundaries since SSE2 operations
234 DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64]; ///< qmat[qpi][is_inter][plane]
236 /* This table contains superblock_count * 16 entries. Each set of 16
237 * numbers corresponds to the fragment indexes 0..15 of the superblock.
238 * An entry will be -1 to indicate that no entry corresponds to that
240 int *superblock_fragments;
242 /* This is an array that indicates how a particular macroblock
244 unsigned char *macroblock_coding;
246 uint8_t *edge_emu_buffer;
253 uint32_t huffman_table[80][32][2];
255 uint8_t filter_limit_values[64];
256 DECLARE_ALIGNED(8, int, bounding_values_array)[256 + 2];
259 /************************************************************************
260 * VP3 specific functions
261 ************************************************************************/
263 static void vp3_decode_flush(AVCodecContext *avctx)
265 Vp3DecodeContext *s = avctx->priv_data;
267 if (s->golden_frame.f)
268 ff_thread_release_buffer(avctx, &s->golden_frame);
270 ff_thread_release_buffer(avctx, &s->last_frame);
271 if (s->current_frame.f)
272 ff_thread_release_buffer(avctx, &s->current_frame);
275 static av_cold int vp3_decode_end(AVCodecContext *avctx)
277 Vp3DecodeContext *s = avctx->priv_data;
280 av_freep(&s->superblock_coding);
281 av_freep(&s->all_fragments);
282 av_freep(&s->coded_fragment_list[0]);
283 av_freep(&s->dct_tokens_base);
284 av_freep(&s->superblock_fragments);
285 av_freep(&s->macroblock_coding);
286 av_freep(&s->motion_val[0]);
287 av_freep(&s->motion_val[1]);
288 av_freep(&s->edge_emu_buffer);
290 /* release all frames */
291 vp3_decode_flush(avctx);
292 av_frame_free(&s->current_frame.f);
293 av_frame_free(&s->last_frame.f);
294 av_frame_free(&s->golden_frame.f);
296 if (avctx->internal->is_copy)
299 for (i = 0; i < 16; i++) {
300 ff_free_vlc(&s->dc_vlc[i]);
301 ff_free_vlc(&s->ac_vlc_1[i]);
302 ff_free_vlc(&s->ac_vlc_2[i]);
303 ff_free_vlc(&s->ac_vlc_3[i]);
304 ff_free_vlc(&s->ac_vlc_4[i]);
307 ff_free_vlc(&s->superblock_run_length_vlc);
308 ff_free_vlc(&s->fragment_run_length_vlc);
309 ff_free_vlc(&s->mode_code_vlc);
310 ff_free_vlc(&s->motion_vector_vlc);
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
329 : s->y_superblock_width;
330 int sb_height = plane ? s->c_superblock_height
331 : s->y_superblock_height;
332 int frag_width = s->fragment_width[!!plane];
333 int frag_height = s->fragment_height[!!plane];
335 for (sb_y = 0; sb_y < sb_height; sb_y++)
336 for (sb_x = 0; sb_x < sb_width; sb_x++)
337 for (i = 0; i < 16; i++) {
338 x = 4 * sb_x + hilbert_offset[i][0];
339 y = 4 * sb_y + hilbert_offset[i][1];
341 if (x < frag_width && y < frag_height)
342 s->superblock_fragments[j++] = s->fragment_start[plane] +
345 s->superblock_fragments[j++] = -1;
349 return 0; /* successful path out */
353 * This function sets up the dequantization tables used for a particular
356 static void init_dequantizer(Vp3DecodeContext *s, int qpi)
358 int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
359 int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
360 int i, plane, inter, qri, bmi, bmj, qistart;
362 for (inter = 0; inter < 2; inter++) {
363 for (plane = 0; plane < 3; plane++) {
365 for (qri = 0; qri < s->qr_count[inter][plane]; qri++) {
366 sum += s->qr_size[inter][plane][qri];
367 if (s->qps[qpi] <= sum)
370 qistart = sum - s->qr_size[inter][plane][qri];
371 bmi = s->qr_base[inter][plane][qri];
372 bmj = s->qr_base[inter][plane][qri + 1];
373 for (i = 0; i < 64; i++) {
374 int coeff = (2 * (sum - s->qps[qpi]) * s->base_matrix[bmi][i] -
375 2 * (qistart - s->qps[qpi]) * s->base_matrix[bmj][i] +
376 s->qr_size[inter][plane][qri]) /
377 (2 * s->qr_size[inter][plane][qri]);
379 int qmin = 8 << (inter + !i);
380 int qscale = i ? ac_scale_factor : dc_scale_factor;
382 s->qmat[qpi][inter][plane][s->idct_permutation[i]] =
383 av_clip((qscale * coeff) / 100 * 4, qmin, 4096);
385 /* all DC coefficients use the same quant so as not to interfere
386 * with DC prediction */
387 s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
393 * This function initializes the loop filter boundary limits if the frame's
394 * quality index is different from the previous frame's.
396 * The filter_limit_values may not be larger than 127.
398 static void init_loop_filter(Vp3DecodeContext *s)
400 int *bounding_values = s->bounding_values_array + 127;
405 filter_limit = s->filter_limit_values[s->qps[0]];
406 assert(filter_limit < 128);
408 /* set up the bounding values */
409 memset(s->bounding_values_array, 0, 256 * sizeof(int));
410 for (x = 0; x < filter_limit; x++) {
411 bounding_values[-x] = -x;
412 bounding_values[x] = x;
414 for (x = value = filter_limit; x < 128 && value; x++, value--) {
415 bounding_values[ x] = value;
416 bounding_values[-x] = -value;
419 bounding_values[128] = value;
420 bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
424 * This function unpacks all of the superblock/macroblock/fragment coding
425 * information from the bitstream.
427 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
429 int superblock_starts[3] = {
430 0, s->u_superblock_start, s->v_superblock_start
433 int current_superblock = 0;
435 int num_partial_superblocks = 0;
438 int current_fragment;
442 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
444 /* unpack the list of partially-coded superblocks */
445 bit = get_bits1(gb) ^ 1;
448 while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) {
449 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
454 current_run = get_vlc2(gb, s->superblock_run_length_vlc.table,
456 if (current_run == 34)
457 current_run += get_bits(gb, 12);
459 if (current_superblock + current_run > s->superblock_count) {
460 av_log(s->avctx, AV_LOG_ERROR,
461 "Invalid partially coded superblock run length\n");
465 memset(s->superblock_coding + current_superblock, bit, current_run);
467 current_superblock += current_run;
469 num_partial_superblocks += current_run;
472 /* unpack the list of fully coded superblocks if any of the blocks were
473 * not marked as partially coded in the previous step */
474 if (num_partial_superblocks < s->superblock_count) {
475 int superblocks_decoded = 0;
477 current_superblock = 0;
478 bit = get_bits1(gb) ^ 1;
481 while (superblocks_decoded < s->superblock_count - num_partial_superblocks &&
482 get_bits_left(gb) > 0) {
483 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
488 current_run = get_vlc2(gb, s->superblock_run_length_vlc.table,
490 if (current_run == 34)
491 current_run += get_bits(gb, 12);
493 for (j = 0; j < current_run; current_superblock++) {
494 if (current_superblock >= s->superblock_count) {
495 av_log(s->avctx, AV_LOG_ERROR,
496 "Invalid fully coded superblock run length\n");
500 /* skip any superblocks already marked as partially coded */
501 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
502 s->superblock_coding[current_superblock] = 2 * bit;
506 superblocks_decoded += current_run;
510 /* if there were partial blocks, initialize bitstream for
511 * unpacking fragment codings */
512 if (num_partial_superblocks) {
515 /* toggle the bit because as soon as the first run length is
516 * fetched the bit will be toggled again */
521 /* figure out which fragments are coded; iterate through each
522 * superblock (all planes) */
523 s->total_num_coded_frags = 0;
524 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
526 for (plane = 0; plane < 3; plane++) {
527 int sb_start = superblock_starts[plane];
528 int sb_end = sb_start + (plane ? s->c_superblock_count
529 : s->y_superblock_count);
530 int num_coded_frags = 0;
532 for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) {
533 /* iterate through all 16 fragments in a superblock */
534 for (j = 0; j < 16; j++) {
535 /* if the fragment is in bounds, check its coding status */
536 current_fragment = s->superblock_fragments[i * 16 + j];
537 if (current_fragment != -1) {
538 int coded = s->superblock_coding[i];
540 if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
541 /* fragment may or may not be coded; this is the case
542 * that cares about the fragment coding runs */
543 if (current_run-- == 0) {
545 current_run = get_vlc2(gb, s->fragment_run_length_vlc.table, 5, 2);
551 /* default mode; actual mode will be decoded in
553 s->all_fragments[current_fragment].coding_method =
555 s->coded_fragment_list[plane][num_coded_frags++] =
558 /* not coded; copy this fragment from the prior frame */
559 s->all_fragments[current_fragment].coding_method =
565 s->total_num_coded_frags += num_coded_frags;
566 for (i = 0; i < 64; i++)
567 s->num_coded_frags[plane][i] = num_coded_frags;
569 s->coded_fragment_list[plane + 1] = s->coded_fragment_list[plane] +
576 * This function unpacks all the coding mode data for individual macroblocks
577 * from the bitstream.
579 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
581 int i, j, k, sb_x, sb_y;
583 int current_macroblock;
584 int current_fragment;
586 int custom_mode_alphabet[CODING_MODE_COUNT];
591 for (i = 0; i < s->fragment_count; i++)
592 s->all_fragments[i].coding_method = MODE_INTRA;
594 /* fetch the mode coding scheme for this frame */
595 scheme = get_bits(gb, 3);
597 /* is it a custom coding scheme? */
599 for (i = 0; i < 8; i++)
600 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
601 for (i = 0; i < 8; i++)
602 custom_mode_alphabet[get_bits(gb, 3)] = i;
603 alphabet = custom_mode_alphabet;
605 alphabet = ModeAlphabet[scheme - 1];
607 /* iterate through all of the macroblocks that contain 1 or more
609 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
610 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
611 if (get_bits_left(gb) <= 0)
614 for (j = 0; j < 4; j++) {
615 int mb_x = 2 * sb_x + (j >> 1);
616 int mb_y = 2 * sb_y + (((j >> 1) + j) & 1);
617 current_macroblock = mb_y * s->macroblock_width + mb_x;
619 if (mb_x >= s->macroblock_width ||
620 mb_y >= s->macroblock_height)
623 #define BLOCK_X (2 * mb_x + (k & 1))
624 #define BLOCK_Y (2 * mb_y + (k >> 1))
625 /* coding modes are only stored if the macroblock has
626 * at least one luma block coded, otherwise it must be
628 for (k = 0; k < 4; k++) {
629 current_fragment = BLOCK_Y *
630 s->fragment_width[0] + BLOCK_X;
631 if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
635 s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
639 /* mode 7 means get 3 bits for each coding mode */
641 coding_mode = get_bits(gb, 3);
643 coding_mode = alphabet[get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
645 s->macroblock_coding[current_macroblock] = coding_mode;
646 for (k = 0; k < 4; k++) {
647 frag = s->all_fragments + BLOCK_Y * s->fragment_width[0] + BLOCK_X;
648 if (frag->coding_method != MODE_COPY)
649 frag->coding_method = coding_mode;
652 #define SET_CHROMA_MODES \
653 if (frag[s->fragment_start[1]].coding_method != MODE_COPY) \
654 frag[s->fragment_start[1]].coding_method = coding_mode; \
655 if (frag[s->fragment_start[2]].coding_method != MODE_COPY) \
656 frag[s->fragment_start[2]].coding_method = coding_mode;
658 if (s->chroma_y_shift) {
659 frag = s->all_fragments + mb_y *
660 s->fragment_width[1] + mb_x;
662 } else if (s->chroma_x_shift) {
663 frag = s->all_fragments +
664 2 * mb_y * s->fragment_width[1] + mb_x;
665 for (k = 0; k < 2; k++) {
667 frag += s->fragment_width[1];
670 for (k = 0; k < 4; k++) {
671 frag = s->all_fragments +
672 BLOCK_Y * s->fragment_width[1] + BLOCK_X;
685 * This function unpacks all the motion vectors for the individual
686 * macroblocks from the bitstream.
688 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
690 int j, k, sb_x, sb_y;
694 int last_motion_x = 0;
695 int last_motion_y = 0;
696 int prior_last_motion_x = 0;
697 int prior_last_motion_y = 0;
698 int current_macroblock;
699 int current_fragment;
705 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
706 coding_mode = get_bits1(gb);
708 /* iterate through all of the macroblocks that contain 1 or more
710 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
711 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
712 if (get_bits_left(gb) <= 0)
715 for (j = 0; j < 4; j++) {
716 int mb_x = 2 * sb_x + (j >> 1);
717 int mb_y = 2 * sb_y + (((j >> 1) + j) & 1);
718 current_macroblock = mb_y * s->macroblock_width + mb_x;
720 if (mb_x >= s->macroblock_width ||
721 mb_y >= s->macroblock_height ||
722 s->macroblock_coding[current_macroblock] == MODE_COPY)
725 switch (s->macroblock_coding[current_macroblock]) {
726 case MODE_INTER_PLUS_MV:
728 /* all 6 fragments use the same motion vector */
729 if (coding_mode == 0) {
730 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
731 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
733 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
734 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
737 /* vector maintenance, only on MODE_INTER_PLUS_MV */
738 if (s->macroblock_coding[current_macroblock] == MODE_INTER_PLUS_MV) {
739 prior_last_motion_x = last_motion_x;
740 prior_last_motion_y = last_motion_y;
741 last_motion_x = motion_x[0];
742 last_motion_y = motion_y[0];
746 case MODE_INTER_FOURMV:
747 /* vector maintenance */
748 prior_last_motion_x = last_motion_x;
749 prior_last_motion_y = last_motion_y;
751 /* fetch 4 vectors from the bitstream, one for each
752 * Y fragment, then average for the C fragment vectors */
753 for (k = 0; k < 4; k++) {
754 current_fragment = BLOCK_Y * s->fragment_width[0] + BLOCK_X;
755 if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
756 if (coding_mode == 0) {
757 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
758 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
760 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
761 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
763 last_motion_x = motion_x[k];
764 last_motion_y = motion_y[k];
772 case MODE_INTER_LAST_MV:
773 /* all 6 fragments use the last motion vector */
774 motion_x[0] = last_motion_x;
775 motion_y[0] = last_motion_y;
777 /* no vector maintenance (last vector remains the
781 case MODE_INTER_PRIOR_LAST:
782 /* all 6 fragments use the motion vector prior to the
783 * last motion vector */
784 motion_x[0] = prior_last_motion_x;
785 motion_y[0] = prior_last_motion_y;
787 /* vector maintenance */
788 prior_last_motion_x = last_motion_x;
789 prior_last_motion_y = last_motion_y;
790 last_motion_x = motion_x[0];
791 last_motion_y = motion_y[0];
795 /* covers intra, inter without MV, golden without MV */
799 /* no vector maintenance */
803 /* assign the motion vectors to the correct fragments */
804 for (k = 0; k < 4; k++) {
806 BLOCK_Y * s->fragment_width[0] + BLOCK_X;
807 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
808 s->motion_val[0][current_fragment][0] = motion_x[k];
809 s->motion_val[0][current_fragment][1] = motion_y[k];
811 s->motion_val[0][current_fragment][0] = motion_x[0];
812 s->motion_val[0][current_fragment][1] = motion_y[0];
816 if (s->chroma_y_shift) {
817 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
818 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] +
819 motion_x[2] + motion_x[3], 2);
820 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] +
821 motion_y[2] + motion_y[3], 2);
823 motion_x[0] = (motion_x[0] >> 1) | (motion_x[0] & 1);
824 motion_y[0] = (motion_y[0] >> 1) | (motion_y[0] & 1);
825 frag = mb_y * s->fragment_width[1] + mb_x;
826 s->motion_val[1][frag][0] = motion_x[0];
827 s->motion_val[1][frag][1] = motion_y[0];
828 } else if (s->chroma_x_shift) {
829 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
830 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1);
831 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1);
832 motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1);
833 motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1);
835 motion_x[1] = motion_x[0];
836 motion_y[1] = motion_y[0];
838 motion_x[0] = (motion_x[0] >> 1) | (motion_x[0] & 1);
839 motion_x[1] = (motion_x[1] >> 1) | (motion_x[1] & 1);
841 frag = 2 * mb_y * s->fragment_width[1] + mb_x;
842 for (k = 0; k < 2; k++) {
843 s->motion_val[1][frag][0] = motion_x[k];
844 s->motion_val[1][frag][1] = motion_y[k];
845 frag += s->fragment_width[1];
848 for (k = 0; k < 4; k++) {
849 frag = BLOCK_Y * s->fragment_width[1] + BLOCK_X;
850 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
851 s->motion_val[1][frag][0] = motion_x[k];
852 s->motion_val[1][frag][1] = motion_y[k];
854 s->motion_val[1][frag][0] = motion_x[0];
855 s->motion_val[1][frag][1] = motion_y[0];
866 static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
868 int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
869 int num_blocks = s->total_num_coded_frags;
871 for (qpi = 0; qpi < s->nqps - 1 && num_blocks > 0; qpi++) {
872 i = blocks_decoded = num_blocks_at_qpi = 0;
874 bit = get_bits1(gb) ^ 1;
878 if (run_length == MAXIMUM_LONG_BIT_RUN)
883 run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
884 if (run_length == 34)
885 run_length += get_bits(gb, 12);
886 blocks_decoded += run_length;
889 num_blocks_at_qpi += run_length;
891 for (j = 0; j < run_length; i++) {
892 if (i >= s->total_num_coded_frags)
895 if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
896 s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
900 } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0);
902 num_blocks -= num_blocks_at_qpi;
909 * This function is called by unpack_dct_coeffs() to extract the VLCs from
910 * the bitstream. The VLCs encode tokens which are used to unpack DCT
911 * data. This function unpacks all the VLCs for either the Y plane or both
912 * C planes, and is called for DC coefficients or different AC coefficient
913 * levels (since different coefficient types require different VLC tables.
915 * This function returns a residual eob run. E.g, if a particular token gave
916 * instructions to EOB the next 5 fragments and there were only 2 fragments
917 * left in the current fragment range, 3 would be returned so that it could
918 * be passed into the next call to this same function.
920 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
921 VLC *table, int coeff_index,
932 int num_coeffs = s->num_coded_frags[plane][coeff_index];
933 int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
935 /* local references to structure members to avoid repeated deferences */
936 int *coded_fragment_list = s->coded_fragment_list[plane];
937 Vp3Fragment *all_fragments = s->all_fragments;
938 VLC_TYPE(*vlc_table)[2] = table->table;
941 av_log(s->avctx, AV_LOG_ERROR,
942 "Invalid number of coefficents at level %d\n", coeff_index);
944 if (eob_run > num_coeffs) {
946 blocks_ended = num_coeffs;
947 eob_run -= num_coeffs;
950 blocks_ended = eob_run;
954 // insert fake EOB token to cover the split between planes or zzi
956 dct_tokens[j++] = blocks_ended << 2;
958 while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
959 /* decode a VLC into a token */
960 token = get_vlc2(gb, vlc_table, 11, 3);
961 /* use the token to get a zero run, a coefficient, and an eob run */
962 if ((unsigned) token <= 6U) {
963 eob_run = eob_run_base[token];
964 if (eob_run_get_bits[token])
965 eob_run += get_bits(gb, eob_run_get_bits[token]);
967 // record only the number of blocks ended in this plane,
968 // any spill will be recorded in the next plane.
969 if (eob_run > num_coeffs - coeff_i) {
970 dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
971 blocks_ended += num_coeffs - coeff_i;
972 eob_run -= num_coeffs - coeff_i;
973 coeff_i = num_coeffs;
975 dct_tokens[j++] = TOKEN_EOB(eob_run);
976 blocks_ended += eob_run;
980 } else if (token >= 0) {
981 bits_to_get = coeff_get_bits[token];
983 bits_to_get = get_bits(gb, bits_to_get);
984 coeff = coeff_tables[token][bits_to_get];
986 zero_run = zero_run_base[token];
987 if (zero_run_get_bits[token])
988 zero_run += get_bits(gb, zero_run_get_bits[token]);
991 dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
993 // Save DC into the fragment structure. DC prediction is
994 // done in raster order, so the actual DC can't be in with
995 // other tokens. We still need the token in dct_tokens[]
996 // however, or else the structure collapses on itself.
998 all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
1000 dct_tokens[j++] = TOKEN_COEFF(coeff);
1003 if (coeff_index + zero_run > 64) {
1004 av_log(s->avctx, AV_LOG_DEBUG,
1005 "Invalid zero run of %d with %d coeffs left\n",
1006 zero_run, 64 - coeff_index);
1007 zero_run = 64 - coeff_index;
1010 // zero runs code multiple coefficients,
1011 // so don't try to decode coeffs for those higher levels
1012 for (i = coeff_index + 1; i <= coeff_index + zero_run; i++)
1013 s->num_coded_frags[plane][i]--;
1016 av_log(s->avctx, AV_LOG_ERROR, "Invalid token %d\n", token);
1021 if (blocks_ended > s->num_coded_frags[plane][coeff_index])
1022 av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
1024 // decrement the number of blocks that have higher coeffecients for each
1025 // EOB run at this level
1027 for (i = coeff_index + 1; i < 64; i++)
1028 s->num_coded_frags[plane][i] -= blocks_ended;
1030 // setup the next buffer
1032 s->dct_tokens[plane + 1][coeff_index] = dct_tokens + j;
1033 else if (coeff_index < 63)
1034 s->dct_tokens[0][coeff_index + 1] = dct_tokens + j;
1039 static void reverse_dc_prediction(Vp3DecodeContext *s,
1042 int fragment_height);
1044 * This function unpacks all of the DCT coefficient data from the
1047 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1054 int residual_eob_run = 0;
1058 s->dct_tokens[0][0] = s->dct_tokens_base;
1060 /* fetch the DC table indexes */
1061 dc_y_table = get_bits(gb, 4);
1062 dc_c_table = get_bits(gb, 4);
1064 /* unpack the Y plane DC coefficients */
1065 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1066 0, residual_eob_run);
1067 if (residual_eob_run < 0)
1068 return residual_eob_run;
1070 /* reverse prediction of the Y-plane DC coefficients */
1071 reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]);
1073 /* unpack the C plane DC coefficients */
1074 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1075 1, residual_eob_run);
1076 if (residual_eob_run < 0)
1077 return residual_eob_run;
1078 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1079 2, residual_eob_run);
1080 if (residual_eob_run < 0)
1081 return residual_eob_run;
1083 /* reverse prediction of the C-plane DC coefficients */
1084 if (!(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
1085 reverse_dc_prediction(s, s->fragment_start[1],
1086 s->fragment_width[1], s->fragment_height[1]);
1087 reverse_dc_prediction(s, s->fragment_start[2],
1088 s->fragment_width[1], s->fragment_height[1]);
1091 /* fetch the AC table indexes */
1092 ac_y_table = get_bits(gb, 4);
1093 ac_c_table = get_bits(gb, 4);
1095 /* build tables of AC VLC tables */
1096 for (i = 1; i <= 5; i++) {
1097 y_tables[i] = &s->ac_vlc_1[ac_y_table];
1098 c_tables[i] = &s->ac_vlc_1[ac_c_table];
1100 for (i = 6; i <= 14; i++) {
1101 y_tables[i] = &s->ac_vlc_2[ac_y_table];
1102 c_tables[i] = &s->ac_vlc_2[ac_c_table];
1104 for (i = 15; i <= 27; i++) {
1105 y_tables[i] = &s->ac_vlc_3[ac_y_table];
1106 c_tables[i] = &s->ac_vlc_3[ac_c_table];
1108 for (i = 28; i <= 63; i++) {
1109 y_tables[i] = &s->ac_vlc_4[ac_y_table];
1110 c_tables[i] = &s->ac_vlc_4[ac_c_table];
1113 /* decode all AC coefficents */
1114 for (i = 1; i <= 63; i++) {
1115 residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1116 0, residual_eob_run);
1117 if (residual_eob_run < 0)
1118 return residual_eob_run;
1120 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1121 1, residual_eob_run);
1122 if (residual_eob_run < 0)
1123 return residual_eob_run;
1124 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1125 2, residual_eob_run);
1126 if (residual_eob_run < 0)
1127 return residual_eob_run;
1134 * This function reverses the DC prediction for each coded fragment in
1135 * the frame. Much of this function is adapted directly from the original
1138 #define COMPATIBLE_FRAME(x) \
1139 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1140 #define DC_COEFF(u) s->all_fragments[u].dc
1142 static void reverse_dc_prediction(Vp3DecodeContext *s,
1145 int fragment_height)
1153 int i = first_fragment;
1157 /* DC values for the left, up-left, up, and up-right fragments */
1158 int vl, vul, vu, vur;
1160 /* indexes for the left, up-left, up, and up-right fragments */
1164 * The 6 fields mean:
1165 * 0: up-left multiplier
1167 * 2: up-right multiplier
1168 * 3: left multiplier
1170 static const int predictor_transform[16][4] = {
1172 { 0, 0, 0, 128 }, // PL
1173 { 0, 0, 128, 0 }, // PUR
1174 { 0, 0, 53, 75 }, // PUR|PL
1175 { 0, 128, 0, 0 }, // PU
1176 { 0, 64, 0, 64 }, // PU |PL
1177 { 0, 128, 0, 0 }, // PU |PUR
1178 { 0, 0, 53, 75 }, // PU |PUR|PL
1179 { 128, 0, 0, 0 }, // PUL
1180 { 0, 0, 0, 128 }, // PUL|PL
1181 { 64, 0, 64, 0 }, // PUL|PUR
1182 { 0, 0, 53, 75 }, // PUL|PUR|PL
1183 { 0, 128, 0, 0 }, // PUL|PU
1184 { -104, 116, 0, 116 }, // PUL|PU |PL
1185 { 24, 80, 24, 0 }, // PUL|PU |PUR
1186 { -104, 116, 0, 116 } // PUL|PU |PUR|PL
1189 /* This table shows which types of blocks can use other blocks for
1190 * prediction. For example, INTRA is the only mode in this table to
1191 * have a frame number of 0. That means INTRA blocks can only predict
1192 * from other INTRA blocks. There are 2 golden frame coding types;
1193 * blocks encoding in these modes can only predict from other blocks
1194 * that were encoded with these 1 of these 2 modes. */
1195 static const unsigned char compatible_frame[9] = {
1196 1, /* MODE_INTER_NO_MV */
1198 1, /* MODE_INTER_PLUS_MV */
1199 1, /* MODE_INTER_LAST_MV */
1200 1, /* MODE_INTER_PRIOR_MV */
1201 2, /* MODE_USING_GOLDEN */
1202 2, /* MODE_GOLDEN_MV */
1203 1, /* MODE_INTER_FOUR_MV */
1206 int current_frame_type;
1208 /* there is a last DC predictor for each of the 3 frame types */
1221 /* for each fragment row... */
1222 for (y = 0; y < fragment_height; y++) {
1223 /* for each fragment in a row... */
1224 for (x = 0; x < fragment_width; x++, i++) {
1226 /* reverse prediction if this block was coded */
1227 if (s->all_fragments[i].coding_method != MODE_COPY) {
1228 current_frame_type =
1229 compatible_frame[s->all_fragments[i].coding_method];
1235 if (COMPATIBLE_FRAME(l))
1239 u = i - fragment_width;
1241 if (COMPATIBLE_FRAME(u))
1244 ul = i - fragment_width - 1;
1246 if (COMPATIBLE_FRAME(ul))
1249 if (x + 1 < fragment_width) {
1250 ur = i - fragment_width + 1;
1252 if (COMPATIBLE_FRAME(ur))
1257 if (transform == 0) {
1258 /* if there were no fragments to predict from, use last
1260 predicted_dc = last_dc[current_frame_type];
1262 /* apply the appropriate predictor transform */
1264 (predictor_transform[transform][0] * vul) +
1265 (predictor_transform[transform][1] * vu) +
1266 (predictor_transform[transform][2] * vur) +
1267 (predictor_transform[transform][3] * vl);
1269 predicted_dc /= 128;
1271 /* check for outranging on the [ul u l] and
1272 * [ul u ur l] predictors */
1273 if ((transform == 15) || (transform == 13)) {
1274 if (FFABS(predicted_dc - vu) > 128)
1276 else if (FFABS(predicted_dc - vl) > 128)
1278 else if (FFABS(predicted_dc - vul) > 128)
1283 /* at long last, apply the predictor */
1284 DC_COEFF(i) += predicted_dc;
1286 last_dc[current_frame_type] = DC_COEFF(i);
1292 static void apply_loop_filter(Vp3DecodeContext *s, int plane,
1293 int ystart, int yend)
1296 int *bounding_values = s->bounding_values_array + 127;
1298 int width = s->fragment_width[!!plane];
1299 int height = s->fragment_height[!!plane];
1300 int fragment = s->fragment_start[plane] + ystart * width;
1301 ptrdiff_t stride = s->current_frame.f->linesize[plane];
1302 uint8_t *plane_data = s->current_frame.f->data[plane];
1303 if (!s->flipped_image)
1305 plane_data += s->data_offset[plane] + 8 * ystart * stride;
1307 for (y = ystart; y < yend; y++) {
1308 for (x = 0; x < width; x++) {
1309 /* This code basically just deblocks on the edges of coded blocks.
1310 * However, it has to be much more complicated because of the
1311 * braindamaged deblock ordering used in VP3/Theora. Order matters
1312 * because some pixels get filtered twice. */
1313 if (s->all_fragments[fragment].coding_method != MODE_COPY) {
1314 /* do not perform left edge filter for left columns frags */
1316 s->vp3dsp.h_loop_filter(
1318 stride, bounding_values);
1321 /* do not perform top edge filter for top row fragments */
1323 s->vp3dsp.v_loop_filter(
1325 stride, bounding_values);
1328 /* do not perform right edge filter for right column
1329 * fragments or if right fragment neighbor is also coded
1330 * in this frame (it will be filtered in next iteration) */
1331 if ((x < width - 1) &&
1332 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1333 s->vp3dsp.h_loop_filter(
1334 plane_data + 8 * x + 8,
1335 stride, bounding_values);
1338 /* do not perform bottom edge filter for bottom row
1339 * fragments or if bottom fragment neighbor is also coded
1340 * in this frame (it will be filtered in the next row) */
1341 if ((y < height - 1) &&
1342 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1343 s->vp3dsp.v_loop_filter(
1344 plane_data + 8 * x + 8 * stride,
1345 stride, bounding_values);
1351 plane_data += 8 * stride;
1356 * Pull DCT tokens from the 64 levels to decode and dequant the coefficients
1357 * for the next block in coding order
1359 static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1360 int plane, int inter, int16_t block[64])
1362 int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1363 uint8_t *perm = s->idct_scantable;
1367 int token = *s->dct_tokens[plane][i];
1368 switch (token & 3) {
1370 if (--token < 4) // 0-3 are token types so the EOB run must now be 0
1371 s->dct_tokens[plane][i]++;
1373 *s->dct_tokens[plane][i] = token & ~3;
1376 s->dct_tokens[plane][i]++;
1377 i += (token >> 2) & 0x7f;
1379 av_log(s->avctx, AV_LOG_ERROR, "Coefficient index overflow\n");
1382 block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1386 block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1387 s->dct_tokens[plane][i++]++;
1389 default: // shouldn't happen
1393 // return value is expected to be a valid level
1396 // the actual DC+prediction is in the fragment structure
1397 block[0] = frag->dc * s->qmat[0][inter][plane][0];
1402 * called when all pixels up to row y are complete
1404 static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1407 int offset[AV_NUM_DATA_POINTERS];
1409 if (HAVE_THREADS && s->avctx->active_thread_type & FF_THREAD_FRAME) {
1410 int y_flipped = s->flipped_image ? s->height - y : y;
1412 /* At the end of the frame, report INT_MAX instead of the height of
1413 * the frame. This makes the other threads' ff_thread_await_progress()
1414 * calls cheaper, because they don't have to clip their values. */
1415 ff_thread_report_progress(&s->current_frame,
1416 y_flipped == s->height ? INT_MAX
1421 if (!s->avctx->draw_horiz_band)
1424 h = y - s->last_slice_end;
1425 s->last_slice_end = y;
1428 if (!s->flipped_image)
1429 y = s->height - y - h;
1431 cy = y >> s->chroma_y_shift;
1432 offset[0] = s->current_frame.f->linesize[0] * y;
1433 offset[1] = s->current_frame.f->linesize[1] * cy;
1434 offset[2] = s->current_frame.f->linesize[2] * cy;
1435 for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
1439 s->avctx->draw_horiz_band(s->avctx, s->current_frame.f, offset, y, 3, h);
1443 * Wait for the reference frame of the current fragment.
1444 * The progress value is in luma pixel rows.
1446 static void await_reference_row(Vp3DecodeContext *s, Vp3Fragment *fragment,
1447 int motion_y, int y)
1449 ThreadFrame *ref_frame;
1451 int border = motion_y & 1;
1453 if (fragment->coding_method == MODE_USING_GOLDEN ||
1454 fragment->coding_method == MODE_GOLDEN_MV)
1455 ref_frame = &s->golden_frame;
1457 ref_frame = &s->last_frame;
1459 ref_row = y + (motion_y >> 1);
1460 ref_row = FFMAX(FFABS(ref_row), ref_row + 8 + border);
1462 ff_thread_await_progress(ref_frame, ref_row, 0);
1466 * Perform the final rendering for a particular slice of data.
1467 * The slice number ranges from 0..(c_superblock_height - 1).
1469 static void render_slice(Vp3DecodeContext *s, int slice)
1471 int x, y, i, j, fragment;
1472 int16_t *block = s->block;
1473 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1474 int motion_halfpel_index;
1475 uint8_t *motion_source;
1476 int plane, first_pixel;
1478 if (slice >= s->c_superblock_height)
1481 for (plane = 0; plane < 3; plane++) {
1482 uint8_t *output_plane = s->current_frame.f->data[plane] +
1483 s->data_offset[plane];
1484 uint8_t *last_plane = s->last_frame.f->data[plane] +
1485 s->data_offset[plane];
1486 uint8_t *golden_plane = s->golden_frame.f->data[plane] +
1487 s->data_offset[plane];
1488 ptrdiff_t stride = s->current_frame.f->linesize[plane];
1489 int plane_width = s->width >> (plane && s->chroma_x_shift);
1490 int plane_height = s->height >> (plane && s->chroma_y_shift);
1491 int8_t(*motion_val)[2] = s->motion_val[!!plane];
1493 int sb_x, sb_y = slice << (!plane && s->chroma_y_shift);
1494 int slice_height = sb_y + 1 + (!plane && s->chroma_y_shift);
1495 int slice_width = plane ? s->c_superblock_width
1496 : s->y_superblock_width;
1498 int fragment_width = s->fragment_width[!!plane];
1499 int fragment_height = s->fragment_height[!!plane];
1500 int fragment_start = s->fragment_start[plane];
1502 int do_await = !plane && HAVE_THREADS &&
1503 (s->avctx->active_thread_type & FF_THREAD_FRAME);
1505 if (!s->flipped_image)
1507 if (CONFIG_GRAY && plane && (s->avctx->flags & AV_CODEC_FLAG_GRAY))
1510 /* for each superblock row in the slice (both of them)... */
1511 for (; sb_y < slice_height; sb_y++) {
1512 /* for each superblock in a row... */
1513 for (sb_x = 0; sb_x < slice_width; sb_x++) {
1514 /* for each block in a superblock... */
1515 for (j = 0; j < 16; j++) {
1516 x = 4 * sb_x + hilbert_offset[j][0];
1517 y = 4 * sb_y + hilbert_offset[j][1];
1518 fragment = y * fragment_width + x;
1520 i = fragment_start + fragment;
1523 if (x >= fragment_width || y >= fragment_height)
1526 first_pixel = 8 * y * stride + 8 * x;
1529 s->all_fragments[i].coding_method != MODE_INTRA)
1530 await_reference_row(s, &s->all_fragments[i],
1531 motion_val[fragment][1],
1532 (16 * y) >> s->chroma_y_shift);
1534 /* transform if this block was coded */
1535 if (s->all_fragments[i].coding_method != MODE_COPY) {
1536 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1537 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1538 motion_source = golden_plane;
1540 motion_source = last_plane;
1542 motion_source += first_pixel;
1543 motion_halfpel_index = 0;
1545 /* sort out the motion vector if this fragment is coded
1546 * using a motion vector method */
1547 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1548 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1550 motion_x = motion_val[fragment][0];
1551 motion_y = motion_val[fragment][1];
1553 src_x = (motion_x >> 1) + 8 * x;
1554 src_y = (motion_y >> 1) + 8 * y;
1556 motion_halfpel_index = motion_x & 0x01;
1557 motion_source += (motion_x >> 1);
1559 motion_halfpel_index |= (motion_y & 0x01) << 1;
1560 motion_source += ((motion_y >> 1) * stride);
1562 if (src_x < 0 || src_y < 0 ||
1563 src_x + 9 >= plane_width ||
1564 src_y + 9 >= plane_height) {
1565 uint8_t *temp = s->edge_emu_buffer;
1569 s->vdsp.emulated_edge_mc(temp, motion_source,
1574 motion_source = temp;
1578 /* first, take care of copying a block from either the
1579 * previous or the golden frame */
1580 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1581 /* Note, it is possible to implement all MC cases
1582 * with put_no_rnd_pixels_l2 which would look more
1583 * like the VP3 source but this would be slower as
1584 * put_no_rnd_pixels_tab is better optimzed */
1585 if (motion_halfpel_index != 3) {
1586 s->hdsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1587 output_plane + first_pixel,
1588 motion_source, stride, 8);
1590 /* d is 0 if motion_x and _y have the same sign,
1592 int d = (motion_x ^ motion_y) >> 31;
1593 s->vp3dsp.put_no_rnd_pixels_l2(output_plane + first_pixel,
1595 motion_source + stride + 1 + d,
1600 /* invert DCT and place (or add) in final output */
1602 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1604 index = vp3_dequant(s, s->all_fragments + i,
1608 s->vp3dsp.idct_put(output_plane + first_pixel,
1612 int index = vp3_dequant(s, s->all_fragments + i,
1617 s->vp3dsp.idct_add(output_plane + first_pixel,
1621 s->vp3dsp.idct_dc_add(output_plane + first_pixel,
1626 /* copy directly from the previous frame */
1627 s->hdsp.put_pixels_tab[1][0](
1628 output_plane + first_pixel,
1629 last_plane + first_pixel,
1635 // Filter up to the last row in the superblock row
1636 if (!s->skip_loop_filter)
1637 apply_loop_filter(s, plane, 4 * sb_y - !!sb_y,
1638 FFMIN(4 * sb_y + 3, fragment_height - 1));
1642 /* this looks like a good place for slice dispatch... */
1644 * if (slice == s->macroblock_height - 1)
1645 * dispatch (both last slice & 2nd-to-last slice);
1646 * else if (slice > 0)
1647 * dispatch (slice - 1);
1650 vp3_draw_horiz_band(s, FFMIN((32 << s->chroma_y_shift) * (slice + 1) - 16,
1654 /// Allocate tables for per-frame data in Vp3DecodeContext
1655 static av_cold int allocate_tables(AVCodecContext *avctx)
1657 Vp3DecodeContext *s = avctx->priv_data;
1658 int y_fragment_count, c_fragment_count;
1660 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1661 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1663 s->superblock_coding = av_malloc(s->superblock_count);
1664 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1666 s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int));
1668 s->dct_tokens_base = av_malloc(64 * s->fragment_count *
1669 sizeof(*s->dct_tokens_base));
1670 s->motion_val[0] = av_malloc(y_fragment_count * sizeof(*s->motion_val[0]));
1671 s->motion_val[1] = av_malloc(c_fragment_count * sizeof(*s->motion_val[1]));
1673 /* work out the block mapping tables */
1674 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1675 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1677 if (!s->superblock_coding || !s->all_fragments ||
1678 !s->dct_tokens_base || !s->coded_fragment_list[0] ||
1679 !s->superblock_fragments || !s->macroblock_coding ||
1680 !s->motion_val[0] || !s->motion_val[1]) {
1681 vp3_decode_end(avctx);
1685 init_block_mapping(s);
1690 static av_cold int init_frames(Vp3DecodeContext *s)
1692 s->current_frame.f = av_frame_alloc();
1693 s->last_frame.f = av_frame_alloc();
1694 s->golden_frame.f = av_frame_alloc();
1696 if (!s->current_frame.f || !s->last_frame.f || !s->golden_frame.f) {
1697 av_frame_free(&s->current_frame.f);
1698 av_frame_free(&s->last_frame.f);
1699 av_frame_free(&s->golden_frame.f);
1700 return AVERROR(ENOMEM);
1706 static av_cold int vp3_decode_init(AVCodecContext *avctx)
1708 Vp3DecodeContext *s = avctx->priv_data;
1709 int i, inter, plane, ret;
1712 int y_fragment_count, c_fragment_count;
1714 ret = init_frames(s);
1718 avctx->internal->allocate_progress = 1;
1720 if (avctx->codec_tag == MKTAG('V', 'P', '3', '0'))
1726 s->width = FFALIGN(avctx->coded_width, 16);
1727 s->height = FFALIGN(avctx->coded_height, 16);
1728 if (avctx->pix_fmt == AV_PIX_FMT_NONE)
1729 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
1730 avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1731 ff_hpeldsp_init(&s->hdsp, avctx->flags | AV_CODEC_FLAG_BITEXACT);
1732 ff_videodsp_init(&s->vdsp, 8);
1733 ff_vp3dsp_init(&s->vp3dsp, avctx->flags);
1735 for (i = 0; i < 64; i++) {
1736 #define TRANSPOSE(x) (x >> 3) | ((x & 7) << 3)
1737 s->idct_permutation[i] = TRANSPOSE(i);
1738 s->idct_scantable[i] = TRANSPOSE(ff_zigzag_direct[i]);
1742 /* initialize to an impossible value which will force a recalculation
1743 * in the first frame decode */
1744 for (i = 0; i < 3; i++)
1747 av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift,
1748 &s->chroma_y_shift);
1750 s->y_superblock_width = (s->width + 31) / 32;
1751 s->y_superblock_height = (s->height + 31) / 32;
1752 s->y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1754 /* work out the dimensions for the C planes */
1755 c_width = s->width >> s->chroma_x_shift;
1756 c_height = s->height >> s->chroma_y_shift;
1757 s->c_superblock_width = (c_width + 31) / 32;
1758 s->c_superblock_height = (c_height + 31) / 32;
1759 s->c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1761 s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
1762 s->u_superblock_start = s->y_superblock_count;
1763 s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
1765 s->macroblock_width = (s->width + 15) / 16;
1766 s->macroblock_height = (s->height + 15) / 16;
1767 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1769 s->fragment_width[0] = s->width / FRAGMENT_PIXELS;
1770 s->fragment_height[0] = s->height / FRAGMENT_PIXELS;
1771 s->fragment_width[1] = s->fragment_width[0] >> s->chroma_x_shift;
1772 s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift;
1774 /* fragment count covers all 8x8 blocks for all 3 planes */
1775 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1776 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1777 s->fragment_count = y_fragment_count + 2 * c_fragment_count;
1778 s->fragment_start[1] = y_fragment_count;
1779 s->fragment_start[2] = y_fragment_count + c_fragment_count;
1781 if (!s->theora_tables) {
1782 for (i = 0; i < 64; i++) {
1783 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1784 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1785 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1786 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1787 s->base_matrix[2][i] = vp31_inter_dequant[i];
1788 s->filter_limit_values[i] = vp31_filter_limit_values[i];
1791 for (inter = 0; inter < 2; inter++) {
1792 for (plane = 0; plane < 3; plane++) {
1793 s->qr_count[inter][plane] = 1;
1794 s->qr_size[inter][plane][0] = 63;
1795 s->qr_base[inter][plane][0] =
1796 s->qr_base[inter][plane][1] = 2 * inter + (!!plane) * !inter;
1800 /* init VLC tables */
1801 for (i = 0; i < 16; i++) {
1803 init_vlc(&s->dc_vlc[i], 11, 32,
1804 &dc_bias[i][0][1], 4, 2,
1805 &dc_bias[i][0][0], 4, 2, 0);
1807 /* group 1 AC histograms */
1808 init_vlc(&s->ac_vlc_1[i], 11, 32,
1809 &ac_bias_0[i][0][1], 4, 2,
1810 &ac_bias_0[i][0][0], 4, 2, 0);
1812 /* group 2 AC histograms */
1813 init_vlc(&s->ac_vlc_2[i], 11, 32,
1814 &ac_bias_1[i][0][1], 4, 2,
1815 &ac_bias_1[i][0][0], 4, 2, 0);
1817 /* group 3 AC histograms */
1818 init_vlc(&s->ac_vlc_3[i], 11, 32,
1819 &ac_bias_2[i][0][1], 4, 2,
1820 &ac_bias_2[i][0][0], 4, 2, 0);
1822 /* group 4 AC histograms */
1823 init_vlc(&s->ac_vlc_4[i], 11, 32,
1824 &ac_bias_3[i][0][1], 4, 2,
1825 &ac_bias_3[i][0][0], 4, 2, 0);
1828 for (i = 0; i < 16; i++) {
1830 if (init_vlc(&s->dc_vlc[i], 11, 32,
1831 &s->huffman_table[i][0][1], 8, 4,
1832 &s->huffman_table[i][0][0], 8, 4, 0) < 0)
1835 /* group 1 AC histograms */
1836 if (init_vlc(&s->ac_vlc_1[i], 11, 32,
1837 &s->huffman_table[i + 16][0][1], 8, 4,
1838 &s->huffman_table[i + 16][0][0], 8, 4, 0) < 0)
1841 /* group 2 AC histograms */
1842 if (init_vlc(&s->ac_vlc_2[i], 11, 32,
1843 &s->huffman_table[i + 16 * 2][0][1], 8, 4,
1844 &s->huffman_table[i + 16 * 2][0][0], 8, 4, 0) < 0)
1847 /* group 3 AC histograms */
1848 if (init_vlc(&s->ac_vlc_3[i], 11, 32,
1849 &s->huffman_table[i + 16 * 3][0][1], 8, 4,
1850 &s->huffman_table[i + 16 * 3][0][0], 8, 4, 0) < 0)
1853 /* group 4 AC histograms */
1854 if (init_vlc(&s->ac_vlc_4[i], 11, 32,
1855 &s->huffman_table[i + 16 * 4][0][1], 8, 4,
1856 &s->huffman_table[i + 16 * 4][0][0], 8, 4, 0) < 0)
1861 init_vlc(&s->superblock_run_length_vlc, 6, 34,
1862 &superblock_run_length_vlc_table[0][1], 4, 2,
1863 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1865 init_vlc(&s->fragment_run_length_vlc, 5, 30,
1866 &fragment_run_length_vlc_table[0][1], 4, 2,
1867 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1869 init_vlc(&s->mode_code_vlc, 3, 8,
1870 &mode_code_vlc_table[0][1], 2, 1,
1871 &mode_code_vlc_table[0][0], 2, 1, 0);
1873 init_vlc(&s->motion_vector_vlc, 6, 63,
1874 &motion_vector_vlc_table[0][1], 2, 1,
1875 &motion_vector_vlc_table[0][0], 2, 1, 0);
1877 return allocate_tables(avctx);
1880 av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1884 /// Release and shuffle frames after decode finishes
1885 static int update_frames(AVCodecContext *avctx)
1887 Vp3DecodeContext *s = avctx->priv_data;
1890 /* shuffle frames (last = current) */
1891 ff_thread_release_buffer(avctx, &s->last_frame);
1892 ret = ff_thread_ref_frame(&s->last_frame, &s->current_frame);
1897 ff_thread_release_buffer(avctx, &s->golden_frame);
1898 ret = ff_thread_ref_frame(&s->golden_frame, &s->current_frame);
1902 ff_thread_release_buffer(avctx, &s->current_frame);
1906 static int ref_frame(Vp3DecodeContext *s, ThreadFrame *dst, ThreadFrame *src)
1908 ff_thread_release_buffer(s->avctx, dst);
1909 if (src->f->data[0])
1910 return ff_thread_ref_frame(dst, src);
1914 static int ref_frames(Vp3DecodeContext *dst, Vp3DecodeContext *src)
1917 if ((ret = ref_frame(dst, &dst->current_frame, &src->current_frame)) < 0 ||
1918 (ret = ref_frame(dst, &dst->golden_frame, &src->golden_frame)) < 0 ||
1919 (ret = ref_frame(dst, &dst->last_frame, &src->last_frame)) < 0)
1924 static int vp3_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
1926 Vp3DecodeContext *s = dst->priv_data, *s1 = src->priv_data;
1927 int qps_changed = 0, i, err;
1929 #define copy_fields(to, from, start_field, end_field) \
1930 memcpy(&to->start_field, &from->start_field, \
1931 (char *) &to->end_field - (char *) &to->start_field)
1933 if (!s1->current_frame.f->data[0] ||
1934 s->width != s1->width || s->height != s1->height) {
1941 // init tables if the first frame hasn't been decoded
1942 if (!s->current_frame.f->data[0]) {
1943 int y_fragment_count, c_fragment_count;
1945 err = allocate_tables(dst);
1948 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1949 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1950 memcpy(s->motion_val[0], s1->motion_val[0],
1951 y_fragment_count * sizeof(*s->motion_val[0]));
1952 memcpy(s->motion_val[1], s1->motion_val[1],
1953 c_fragment_count * sizeof(*s->motion_val[1]));
1956 // copy previous frame data
1957 if ((err = ref_frames(s, s1)) < 0)
1960 s->keyframe = s1->keyframe;
1962 // copy qscale data if necessary
1963 for (i = 0; i < 3; i++) {
1964 if (s->qps[i] != s1->qps[1]) {
1966 memcpy(&s->qmat[i], &s1->qmat[i], sizeof(s->qmat[i]));
1970 if (s->qps[0] != s1->qps[0])
1971 memcpy(&s->bounding_values_array, &s1->bounding_values_array,
1972 sizeof(s->bounding_values_array));
1975 copy_fields(s, s1, qps, superblock_count);
1979 return update_frames(dst);
1982 static int vp3_decode_frame(AVCodecContext *avctx,
1983 void *data, int *got_frame,
1986 const uint8_t *buf = avpkt->data;
1987 int buf_size = avpkt->size;
1988 Vp3DecodeContext *s = avctx->priv_data;
1992 init_get_bits(&gb, buf, buf_size * 8);
1994 if (s->theora && get_bits1(&gb)) {
1995 av_log(avctx, AV_LOG_ERROR,
1996 "Header packet passed to frame decoder, skipping\n");
2000 s->keyframe = !get_bits1(&gb);
2003 for (i = 0; i < 3; i++)
2004 s->last_qps[i] = s->qps[i];
2008 s->qps[s->nqps++] = get_bits(&gb, 6);
2009 } while (s->theora >= 0x030200 && s->nqps < 3 && get_bits1(&gb));
2010 for (i = s->nqps; i < 3; i++)
2013 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
2014 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
2015 s->keyframe ? "key" : "", avctx->frame_number + 1, s->qps[0]);
2017 s->skip_loop_filter = !s->filter_limit_values[s->qps[0]] ||
2018 avctx->skip_loop_filter >= (s->keyframe ? AVDISCARD_ALL
2019 : AVDISCARD_NONKEY);
2021 if (s->qps[0] != s->last_qps[0])
2022 init_loop_filter(s);
2024 for (i = 0; i < s->nqps; i++)
2025 // reinit all dequantizers if the first one changed, because
2026 // the DC of the first quantizer must be used for all matrices
2027 if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
2028 init_dequantizer(s, i);
2030 if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
2033 s->current_frame.f->pict_type = s->keyframe ? AV_PICTURE_TYPE_I
2034 : AV_PICTURE_TYPE_P;
2035 if (ff_thread_get_buffer(avctx, &s->current_frame, AV_GET_BUFFER_FLAG_REF) < 0) {
2036 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
2040 if (!s->edge_emu_buffer)
2041 s->edge_emu_buffer = av_malloc(9 * FFABS(s->current_frame.f->linesize[0]));
2045 skip_bits(&gb, 4); /* width code */
2046 skip_bits(&gb, 4); /* height code */
2048 s->version = get_bits(&gb, 5);
2049 if (avctx->frame_number == 0)
2050 av_log(s->avctx, AV_LOG_DEBUG,
2051 "VP version: %d\n", s->version);
2054 if (s->version || s->theora) {
2056 av_log(s->avctx, AV_LOG_ERROR,
2057 "Warning, unsupported keyframe coding type?!\n");
2058 skip_bits(&gb, 2); /* reserved? */
2061 if (!s->golden_frame.f->data[0]) {
2062 av_log(s->avctx, AV_LOG_WARNING,
2063 "vp3: first frame not a keyframe\n");
2065 s->golden_frame.f->pict_type = AV_PICTURE_TYPE_I;
2066 if (ff_thread_get_buffer(avctx, &s->golden_frame,
2067 AV_GET_BUFFER_FLAG_REF) < 0) {
2068 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
2071 ff_thread_release_buffer(avctx, &s->last_frame);
2072 if ((ret = ff_thread_ref_frame(&s->last_frame,
2073 &s->golden_frame)) < 0)
2075 ff_thread_report_progress(&s->last_frame, INT_MAX, 0);
2079 memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
2080 ff_thread_finish_setup(avctx);
2082 if (unpack_superblocks(s, &gb)) {
2083 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
2086 if (unpack_modes(s, &gb)) {
2087 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
2090 if (unpack_vectors(s, &gb)) {
2091 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2094 if (unpack_block_qpis(s, &gb)) {
2095 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
2098 if (unpack_dct_coeffs(s, &gb)) {
2099 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
2103 for (i = 0; i < 3; i++) {
2104 int height = s->height >> (i && s->chroma_y_shift);
2105 if (s->flipped_image)
2106 s->data_offset[i] = 0;
2108 s->data_offset[i] = (height - 1) * s->current_frame.f->linesize[i];
2111 s->last_slice_end = 0;
2112 for (i = 0; i < s->c_superblock_height; i++)
2115 // filter the last row
2116 for (i = 0; i < 3; i++) {
2117 int row = (s->height >> (3 + (i && s->chroma_y_shift))) - 1;
2118 apply_loop_filter(s, i, row, row + 1);
2120 vp3_draw_horiz_band(s, s->height);
2122 /* output frame, offset as needed */
2123 if ((ret = av_frame_ref(data, s->current_frame.f)) < 0)
2125 for (i = 0; i < 3; i++) {
2126 AVFrame *dst = data;
2127 int off = (s->offset_x >> (i && s->chroma_y_shift)) +
2128 (s->offset_y >> (i && s->chroma_y_shift)) * dst->linesize[i];
2129 dst->data[i] += off;
2133 if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_FRAME)) {
2134 ret = update_frames(avctx);
2142 ff_thread_report_progress(&s->current_frame, INT_MAX, 0);
2144 if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_FRAME))
2145 av_frame_unref(s->current_frame.f);
2150 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2152 Vp3DecodeContext *s = avctx->priv_data;
2154 if (get_bits1(gb)) {
2156 if (s->entries >= 32) { /* overflow */
2157 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2160 token = get_bits(gb, 5);
2161 ff_dlog(avctx, "hti %d hbits %x token %d entry : %d size %d\n",
2162 s->hti, s->hbits, token, s->entries, s->huff_code_size);
2163 s->huffman_table[s->hti][token][0] = s->hbits;
2164 s->huffman_table[s->hti][token][1] = s->huff_code_size;
2167 if (s->huff_code_size >= 32) { /* overflow */
2168 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2171 s->huff_code_size++;
2173 if (read_huffman_tree(avctx, gb))
2176 if (read_huffman_tree(avctx, gb))
2179 s->huff_code_size--;
2184 static int vp3_init_thread_copy(AVCodecContext *avctx)
2186 Vp3DecodeContext *s = avctx->priv_data;
2188 s->superblock_coding = NULL;
2189 s->all_fragments = NULL;
2190 s->coded_fragment_list[0] = NULL;
2191 s->dct_tokens_base = NULL;
2192 s->superblock_fragments = NULL;
2193 s->macroblock_coding = NULL;
2194 s->motion_val[0] = NULL;
2195 s->motion_val[1] = NULL;
2196 s->edge_emu_buffer = NULL;
2198 return init_frames(s);
2201 #if CONFIG_THEORA_DECODER
2202 static const enum AVPixelFormat theora_pix_fmts[4] = {
2203 AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P
2206 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2208 Vp3DecodeContext *s = avctx->priv_data;
2209 int visible_width, visible_height, colorspace;
2210 uint8_t offset_x = 0, offset_y = 0;
2212 AVRational fps, aspect;
2214 s->theora = get_bits_long(gb, 24);
2215 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2217 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3
2218 * but previous versions have the image flipped relative to vp3 */
2219 if (s->theora < 0x030200) {
2220 s->flipped_image = 1;
2221 av_log(avctx, AV_LOG_DEBUG,
2222 "Old (<alpha3) Theora bitstream, flipped image\n");
2226 s->width = get_bits(gb, 16) << 4;
2228 s->height = get_bits(gb, 16) << 4;
2230 if (s->theora >= 0x030200) {
2231 visible_width = get_bits_long(gb, 24);
2232 visible_height = get_bits_long(gb, 24);
2234 offset_x = get_bits(gb, 8); /* offset x */
2235 offset_y = get_bits(gb, 8); /* offset y, from bottom */
2239 if (av_image_check_size(visible_width, visible_height, 0, avctx) < 0 ||
2240 visible_width + offset_x > s->width ||
2241 visible_height + offset_y > s->height) {
2242 av_log(s, AV_LOG_ERROR,
2243 "Invalid frame dimensions - w:%d h:%d x:%d y:%d (%dx%d).\n",
2244 visible_width, visible_height, offset_x, offset_y,
2245 s->width, s->height);
2246 return AVERROR_INVALIDDATA;
2249 fps.num = get_bits_long(gb, 32);
2250 fps.den = get_bits_long(gb, 32);
2251 if (fps.num && fps.den) {
2252 if (fps.num < 0 || fps.den < 0) {
2253 av_log(avctx, AV_LOG_ERROR, "Invalid framerate\n");
2254 return AVERROR_INVALIDDATA;
2256 av_reduce(&avctx->framerate.den, &avctx->framerate.num,
2257 fps.den, fps.num, 1 << 30);
2260 aspect.num = get_bits_long(gb, 24);
2261 aspect.den = get_bits_long(gb, 24);
2262 if (aspect.num && aspect.den) {
2263 av_reduce(&avctx->sample_aspect_ratio.num,
2264 &avctx->sample_aspect_ratio.den,
2265 aspect.num, aspect.den, 1 << 30);
2266 ff_set_sar(avctx, avctx->sample_aspect_ratio);
2269 if (s->theora < 0x030200)
2270 skip_bits(gb, 5); /* keyframe frequency force */
2271 colorspace = get_bits(gb, 8);
2272 skip_bits(gb, 24); /* bitrate */
2274 skip_bits(gb, 6); /* quality hint */
2276 if (s->theora >= 0x030200) {
2277 skip_bits(gb, 5); /* keyframe frequency force */
2278 avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)];
2279 skip_bits(gb, 3); /* reserved */
2282 ret = ff_set_dimensions(avctx, s->width, s->height);
2285 if (!(avctx->flags2 & AV_CODEC_FLAG2_IGNORE_CROP) &&
2286 (visible_width != s->width || visible_height != s->height)) {
2287 avctx->width = visible_width;
2288 avctx->height = visible_height;
2289 // translate offsets from theora axis ([0,0] lower left)
2290 // to normal axis ([0,0] upper left)
2291 s->offset_x = offset_x;
2292 s->offset_y = s->height - visible_height - offset_y;
2294 if ((s->offset_x & 0x1F) && !(avctx->flags & AV_CODEC_FLAG_UNALIGNED)) {
2295 s->offset_x &= ~0x1F;
2296 av_log(avctx, AV_LOG_WARNING, "Reducing offset_x from %d to %d"
2297 "chroma samples to preserve alignment.\n",
2298 offset_x, s->offset_x);
2302 if (colorspace == 1)
2303 avctx->color_primaries = AVCOL_PRI_BT470M;
2304 else if (colorspace == 2)
2305 avctx->color_primaries = AVCOL_PRI_BT470BG;
2307 if (colorspace == 1 || colorspace == 2) {
2308 avctx->colorspace = AVCOL_SPC_BT470BG;
2309 avctx->color_trc = AVCOL_TRC_BT709;
2315 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2317 Vp3DecodeContext *s = avctx->priv_data;
2318 int i, n, matrices, inter, plane;
2320 if (s->theora >= 0x030200) {
2321 n = get_bits(gb, 3);
2322 /* loop filter limit values table */
2324 for (i = 0; i < 64; i++)
2325 s->filter_limit_values[i] = get_bits(gb, n);
2328 if (s->theora >= 0x030200)
2329 n = get_bits(gb, 4) + 1;
2332 /* quality threshold table */
2333 for (i = 0; i < 64; i++)
2334 s->coded_ac_scale_factor[i] = get_bits(gb, n);
2336 if (s->theora >= 0x030200)
2337 n = get_bits(gb, 4) + 1;
2340 /* dc scale factor table */
2341 for (i = 0; i < 64; i++)
2342 s->coded_dc_scale_factor[i] = get_bits(gb, n);
2344 if (s->theora >= 0x030200)
2345 matrices = get_bits(gb, 9) + 1;
2349 if (matrices > 384) {
2350 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2354 for (n = 0; n < matrices; n++)
2355 for (i = 0; i < 64; i++)
2356 s->base_matrix[n][i] = get_bits(gb, 8);
2358 for (inter = 0; inter <= 1; inter++) {
2359 for (plane = 0; plane <= 2; plane++) {
2361 if (inter || plane > 0)
2362 newqr = get_bits1(gb);
2365 if (inter && get_bits1(gb)) {
2369 qtj = (3 * inter + plane - 1) / 3;
2370 plj = (plane + 2) % 3;
2372 s->qr_count[inter][plane] = s->qr_count[qtj][plj];
2373 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj],
2374 sizeof(s->qr_size[0][0]));
2375 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj],
2376 sizeof(s->qr_base[0][0]));
2382 i = get_bits(gb, av_log2(matrices - 1) + 1);
2383 if (i >= matrices) {
2384 av_log(avctx, AV_LOG_ERROR,
2385 "invalid base matrix index\n");
2388 s->qr_base[inter][plane][qri] = i;
2391 i = get_bits(gb, av_log2(63 - qi) + 1) + 1;
2392 s->qr_size[inter][plane][qri++] = i;
2397 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2400 s->qr_count[inter][plane] = qri;
2405 /* Huffman tables */
2406 for (s->hti = 0; s->hti < 80; s->hti++) {
2408 s->huff_code_size = 1;
2409 if (!get_bits1(gb)) {
2411 if (read_huffman_tree(avctx, gb))
2414 if (read_huffman_tree(avctx, gb))
2419 s->theora_tables = 1;
2424 static av_cold int theora_decode_init(AVCodecContext *avctx)
2426 Vp3DecodeContext *s = avctx->priv_data;
2429 uint8_t *header_start[3];
2435 if (!avctx->extradata_size) {
2436 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2440 if (avpriv_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2441 42, header_start, header_len) < 0) {
2442 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2446 for (i = 0; i < 3; i++) {
2447 if (header_len[i] <= 0)
2449 init_get_bits(&gb, header_start[i], header_len[i] * 8);
2451 ptype = get_bits(&gb, 8);
2453 if (!(ptype & 0x80)) {
2454 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2458 // FIXME: Check for this as well.
2459 skip_bits_long(&gb, 6 * 8); /* "theora" */
2463 theora_decode_header(avctx, &gb);
2466 // FIXME: is this needed? it breaks sometimes
2467 // theora_decode_comments(avctx, gb);
2470 if (theora_decode_tables(avctx, &gb))
2474 av_log(avctx, AV_LOG_ERROR,
2475 "Unknown Theora config packet: %d\n", ptype & ~0x80);
2478 if (ptype != 0x81 && 8 * header_len[i] != get_bits_count(&gb))
2479 av_log(avctx, AV_LOG_WARNING,
2480 "%d bits left in packet %X\n",
2481 8 * header_len[i] - get_bits_count(&gb), ptype);
2482 if (s->theora < 0x030200)
2486 return vp3_decode_init(avctx);
2489 AVCodec ff_theora_decoder = {
2491 .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2492 .type = AVMEDIA_TYPE_VIDEO,
2493 .id = AV_CODEC_ID_THEORA,
2494 .priv_data_size = sizeof(Vp3DecodeContext),
2495 .init = theora_decode_init,
2496 .close = vp3_decode_end,
2497 .decode = vp3_decode_frame,
2498 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DRAW_HORIZ_BAND |
2499 AV_CODEC_CAP_FRAME_THREADS,
2500 .flush = vp3_decode_flush,
2501 .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
2502 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context)
2506 AVCodec ff_vp3_decoder = {
2508 .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2509 .type = AVMEDIA_TYPE_VIDEO,
2510 .id = AV_CODEC_ID_VP3,
2511 .priv_data_size = sizeof(Vp3DecodeContext),
2512 .init = vp3_decode_init,
2513 .close = vp3_decode_end,
2514 .decode = vp3_decode_frame,
2515 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DRAW_HORIZ_BAND |
2516 AV_CODEC_CAP_FRAME_THREADS,
2517 .flush = vp3_decode_flush,
2518 .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
2519 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context),