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];
175 int8_t (*motion_val[2])[2];
178 uint16_t coded_dc_scale_factor[64];
179 uint32_t coded_ac_scale_factor[64];
180 uint8_t base_matrix[384][64];
181 uint8_t qr_count[2][3];
182 uint8_t qr_size[2][3][64];
183 uint16_t qr_base[2][3][64];
186 * This is a list of all tokens in bitstream order. Reordering takes place
187 * by pulling from each level during IDCT. As a consequence, IDCT must be
188 * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32
189 * otherwise. The 32 different tokens with up to 12 bits of extradata are
190 * collapsed into 3 types, packed as follows:
191 * (from the low to high bits)
193 * 2 bits: type (0,1,2)
194 * 0: EOB run, 14 bits for run length (12 needed)
195 * 1: zero run, 7 bits for run length
196 * 7 bits for the next coefficient (3 needed)
197 * 2: coefficient, 14 bits (11 needed)
199 * Coefficients are signed, so are packed in the highest bits for automatic
202 int16_t *dct_tokens[3][64];
203 int16_t *dct_tokens_base;
204 #define TOKEN_EOB(eob_run) ((eob_run) << 2)
205 #define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) << 9) + ((zero_run) << 2) + 1)
206 #define TOKEN_COEFF(coeff) (((coeff) << 2) + 2)
209 * number of blocks that contain DCT coefficients at
210 * the given level or higher
212 int num_coded_frags[3][64];
213 int total_num_coded_frags;
215 /* this is a list of indexes into the all_fragments array indicating
216 * which of the fragments are coded */
217 int *coded_fragment_list[3];
225 VLC superblock_run_length_vlc;
226 VLC fragment_run_length_vlc;
228 VLC motion_vector_vlc;
230 /* these arrays need to be on 16-byte boundaries since SSE2 operations
232 DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64]; ///< qmat[qpi][is_inter][plane]
234 /* This table contains superblock_count * 16 entries. Each set of 16
235 * numbers corresponds to the fragment indexes 0..15 of the superblock.
236 * An entry will be -1 to indicate that no entry corresponds to that
238 int *superblock_fragments;
240 /* This is an array that indicates how a particular macroblock
242 unsigned char *macroblock_coding;
244 uint8_t *edge_emu_buffer;
251 uint32_t huffman_table[80][32][2];
253 uint8_t filter_limit_values[64];
254 DECLARE_ALIGNED(8, int, bounding_values_array)[256 + 2];
257 /************************************************************************
258 * VP3 specific functions
259 ************************************************************************/
261 static void vp3_decode_flush(AVCodecContext *avctx)
263 Vp3DecodeContext *s = avctx->priv_data;
265 if (s->golden_frame.f)
266 ff_thread_release_buffer(avctx, &s->golden_frame);
268 ff_thread_release_buffer(avctx, &s->last_frame);
269 if (s->current_frame.f)
270 ff_thread_release_buffer(avctx, &s->current_frame);
273 static av_cold int vp3_decode_end(AVCodecContext *avctx)
275 Vp3DecodeContext *s = avctx->priv_data;
278 av_freep(&s->superblock_coding);
279 av_freep(&s->all_fragments);
280 av_freep(&s->coded_fragment_list[0]);
281 av_freep(&s->dct_tokens_base);
282 av_freep(&s->superblock_fragments);
283 av_freep(&s->macroblock_coding);
284 av_freep(&s->motion_val[0]);
285 av_freep(&s->motion_val[1]);
286 av_freep(&s->edge_emu_buffer);
288 /* release all frames */
289 vp3_decode_flush(avctx);
290 av_frame_free(&s->current_frame.f);
291 av_frame_free(&s->last_frame.f);
292 av_frame_free(&s->golden_frame.f);
294 if (avctx->internal->is_copy)
297 for (i = 0; i < 16; i++) {
298 ff_free_vlc(&s->dc_vlc[i]);
299 ff_free_vlc(&s->ac_vlc_1[i]);
300 ff_free_vlc(&s->ac_vlc_2[i]);
301 ff_free_vlc(&s->ac_vlc_3[i]);
302 ff_free_vlc(&s->ac_vlc_4[i]);
305 ff_free_vlc(&s->superblock_run_length_vlc);
306 ff_free_vlc(&s->fragment_run_length_vlc);
307 ff_free_vlc(&s->mode_code_vlc);
308 ff_free_vlc(&s->motion_vector_vlc);
314 * This function sets up all of the various blocks mappings:
315 * superblocks <-> fragments, macroblocks <-> fragments,
316 * superblocks <-> macroblocks
318 * @return 0 is successful; returns 1 if *anything* went wrong.
320 static int init_block_mapping(Vp3DecodeContext *s)
322 int sb_x, sb_y, plane;
325 for (plane = 0; plane < 3; plane++) {
326 int sb_width = plane ? s->c_superblock_width
327 : s->y_superblock_width;
328 int sb_height = plane ? s->c_superblock_height
329 : 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] +
343 s->superblock_fragments[j++] = -1;
347 return 0; /* successful path out */
351 * This function sets up the dequantization tables used for a particular
354 static void init_dequantizer(Vp3DecodeContext *s, int qpi)
356 int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
357 int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
358 int i, plane, inter, qri, bmi, bmj, qistart;
360 for (inter = 0; inter < 2; inter++) {
361 for (plane = 0; plane < 3; plane++) {
363 for (qri = 0; qri < s->qr_count[inter][plane]; qri++) {
364 sum += s->qr_size[inter][plane][qri];
365 if (s->qps[qpi] <= sum)
368 qistart = sum - s->qr_size[inter][plane][qri];
369 bmi = s->qr_base[inter][plane][qri];
370 bmj = s->qr_base[inter][plane][qri + 1];
371 for (i = 0; i < 64; i++) {
372 int coeff = (2 * (sum - s->qps[qpi]) * s->base_matrix[bmi][i] -
373 2 * (qistart - s->qps[qpi]) * s->base_matrix[bmj][i] +
374 s->qr_size[inter][plane][qri]) /
375 (2 * s->qr_size[inter][plane][qri]);
377 int qmin = 8 << (inter + !i);
378 int qscale = i ? ac_scale_factor : dc_scale_factor;
380 s->qmat[qpi][inter][plane][s->idct_permutation[i]] =
381 av_clip((qscale * coeff) / 100 * 4, qmin, 4096);
383 /* all DC coefficients use the same quant so as not to interfere
384 * with DC prediction */
385 s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
391 * This function initializes the loop filter boundary limits if the frame's
392 * quality index is different from the previous frame's.
394 * The filter_limit_values may not be larger than 127.
396 static void init_loop_filter(Vp3DecodeContext *s)
398 int *bounding_values = s->bounding_values_array + 127;
403 filter_limit = s->filter_limit_values[s->qps[0]];
404 assert(filter_limit < 128);
406 /* set up the bounding values */
407 memset(s->bounding_values_array, 0, 256 * sizeof(int));
408 for (x = 0; x < filter_limit; x++) {
409 bounding_values[-x] = -x;
410 bounding_values[x] = x;
412 for (x = value = filter_limit; x < 128 && value; x++, value--) {
413 bounding_values[ x] = value;
414 bounding_values[-x] = -value;
417 bounding_values[128] = value;
418 bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
422 * This function unpacks all of the superblock/macroblock/fragment coding
423 * information from the bitstream.
425 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
427 int superblock_starts[3] = {
428 0, s->u_superblock_start, s->v_superblock_start
431 int current_superblock = 0;
433 int num_partial_superblocks = 0;
436 int current_fragment;
440 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
442 /* unpack the list of partially-coded superblocks */
443 bit = get_bits1(gb) ^ 1;
446 while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) {
447 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
452 current_run = get_vlc2(gb, s->superblock_run_length_vlc.table,
454 if (current_run == 34)
455 current_run += get_bits(gb, 12);
457 if (current_superblock + current_run > s->superblock_count) {
458 av_log(s->avctx, AV_LOG_ERROR,
459 "Invalid partially coded superblock run length\n");
463 memset(s->superblock_coding + current_superblock, bit, current_run);
465 current_superblock += current_run;
467 num_partial_superblocks += current_run;
470 /* unpack the list of fully coded superblocks if any of the blocks were
471 * not marked as partially coded in the previous step */
472 if (num_partial_superblocks < s->superblock_count) {
473 int superblocks_decoded = 0;
475 current_superblock = 0;
476 bit = get_bits1(gb) ^ 1;
479 while (superblocks_decoded < s->superblock_count - num_partial_superblocks &&
480 get_bits_left(gb) > 0) {
481 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
486 current_run = get_vlc2(gb, s->superblock_run_length_vlc.table,
488 if (current_run == 34)
489 current_run += get_bits(gb, 12);
491 for (j = 0; j < current_run; current_superblock++) {
492 if (current_superblock >= s->superblock_count) {
493 av_log(s->avctx, AV_LOG_ERROR,
494 "Invalid fully coded superblock run length\n");
498 /* skip any superblocks already marked as partially coded */
499 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
500 s->superblock_coding[current_superblock] = 2 * bit;
504 superblocks_decoded += current_run;
508 /* if there were partial blocks, initialize bitstream for
509 * unpacking fragment codings */
510 if (num_partial_superblocks) {
513 /* toggle the bit because as soon as the first run length is
514 * fetched the bit will be toggled again */
519 /* figure out which fragments are coded; iterate through each
520 * superblock (all planes) */
521 s->total_num_coded_frags = 0;
522 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
524 for (plane = 0; plane < 3; plane++) {
525 int sb_start = superblock_starts[plane];
526 int sb_end = sb_start + (plane ? s->c_superblock_count
527 : s->y_superblock_count);
528 int num_coded_frags = 0;
530 for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) {
531 /* iterate through all 16 fragments in a superblock */
532 for (j = 0; j < 16; j++) {
533 /* if the fragment is in bounds, check its coding status */
534 current_fragment = s->superblock_fragments[i * 16 + j];
535 if (current_fragment != -1) {
536 int coded = s->superblock_coding[i];
538 if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
539 /* fragment may or may not be coded; this is the case
540 * that cares about the fragment coding runs */
541 if (current_run-- == 0) {
543 current_run = get_vlc2(gb, 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] +
574 * This function unpacks all the coding mode data for individual macroblocks
575 * from the bitstream.
577 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
579 int i, j, k, sb_x, sb_y;
581 int current_macroblock;
582 int current_fragment;
584 int custom_mode_alphabet[CODING_MODE_COUNT];
589 for (i = 0; i < s->fragment_count; i++)
590 s->all_fragments[i].coding_method = MODE_INTRA;
592 /* fetch the mode coding scheme for this frame */
593 scheme = get_bits(gb, 3);
595 /* is it a custom coding scheme? */
597 for (i = 0; i < 8; i++)
598 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
599 for (i = 0; i < 8; i++)
600 custom_mode_alphabet[get_bits(gb, 3)] = i;
601 alphabet = custom_mode_alphabet;
603 alphabet = ModeAlphabet[scheme - 1];
605 /* iterate through all of the macroblocks that contain 1 or more
607 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
608 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
609 if (get_bits_left(gb) <= 0)
612 for (j = 0; j < 4; j++) {
613 int mb_x = 2 * sb_x + (j >> 1);
614 int mb_y = 2 * sb_y + (((j >> 1) + j) & 1);
615 current_macroblock = mb_y * s->macroblock_width + mb_x;
617 if (mb_x >= s->macroblock_width ||
618 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
624 * at least one luma block coded, otherwise it must be
626 for (k = 0; k < 4; k++) {
627 current_fragment = BLOCK_Y *
628 s->fragment_width[0] + BLOCK_X;
629 if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
633 s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
637 /* mode 7 means get 3 bits for each coding mode */
639 coding_mode = get_bits(gb, 3);
641 coding_mode = alphabet[get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
643 s->macroblock_coding[current_macroblock] = coding_mode;
644 for (k = 0; k < 4; k++) {
645 frag = s->all_fragments + BLOCK_Y * s->fragment_width[0] + BLOCK_X;
646 if (frag->coding_method != MODE_COPY)
647 frag->coding_method = coding_mode;
650 #define SET_CHROMA_MODES \
651 if (frag[s->fragment_start[1]].coding_method != MODE_COPY) \
652 frag[s->fragment_start[1]].coding_method = coding_mode; \
653 if (frag[s->fragment_start[2]].coding_method != MODE_COPY) \
654 frag[s->fragment_start[2]].coding_method = coding_mode;
656 if (s->chroma_y_shift) {
657 frag = s->all_fragments + mb_y *
658 s->fragment_width[1] + mb_x;
660 } else if (s->chroma_x_shift) {
661 frag = s->all_fragments +
662 2 * mb_y * s->fragment_width[1] + mb_x;
663 for (k = 0; k < 2; k++) {
665 frag += s->fragment_width[1];
668 for (k = 0; k < 4; k++) {
669 frag = s->all_fragments +
670 BLOCK_Y * s->fragment_width[1] + BLOCK_X;
683 * This function unpacks all the motion vectors for the individual
684 * macroblocks from the bitstream.
686 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
688 int j, k, sb_x, sb_y;
692 int last_motion_x = 0;
693 int last_motion_y = 0;
694 int prior_last_motion_x = 0;
695 int prior_last_motion_y = 0;
696 int current_macroblock;
697 int current_fragment;
703 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
704 coding_mode = get_bits1(gb);
706 /* iterate through all of the macroblocks that contain 1 or more
708 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
709 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
710 if (get_bits_left(gb) <= 0)
713 for (j = 0; j < 4; j++) {
714 int mb_x = 2 * sb_x + (j >> 1);
715 int mb_y = 2 * sb_y + (((j >> 1) + j) & 1);
716 current_macroblock = mb_y * s->macroblock_width + mb_x;
718 if (mb_x >= s->macroblock_width ||
719 mb_y >= s->macroblock_height ||
720 s->macroblock_coding[current_macroblock] == MODE_COPY)
723 switch (s->macroblock_coding[current_macroblock]) {
724 case MODE_INTER_PLUS_MV:
726 /* all 6 fragments use the same motion vector */
727 if (coding_mode == 0) {
728 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
729 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
731 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
732 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
735 /* vector maintenance, only on MODE_INTER_PLUS_MV */
736 if (s->macroblock_coding[current_macroblock] == MODE_INTER_PLUS_MV) {
737 prior_last_motion_x = last_motion_x;
738 prior_last_motion_y = last_motion_y;
739 last_motion_x = motion_x[0];
740 last_motion_y = motion_y[0];
744 case MODE_INTER_FOURMV:
745 /* vector maintenance */
746 prior_last_motion_x = last_motion_x;
747 prior_last_motion_y = last_motion_y;
749 /* fetch 4 vectors from the bitstream, one for each
750 * Y fragment, then average for the C fragment vectors */
751 for (k = 0; k < 4; k++) {
752 current_fragment = BLOCK_Y * s->fragment_width[0] + BLOCK_X;
753 if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
754 if (coding_mode == 0) {
755 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
756 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
758 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
759 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
761 last_motion_x = motion_x[k];
762 last_motion_y = motion_y[k];
770 case MODE_INTER_LAST_MV:
771 /* all 6 fragments use the last motion vector */
772 motion_x[0] = last_motion_x;
773 motion_y[0] = last_motion_y;
775 /* no vector maintenance (last vector remains the
779 case MODE_INTER_PRIOR_LAST:
780 /* all 6 fragments use the motion vector prior to the
781 * last motion vector */
782 motion_x[0] = prior_last_motion_x;
783 motion_y[0] = prior_last_motion_y;
785 /* vector maintenance */
786 prior_last_motion_x = last_motion_x;
787 prior_last_motion_y = last_motion_y;
788 last_motion_x = motion_x[0];
789 last_motion_y = motion_y[0];
793 /* covers intra, inter without MV, golden without MV */
797 /* no vector maintenance */
801 /* assign the motion vectors to the correct fragments */
802 for (k = 0; k < 4; k++) {
804 BLOCK_Y * s->fragment_width[0] + BLOCK_X;
805 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
806 s->motion_val[0][current_fragment][0] = motion_x[k];
807 s->motion_val[0][current_fragment][1] = motion_y[k];
809 s->motion_val[0][current_fragment][0] = motion_x[0];
810 s->motion_val[0][current_fragment][1] = motion_y[0];
814 if (s->chroma_y_shift) {
815 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
816 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] +
817 motion_x[2] + motion_x[3], 2);
818 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] +
819 motion_y[2] + motion_y[3], 2);
821 motion_x[0] = (motion_x[0] >> 1) | (motion_x[0] & 1);
822 motion_y[0] = (motion_y[0] >> 1) | (motion_y[0] & 1);
823 frag = mb_y * s->fragment_width[1] + mb_x;
824 s->motion_val[1][frag][0] = motion_x[0];
825 s->motion_val[1][frag][1] = motion_y[0];
826 } else if (s->chroma_x_shift) {
827 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
828 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1);
829 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1);
830 motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1);
831 motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1);
833 motion_x[1] = motion_x[0];
834 motion_y[1] = motion_y[0];
836 motion_x[0] = (motion_x[0] >> 1) | (motion_x[0] & 1);
837 motion_x[1] = (motion_x[1] >> 1) | (motion_x[1] & 1);
839 frag = 2 * mb_y * s->fragment_width[1] + mb_x;
840 for (k = 0; k < 2; k++) {
841 s->motion_val[1][frag][0] = motion_x[k];
842 s->motion_val[1][frag][1] = motion_y[k];
843 frag += s->fragment_width[1];
846 for (k = 0; k < 4; k++) {
847 frag = BLOCK_Y * s->fragment_width[1] + BLOCK_X;
848 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
849 s->motion_val[1][frag][0] = motion_x[k];
850 s->motion_val[1][frag][1] = motion_y[k];
852 s->motion_val[1][frag][0] = motion_x[0];
853 s->motion_val[1][frag][1] = motion_y[0];
864 static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
866 int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
867 int num_blocks = s->total_num_coded_frags;
869 for (qpi = 0; qpi < s->nqps - 1 && num_blocks > 0; qpi++) {
870 i = blocks_decoded = num_blocks_at_qpi = 0;
872 bit = get_bits1(gb) ^ 1;
876 if (run_length == MAXIMUM_LONG_BIT_RUN)
881 run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
882 if (run_length == 34)
883 run_length += get_bits(gb, 12);
884 blocks_decoded += run_length;
887 num_blocks_at_qpi += run_length;
889 for (j = 0; j < run_length; i++) {
890 if (i >= s->total_num_coded_frags)
893 if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
894 s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
898 } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0);
900 num_blocks -= num_blocks_at_qpi;
907 * This function is called by unpack_dct_coeffs() to extract the VLCs from
908 * the bitstream. The VLCs encode tokens which are used to unpack DCT
909 * data. This function unpacks all the VLCs for either the Y plane or both
910 * C planes, and is called for DC coefficients or different AC coefficient
911 * levels (since different coefficient types require different VLC tables.
913 * This function returns a residual eob run. E.g, if a particular token gave
914 * instructions to EOB the next 5 fragments and there were only 2 fragments
915 * left in the current fragment range, 3 would be returned so that it could
916 * be passed into the next call to this same function.
918 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
919 VLC *table, int coeff_index,
930 int num_coeffs = s->num_coded_frags[plane][coeff_index];
931 int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
933 /* local references to structure members to avoid repeated deferences */
934 int *coded_fragment_list = s->coded_fragment_list[plane];
935 Vp3Fragment *all_fragments = s->all_fragments;
936 VLC_TYPE(*vlc_table)[2] = table->table;
939 av_log(s->avctx, AV_LOG_ERROR,
940 "Invalid number of coefficents at level %d\n", coeff_index);
942 if (eob_run > num_coeffs) {
944 blocks_ended = num_coeffs;
945 eob_run -= num_coeffs;
948 blocks_ended = eob_run;
952 // insert fake EOB token to cover the split between planes or zzi
954 dct_tokens[j++] = blocks_ended << 2;
956 while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
957 /* decode a VLC into a token */
958 token = get_vlc2(gb, vlc_table, 11, 3);
959 /* use the token to get a zero run, a coefficient, and an eob run */
960 if ((unsigned) token <= 6U) {
961 eob_run = eob_run_base[token];
962 if (eob_run_get_bits[token])
963 eob_run += get_bits(gb, eob_run_get_bits[token]);
965 // record only the number of blocks ended in this plane,
966 // any spill will be recorded in the next plane.
967 if (eob_run > num_coeffs - coeff_i) {
968 dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
969 blocks_ended += num_coeffs - coeff_i;
970 eob_run -= num_coeffs - coeff_i;
971 coeff_i = num_coeffs;
973 dct_tokens[j++] = TOKEN_EOB(eob_run);
974 blocks_ended += eob_run;
978 } else if (token >= 0) {
979 bits_to_get = coeff_get_bits[token];
981 bits_to_get = get_bits(gb, bits_to_get);
982 coeff = coeff_tables[token][bits_to_get];
984 zero_run = zero_run_base[token];
985 if (zero_run_get_bits[token])
986 zero_run += get_bits(gb, zero_run_get_bits[token]);
989 dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
991 // Save DC into the fragment structure. DC prediction is
992 // done in raster order, so the actual DC can't be in with
993 // other tokens. We still need the token in dct_tokens[]
994 // however, or else the structure collapses on itself.
996 all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
998 dct_tokens[j++] = TOKEN_COEFF(coeff);
1001 if (coeff_index + zero_run > 64) {
1002 av_log(s->avctx, AV_LOG_DEBUG,
1003 "Invalid zero run of %d with %d coeffs left\n",
1004 zero_run, 64 - coeff_index);
1005 zero_run = 64 - coeff_index;
1008 // zero runs code multiple coefficients,
1009 // so don't try to decode coeffs for those higher levels
1010 for (i = coeff_index + 1; i <= coeff_index + zero_run; i++)
1011 s->num_coded_frags[plane][i]--;
1014 av_log(s->avctx, AV_LOG_ERROR, "Invalid token %d\n", token);
1019 if (blocks_ended > s->num_coded_frags[plane][coeff_index])
1020 av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
1022 // decrement the number of blocks that have higher coeffecients for each
1023 // EOB run at this level
1025 for (i = coeff_index + 1; i < 64; i++)
1026 s->num_coded_frags[plane][i] -= blocks_ended;
1028 // setup the next buffer
1030 s->dct_tokens[plane + 1][coeff_index] = dct_tokens + j;
1031 else if (coeff_index < 63)
1032 s->dct_tokens[0][coeff_index + 1] = dct_tokens + j;
1037 static void reverse_dc_prediction(Vp3DecodeContext *s,
1040 int fragment_height);
1042 * This function unpacks all of the DCT coefficient data from the
1045 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1052 int residual_eob_run = 0;
1056 s->dct_tokens[0][0] = s->dct_tokens_base;
1058 /* fetch the DC table indexes */
1059 dc_y_table = get_bits(gb, 4);
1060 dc_c_table = get_bits(gb, 4);
1062 /* unpack the Y plane DC coefficients */
1063 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1064 0, residual_eob_run);
1065 if (residual_eob_run < 0)
1066 return residual_eob_run;
1068 /* reverse prediction of the Y-plane DC coefficients */
1069 reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]);
1071 /* unpack the C plane DC coefficients */
1072 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1073 1, residual_eob_run);
1074 if (residual_eob_run < 0)
1075 return residual_eob_run;
1076 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1077 2, residual_eob_run);
1078 if (residual_eob_run < 0)
1079 return residual_eob_run;
1081 /* reverse prediction of the C-plane DC coefficients */
1082 if (!(s->avctx->flags & CODEC_FLAG_GRAY)) {
1083 reverse_dc_prediction(s, s->fragment_start[1],
1084 s->fragment_width[1], s->fragment_height[1]);
1085 reverse_dc_prediction(s, s->fragment_start[2],
1086 s->fragment_width[1], s->fragment_height[1]);
1089 /* fetch the AC table indexes */
1090 ac_y_table = get_bits(gb, 4);
1091 ac_c_table = get_bits(gb, 4);
1093 /* build tables of AC VLC tables */
1094 for (i = 1; i <= 5; i++) {
1095 y_tables[i] = &s->ac_vlc_1[ac_y_table];
1096 c_tables[i] = &s->ac_vlc_1[ac_c_table];
1098 for (i = 6; i <= 14; i++) {
1099 y_tables[i] = &s->ac_vlc_2[ac_y_table];
1100 c_tables[i] = &s->ac_vlc_2[ac_c_table];
1102 for (i = 15; i <= 27; i++) {
1103 y_tables[i] = &s->ac_vlc_3[ac_y_table];
1104 c_tables[i] = &s->ac_vlc_3[ac_c_table];
1106 for (i = 28; i <= 63; i++) {
1107 y_tables[i] = &s->ac_vlc_4[ac_y_table];
1108 c_tables[i] = &s->ac_vlc_4[ac_c_table];
1111 /* decode all AC coefficents */
1112 for (i = 1; i <= 63; i++) {
1113 residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1114 0, residual_eob_run);
1115 if (residual_eob_run < 0)
1116 return residual_eob_run;
1118 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1119 1, residual_eob_run);
1120 if (residual_eob_run < 0)
1121 return residual_eob_run;
1122 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1123 2, residual_eob_run);
1124 if (residual_eob_run < 0)
1125 return residual_eob_run;
1132 * This function reverses the DC prediction for each coded fragment in
1133 * the frame. Much of this function is adapted directly from the original
1136 #define COMPATIBLE_FRAME(x) \
1137 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1138 #define DC_COEFF(u) s->all_fragments[u].dc
1140 static void reverse_dc_prediction(Vp3DecodeContext *s,
1143 int fragment_height)
1151 int i = first_fragment;
1155 /* DC values for the left, up-left, up, and up-right fragments */
1156 int vl, vul, vu, vur;
1158 /* indexes for the left, up-left, up, and up-right fragments */
1162 * The 6 fields mean:
1163 * 0: up-left multiplier
1165 * 2: up-right multiplier
1166 * 3: left multiplier
1168 static const int predictor_transform[16][4] = {
1170 { 0, 0, 0, 128 }, // PL
1171 { 0, 0, 128, 0 }, // PUR
1172 { 0, 0, 53, 75 }, // PUR|PL
1173 { 0, 128, 0, 0 }, // PU
1174 { 0, 64, 0, 64 }, // PU |PL
1175 { 0, 128, 0, 0 }, // PU |PUR
1176 { 0, 0, 53, 75 }, // PU |PUR|PL
1177 { 128, 0, 0, 0 }, // PUL
1178 { 0, 0, 0, 128 }, // PUL|PL
1179 { 64, 0, 64, 0 }, // PUL|PUR
1180 { 0, 0, 53, 75 }, // PUL|PUR|PL
1181 { 0, 128, 0, 0 }, // PUL|PU
1182 { -104, 116, 0, 116 }, // PUL|PU |PL
1183 { 24, 80, 24, 0 }, // PUL|PU |PUR
1184 { -104, 116, 0, 116 } // PUL|PU |PUR|PL
1187 /* This table shows which types of blocks can use other blocks for
1188 * prediction. For example, INTRA is the only mode in this table to
1189 * have a frame number of 0. That means INTRA blocks can only predict
1190 * from other INTRA blocks. There are 2 golden frame coding types;
1191 * blocks encoding in these modes can only predict from other blocks
1192 * that were encoded with these 1 of these 2 modes. */
1193 static const unsigned char compatible_frame[9] = {
1194 1, /* MODE_INTER_NO_MV */
1196 1, /* MODE_INTER_PLUS_MV */
1197 1, /* MODE_INTER_LAST_MV */
1198 1, /* MODE_INTER_PRIOR_MV */
1199 2, /* MODE_USING_GOLDEN */
1200 2, /* MODE_GOLDEN_MV */
1201 1, /* MODE_INTER_FOUR_MV */
1204 int current_frame_type;
1206 /* there is a last DC predictor for each of the 3 frame types */
1219 /* for each fragment row... */
1220 for (y = 0; y < fragment_height; y++) {
1221 /* for each fragment in a row... */
1222 for (x = 0; x < fragment_width; x++, i++) {
1224 /* reverse prediction if this block was coded */
1225 if (s->all_fragments[i].coding_method != MODE_COPY) {
1226 current_frame_type =
1227 compatible_frame[s->all_fragments[i].coding_method];
1233 if (COMPATIBLE_FRAME(l))
1237 u = i - fragment_width;
1239 if (COMPATIBLE_FRAME(u))
1242 ul = i - fragment_width - 1;
1244 if (COMPATIBLE_FRAME(ul))
1247 if (x + 1 < fragment_width) {
1248 ur = i - fragment_width + 1;
1250 if (COMPATIBLE_FRAME(ur))
1255 if (transform == 0) {
1256 /* if there were no fragments to predict from, use last
1258 predicted_dc = last_dc[current_frame_type];
1260 /* apply the appropriate predictor transform */
1262 (predictor_transform[transform][0] * vul) +
1263 (predictor_transform[transform][1] * vu) +
1264 (predictor_transform[transform][2] * vur) +
1265 (predictor_transform[transform][3] * vl);
1267 predicted_dc /= 128;
1269 /* check for outranging on the [ul u l] and
1270 * [ul u ur l] predictors */
1271 if ((transform == 15) || (transform == 13)) {
1272 if (FFABS(predicted_dc - vu) > 128)
1274 else if (FFABS(predicted_dc - vl) > 128)
1276 else if (FFABS(predicted_dc - vul) > 128)
1281 /* at long last, apply the predictor */
1282 DC_COEFF(i) += predicted_dc;
1284 last_dc[current_frame_type] = DC_COEFF(i);
1290 static void apply_loop_filter(Vp3DecodeContext *s, int plane,
1291 int ystart, int yend)
1294 int *bounding_values = s->bounding_values_array + 127;
1296 int width = s->fragment_width[!!plane];
1297 int height = s->fragment_height[!!plane];
1298 int fragment = s->fragment_start[plane] + ystart * width;
1299 ptrdiff_t stride = s->current_frame.f->linesize[plane];
1300 uint8_t *plane_data = s->current_frame.f->data[plane];
1301 if (!s->flipped_image)
1303 plane_data += s->data_offset[plane] + 8 * ystart * stride;
1305 for (y = ystart; y < yend; y++) {
1306 for (x = 0; x < width; x++) {
1307 /* This code basically just deblocks on the edges of coded blocks.
1308 * However, it has to be much more complicated because of the
1309 * braindamaged deblock ordering used in VP3/Theora. Order matters
1310 * because some pixels get filtered twice. */
1311 if (s->all_fragments[fragment].coding_method != MODE_COPY) {
1312 /* do not perform left edge filter for left columns frags */
1314 s->vp3dsp.h_loop_filter(
1316 stride, bounding_values);
1319 /* do not perform top edge filter for top row fragments */
1321 s->vp3dsp.v_loop_filter(
1323 stride, bounding_values);
1326 /* do not perform right edge filter for right column
1327 * fragments or if right fragment neighbor is also coded
1328 * in this frame (it will be filtered in next iteration) */
1329 if ((x < width - 1) &&
1330 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1331 s->vp3dsp.h_loop_filter(
1332 plane_data + 8 * x + 8,
1333 stride, bounding_values);
1336 /* do not perform bottom edge filter for bottom row
1337 * fragments or if bottom fragment neighbor is also coded
1338 * in this frame (it will be filtered in the next row) */
1339 if ((y < height - 1) &&
1340 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1341 s->vp3dsp.v_loop_filter(
1342 plane_data + 8 * x + 8 * stride,
1343 stride, bounding_values);
1349 plane_data += 8 * stride;
1354 * Pull DCT tokens from the 64 levels to decode and dequant the coefficients
1355 * for the next block in coding order
1357 static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1358 int plane, int inter, int16_t block[64])
1360 int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1361 uint8_t *perm = s->idct_scantable;
1365 int token = *s->dct_tokens[plane][i];
1366 switch (token & 3) {
1368 if (--token < 4) // 0-3 are token types so the EOB run must now be 0
1369 s->dct_tokens[plane][i]++;
1371 *s->dct_tokens[plane][i] = token & ~3;
1374 s->dct_tokens[plane][i]++;
1375 i += (token >> 2) & 0x7f;
1377 av_log(s->avctx, AV_LOG_ERROR, "Coefficient index overflow\n");
1380 block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1384 block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1385 s->dct_tokens[plane][i++]++;
1387 default: // shouldn't happen
1391 // return value is expected to be a valid level
1394 // the actual DC+prediction is in the fragment structure
1395 block[0] = frag->dc * s->qmat[0][inter][plane][0];
1400 * called when all pixels up to row y are complete
1402 static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1405 int offset[AV_NUM_DATA_POINTERS];
1407 if (HAVE_THREADS && s->avctx->active_thread_type & FF_THREAD_FRAME) {
1408 int y_flipped = s->flipped_image ? s->avctx->height - y : y;
1410 /* At the end of the frame, report INT_MAX instead of the height of
1411 * the frame. This makes the other threads' ff_thread_await_progress()
1412 * calls cheaper, because they don't have to clip their values. */
1413 ff_thread_report_progress(&s->current_frame,
1414 y_flipped == s->avctx->height ? INT_MAX
1419 if (s->avctx->draw_horiz_band == NULL)
1422 h = y - s->last_slice_end;
1423 s->last_slice_end = y;
1426 if (!s->flipped_image)
1427 y = s->avctx->height - y - h;
1429 cy = y >> s->chroma_y_shift;
1430 offset[0] = s->current_frame.f->linesize[0] * y;
1431 offset[1] = s->current_frame.f->linesize[1] * cy;
1432 offset[2] = s->current_frame.f->linesize[2] * cy;
1433 for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
1437 s->avctx->draw_horiz_band(s->avctx, s->current_frame.f, offset, y, 3, h);
1441 * Wait for the reference frame of the current fragment.
1442 * The progress value is in luma pixel rows.
1444 static void await_reference_row(Vp3DecodeContext *s, Vp3Fragment *fragment,
1445 int motion_y, int y)
1447 ThreadFrame *ref_frame;
1449 int border = motion_y & 1;
1451 if (fragment->coding_method == MODE_USING_GOLDEN ||
1452 fragment->coding_method == MODE_GOLDEN_MV)
1453 ref_frame = &s->golden_frame;
1455 ref_frame = &s->last_frame;
1457 ref_row = y + (motion_y >> 1);
1458 ref_row = FFMAX(FFABS(ref_row), ref_row + 8 + border);
1460 ff_thread_await_progress(ref_frame, ref_row, 0);
1464 * Perform the final rendering for a particular slice of data.
1465 * The slice number ranges from 0..(c_superblock_height - 1).
1467 static void render_slice(Vp3DecodeContext *s, int slice)
1469 int x, y, i, j, fragment;
1470 int16_t *block = s->block;
1471 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1472 int motion_halfpel_index;
1473 uint8_t *motion_source;
1474 int plane, first_pixel;
1476 if (slice >= s->c_superblock_height)
1479 for (plane = 0; plane < 3; plane++) {
1480 uint8_t *output_plane = s->current_frame.f->data[plane] +
1481 s->data_offset[plane];
1482 uint8_t *last_plane = s->last_frame.f->data[plane] +
1483 s->data_offset[plane];
1484 uint8_t *golden_plane = s->golden_frame.f->data[plane] +
1485 s->data_offset[plane];
1486 ptrdiff_t stride = s->current_frame.f->linesize[plane];
1487 int plane_width = s->width >> (plane && s->chroma_x_shift);
1488 int plane_height = s->height >> (plane && s->chroma_y_shift);
1489 int8_t(*motion_val)[2] = s->motion_val[!!plane];
1491 int sb_x, sb_y = slice << (!plane && s->chroma_y_shift);
1492 int slice_height = sb_y + 1 + (!plane && s->chroma_y_shift);
1493 int slice_width = plane ? s->c_superblock_width
1494 : s->y_superblock_width;
1496 int fragment_width = s->fragment_width[!!plane];
1497 int fragment_height = s->fragment_height[!!plane];
1498 int fragment_start = s->fragment_start[plane];
1500 int do_await = !plane && HAVE_THREADS &&
1501 (s->avctx->active_thread_type & FF_THREAD_FRAME);
1503 if (!s->flipped_image)
1505 if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
1508 /* for each superblock row in the slice (both of them)... */
1509 for (; sb_y < slice_height; sb_y++) {
1510 /* for each superblock in a row... */
1511 for (sb_x = 0; sb_x < slice_width; sb_x++) {
1512 /* for each block in a superblock... */
1513 for (j = 0; j < 16; j++) {
1514 x = 4 * sb_x + hilbert_offset[j][0];
1515 y = 4 * sb_y + hilbert_offset[j][1];
1516 fragment = y * fragment_width + x;
1518 i = fragment_start + fragment;
1521 if (x >= fragment_width || y >= fragment_height)
1524 first_pixel = 8 * y * stride + 8 * x;
1527 s->all_fragments[i].coding_method != MODE_INTRA)
1528 await_reference_row(s, &s->all_fragments[i],
1529 motion_val[fragment][1],
1530 (16 * y) >> s->chroma_y_shift);
1532 /* transform if this block was coded */
1533 if (s->all_fragments[i].coding_method != MODE_COPY) {
1534 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1535 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1536 motion_source = golden_plane;
1538 motion_source = last_plane;
1540 motion_source += first_pixel;
1541 motion_halfpel_index = 0;
1543 /* sort out the motion vector if this fragment is coded
1544 * using a motion vector method */
1545 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1546 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1548 motion_x = motion_val[fragment][0];
1549 motion_y = motion_val[fragment][1];
1551 src_x = (motion_x >> 1) + 8 * x;
1552 src_y = (motion_y >> 1) + 8 * y;
1554 motion_halfpel_index = motion_x & 0x01;
1555 motion_source += (motion_x >> 1);
1557 motion_halfpel_index |= (motion_y & 0x01) << 1;
1558 motion_source += ((motion_y >> 1) * stride);
1560 if (src_x < 0 || src_y < 0 ||
1561 src_x + 9 >= plane_width ||
1562 src_y + 9 >= plane_height) {
1563 uint8_t *temp = s->edge_emu_buffer;
1567 s->vdsp.emulated_edge_mc(temp, motion_source,
1572 motion_source = temp;
1576 /* first, take care of copying a block from either the
1577 * previous or the golden frame */
1578 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1579 /* Note, it is possible to implement all MC cases
1580 * with put_no_rnd_pixels_l2 which would look more
1581 * like the VP3 source but this would be slower as
1582 * put_no_rnd_pixels_tab is better optimzed */
1583 if (motion_halfpel_index != 3) {
1584 s->hdsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1585 output_plane + first_pixel,
1586 motion_source, stride, 8);
1588 /* d is 0 if motion_x and _y have the same sign,
1590 int d = (motion_x ^ motion_y) >> 31;
1591 s->vp3dsp.put_no_rnd_pixels_l2(output_plane + first_pixel,
1593 motion_source + stride + 1 + d,
1598 /* invert DCT and place (or add) in final output */
1600 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1602 index = vp3_dequant(s, s->all_fragments + i,
1606 s->vp3dsp.idct_put(output_plane + first_pixel,
1610 int index = vp3_dequant(s, s->all_fragments + i,
1615 s->vp3dsp.idct_add(output_plane + first_pixel,
1619 s->vp3dsp.idct_dc_add(output_plane + first_pixel,
1624 /* copy directly from the previous frame */
1625 s->hdsp.put_pixels_tab[1][0](
1626 output_plane + first_pixel,
1627 last_plane + first_pixel,
1633 // Filter up to the last row in the superblock row
1634 if (!s->skip_loop_filter)
1635 apply_loop_filter(s, plane, 4 * sb_y - !!sb_y,
1636 FFMIN(4 * sb_y + 3, fragment_height - 1));
1640 /* this looks like a good place for slice dispatch... */
1642 * if (slice == s->macroblock_height - 1)
1643 * dispatch (both last slice & 2nd-to-last slice);
1644 * else if (slice > 0)
1645 * dispatch (slice - 1);
1648 vp3_draw_horiz_band(s, FFMIN((32 << s->chroma_y_shift) * (slice + 1) - 16,
1652 /// Allocate tables for per-frame data in Vp3DecodeContext
1653 static av_cold int allocate_tables(AVCodecContext *avctx)
1655 Vp3DecodeContext *s = avctx->priv_data;
1656 int y_fragment_count, c_fragment_count;
1658 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1659 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1661 s->superblock_coding = av_malloc(s->superblock_count);
1662 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1664 s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int));
1666 s->dct_tokens_base = av_malloc(64 * s->fragment_count *
1667 sizeof(*s->dct_tokens_base));
1668 s->motion_val[0] = av_malloc(y_fragment_count * sizeof(*s->motion_val[0]));
1669 s->motion_val[1] = av_malloc(c_fragment_count * sizeof(*s->motion_val[1]));
1671 /* work out the block mapping tables */
1672 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1673 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1675 if (!s->superblock_coding || !s->all_fragments ||
1676 !s->dct_tokens_base || !s->coded_fragment_list[0] ||
1677 !s->superblock_fragments || !s->macroblock_coding ||
1678 !s->motion_val[0] || !s->motion_val[1]) {
1679 vp3_decode_end(avctx);
1683 init_block_mapping(s);
1688 static av_cold int init_frames(Vp3DecodeContext *s)
1690 s->current_frame.f = av_frame_alloc();
1691 s->last_frame.f = av_frame_alloc();
1692 s->golden_frame.f = av_frame_alloc();
1694 if (!s->current_frame.f || !s->last_frame.f || !s->golden_frame.f) {
1695 av_frame_free(&s->current_frame.f);
1696 av_frame_free(&s->last_frame.f);
1697 av_frame_free(&s->golden_frame.f);
1698 return AVERROR(ENOMEM);
1704 static av_cold int vp3_decode_init(AVCodecContext *avctx)
1706 Vp3DecodeContext *s = avctx->priv_data;
1707 int i, inter, plane, ret;
1710 int y_fragment_count, c_fragment_count;
1712 ret = init_frames(s);
1716 avctx->internal->allocate_progress = 1;
1718 if (avctx->codec_tag == MKTAG('V', 'P', '3', '0'))
1724 s->width = FFALIGN(avctx->width, 16);
1725 s->height = FFALIGN(avctx->height, 16);
1726 if (avctx->pix_fmt == AV_PIX_FMT_NONE)
1727 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
1728 avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1729 ff_hpeldsp_init(&s->hdsp, avctx->flags | CODEC_FLAG_BITEXACT);
1730 ff_videodsp_init(&s->vdsp, 8);
1731 ff_vp3dsp_init(&s->vp3dsp, avctx->flags);
1733 for (i = 0; i < 64; i++) {
1734 #define TRANSPOSE(x) (x >> 3) | ((x & 7) << 3)
1735 s->idct_permutation[i] = TRANSPOSE(i);
1736 s->idct_scantable[i] = TRANSPOSE(ff_zigzag_direct[i]);
1740 /* initialize to an impossible value which will force a recalculation
1741 * in the first frame decode */
1742 for (i = 0; i < 3; i++)
1745 av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift,
1746 &s->chroma_y_shift);
1748 s->y_superblock_width = (s->width + 31) / 32;
1749 s->y_superblock_height = (s->height + 31) / 32;
1750 s->y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1752 /* work out the dimensions for the C planes */
1753 c_width = s->width >> s->chroma_x_shift;
1754 c_height = s->height >> s->chroma_y_shift;
1755 s->c_superblock_width = (c_width + 31) / 32;
1756 s->c_superblock_height = (c_height + 31) / 32;
1757 s->c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1759 s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
1760 s->u_superblock_start = s->y_superblock_count;
1761 s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
1763 s->macroblock_width = (s->width + 15) / 16;
1764 s->macroblock_height = (s->height + 15) / 16;
1765 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1767 s->fragment_width[0] = s->width / FRAGMENT_PIXELS;
1768 s->fragment_height[0] = s->height / FRAGMENT_PIXELS;
1769 s->fragment_width[1] = s->fragment_width[0] >> s->chroma_x_shift;
1770 s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift;
1772 /* fragment count covers all 8x8 blocks for all 3 planes */
1773 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1774 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1775 s->fragment_count = y_fragment_count + 2 * c_fragment_count;
1776 s->fragment_start[1] = y_fragment_count;
1777 s->fragment_start[2] = y_fragment_count + c_fragment_count;
1779 if (!s->theora_tables) {
1780 for (i = 0; i < 64; i++) {
1781 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1782 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1783 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1784 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1785 s->base_matrix[2][i] = vp31_inter_dequant[i];
1786 s->filter_limit_values[i] = vp31_filter_limit_values[i];
1789 for (inter = 0; inter < 2; inter++) {
1790 for (plane = 0; plane < 3; plane++) {
1791 s->qr_count[inter][plane] = 1;
1792 s->qr_size[inter][plane][0] = 63;
1793 s->qr_base[inter][plane][0] =
1794 s->qr_base[inter][plane][1] = 2 * inter + (!!plane) * !inter;
1798 /* init VLC tables */
1799 for (i = 0; i < 16; i++) {
1801 init_vlc(&s->dc_vlc[i], 11, 32,
1802 &dc_bias[i][0][1], 4, 2,
1803 &dc_bias[i][0][0], 4, 2, 0);
1805 /* group 1 AC histograms */
1806 init_vlc(&s->ac_vlc_1[i], 11, 32,
1807 &ac_bias_0[i][0][1], 4, 2,
1808 &ac_bias_0[i][0][0], 4, 2, 0);
1810 /* group 2 AC histograms */
1811 init_vlc(&s->ac_vlc_2[i], 11, 32,
1812 &ac_bias_1[i][0][1], 4, 2,
1813 &ac_bias_1[i][0][0], 4, 2, 0);
1815 /* group 3 AC histograms */
1816 init_vlc(&s->ac_vlc_3[i], 11, 32,
1817 &ac_bias_2[i][0][1], 4, 2,
1818 &ac_bias_2[i][0][0], 4, 2, 0);
1820 /* group 4 AC histograms */
1821 init_vlc(&s->ac_vlc_4[i], 11, 32,
1822 &ac_bias_3[i][0][1], 4, 2,
1823 &ac_bias_3[i][0][0], 4, 2, 0);
1826 for (i = 0; i < 16; i++) {
1828 if (init_vlc(&s->dc_vlc[i], 11, 32,
1829 &s->huffman_table[i][0][1], 8, 4,
1830 &s->huffman_table[i][0][0], 8, 4, 0) < 0)
1833 /* group 1 AC histograms */
1834 if (init_vlc(&s->ac_vlc_1[i], 11, 32,
1835 &s->huffman_table[i + 16][0][1], 8, 4,
1836 &s->huffman_table[i + 16][0][0], 8, 4, 0) < 0)
1839 /* group 2 AC histograms */
1840 if (init_vlc(&s->ac_vlc_2[i], 11, 32,
1841 &s->huffman_table[i + 16 * 2][0][1], 8, 4,
1842 &s->huffman_table[i + 16 * 2][0][0], 8, 4, 0) < 0)
1845 /* group 3 AC histograms */
1846 if (init_vlc(&s->ac_vlc_3[i], 11, 32,
1847 &s->huffman_table[i + 16 * 3][0][1], 8, 4,
1848 &s->huffman_table[i + 16 * 3][0][0], 8, 4, 0) < 0)
1851 /* group 4 AC histograms */
1852 if (init_vlc(&s->ac_vlc_4[i], 11, 32,
1853 &s->huffman_table[i + 16 * 4][0][1], 8, 4,
1854 &s->huffman_table[i + 16 * 4][0][0], 8, 4, 0) < 0)
1859 init_vlc(&s->superblock_run_length_vlc, 6, 34,
1860 &superblock_run_length_vlc_table[0][1], 4, 2,
1861 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1863 init_vlc(&s->fragment_run_length_vlc, 5, 30,
1864 &fragment_run_length_vlc_table[0][1], 4, 2,
1865 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1867 init_vlc(&s->mode_code_vlc, 3, 8,
1868 &mode_code_vlc_table[0][1], 2, 1,
1869 &mode_code_vlc_table[0][0], 2, 1, 0);
1871 init_vlc(&s->motion_vector_vlc, 6, 63,
1872 &motion_vector_vlc_table[0][1], 2, 1,
1873 &motion_vector_vlc_table[0][0], 2, 1, 0);
1875 return allocate_tables(avctx);
1878 av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1882 /// Release and shuffle frames after decode finishes
1883 static int update_frames(AVCodecContext *avctx)
1885 Vp3DecodeContext *s = avctx->priv_data;
1888 /* shuffle frames (last = current) */
1889 ff_thread_release_buffer(avctx, &s->last_frame);
1890 ret = ff_thread_ref_frame(&s->last_frame, &s->current_frame);
1895 ff_thread_release_buffer(avctx, &s->golden_frame);
1896 ret = ff_thread_ref_frame(&s->golden_frame, &s->current_frame);
1900 ff_thread_release_buffer(avctx, &s->current_frame);
1904 static int ref_frame(Vp3DecodeContext *s, ThreadFrame *dst, ThreadFrame *src)
1906 ff_thread_release_buffer(s->avctx, dst);
1907 if (src->f->data[0])
1908 return ff_thread_ref_frame(dst, src);
1912 static int ref_frames(Vp3DecodeContext *dst, Vp3DecodeContext *src)
1915 if ((ret = ref_frame(dst, &dst->current_frame, &src->current_frame)) < 0 ||
1916 (ret = ref_frame(dst, &dst->golden_frame, &src->golden_frame)) < 0 ||
1917 (ret = ref_frame(dst, &dst->last_frame, &src->last_frame)) < 0)
1922 static int vp3_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
1924 Vp3DecodeContext *s = dst->priv_data, *s1 = src->priv_data;
1925 int qps_changed = 0, i, err;
1927 #define copy_fields(to, from, start_field, end_field) \
1928 memcpy(&to->start_field, &from->start_field, \
1929 (char *) &to->end_field - (char *) &to->start_field)
1931 if (!s1->current_frame.f->data[0] ||
1932 s->width != s1->width || s->height != s1->height) {
1939 // init tables if the first frame hasn't been decoded
1940 if (!s->current_frame.f->data[0]) {
1941 int y_fragment_count, c_fragment_count;
1943 err = allocate_tables(dst);
1946 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1947 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1948 memcpy(s->motion_val[0], s1->motion_val[0],
1949 y_fragment_count * sizeof(*s->motion_val[0]));
1950 memcpy(s->motion_val[1], s1->motion_val[1],
1951 c_fragment_count * sizeof(*s->motion_val[1]));
1954 // copy previous frame data
1955 if ((err = ref_frames(s, s1)) < 0)
1958 s->keyframe = s1->keyframe;
1960 // copy qscale data if necessary
1961 for (i = 0; i < 3; i++) {
1962 if (s->qps[i] != s1->qps[1]) {
1964 memcpy(&s->qmat[i], &s1->qmat[i], sizeof(s->qmat[i]));
1968 if (s->qps[0] != s1->qps[0])
1969 memcpy(&s->bounding_values_array, &s1->bounding_values_array,
1970 sizeof(s->bounding_values_array));
1973 copy_fields(s, s1, qps, superblock_count);
1977 return update_frames(dst);
1980 static int vp3_decode_frame(AVCodecContext *avctx,
1981 void *data, int *got_frame,
1984 const uint8_t *buf = avpkt->data;
1985 int buf_size = avpkt->size;
1986 Vp3DecodeContext *s = avctx->priv_data;
1990 init_get_bits(&gb, buf, buf_size * 8);
1992 if (s->theora && get_bits1(&gb)) {
1993 av_log(avctx, AV_LOG_ERROR,
1994 "Header packet passed to frame decoder, skipping\n");
1998 s->keyframe = !get_bits1(&gb);
2001 for (i = 0; i < 3; i++)
2002 s->last_qps[i] = s->qps[i];
2006 s->qps[s->nqps++] = get_bits(&gb, 6);
2007 } while (s->theora >= 0x030200 && s->nqps < 3 && get_bits1(&gb));
2008 for (i = s->nqps; i < 3; i++)
2011 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
2012 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
2013 s->keyframe ? "key" : "", avctx->frame_number + 1, s->qps[0]);
2015 s->skip_loop_filter = !s->filter_limit_values[s->qps[0]] ||
2016 avctx->skip_loop_filter >= (s->keyframe ? AVDISCARD_ALL
2017 : AVDISCARD_NONKEY);
2019 if (s->qps[0] != s->last_qps[0])
2020 init_loop_filter(s);
2022 for (i = 0; i < s->nqps; i++)
2023 // reinit all dequantizers if the first one changed, because
2024 // the DC of the first quantizer must be used for all matrices
2025 if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
2026 init_dequantizer(s, i);
2028 if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
2031 s->current_frame.f->pict_type = s->keyframe ? AV_PICTURE_TYPE_I
2032 : AV_PICTURE_TYPE_P;
2033 if (ff_thread_get_buffer(avctx, &s->current_frame, AV_GET_BUFFER_FLAG_REF) < 0) {
2034 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
2038 if (!s->edge_emu_buffer)
2039 s->edge_emu_buffer = av_malloc(9 * FFABS(s->current_frame.f->linesize[0]));
2043 skip_bits(&gb, 4); /* width code */
2044 skip_bits(&gb, 4); /* height code */
2046 s->version = get_bits(&gb, 5);
2047 if (avctx->frame_number == 0)
2048 av_log(s->avctx, AV_LOG_DEBUG,
2049 "VP version: %d\n", s->version);
2052 if (s->version || s->theora) {
2054 av_log(s->avctx, AV_LOG_ERROR,
2055 "Warning, unsupported keyframe coding type?!\n");
2056 skip_bits(&gb, 2); /* reserved? */
2059 if (!s->golden_frame.f->data[0]) {
2060 av_log(s->avctx, AV_LOG_WARNING,
2061 "vp3: first frame not a keyframe\n");
2063 s->golden_frame.f->pict_type = AV_PICTURE_TYPE_I;
2064 if (ff_thread_get_buffer(avctx, &s->golden_frame,
2065 AV_GET_BUFFER_FLAG_REF) < 0) {
2066 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
2069 ff_thread_release_buffer(avctx, &s->last_frame);
2070 if ((ret = ff_thread_ref_frame(&s->last_frame,
2071 &s->golden_frame)) < 0)
2073 ff_thread_report_progress(&s->last_frame, INT_MAX, 0);
2077 memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
2078 ff_thread_finish_setup(avctx);
2080 if (unpack_superblocks(s, &gb)) {
2081 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
2084 if (unpack_modes(s, &gb)) {
2085 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
2088 if (unpack_vectors(s, &gb)) {
2089 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2092 if (unpack_block_qpis(s, &gb)) {
2093 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
2096 if (unpack_dct_coeffs(s, &gb)) {
2097 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
2101 for (i = 0; i < 3; i++) {
2102 int height = s->height >> (i && s->chroma_y_shift);
2103 if (s->flipped_image)
2104 s->data_offset[i] = 0;
2106 s->data_offset[i] = (height - 1) * s->current_frame.f->linesize[i];
2109 s->last_slice_end = 0;
2110 for (i = 0; i < s->c_superblock_height; i++)
2113 // filter the last row
2114 for (i = 0; i < 3; i++) {
2115 int row = (s->height >> (3 + (i && s->chroma_y_shift))) - 1;
2116 apply_loop_filter(s, i, row, row + 1);
2118 vp3_draw_horiz_band(s, s->avctx->height);
2120 if ((ret = av_frame_ref(data, s->current_frame.f)) < 0)
2124 if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_FRAME)) {
2125 ret = update_frames(avctx);
2133 ff_thread_report_progress(&s->current_frame, INT_MAX, 0);
2135 if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_FRAME))
2136 av_frame_unref(s->current_frame.f);
2141 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2143 Vp3DecodeContext *s = avctx->priv_data;
2145 if (get_bits1(gb)) {
2147 if (s->entries >= 32) { /* overflow */
2148 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2151 token = get_bits(gb, 5);
2152 av_dlog(avctx, "hti %d hbits %x token %d entry : %d size %d\n",
2153 s->hti, s->hbits, token, s->entries, s->huff_code_size);
2154 s->huffman_table[s->hti][token][0] = s->hbits;
2155 s->huffman_table[s->hti][token][1] = s->huff_code_size;
2158 if (s->huff_code_size >= 32) { /* overflow */
2159 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2162 s->huff_code_size++;
2164 if (read_huffman_tree(avctx, gb))
2167 if (read_huffman_tree(avctx, gb))
2170 s->huff_code_size--;
2175 static int vp3_init_thread_copy(AVCodecContext *avctx)
2177 Vp3DecodeContext *s = avctx->priv_data;
2179 s->superblock_coding = NULL;
2180 s->all_fragments = NULL;
2181 s->coded_fragment_list[0] = NULL;
2182 s->dct_tokens_base = NULL;
2183 s->superblock_fragments = NULL;
2184 s->macroblock_coding = NULL;
2185 s->motion_val[0] = NULL;
2186 s->motion_val[1] = NULL;
2187 s->edge_emu_buffer = NULL;
2189 return init_frames(s);
2192 #if CONFIG_THEORA_DECODER
2193 static const enum AVPixelFormat theora_pix_fmts[4] = {
2194 AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P
2197 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2199 Vp3DecodeContext *s = avctx->priv_data;
2200 int visible_width, visible_height, colorspace;
2201 int offset_x = 0, offset_y = 0;
2203 AVRational fps, aspect;
2205 s->theora = get_bits_long(gb, 24);
2206 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2208 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3
2209 * but previous versions have the image flipped relative to vp3 */
2210 if (s->theora < 0x030200) {
2211 s->flipped_image = 1;
2212 av_log(avctx, AV_LOG_DEBUG,
2213 "Old (<alpha3) Theora bitstream, flipped image\n");
2217 s->width = get_bits(gb, 16) << 4;
2219 s->height = get_bits(gb, 16) << 4;
2221 if (s->theora >= 0x030200) {
2222 visible_width = get_bits_long(gb, 24);
2223 visible_height = get_bits_long(gb, 24);
2225 offset_x = get_bits(gb, 8); /* offset x */
2226 offset_y = get_bits(gb, 8); /* offset y, from bottom */
2229 fps.num = get_bits_long(gb, 32);
2230 fps.den = get_bits_long(gb, 32);
2231 if (fps.num && fps.den) {
2232 if (fps.num < 0 || fps.den < 0) {
2233 av_log(avctx, AV_LOG_ERROR, "Invalid framerate\n");
2234 return AVERROR_INVALIDDATA;
2236 av_reduce(&avctx->time_base.num, &avctx->time_base.den,
2237 fps.den, fps.num, 1 << 30);
2240 aspect.num = get_bits_long(gb, 24);
2241 aspect.den = get_bits_long(gb, 24);
2242 if (aspect.num && aspect.den) {
2243 av_reduce(&avctx->sample_aspect_ratio.num,
2244 &avctx->sample_aspect_ratio.den,
2245 aspect.num, aspect.den, 1 << 30);
2248 if (s->theora < 0x030200)
2249 skip_bits(gb, 5); /* keyframe frequency force */
2250 colorspace = get_bits(gb, 8);
2251 skip_bits(gb, 24); /* bitrate */
2253 skip_bits(gb, 6); /* quality hint */
2255 if (s->theora >= 0x030200) {
2256 skip_bits(gb, 5); /* keyframe frequency force */
2257 avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)];
2258 skip_bits(gb, 3); /* reserved */
2261 // align_get_bits(gb);
2263 if (visible_width <= s->width && visible_width > s->width - 16 &&
2264 visible_height <= s->height && visible_height > s->height - 16 &&
2265 !offset_x && (offset_y == s->height - visible_height))
2266 ret = ff_set_dimensions(avctx, visible_width, visible_height);
2268 ret = ff_set_dimensions(avctx, s->width, s->height);
2272 if (colorspace == 1)
2273 avctx->color_primaries = AVCOL_PRI_BT470M;
2274 else if (colorspace == 2)
2275 avctx->color_primaries = AVCOL_PRI_BT470BG;
2277 if (colorspace == 1 || colorspace == 2) {
2278 avctx->colorspace = AVCOL_SPC_BT470BG;
2279 avctx->color_trc = AVCOL_TRC_BT709;
2285 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2287 Vp3DecodeContext *s = avctx->priv_data;
2288 int i, n, matrices, inter, plane;
2290 if (s->theora >= 0x030200) {
2291 n = get_bits(gb, 3);
2292 /* loop filter limit values table */
2294 for (i = 0; i < 64; i++)
2295 s->filter_limit_values[i] = get_bits(gb, n);
2298 if (s->theora >= 0x030200)
2299 n = get_bits(gb, 4) + 1;
2302 /* quality threshold table */
2303 for (i = 0; i < 64; i++)
2304 s->coded_ac_scale_factor[i] = get_bits(gb, n);
2306 if (s->theora >= 0x030200)
2307 n = get_bits(gb, 4) + 1;
2310 /* dc scale factor table */
2311 for (i = 0; i < 64; i++)
2312 s->coded_dc_scale_factor[i] = get_bits(gb, n);
2314 if (s->theora >= 0x030200)
2315 matrices = get_bits(gb, 9) + 1;
2319 if (matrices > 384) {
2320 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2324 for (n = 0; n < matrices; n++)
2325 for (i = 0; i < 64; i++)
2326 s->base_matrix[n][i] = get_bits(gb, 8);
2328 for (inter = 0; inter <= 1; inter++) {
2329 for (plane = 0; plane <= 2; plane++) {
2331 if (inter || plane > 0)
2332 newqr = get_bits1(gb);
2335 if (inter && get_bits1(gb)) {
2339 qtj = (3 * inter + plane - 1) / 3;
2340 plj = (plane + 2) % 3;
2342 s->qr_count[inter][plane] = s->qr_count[qtj][plj];
2343 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj],
2344 sizeof(s->qr_size[0][0]));
2345 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj],
2346 sizeof(s->qr_base[0][0]));
2352 i = get_bits(gb, av_log2(matrices - 1) + 1);
2353 if (i >= matrices) {
2354 av_log(avctx, AV_LOG_ERROR,
2355 "invalid base matrix index\n");
2358 s->qr_base[inter][plane][qri] = i;
2361 i = get_bits(gb, av_log2(63 - qi) + 1) + 1;
2362 s->qr_size[inter][plane][qri++] = i;
2367 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2370 s->qr_count[inter][plane] = qri;
2375 /* Huffman tables */
2376 for (s->hti = 0; s->hti < 80; s->hti++) {
2378 s->huff_code_size = 1;
2379 if (!get_bits1(gb)) {
2381 if (read_huffman_tree(avctx, gb))
2384 if (read_huffman_tree(avctx, gb))
2389 s->theora_tables = 1;
2394 static av_cold int theora_decode_init(AVCodecContext *avctx)
2396 Vp3DecodeContext *s = avctx->priv_data;
2399 uint8_t *header_start[3];
2405 if (!avctx->extradata_size) {
2406 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2410 if (avpriv_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2411 42, header_start, header_len) < 0) {
2412 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2416 for (i = 0; i < 3; i++) {
2417 if (header_len[i] <= 0)
2419 init_get_bits(&gb, header_start[i], header_len[i] * 8);
2421 ptype = get_bits(&gb, 8);
2423 if (!(ptype & 0x80)) {
2424 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2428 // FIXME: Check for this as well.
2429 skip_bits_long(&gb, 6 * 8); /* "theora" */
2433 theora_decode_header(avctx, &gb);
2436 // FIXME: is this needed? it breaks sometimes
2437 // theora_decode_comments(avctx, gb);
2440 if (theora_decode_tables(avctx, &gb))
2444 av_log(avctx, AV_LOG_ERROR,
2445 "Unknown Theora config packet: %d\n", ptype & ~0x80);
2448 if (ptype != 0x81 && 8 * header_len[i] != get_bits_count(&gb))
2449 av_log(avctx, AV_LOG_WARNING,
2450 "%d bits left in packet %X\n",
2451 8 * header_len[i] - get_bits_count(&gb), ptype);
2452 if (s->theora < 0x030200)
2456 return vp3_decode_init(avctx);
2459 AVCodec ff_theora_decoder = {
2461 .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2462 .type = AVMEDIA_TYPE_VIDEO,
2463 .id = AV_CODEC_ID_THEORA,
2464 .priv_data_size = sizeof(Vp3DecodeContext),
2465 .init = theora_decode_init,
2466 .close = vp3_decode_end,
2467 .decode = vp3_decode_frame,
2468 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND |
2469 CODEC_CAP_FRAME_THREADS,
2470 .flush = vp3_decode_flush,
2471 .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
2472 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context)
2476 AVCodec ff_vp3_decoder = {
2478 .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2479 .type = AVMEDIA_TYPE_VIDEO,
2480 .id = AV_CODEC_ID_VP3,
2481 .priv_data_size = sizeof(Vp3DecodeContext),
2482 .init = vp3_decode_init,
2483 .close = vp3_decode_end,
2484 .decode = vp3_decode_frame,
2485 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND |
2486 CODEC_CAP_FRAME_THREADS,
2487 .flush = vp3_decode_flush,
2488 .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
2489 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context),