2 * Copyright (C) 2003-2004 The FFmpeg project
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 * On2 VP3 Video Decoder
25 * VP3 Video Decoder by Mike Melanson (mike at multimedia.cx)
26 * For more information about the VP3 coding process, visit:
27 * http://wiki.multimedia.cx/index.php?title=On2_VP3
29 * Theora decoder by Alex Beregszaszi
36 #include "libavutil/imgutils.h"
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 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb);
81 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb);
84 /* There are 6 preset schemes, plus a free-form scheme */
85 static const int ModeAlphabet[6][CODING_MODE_COUNT] = {
86 /* scheme 1: Last motion vector dominates */
87 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
88 MODE_INTER_PLUS_MV, MODE_INTER_NO_MV,
89 MODE_INTRA, MODE_USING_GOLDEN,
90 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
93 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
94 MODE_INTER_NO_MV, MODE_INTER_PLUS_MV,
95 MODE_INTRA, MODE_USING_GOLDEN,
96 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
99 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
100 MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
101 MODE_INTRA, MODE_USING_GOLDEN,
102 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
105 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
106 MODE_INTER_NO_MV, MODE_INTER_PRIOR_LAST,
107 MODE_INTRA, MODE_USING_GOLDEN,
108 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
110 /* scheme 5: No motion vector dominates */
111 { MODE_INTER_NO_MV, MODE_INTER_LAST_MV,
112 MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
113 MODE_INTRA, MODE_USING_GOLDEN,
114 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
117 { MODE_INTER_NO_MV, MODE_USING_GOLDEN,
118 MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
119 MODE_INTER_PLUS_MV, MODE_INTRA,
120 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
123 static const uint8_t hilbert_offset[16][2] = {
124 { 0, 0 }, { 1, 0 }, { 1, 1 }, { 0, 1 },
125 { 0, 2 }, { 0, 3 }, { 1, 3 }, { 1, 2 },
126 { 2, 2 }, { 2, 3 }, { 3, 3 }, { 3, 2 },
127 { 3, 1 }, { 2, 1 }, { 2, 0 }, { 3, 0 }
130 #define MIN_DEQUANT_VAL 2
132 typedef struct Vp3DecodeContext {
133 AVCodecContext *avctx;
134 int theora, theora_tables, theora_header;
137 int chroma_x_shift, chroma_y_shift;
138 ThreadFrame golden_frame;
139 ThreadFrame last_frame;
140 ThreadFrame current_frame;
142 uint8_t idct_permutation[64];
143 uint8_t idct_scantable[64];
145 VideoDSPContext vdsp;
146 VP3DSPContext vp3dsp;
147 DECLARE_ALIGNED(16, int16_t, block)[64];
150 int skip_loop_filter;
156 int superblock_count;
157 int y_superblock_width;
158 int y_superblock_height;
159 int y_superblock_count;
160 int c_superblock_width;
161 int c_superblock_height;
162 int c_superblock_count;
163 int u_superblock_start;
164 int v_superblock_start;
165 unsigned char *superblock_coding;
167 int macroblock_count;
168 int macroblock_width;
169 int macroblock_height;
172 int fragment_width[2];
173 int fragment_height[2];
175 Vp3Fragment *all_fragments;
176 int fragment_start[3];
182 int8_t (*motion_val[2])[2];
185 uint16_t coded_dc_scale_factor[64];
186 uint32_t coded_ac_scale_factor[64];
187 uint8_t base_matrix[384][64];
188 uint8_t qr_count[2][3];
189 uint8_t qr_size[2][3][64];
190 uint16_t qr_base[2][3][64];
193 * This is a list of all tokens in bitstream order. Reordering takes place
194 * by pulling from each level during IDCT. As a consequence, IDCT must be
195 * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32
196 * otherwise. The 32 different tokens with up to 12 bits of extradata are
197 * collapsed into 3 types, packed as follows:
198 * (from the low to high bits)
200 * 2 bits: type (0,1,2)
201 * 0: EOB run, 14 bits for run length (12 needed)
202 * 1: zero run, 7 bits for run length
203 * 7 bits for the next coefficient (3 needed)
204 * 2: coefficient, 14 bits (11 needed)
206 * Coefficients are signed, so are packed in the highest bits for automatic
209 int16_t *dct_tokens[3][64];
210 int16_t *dct_tokens_base;
211 #define TOKEN_EOB(eob_run) ((eob_run) << 2)
212 #define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) * 512) + ((zero_run) << 2) + 1)
213 #define TOKEN_COEFF(coeff) (((coeff) * 4) + 2)
216 * number of blocks that contain DCT coefficients at
217 * the given level or higher
219 int num_coded_frags[3][64];
220 int total_num_coded_frags;
222 /* this is a list of indexes into the all_fragments array indicating
223 * which of the fragments are coded */
224 int *coded_fragment_list[3];
232 VLC superblock_run_length_vlc;
233 VLC fragment_run_length_vlc;
235 VLC motion_vector_vlc;
237 /* these arrays need to be on 16-byte boundaries since SSE2 operations
239 DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64]; ///< qmat[qpi][is_inter][plane]
241 /* This table contains superblock_count * 16 entries. Each set of 16
242 * numbers corresponds to the fragment indexes 0..15 of the superblock.
243 * An entry will be -1 to indicate that no entry corresponds to that
245 int *superblock_fragments;
247 /* This is an array that indicates how a particular macroblock
249 unsigned char *macroblock_coding;
251 uint8_t *edge_emu_buffer;
258 uint32_t huffman_table[80][32][2];
260 uint8_t filter_limit_values[64];
261 DECLARE_ALIGNED(8, int, bounding_values_array)[256 + 2];
264 /************************************************************************
265 * VP3 specific functions
266 ************************************************************************/
268 static av_cold void free_tables(AVCodecContext *avctx)
270 Vp3DecodeContext *s = avctx->priv_data;
272 av_freep(&s->superblock_coding);
273 av_freep(&s->all_fragments);
274 av_freep(&s->coded_fragment_list[0]);
275 av_freep(&s->dct_tokens_base);
276 av_freep(&s->superblock_fragments);
277 av_freep(&s->macroblock_coding);
278 av_freep(&s->motion_val[0]);
279 av_freep(&s->motion_val[1]);
282 static void vp3_decode_flush(AVCodecContext *avctx)
284 Vp3DecodeContext *s = avctx->priv_data;
286 if (s->golden_frame.f)
287 ff_thread_release_buffer(avctx, &s->golden_frame);
289 ff_thread_release_buffer(avctx, &s->last_frame);
290 if (s->current_frame.f)
291 ff_thread_release_buffer(avctx, &s->current_frame);
294 static av_cold int vp3_decode_end(AVCodecContext *avctx)
296 Vp3DecodeContext *s = avctx->priv_data;
300 av_freep(&s->edge_emu_buffer);
302 s->theora_tables = 0;
304 /* release all frames */
305 vp3_decode_flush(avctx);
306 av_frame_free(&s->current_frame.f);
307 av_frame_free(&s->last_frame.f);
308 av_frame_free(&s->golden_frame.f);
310 if (avctx->internal->is_copy)
313 for (i = 0; i < 16; i++) {
314 ff_free_vlc(&s->dc_vlc[i]);
315 ff_free_vlc(&s->ac_vlc_1[i]);
316 ff_free_vlc(&s->ac_vlc_2[i]);
317 ff_free_vlc(&s->ac_vlc_3[i]);
318 ff_free_vlc(&s->ac_vlc_4[i]);
321 ff_free_vlc(&s->superblock_run_length_vlc);
322 ff_free_vlc(&s->fragment_run_length_vlc);
323 ff_free_vlc(&s->mode_code_vlc);
324 ff_free_vlc(&s->motion_vector_vlc);
330 * This function sets up all of the various blocks mappings:
331 * superblocks <-> fragments, macroblocks <-> fragments,
332 * superblocks <-> macroblocks
334 * @return 0 is successful; returns 1 if *anything* went wrong.
336 static int init_block_mapping(Vp3DecodeContext *s)
338 int sb_x, sb_y, plane;
341 for (plane = 0; plane < 3; plane++) {
342 int sb_width = plane ? s->c_superblock_width
343 : s->y_superblock_width;
344 int sb_height = plane ? s->c_superblock_height
345 : s->y_superblock_height;
346 int frag_width = s->fragment_width[!!plane];
347 int frag_height = s->fragment_height[!!plane];
349 for (sb_y = 0; sb_y < sb_height; sb_y++)
350 for (sb_x = 0; sb_x < sb_width; sb_x++)
351 for (i = 0; i < 16; i++) {
352 x = 4 * sb_x + hilbert_offset[i][0];
353 y = 4 * sb_y + hilbert_offset[i][1];
355 if (x < frag_width && y < frag_height)
356 s->superblock_fragments[j++] = s->fragment_start[plane] +
359 s->superblock_fragments[j++] = -1;
363 return 0; /* successful path out */
367 * This function sets up the dequantization tables used for a particular
370 static void init_dequantizer(Vp3DecodeContext *s, int qpi)
372 int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
373 int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
374 int i, plane, inter, qri, bmi, bmj, qistart;
376 for (inter = 0; inter < 2; inter++) {
377 for (plane = 0; plane < 3; plane++) {
379 for (qri = 0; qri < s->qr_count[inter][plane]; qri++) {
380 sum += s->qr_size[inter][plane][qri];
381 if (s->qps[qpi] <= sum)
384 qistart = sum - s->qr_size[inter][plane][qri];
385 bmi = s->qr_base[inter][plane][qri];
386 bmj = s->qr_base[inter][plane][qri + 1];
387 for (i = 0; i < 64; i++) {
388 int coeff = (2 * (sum - s->qps[qpi]) * s->base_matrix[bmi][i] -
389 2 * (qistart - s->qps[qpi]) * s->base_matrix[bmj][i] +
390 s->qr_size[inter][plane][qri]) /
391 (2 * s->qr_size[inter][plane][qri]);
393 int qmin = 8 << (inter + !i);
394 int qscale = i ? ac_scale_factor : dc_scale_factor;
396 s->qmat[qpi][inter][plane][s->idct_permutation[i]] =
397 av_clip((qscale * coeff) / 100 * 4, qmin, 4096);
399 /* all DC coefficients use the same quant so as not to interfere
400 * with DC prediction */
401 s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
407 * This function initializes the loop filter boundary limits if the frame's
408 * quality index is different from the previous frame's.
410 * The filter_limit_values may not be larger than 127.
412 static void init_loop_filter(Vp3DecodeContext *s)
414 int *bounding_values = s->bounding_values_array + 127;
419 filter_limit = s->filter_limit_values[s->qps[0]];
420 av_assert0(filter_limit < 128U);
422 /* set up the bounding values */
423 memset(s->bounding_values_array, 0, 256 * sizeof(int));
424 for (x = 0; x < filter_limit; x++) {
425 bounding_values[-x] = -x;
426 bounding_values[x] = x;
428 for (x = value = filter_limit; x < 128 && value; x++, value--) {
429 bounding_values[ x] = value;
430 bounding_values[-x] = -value;
433 bounding_values[128] = value;
434 bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
438 * This function unpacks all of the superblock/macroblock/fragment coding
439 * information from the bitstream.
441 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
443 int superblock_starts[3] = {
444 0, s->u_superblock_start, s->v_superblock_start
447 int current_superblock = 0;
449 int num_partial_superblocks = 0;
452 int current_fragment;
454 int plane0_num_coded_frags = 0;
457 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
459 /* unpack the list of partially-coded superblocks */
460 bit = get_bits1(gb) ^ 1;
463 while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) {
464 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
469 current_run = get_vlc2(gb, s->superblock_run_length_vlc.table,
471 if (current_run == 34)
472 current_run += get_bits(gb, 12);
474 if (current_run > s->superblock_count - current_superblock) {
475 av_log(s->avctx, AV_LOG_ERROR,
476 "Invalid partially coded superblock run length\n");
480 memset(s->superblock_coding + current_superblock, bit, current_run);
482 current_superblock += current_run;
484 num_partial_superblocks += current_run;
487 /* unpack the list of fully coded superblocks if any of the blocks were
488 * not marked as partially coded in the previous step */
489 if (num_partial_superblocks < s->superblock_count) {
490 int superblocks_decoded = 0;
492 current_superblock = 0;
493 bit = get_bits1(gb) ^ 1;
496 while (superblocks_decoded < s->superblock_count - num_partial_superblocks &&
497 get_bits_left(gb) > 0) {
498 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
503 current_run = get_vlc2(gb, s->superblock_run_length_vlc.table,
505 if (current_run == 34)
506 current_run += get_bits(gb, 12);
508 for (j = 0; j < current_run; current_superblock++) {
509 if (current_superblock >= s->superblock_count) {
510 av_log(s->avctx, AV_LOG_ERROR,
511 "Invalid fully coded superblock run length\n");
515 /* skip any superblocks already marked as partially coded */
516 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
517 s->superblock_coding[current_superblock] = 2 * bit;
521 superblocks_decoded += current_run;
525 /* if there were partial blocks, initialize bitstream for
526 * unpacking fragment codings */
527 if (num_partial_superblocks) {
530 /* toggle the bit because as soon as the first run length is
531 * fetched the bit will be toggled again */
536 /* figure out which fragments are coded; iterate through each
537 * superblock (all planes) */
538 s->total_num_coded_frags = 0;
539 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
541 for (plane = 0; plane < 3; plane++) {
542 int sb_start = superblock_starts[plane];
543 int sb_end = sb_start + (plane ? s->c_superblock_count
544 : s->y_superblock_count);
545 int num_coded_frags = 0;
547 for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) {
548 if (s->keyframe == 0 && get_bits_left(gb) < plane0_num_coded_frags >> 2) {
549 return AVERROR_INVALIDDATA;
551 /* iterate through all 16 fragments in a superblock */
552 for (j = 0; j < 16; j++) {
553 /* if the fragment is in bounds, check its coding status */
554 current_fragment = s->superblock_fragments[i * 16 + j];
555 if (current_fragment != -1) {
556 int coded = s->superblock_coding[i];
558 if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
559 /* fragment may or may not be coded; this is the case
560 * that cares about the fragment coding runs */
561 if (current_run-- == 0) {
563 current_run = get_vlc2(gb, s->fragment_run_length_vlc.table, 5, 2);
569 /* default mode; actual mode will be decoded in
571 s->all_fragments[current_fragment].coding_method =
573 s->coded_fragment_list[plane][num_coded_frags++] =
576 /* not coded; copy this fragment from the prior frame */
577 s->all_fragments[current_fragment].coding_method =
584 plane0_num_coded_frags = num_coded_frags;
585 s->total_num_coded_frags += num_coded_frags;
586 for (i = 0; i < 64; i++)
587 s->num_coded_frags[plane][i] = num_coded_frags;
589 s->coded_fragment_list[plane + 1] = s->coded_fragment_list[plane] +
596 * This function unpacks all the coding mode data for individual macroblocks
597 * from the bitstream.
599 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
601 int i, j, k, sb_x, sb_y;
603 int current_macroblock;
604 int current_fragment;
606 int custom_mode_alphabet[CODING_MODE_COUNT];
611 for (i = 0; i < s->fragment_count; i++)
612 s->all_fragments[i].coding_method = MODE_INTRA;
614 /* fetch the mode coding scheme for this frame */
615 scheme = get_bits(gb, 3);
617 /* is it a custom coding scheme? */
619 for (i = 0; i < 8; i++)
620 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
621 for (i = 0; i < 8; i++)
622 custom_mode_alphabet[get_bits(gb, 3)] = i;
623 alphabet = custom_mode_alphabet;
625 alphabet = ModeAlphabet[scheme - 1];
627 /* iterate through all of the macroblocks that contain 1 or more
629 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
630 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
631 if (get_bits_left(gb) <= 0)
634 for (j = 0; j < 4; j++) {
635 int mb_x = 2 * sb_x + (j >> 1);
636 int mb_y = 2 * sb_y + (((j >> 1) + j) & 1);
637 current_macroblock = mb_y * s->macroblock_width + mb_x;
639 if (mb_x >= s->macroblock_width ||
640 mb_y >= s->macroblock_height)
643 #define BLOCK_X (2 * mb_x + (k & 1))
644 #define BLOCK_Y (2 * mb_y + (k >> 1))
645 /* coding modes are only stored if the macroblock has
646 * at least one luma block coded, otherwise it must be
648 for (k = 0; k < 4; k++) {
649 current_fragment = BLOCK_Y *
650 s->fragment_width[0] + BLOCK_X;
651 if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
655 s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
659 /* mode 7 means get 3 bits for each coding mode */
661 coding_mode = get_bits(gb, 3);
663 coding_mode = alphabet[get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
665 s->macroblock_coding[current_macroblock] = coding_mode;
666 for (k = 0; k < 4; k++) {
667 frag = s->all_fragments + BLOCK_Y * s->fragment_width[0] + BLOCK_X;
668 if (frag->coding_method != MODE_COPY)
669 frag->coding_method = coding_mode;
672 #define SET_CHROMA_MODES \
673 if (frag[s->fragment_start[1]].coding_method != MODE_COPY) \
674 frag[s->fragment_start[1]].coding_method = coding_mode; \
675 if (frag[s->fragment_start[2]].coding_method != MODE_COPY) \
676 frag[s->fragment_start[2]].coding_method = coding_mode;
678 if (s->chroma_y_shift) {
679 frag = s->all_fragments + mb_y *
680 s->fragment_width[1] + mb_x;
682 } else if (s->chroma_x_shift) {
683 frag = s->all_fragments +
684 2 * mb_y * s->fragment_width[1] + mb_x;
685 for (k = 0; k < 2; k++) {
687 frag += s->fragment_width[1];
690 for (k = 0; k < 4; k++) {
691 frag = s->all_fragments +
692 BLOCK_Y * s->fragment_width[1] + BLOCK_X;
705 * This function unpacks all the motion vectors for the individual
706 * macroblocks from the bitstream.
708 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
710 int j, k, sb_x, sb_y;
714 int last_motion_x = 0;
715 int last_motion_y = 0;
716 int prior_last_motion_x = 0;
717 int prior_last_motion_y = 0;
718 int current_macroblock;
719 int current_fragment;
725 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
726 coding_mode = get_bits1(gb);
728 /* iterate through all of the macroblocks that contain 1 or more
730 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
731 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
732 if (get_bits_left(gb) <= 0)
735 for (j = 0; j < 4; j++) {
736 int mb_x = 2 * sb_x + (j >> 1);
737 int mb_y = 2 * sb_y + (((j >> 1) + j) & 1);
738 current_macroblock = mb_y * s->macroblock_width + mb_x;
740 if (mb_x >= s->macroblock_width ||
741 mb_y >= s->macroblock_height ||
742 s->macroblock_coding[current_macroblock] == MODE_COPY)
745 switch (s->macroblock_coding[current_macroblock]) {
746 case MODE_INTER_PLUS_MV:
748 /* all 6 fragments use the same motion vector */
749 if (coding_mode == 0) {
750 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
751 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
753 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
754 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
757 /* vector maintenance, only on MODE_INTER_PLUS_MV */
758 if (s->macroblock_coding[current_macroblock] == MODE_INTER_PLUS_MV) {
759 prior_last_motion_x = last_motion_x;
760 prior_last_motion_y = last_motion_y;
761 last_motion_x = motion_x[0];
762 last_motion_y = motion_y[0];
766 case MODE_INTER_FOURMV:
767 /* vector maintenance */
768 prior_last_motion_x = last_motion_x;
769 prior_last_motion_y = last_motion_y;
771 /* fetch 4 vectors from the bitstream, one for each
772 * Y fragment, then average for the C fragment vectors */
773 for (k = 0; k < 4; k++) {
774 current_fragment = BLOCK_Y * s->fragment_width[0] + BLOCK_X;
775 if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
776 if (coding_mode == 0) {
777 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
778 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
780 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
781 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
783 last_motion_x = motion_x[k];
784 last_motion_y = motion_y[k];
792 case MODE_INTER_LAST_MV:
793 /* all 6 fragments use the last motion vector */
794 motion_x[0] = last_motion_x;
795 motion_y[0] = last_motion_y;
797 /* no vector maintenance (last vector remains the
801 case MODE_INTER_PRIOR_LAST:
802 /* all 6 fragments use the motion vector prior to the
803 * last motion vector */
804 motion_x[0] = prior_last_motion_x;
805 motion_y[0] = prior_last_motion_y;
807 /* vector maintenance */
808 prior_last_motion_x = last_motion_x;
809 prior_last_motion_y = last_motion_y;
810 last_motion_x = motion_x[0];
811 last_motion_y = motion_y[0];
815 /* covers intra, inter without MV, golden without MV */
819 /* no vector maintenance */
823 /* assign the motion vectors to the correct fragments */
824 for (k = 0; k < 4; k++) {
826 BLOCK_Y * s->fragment_width[0] + BLOCK_X;
827 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
828 s->motion_val[0][current_fragment][0] = motion_x[k];
829 s->motion_val[0][current_fragment][1] = motion_y[k];
831 s->motion_val[0][current_fragment][0] = motion_x[0];
832 s->motion_val[0][current_fragment][1] = motion_y[0];
836 if (s->chroma_y_shift) {
837 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
838 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] +
839 motion_x[2] + motion_x[3], 2);
840 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] +
841 motion_y[2] + motion_y[3], 2);
843 motion_x[0] = (motion_x[0] >> 1) | (motion_x[0] & 1);
844 motion_y[0] = (motion_y[0] >> 1) | (motion_y[0] & 1);
845 frag = mb_y * s->fragment_width[1] + mb_x;
846 s->motion_val[1][frag][0] = motion_x[0];
847 s->motion_val[1][frag][1] = motion_y[0];
848 } else if (s->chroma_x_shift) {
849 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
850 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1);
851 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1);
852 motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1);
853 motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1);
855 motion_x[1] = motion_x[0];
856 motion_y[1] = motion_y[0];
858 motion_x[0] = (motion_x[0] >> 1) | (motion_x[0] & 1);
859 motion_x[1] = (motion_x[1] >> 1) | (motion_x[1] & 1);
861 frag = 2 * mb_y * s->fragment_width[1] + mb_x;
862 for (k = 0; k < 2; k++) {
863 s->motion_val[1][frag][0] = motion_x[k];
864 s->motion_val[1][frag][1] = motion_y[k];
865 frag += s->fragment_width[1];
868 for (k = 0; k < 4; k++) {
869 frag = BLOCK_Y * s->fragment_width[1] + BLOCK_X;
870 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
871 s->motion_val[1][frag][0] = motion_x[k];
872 s->motion_val[1][frag][1] = motion_y[k];
874 s->motion_val[1][frag][0] = motion_x[0];
875 s->motion_val[1][frag][1] = motion_y[0];
886 static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
888 int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
889 int num_blocks = s->total_num_coded_frags;
891 for (qpi = 0; qpi < s->nqps - 1 && num_blocks > 0; qpi++) {
892 i = blocks_decoded = num_blocks_at_qpi = 0;
894 bit = get_bits1(gb) ^ 1;
898 if (run_length == MAXIMUM_LONG_BIT_RUN)
903 run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
904 if (run_length == 34)
905 run_length += get_bits(gb, 12);
906 blocks_decoded += run_length;
909 num_blocks_at_qpi += run_length;
911 for (j = 0; j < run_length; i++) {
912 if (i >= s->total_num_coded_frags)
915 if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
916 s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
920 } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0);
922 num_blocks -= num_blocks_at_qpi;
929 * This function is called by unpack_dct_coeffs() to extract the VLCs from
930 * the bitstream. The VLCs encode tokens which are used to unpack DCT
931 * data. This function unpacks all the VLCs for either the Y plane or both
932 * C planes, and is called for DC coefficients or different AC coefficient
933 * levels (since different coefficient types require different VLC tables.
935 * This function returns a residual eob run. E.g, if a particular token gave
936 * instructions to EOB the next 5 fragments and there were only 2 fragments
937 * left in the current fragment range, 3 would be returned so that it could
938 * be passed into the next call to this same function.
940 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
941 VLC *table, int coeff_index,
952 int num_coeffs = s->num_coded_frags[plane][coeff_index];
953 int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
955 /* local references to structure members to avoid repeated dereferences */
956 int *coded_fragment_list = s->coded_fragment_list[plane];
957 Vp3Fragment *all_fragments = s->all_fragments;
958 VLC_TYPE(*vlc_table)[2] = table->table;
960 if (num_coeffs < 0) {
961 av_log(s->avctx, AV_LOG_ERROR,
962 "Invalid number of coefficients at level %d\n", coeff_index);
963 return AVERROR_INVALIDDATA;
966 if (eob_run > num_coeffs) {
968 blocks_ended = num_coeffs;
969 eob_run -= num_coeffs;
972 blocks_ended = eob_run;
976 // insert fake EOB token to cover the split between planes or zzi
978 dct_tokens[j++] = blocks_ended << 2;
980 while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
981 /* decode a VLC into a token */
982 token = get_vlc2(gb, vlc_table, 11, 3);
983 /* use the token to get a zero run, a coefficient, and an eob run */
984 if ((unsigned) token <= 6U) {
985 eob_run = eob_run_base[token];
986 if (eob_run_get_bits[token])
987 eob_run += get_bits(gb, eob_run_get_bits[token]);
992 // record only the number of blocks ended in this plane,
993 // any spill will be recorded in the next plane.
994 if (eob_run > num_coeffs - coeff_i) {
995 dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
996 blocks_ended += num_coeffs - coeff_i;
997 eob_run -= num_coeffs - coeff_i;
998 coeff_i = num_coeffs;
1000 dct_tokens[j++] = TOKEN_EOB(eob_run);
1001 blocks_ended += eob_run;
1005 } else if (token >= 0) {
1006 bits_to_get = coeff_get_bits[token];
1008 bits_to_get = get_bits(gb, bits_to_get);
1009 coeff = coeff_tables[token][bits_to_get];
1011 zero_run = zero_run_base[token];
1012 if (zero_run_get_bits[token])
1013 zero_run += get_bits(gb, zero_run_get_bits[token]);
1016 dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
1018 // Save DC into the fragment structure. DC prediction is
1019 // done in raster order, so the actual DC can't be in with
1020 // other tokens. We still need the token in dct_tokens[]
1021 // however, or else the structure collapses on itself.
1023 all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
1025 dct_tokens[j++] = TOKEN_COEFF(coeff);
1028 if (coeff_index + zero_run > 64) {
1029 av_log(s->avctx, AV_LOG_DEBUG,
1030 "Invalid zero run of %d with %d coeffs left\n",
1031 zero_run, 64 - coeff_index);
1032 zero_run = 64 - coeff_index;
1035 // zero runs code multiple coefficients,
1036 // so don't try to decode coeffs for those higher levels
1037 for (i = coeff_index + 1; i <= coeff_index + zero_run; i++)
1038 s->num_coded_frags[plane][i]--;
1041 av_log(s->avctx, AV_LOG_ERROR, "Invalid token %d\n", token);
1046 if (blocks_ended > s->num_coded_frags[plane][coeff_index])
1047 av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
1049 // decrement the number of blocks that have higher coefficients for each
1050 // EOB run at this level
1052 for (i = coeff_index + 1; i < 64; i++)
1053 s->num_coded_frags[plane][i] -= blocks_ended;
1055 // setup the next buffer
1057 s->dct_tokens[plane + 1][coeff_index] = dct_tokens + j;
1058 else if (coeff_index < 63)
1059 s->dct_tokens[0][coeff_index + 1] = dct_tokens + j;
1064 static void reverse_dc_prediction(Vp3DecodeContext *s,
1067 int fragment_height);
1069 * This function unpacks all of the DCT coefficient data from the
1072 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1079 int residual_eob_run = 0;
1083 s->dct_tokens[0][0] = s->dct_tokens_base;
1085 if (get_bits_left(gb) < 16)
1086 return AVERROR_INVALIDDATA;
1088 /* fetch the DC table indexes */
1089 dc_y_table = get_bits(gb, 4);
1090 dc_c_table = get_bits(gb, 4);
1092 /* unpack the Y plane DC coefficients */
1093 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1094 0, residual_eob_run);
1095 if (residual_eob_run < 0)
1096 return residual_eob_run;
1097 if (get_bits_left(gb) < 8)
1098 return AVERROR_INVALIDDATA;
1100 /* reverse prediction of the Y-plane DC coefficients */
1101 reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]);
1103 /* unpack the C plane DC coefficients */
1104 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1105 1, residual_eob_run);
1106 if (residual_eob_run < 0)
1107 return residual_eob_run;
1108 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1109 2, residual_eob_run);
1110 if (residual_eob_run < 0)
1111 return residual_eob_run;
1113 /* reverse prediction of the C-plane DC coefficients */
1114 if (!(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
1115 reverse_dc_prediction(s, s->fragment_start[1],
1116 s->fragment_width[1], s->fragment_height[1]);
1117 reverse_dc_prediction(s, s->fragment_start[2],
1118 s->fragment_width[1], s->fragment_height[1]);
1121 if (get_bits_left(gb) < 8)
1122 return AVERROR_INVALIDDATA;
1123 /* fetch the AC table indexes */
1124 ac_y_table = get_bits(gb, 4);
1125 ac_c_table = get_bits(gb, 4);
1127 /* build tables of AC VLC tables */
1128 for (i = 1; i <= 5; i++) {
1129 y_tables[i] = &s->ac_vlc_1[ac_y_table];
1130 c_tables[i] = &s->ac_vlc_1[ac_c_table];
1132 for (i = 6; i <= 14; i++) {
1133 y_tables[i] = &s->ac_vlc_2[ac_y_table];
1134 c_tables[i] = &s->ac_vlc_2[ac_c_table];
1136 for (i = 15; i <= 27; i++) {
1137 y_tables[i] = &s->ac_vlc_3[ac_y_table];
1138 c_tables[i] = &s->ac_vlc_3[ac_c_table];
1140 for (i = 28; i <= 63; i++) {
1141 y_tables[i] = &s->ac_vlc_4[ac_y_table];
1142 c_tables[i] = &s->ac_vlc_4[ac_c_table];
1145 /* decode all AC coefficients */
1146 for (i = 1; i <= 63; i++) {
1147 residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1148 0, residual_eob_run);
1149 if (residual_eob_run < 0)
1150 return residual_eob_run;
1152 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1153 1, residual_eob_run);
1154 if (residual_eob_run < 0)
1155 return residual_eob_run;
1156 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1157 2, residual_eob_run);
1158 if (residual_eob_run < 0)
1159 return residual_eob_run;
1166 * This function reverses the DC prediction for each coded fragment in
1167 * the frame. Much of this function is adapted directly from the original
1170 #define COMPATIBLE_FRAME(x) \
1171 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1172 #define DC_COEFF(u) s->all_fragments[u].dc
1174 static void reverse_dc_prediction(Vp3DecodeContext *s,
1177 int fragment_height)
1185 int i = first_fragment;
1189 /* DC values for the left, up-left, up, and up-right fragments */
1190 int vl, vul, vu, vur;
1192 /* indexes for the left, up-left, up, and up-right fragments */
1196 * The 6 fields mean:
1197 * 0: up-left multiplier
1199 * 2: up-right multiplier
1200 * 3: left multiplier
1202 static const int predictor_transform[16][4] = {
1204 { 0, 0, 0, 128 }, // PL
1205 { 0, 0, 128, 0 }, // PUR
1206 { 0, 0, 53, 75 }, // PUR|PL
1207 { 0, 128, 0, 0 }, // PU
1208 { 0, 64, 0, 64 }, // PU |PL
1209 { 0, 128, 0, 0 }, // PU |PUR
1210 { 0, 0, 53, 75 }, // PU |PUR|PL
1211 { 128, 0, 0, 0 }, // PUL
1212 { 0, 0, 0, 128 }, // PUL|PL
1213 { 64, 0, 64, 0 }, // PUL|PUR
1214 { 0, 0, 53, 75 }, // PUL|PUR|PL
1215 { 0, 128, 0, 0 }, // PUL|PU
1216 { -104, 116, 0, 116 }, // PUL|PU |PL
1217 { 24, 80, 24, 0 }, // PUL|PU |PUR
1218 { -104, 116, 0, 116 } // PUL|PU |PUR|PL
1221 /* This table shows which types of blocks can use other blocks for
1222 * prediction. For example, INTRA is the only mode in this table to
1223 * have a frame number of 0. That means INTRA blocks can only predict
1224 * from other INTRA blocks. There are 2 golden frame coding types;
1225 * blocks encoding in these modes can only predict from other blocks
1226 * that were encoded with these 1 of these 2 modes. */
1227 static const unsigned char compatible_frame[9] = {
1228 1, /* MODE_INTER_NO_MV */
1230 1, /* MODE_INTER_PLUS_MV */
1231 1, /* MODE_INTER_LAST_MV */
1232 1, /* MODE_INTER_PRIOR_MV */
1233 2, /* MODE_USING_GOLDEN */
1234 2, /* MODE_GOLDEN_MV */
1235 1, /* MODE_INTER_FOUR_MV */
1238 int current_frame_type;
1240 /* there is a last DC predictor for each of the 3 frame types */
1253 /* for each fragment row... */
1254 for (y = 0; y < fragment_height; y++) {
1255 /* for each fragment in a row... */
1256 for (x = 0; x < fragment_width; x++, i++) {
1258 /* reverse prediction if this block was coded */
1259 if (s->all_fragments[i].coding_method != MODE_COPY) {
1260 current_frame_type =
1261 compatible_frame[s->all_fragments[i].coding_method];
1267 if (COMPATIBLE_FRAME(l))
1271 u = i - fragment_width;
1273 if (COMPATIBLE_FRAME(u))
1276 ul = i - fragment_width - 1;
1278 if (COMPATIBLE_FRAME(ul))
1281 if (x + 1 < fragment_width) {
1282 ur = i - fragment_width + 1;
1284 if (COMPATIBLE_FRAME(ur))
1289 if (transform == 0) {
1290 /* if there were no fragments to predict from, use last
1292 predicted_dc = last_dc[current_frame_type];
1294 /* apply the appropriate predictor transform */
1296 (predictor_transform[transform][0] * vul) +
1297 (predictor_transform[transform][1] * vu) +
1298 (predictor_transform[transform][2] * vur) +
1299 (predictor_transform[transform][3] * vl);
1301 predicted_dc /= 128;
1303 /* check for outranging on the [ul u l] and
1304 * [ul u ur l] predictors */
1305 if ((transform == 15) || (transform == 13)) {
1306 if (FFABS(predicted_dc - vu) > 128)
1308 else if (FFABS(predicted_dc - vl) > 128)
1310 else if (FFABS(predicted_dc - vul) > 128)
1315 /* at long last, apply the predictor */
1316 DC_COEFF(i) += predicted_dc;
1318 last_dc[current_frame_type] = DC_COEFF(i);
1324 static void apply_loop_filter(Vp3DecodeContext *s, int plane,
1325 int ystart, int yend)
1328 int *bounding_values = s->bounding_values_array + 127;
1330 int width = s->fragment_width[!!plane];
1331 int height = s->fragment_height[!!plane];
1332 int fragment = s->fragment_start[plane] + ystart * width;
1333 ptrdiff_t stride = s->current_frame.f->linesize[plane];
1334 uint8_t *plane_data = s->current_frame.f->data[plane];
1335 if (!s->flipped_image)
1337 plane_data += s->data_offset[plane] + 8 * ystart * stride;
1339 for (y = ystart; y < yend; y++) {
1340 for (x = 0; x < width; x++) {
1341 /* This code basically just deblocks on the edges of coded blocks.
1342 * However, it has to be much more complicated because of the
1343 * brain damaged deblock ordering used in VP3/Theora. Order matters
1344 * because some pixels get filtered twice. */
1345 if (s->all_fragments[fragment].coding_method != MODE_COPY) {
1346 /* do not perform left edge filter for left columns frags */
1348 s->vp3dsp.h_loop_filter(
1350 stride, bounding_values);
1353 /* do not perform top edge filter for top row fragments */
1355 s->vp3dsp.v_loop_filter(
1357 stride, bounding_values);
1360 /* do not perform right edge filter for right column
1361 * fragments or if right fragment neighbor is also coded
1362 * in this frame (it will be filtered in next iteration) */
1363 if ((x < width - 1) &&
1364 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1365 s->vp3dsp.h_loop_filter(
1366 plane_data + 8 * x + 8,
1367 stride, bounding_values);
1370 /* do not perform bottom edge filter for bottom row
1371 * fragments or if bottom fragment neighbor is also coded
1372 * in this frame (it will be filtered in the next row) */
1373 if ((y < height - 1) &&
1374 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1375 s->vp3dsp.v_loop_filter(
1376 plane_data + 8 * x + 8 * stride,
1377 stride, bounding_values);
1383 plane_data += 8 * stride;
1388 * Pull DCT tokens from the 64 levels to decode and dequant the coefficients
1389 * for the next block in coding order
1391 static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1392 int plane, int inter, int16_t block[64])
1394 int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1395 uint8_t *perm = s->idct_scantable;
1399 int token = *s->dct_tokens[plane][i];
1400 switch (token & 3) {
1402 if (--token < 4) // 0-3 are token types so the EOB run must now be 0
1403 s->dct_tokens[plane][i]++;
1405 *s->dct_tokens[plane][i] = token & ~3;
1408 s->dct_tokens[plane][i]++;
1409 i += (token >> 2) & 0x7f;
1411 av_log(s->avctx, AV_LOG_ERROR, "Coefficient index overflow\n");
1414 block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1418 block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1419 s->dct_tokens[plane][i++]++;
1421 default: // shouldn't happen
1425 // return value is expected to be a valid level
1428 // the actual DC+prediction is in the fragment structure
1429 block[0] = frag->dc * s->qmat[0][inter][plane][0];
1434 * called when all pixels up to row y are complete
1436 static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1439 int offset[AV_NUM_DATA_POINTERS];
1441 if (HAVE_THREADS && s->avctx->active_thread_type & FF_THREAD_FRAME) {
1442 int y_flipped = s->flipped_image ? s->height - y : y;
1444 /* At the end of the frame, report INT_MAX instead of the height of
1445 * the frame. This makes the other threads' ff_thread_await_progress()
1446 * calls cheaper, because they don't have to clip their values. */
1447 ff_thread_report_progress(&s->current_frame,
1448 y_flipped == s->height ? INT_MAX
1453 if (!s->avctx->draw_horiz_band)
1456 h = y - s->last_slice_end;
1457 s->last_slice_end = y;
1460 if (!s->flipped_image)
1461 y = s->height - y - h;
1463 cy = y >> s->chroma_y_shift;
1464 offset[0] = s->current_frame.f->linesize[0] * y;
1465 offset[1] = s->current_frame.f->linesize[1] * cy;
1466 offset[2] = s->current_frame.f->linesize[2] * cy;
1467 for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
1471 s->avctx->draw_horiz_band(s->avctx, s->current_frame.f, offset, y, 3, h);
1475 * Wait for the reference frame of the current fragment.
1476 * The progress value is in luma pixel rows.
1478 static void await_reference_row(Vp3DecodeContext *s, Vp3Fragment *fragment,
1479 int motion_y, int y)
1481 ThreadFrame *ref_frame;
1483 int border = motion_y & 1;
1485 if (fragment->coding_method == MODE_USING_GOLDEN ||
1486 fragment->coding_method == MODE_GOLDEN_MV)
1487 ref_frame = &s->golden_frame;
1489 ref_frame = &s->last_frame;
1491 ref_row = y + (motion_y >> 1);
1492 ref_row = FFMAX(FFABS(ref_row), ref_row + 8 + border);
1494 ff_thread_await_progress(ref_frame, ref_row, 0);
1498 * Perform the final rendering for a particular slice of data.
1499 * The slice number ranges from 0..(c_superblock_height - 1).
1501 static void render_slice(Vp3DecodeContext *s, int slice)
1503 int x, y, i, j, fragment;
1504 int16_t *block = s->block;
1505 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1506 int motion_halfpel_index;
1507 uint8_t *motion_source;
1508 int plane, first_pixel;
1510 if (slice >= s->c_superblock_height)
1513 for (plane = 0; plane < 3; plane++) {
1514 uint8_t *output_plane = s->current_frame.f->data[plane] +
1515 s->data_offset[plane];
1516 uint8_t *last_plane = s->last_frame.f->data[plane] +
1517 s->data_offset[plane];
1518 uint8_t *golden_plane = s->golden_frame.f->data[plane] +
1519 s->data_offset[plane];
1520 ptrdiff_t stride = s->current_frame.f->linesize[plane];
1521 int plane_width = s->width >> (plane && s->chroma_x_shift);
1522 int plane_height = s->height >> (plane && s->chroma_y_shift);
1523 int8_t(*motion_val)[2] = s->motion_val[!!plane];
1525 int sb_x, sb_y = slice << (!plane && s->chroma_y_shift);
1526 int slice_height = sb_y + 1 + (!plane && s->chroma_y_shift);
1527 int slice_width = plane ? s->c_superblock_width
1528 : s->y_superblock_width;
1530 int fragment_width = s->fragment_width[!!plane];
1531 int fragment_height = s->fragment_height[!!plane];
1532 int fragment_start = s->fragment_start[plane];
1534 int do_await = !plane && HAVE_THREADS &&
1535 (s->avctx->active_thread_type & FF_THREAD_FRAME);
1537 if (!s->flipped_image)
1539 if (CONFIG_GRAY && plane && (s->avctx->flags & AV_CODEC_FLAG_GRAY))
1542 /* for each superblock row in the slice (both of them)... */
1543 for (; sb_y < slice_height; sb_y++) {
1544 /* for each superblock in a row... */
1545 for (sb_x = 0; sb_x < slice_width; sb_x++) {
1546 /* for each block in a superblock... */
1547 for (j = 0; j < 16; j++) {
1548 x = 4 * sb_x + hilbert_offset[j][0];
1549 y = 4 * sb_y + hilbert_offset[j][1];
1550 fragment = y * fragment_width + x;
1552 i = fragment_start + fragment;
1555 if (x >= fragment_width || y >= fragment_height)
1558 first_pixel = 8 * y * stride + 8 * x;
1561 s->all_fragments[i].coding_method != MODE_INTRA)
1562 await_reference_row(s, &s->all_fragments[i],
1563 motion_val[fragment][1],
1564 (16 * y) >> s->chroma_y_shift);
1566 /* transform if this block was coded */
1567 if (s->all_fragments[i].coding_method != MODE_COPY) {
1568 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1569 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1570 motion_source = golden_plane;
1572 motion_source = last_plane;
1574 motion_source += first_pixel;
1575 motion_halfpel_index = 0;
1577 /* sort out the motion vector if this fragment is coded
1578 * using a motion vector method */
1579 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1580 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1582 motion_x = motion_val[fragment][0];
1583 motion_y = motion_val[fragment][1];
1585 src_x = (motion_x >> 1) + 8 * x;
1586 src_y = (motion_y >> 1) + 8 * y;
1588 motion_halfpel_index = motion_x & 0x01;
1589 motion_source += (motion_x >> 1);
1591 motion_halfpel_index |= (motion_y & 0x01) << 1;
1592 motion_source += ((motion_y >> 1) * stride);
1594 if (src_x < 0 || src_y < 0 ||
1595 src_x + 9 >= plane_width ||
1596 src_y + 9 >= plane_height) {
1597 uint8_t *temp = s->edge_emu_buffer;
1601 s->vdsp.emulated_edge_mc(temp, motion_source,
1606 motion_source = temp;
1610 /* first, take care of copying a block from either the
1611 * previous or the golden frame */
1612 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1613 /* Note, it is possible to implement all MC cases
1614 * with put_no_rnd_pixels_l2 which would look more
1615 * like the VP3 source but this would be slower as
1616 * put_no_rnd_pixels_tab is better optimized */
1617 if (motion_halfpel_index != 3) {
1618 s->hdsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1619 output_plane + first_pixel,
1620 motion_source, stride, 8);
1622 /* d is 0 if motion_x and _y have the same sign,
1624 int d = (motion_x ^ motion_y) >> 31;
1625 s->vp3dsp.put_no_rnd_pixels_l2(output_plane + first_pixel,
1627 motion_source + stride + 1 + d,
1632 /* invert DCT and place (or add) in final output */
1634 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1635 vp3_dequant(s, s->all_fragments + i,
1637 s->vp3dsp.idct_put(output_plane + first_pixel,
1641 if (vp3_dequant(s, s->all_fragments + i,
1643 s->vp3dsp.idct_add(output_plane + first_pixel,
1647 s->vp3dsp.idct_dc_add(output_plane + first_pixel,
1652 /* copy directly from the previous frame */
1653 s->hdsp.put_pixels_tab[1][0](
1654 output_plane + first_pixel,
1655 last_plane + first_pixel,
1661 // Filter up to the last row in the superblock row
1662 if (!s->skip_loop_filter)
1663 apply_loop_filter(s, plane, 4 * sb_y - !!sb_y,
1664 FFMIN(4 * sb_y + 3, fragment_height - 1));
1668 /* this looks like a good place for slice dispatch... */
1670 * if (slice == s->macroblock_height - 1)
1671 * dispatch (both last slice & 2nd-to-last slice);
1672 * else if (slice > 0)
1673 * dispatch (slice - 1);
1676 vp3_draw_horiz_band(s, FFMIN((32 << s->chroma_y_shift) * (slice + 1) - 16,
1680 /// Allocate tables for per-frame data in Vp3DecodeContext
1681 static av_cold int allocate_tables(AVCodecContext *avctx)
1683 Vp3DecodeContext *s = avctx->priv_data;
1684 int y_fragment_count, c_fragment_count;
1688 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1689 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1691 s->superblock_coding = av_mallocz(s->superblock_count);
1692 s->all_fragments = av_mallocz_array(s->fragment_count, sizeof(Vp3Fragment));
1694 s->coded_fragment_list[0] = av_mallocz_array(s->fragment_count, sizeof(int));
1696 s->dct_tokens_base = av_mallocz_array(s->fragment_count,
1697 64 * sizeof(*s->dct_tokens_base));
1698 s->motion_val[0] = av_mallocz_array(y_fragment_count, sizeof(*s->motion_val[0]));
1699 s->motion_val[1] = av_mallocz_array(c_fragment_count, sizeof(*s->motion_val[1]));
1701 /* work out the block mapping tables */
1702 s->superblock_fragments = av_mallocz_array(s->superblock_count, 16 * sizeof(int));
1703 s->macroblock_coding = av_mallocz(s->macroblock_count + 1);
1705 if (!s->superblock_coding || !s->all_fragments ||
1706 !s->dct_tokens_base || !s->coded_fragment_list[0] ||
1707 !s->superblock_fragments || !s->macroblock_coding ||
1708 !s->motion_val[0] || !s->motion_val[1]) {
1709 vp3_decode_end(avctx);
1713 init_block_mapping(s);
1718 static av_cold int init_frames(Vp3DecodeContext *s)
1720 s->current_frame.f = av_frame_alloc();
1721 s->last_frame.f = av_frame_alloc();
1722 s->golden_frame.f = av_frame_alloc();
1724 if (!s->current_frame.f || !s->last_frame.f || !s->golden_frame.f) {
1725 av_frame_free(&s->current_frame.f);
1726 av_frame_free(&s->last_frame.f);
1727 av_frame_free(&s->golden_frame.f);
1728 return AVERROR(ENOMEM);
1734 static av_cold int vp3_decode_init(AVCodecContext *avctx)
1736 Vp3DecodeContext *s = avctx->priv_data;
1737 int i, inter, plane, ret;
1740 int y_fragment_count, c_fragment_count;
1742 ret = init_frames(s);
1746 avctx->internal->allocate_progress = 1;
1748 if (avctx->codec_tag == MKTAG('V', 'P', '3', '0'))
1754 s->width = FFALIGN(avctx->coded_width, 16);
1755 s->height = FFALIGN(avctx->coded_height, 16);
1756 if (avctx->codec_id != AV_CODEC_ID_THEORA)
1757 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
1758 avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1759 ff_hpeldsp_init(&s->hdsp, avctx->flags | AV_CODEC_FLAG_BITEXACT);
1760 ff_videodsp_init(&s->vdsp, 8);
1761 ff_vp3dsp_init(&s->vp3dsp, avctx->flags);
1763 for (i = 0; i < 64; i++) {
1764 #define TRANSPOSE(x) (((x) >> 3) | (((x) & 7) << 3))
1765 s->idct_permutation[i] = TRANSPOSE(i);
1766 s->idct_scantable[i] = TRANSPOSE(ff_zigzag_direct[i]);
1770 /* initialize to an impossible value which will force a recalculation
1771 * in the first frame decode */
1772 for (i = 0; i < 3; i++)
1775 ret = av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift);
1779 s->y_superblock_width = (s->width + 31) / 32;
1780 s->y_superblock_height = (s->height + 31) / 32;
1781 s->y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1783 /* work out the dimensions for the C planes */
1784 c_width = s->width >> s->chroma_x_shift;
1785 c_height = s->height >> s->chroma_y_shift;
1786 s->c_superblock_width = (c_width + 31) / 32;
1787 s->c_superblock_height = (c_height + 31) / 32;
1788 s->c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1790 s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
1791 s->u_superblock_start = s->y_superblock_count;
1792 s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
1794 s->macroblock_width = (s->width + 15) / 16;
1795 s->macroblock_height = (s->height + 15) / 16;
1796 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1798 s->fragment_width[0] = s->width / FRAGMENT_PIXELS;
1799 s->fragment_height[0] = s->height / FRAGMENT_PIXELS;
1800 s->fragment_width[1] = s->fragment_width[0] >> s->chroma_x_shift;
1801 s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift;
1803 /* fragment count covers all 8x8 blocks for all 3 planes */
1804 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1805 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1806 s->fragment_count = y_fragment_count + 2 * c_fragment_count;
1807 s->fragment_start[1] = y_fragment_count;
1808 s->fragment_start[2] = y_fragment_count + c_fragment_count;
1810 if (!s->theora_tables) {
1811 for (i = 0; i < 64; i++) {
1812 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1813 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1814 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1815 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1816 s->base_matrix[2][i] = vp31_inter_dequant[i];
1817 s->filter_limit_values[i] = vp31_filter_limit_values[i];
1820 for (inter = 0; inter < 2; inter++) {
1821 for (plane = 0; plane < 3; plane++) {
1822 s->qr_count[inter][plane] = 1;
1823 s->qr_size[inter][plane][0] = 63;
1824 s->qr_base[inter][plane][0] =
1825 s->qr_base[inter][plane][1] = 2 * inter + (!!plane) * !inter;
1829 /* init VLC tables */
1830 for (i = 0; i < 16; i++) {
1832 init_vlc(&s->dc_vlc[i], 11, 32,
1833 &dc_bias[i][0][1], 4, 2,
1834 &dc_bias[i][0][0], 4, 2, 0);
1836 /* group 1 AC histograms */
1837 init_vlc(&s->ac_vlc_1[i], 11, 32,
1838 &ac_bias_0[i][0][1], 4, 2,
1839 &ac_bias_0[i][0][0], 4, 2, 0);
1841 /* group 2 AC histograms */
1842 init_vlc(&s->ac_vlc_2[i], 11, 32,
1843 &ac_bias_1[i][0][1], 4, 2,
1844 &ac_bias_1[i][0][0], 4, 2, 0);
1846 /* group 3 AC histograms */
1847 init_vlc(&s->ac_vlc_3[i], 11, 32,
1848 &ac_bias_2[i][0][1], 4, 2,
1849 &ac_bias_2[i][0][0], 4, 2, 0);
1851 /* group 4 AC histograms */
1852 init_vlc(&s->ac_vlc_4[i], 11, 32,
1853 &ac_bias_3[i][0][1], 4, 2,
1854 &ac_bias_3[i][0][0], 4, 2, 0);
1857 for (i = 0; i < 16; i++) {
1859 if (init_vlc(&s->dc_vlc[i], 11, 32,
1860 &s->huffman_table[i][0][1], 8, 4,
1861 &s->huffman_table[i][0][0], 8, 4, 0) < 0)
1864 /* group 1 AC histograms */
1865 if (init_vlc(&s->ac_vlc_1[i], 11, 32,
1866 &s->huffman_table[i + 16][0][1], 8, 4,
1867 &s->huffman_table[i + 16][0][0], 8, 4, 0) < 0)
1870 /* group 2 AC histograms */
1871 if (init_vlc(&s->ac_vlc_2[i], 11, 32,
1872 &s->huffman_table[i + 16 * 2][0][1], 8, 4,
1873 &s->huffman_table[i + 16 * 2][0][0], 8, 4, 0) < 0)
1876 /* group 3 AC histograms */
1877 if (init_vlc(&s->ac_vlc_3[i], 11, 32,
1878 &s->huffman_table[i + 16 * 3][0][1], 8, 4,
1879 &s->huffman_table[i + 16 * 3][0][0], 8, 4, 0) < 0)
1882 /* group 4 AC histograms */
1883 if (init_vlc(&s->ac_vlc_4[i], 11, 32,
1884 &s->huffman_table[i + 16 * 4][0][1], 8, 4,
1885 &s->huffman_table[i + 16 * 4][0][0], 8, 4, 0) < 0)
1890 init_vlc(&s->superblock_run_length_vlc, 6, 34,
1891 &superblock_run_length_vlc_table[0][1], 4, 2,
1892 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1894 init_vlc(&s->fragment_run_length_vlc, 5, 30,
1895 &fragment_run_length_vlc_table[0][1], 4, 2,
1896 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1898 init_vlc(&s->mode_code_vlc, 3, 8,
1899 &mode_code_vlc_table[0][1], 2, 1,
1900 &mode_code_vlc_table[0][0], 2, 1, 0);
1902 init_vlc(&s->motion_vector_vlc, 6, 63,
1903 &motion_vector_vlc_table[0][1], 2, 1,
1904 &motion_vector_vlc_table[0][0], 2, 1, 0);
1906 return allocate_tables(avctx);
1909 av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1913 /// Release and shuffle frames after decode finishes
1914 static int update_frames(AVCodecContext *avctx)
1916 Vp3DecodeContext *s = avctx->priv_data;
1919 /* shuffle frames (last = current) */
1920 ff_thread_release_buffer(avctx, &s->last_frame);
1921 ret = ff_thread_ref_frame(&s->last_frame, &s->current_frame);
1926 ff_thread_release_buffer(avctx, &s->golden_frame);
1927 ret = ff_thread_ref_frame(&s->golden_frame, &s->current_frame);
1931 ff_thread_release_buffer(avctx, &s->current_frame);
1935 static int ref_frame(Vp3DecodeContext *s, ThreadFrame *dst, ThreadFrame *src)
1937 ff_thread_release_buffer(s->avctx, dst);
1938 if (src->f->data[0])
1939 return ff_thread_ref_frame(dst, src);
1943 static int ref_frames(Vp3DecodeContext *dst, Vp3DecodeContext *src)
1946 if ((ret = ref_frame(dst, &dst->current_frame, &src->current_frame)) < 0 ||
1947 (ret = ref_frame(dst, &dst->golden_frame, &src->golden_frame)) < 0 ||
1948 (ret = ref_frame(dst, &dst->last_frame, &src->last_frame)) < 0)
1954 static int vp3_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
1956 Vp3DecodeContext *s = dst->priv_data, *s1 = src->priv_data;
1957 int qps_changed = 0, i, err;
1959 #define copy_fields(to, from, start_field, end_field) \
1960 memcpy(&to->start_field, &from->start_field, \
1961 (char *) &to->end_field - (char *) &to->start_field)
1963 if (!s1->current_frame.f->data[0] ||
1964 s->width != s1->width || s->height != s1->height) {
1971 if (!s->current_frame.f)
1972 return AVERROR(ENOMEM);
1973 // init tables if the first frame hasn't been decoded
1974 if (!s->current_frame.f->data[0]) {
1975 int y_fragment_count, c_fragment_count;
1977 err = allocate_tables(dst);
1980 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1981 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1982 memcpy(s->motion_val[0], s1->motion_val[0],
1983 y_fragment_count * sizeof(*s->motion_val[0]));
1984 memcpy(s->motion_val[1], s1->motion_val[1],
1985 c_fragment_count * sizeof(*s->motion_val[1]));
1988 // copy previous frame data
1989 if ((err = ref_frames(s, s1)) < 0)
1992 s->keyframe = s1->keyframe;
1994 // copy qscale data if necessary
1995 for (i = 0; i < 3; i++) {
1996 if (s->qps[i] != s1->qps[1]) {
1998 memcpy(&s->qmat[i], &s1->qmat[i], sizeof(s->qmat[i]));
2002 if (s->qps[0] != s1->qps[0])
2003 memcpy(&s->bounding_values_array, &s1->bounding_values_array,
2004 sizeof(s->bounding_values_array));
2007 copy_fields(s, s1, qps, superblock_count);
2011 return update_frames(dst);
2015 static int vp3_decode_frame(AVCodecContext *avctx,
2016 void *data, int *got_frame,
2019 AVFrame *frame = data;
2020 const uint8_t *buf = avpkt->data;
2021 int buf_size = avpkt->size;
2022 Vp3DecodeContext *s = avctx->priv_data;
2026 if ((ret = init_get_bits8(&gb, buf, buf_size)) < 0)
2029 #if CONFIG_THEORA_DECODER
2030 if (s->theora && get_bits1(&gb)) {
2031 int type = get_bits(&gb, 7);
2032 skip_bits_long(&gb, 6*8); /* "theora" */
2034 if (s->avctx->active_thread_type&FF_THREAD_FRAME) {
2035 av_log(avctx, AV_LOG_ERROR, "midstream reconfiguration with multithreading is unsupported, try -threads 1\n");
2036 return AVERROR_PATCHWELCOME;
2039 vp3_decode_end(avctx);
2040 ret = theora_decode_header(avctx, &gb);
2043 ret = vp3_decode_init(avctx);
2045 vp3_decode_end(avctx);
2049 } else if (type == 2) {
2050 vp3_decode_end(avctx);
2051 ret = theora_decode_tables(avctx, &gb);
2053 ret = vp3_decode_init(avctx);
2055 vp3_decode_end(avctx);
2061 av_log(avctx, AV_LOG_ERROR,
2062 "Header packet passed to frame decoder, skipping\n");
2067 s->keyframe = !get_bits1(&gb);
2068 if (!s->all_fragments) {
2069 av_log(avctx, AV_LOG_ERROR, "Data packet without prior valid headers\n");
2074 for (i = 0; i < 3; i++)
2075 s->last_qps[i] = s->qps[i];
2079 s->qps[s->nqps++] = get_bits(&gb, 6);
2080 } while (s->theora >= 0x030200 && s->nqps < 3 && get_bits1(&gb));
2081 for (i = s->nqps; i < 3; i++)
2084 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
2085 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
2086 s->keyframe ? "key" : "", avctx->frame_number + 1, s->qps[0]);
2088 s->skip_loop_filter = !s->filter_limit_values[s->qps[0]] ||
2089 avctx->skip_loop_filter >= (s->keyframe ? AVDISCARD_ALL
2090 : AVDISCARD_NONKEY);
2092 if (s->qps[0] != s->last_qps[0])
2093 init_loop_filter(s);
2095 for (i = 0; i < s->nqps; i++)
2096 // reinit all dequantizers if the first one changed, because
2097 // the DC of the first quantizer must be used for all matrices
2098 if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
2099 init_dequantizer(s, i);
2101 if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
2104 s->current_frame.f->pict_type = s->keyframe ? AV_PICTURE_TYPE_I
2105 : AV_PICTURE_TYPE_P;
2106 s->current_frame.f->key_frame = s->keyframe;
2107 if (ff_thread_get_buffer(avctx, &s->current_frame, AV_GET_BUFFER_FLAG_REF) < 0)
2110 if (!s->edge_emu_buffer)
2111 s->edge_emu_buffer = av_malloc(9 * FFABS(s->current_frame.f->linesize[0]));
2115 skip_bits(&gb, 4); /* width code */
2116 skip_bits(&gb, 4); /* height code */
2118 s->version = get_bits(&gb, 5);
2119 if (avctx->frame_number == 0)
2120 av_log(s->avctx, AV_LOG_DEBUG,
2121 "VP version: %d\n", s->version);
2124 if (s->version || s->theora) {
2126 av_log(s->avctx, AV_LOG_ERROR,
2127 "Warning, unsupported keyframe coding type?!\n");
2128 skip_bits(&gb, 2); /* reserved? */
2131 if (!s->golden_frame.f->data[0]) {
2132 av_log(s->avctx, AV_LOG_WARNING,
2133 "vp3: first frame not a keyframe\n");
2135 s->golden_frame.f->pict_type = AV_PICTURE_TYPE_I;
2136 if (ff_thread_get_buffer(avctx, &s->golden_frame,
2137 AV_GET_BUFFER_FLAG_REF) < 0)
2139 ff_thread_release_buffer(avctx, &s->last_frame);
2140 if ((ret = ff_thread_ref_frame(&s->last_frame,
2141 &s->golden_frame)) < 0)
2143 ff_thread_report_progress(&s->last_frame, INT_MAX, 0);
2147 memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
2148 ff_thread_finish_setup(avctx);
2150 if (unpack_superblocks(s, &gb)) {
2151 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
2154 if (unpack_modes(s, &gb)) {
2155 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
2158 if (unpack_vectors(s, &gb)) {
2159 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2162 if (unpack_block_qpis(s, &gb)) {
2163 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
2166 if (unpack_dct_coeffs(s, &gb)) {
2167 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
2171 for (i = 0; i < 3; i++) {
2172 int height = s->height >> (i && s->chroma_y_shift);
2173 if (s->flipped_image)
2174 s->data_offset[i] = 0;
2176 s->data_offset[i] = (height - 1) * s->current_frame.f->linesize[i];
2179 s->last_slice_end = 0;
2180 for (i = 0; i < s->c_superblock_height; i++)
2183 // filter the last row
2184 for (i = 0; i < 3; i++) {
2185 int row = (s->height >> (3 + (i && s->chroma_y_shift))) - 1;
2186 apply_loop_filter(s, i, row, row + 1);
2188 vp3_draw_horiz_band(s, s->height);
2190 /* output frame, offset as needed */
2191 if ((ret = av_frame_ref(data, s->current_frame.f)) < 0)
2194 frame->crop_left = s->offset_x;
2195 frame->crop_right = avctx->coded_width - avctx->width - s->offset_x;
2196 frame->crop_top = s->offset_y;
2197 frame->crop_bottom = avctx->coded_height - avctx->height - s->offset_y;
2201 if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_FRAME)) {
2202 ret = update_frames(avctx);
2210 ff_thread_report_progress(&s->current_frame, INT_MAX, 0);
2212 if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_FRAME))
2213 av_frame_unref(s->current_frame.f);
2218 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2220 Vp3DecodeContext *s = avctx->priv_data;
2222 if (get_bits1(gb)) {
2224 if (s->entries >= 32) { /* overflow */
2225 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2228 token = get_bits(gb, 5);
2229 ff_dlog(avctx, "hti %d hbits %x token %d entry : %d size %d\n",
2230 s->hti, s->hbits, token, s->entries, s->huff_code_size);
2231 s->huffman_table[s->hti][token][0] = s->hbits;
2232 s->huffman_table[s->hti][token][1] = s->huff_code_size;
2235 if (s->huff_code_size >= 32) { /* overflow */
2236 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2239 s->huff_code_size++;
2241 if (read_huffman_tree(avctx, gb))
2244 if (read_huffman_tree(avctx, gb))
2247 s->huff_code_size--;
2253 static int vp3_init_thread_copy(AVCodecContext *avctx)
2255 Vp3DecodeContext *s = avctx->priv_data;
2257 s->superblock_coding = NULL;
2258 s->all_fragments = NULL;
2259 s->coded_fragment_list[0] = NULL;
2260 s->dct_tokens_base = NULL;
2261 s->superblock_fragments = NULL;
2262 s->macroblock_coding = NULL;
2263 s->motion_val[0] = NULL;
2264 s->motion_val[1] = NULL;
2265 s->edge_emu_buffer = NULL;
2267 return init_frames(s);
2271 #if CONFIG_THEORA_DECODER
2272 static const enum AVPixelFormat theora_pix_fmts[4] = {
2273 AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P
2276 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2278 Vp3DecodeContext *s = avctx->priv_data;
2279 int visible_width, visible_height, colorspace;
2280 uint8_t offset_x = 0, offset_y = 0;
2282 AVRational fps, aspect;
2284 s->theora_header = 0;
2285 s->theora = get_bits_long(gb, 24);
2286 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2288 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3
2289 * but previous versions have the image flipped relative to vp3 */
2290 if (s->theora < 0x030200) {
2291 s->flipped_image = 1;
2292 av_log(avctx, AV_LOG_DEBUG,
2293 "Old (<alpha3) Theora bitstream, flipped image\n");
2297 s->width = get_bits(gb, 16) << 4;
2299 s->height = get_bits(gb, 16) << 4;
2301 if (s->theora >= 0x030200) {
2302 visible_width = get_bits_long(gb, 24);
2303 visible_height = get_bits_long(gb, 24);
2305 offset_x = get_bits(gb, 8); /* offset x */
2306 offset_y = get_bits(gb, 8); /* offset y, from bottom */
2310 if (av_image_check_size(visible_width, visible_height, 0, avctx) < 0 ||
2311 visible_width + offset_x > s->width ||
2312 visible_height + offset_y > s->height) {
2313 av_log(avctx, AV_LOG_ERROR,
2314 "Invalid frame dimensions - w:%d h:%d x:%d y:%d (%dx%d).\n",
2315 visible_width, visible_height, offset_x, offset_y,
2316 s->width, s->height);
2317 return AVERROR_INVALIDDATA;
2320 fps.num = get_bits_long(gb, 32);
2321 fps.den = get_bits_long(gb, 32);
2322 if (fps.num && fps.den) {
2323 if (fps.num < 0 || fps.den < 0) {
2324 av_log(avctx, AV_LOG_ERROR, "Invalid framerate\n");
2325 return AVERROR_INVALIDDATA;
2327 av_reduce(&avctx->framerate.den, &avctx->framerate.num,
2328 fps.den, fps.num, 1 << 30);
2331 aspect.num = get_bits_long(gb, 24);
2332 aspect.den = get_bits_long(gb, 24);
2333 if (aspect.num && aspect.den) {
2334 av_reduce(&avctx->sample_aspect_ratio.num,
2335 &avctx->sample_aspect_ratio.den,
2336 aspect.num, aspect.den, 1 << 30);
2337 ff_set_sar(avctx, avctx->sample_aspect_ratio);
2340 if (s->theora < 0x030200)
2341 skip_bits(gb, 5); /* keyframe frequency force */
2342 colorspace = get_bits(gb, 8);
2343 skip_bits(gb, 24); /* bitrate */
2345 skip_bits(gb, 6); /* quality hint */
2347 if (s->theora >= 0x030200) {
2348 skip_bits(gb, 5); /* keyframe frequency force */
2349 avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)];
2350 if (avctx->pix_fmt == AV_PIX_FMT_NONE) {
2351 av_log(avctx, AV_LOG_ERROR, "Invalid pixel format\n");
2352 return AVERROR_INVALIDDATA;
2354 skip_bits(gb, 3); /* reserved */
2356 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
2358 ret = ff_set_dimensions(avctx, s->width, s->height);
2361 if (!(avctx->flags2 & AV_CODEC_FLAG2_IGNORE_CROP)) {
2362 avctx->width = visible_width;
2363 avctx->height = visible_height;
2364 // translate offsets from theora axis ([0,0] lower left)
2365 // to normal axis ([0,0] upper left)
2366 s->offset_x = offset_x;
2367 s->offset_y = s->height - visible_height - offset_y;
2370 if (colorspace == 1)
2371 avctx->color_primaries = AVCOL_PRI_BT470M;
2372 else if (colorspace == 2)
2373 avctx->color_primaries = AVCOL_PRI_BT470BG;
2375 if (colorspace == 1 || colorspace == 2) {
2376 avctx->colorspace = AVCOL_SPC_BT470BG;
2377 avctx->color_trc = AVCOL_TRC_BT709;
2380 s->theora_header = 1;
2384 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2386 Vp3DecodeContext *s = avctx->priv_data;
2387 int i, n, matrices, inter, plane;
2389 if (!s->theora_header)
2390 return AVERROR_INVALIDDATA;
2392 if (s->theora >= 0x030200) {
2393 n = get_bits(gb, 3);
2394 /* loop filter limit values table */
2396 for (i = 0; i < 64; i++)
2397 s->filter_limit_values[i] = get_bits(gb, n);
2400 if (s->theora >= 0x030200)
2401 n = get_bits(gb, 4) + 1;
2404 /* quality threshold table */
2405 for (i = 0; i < 64; i++)
2406 s->coded_ac_scale_factor[i] = get_bits(gb, n);
2408 if (s->theora >= 0x030200)
2409 n = get_bits(gb, 4) + 1;
2412 /* dc scale factor table */
2413 for (i = 0; i < 64; i++)
2414 s->coded_dc_scale_factor[i] = get_bits(gb, n);
2416 if (s->theora >= 0x030200)
2417 matrices = get_bits(gb, 9) + 1;
2421 if (matrices > 384) {
2422 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2426 for (n = 0; n < matrices; n++)
2427 for (i = 0; i < 64; i++)
2428 s->base_matrix[n][i] = get_bits(gb, 8);
2430 for (inter = 0; inter <= 1; inter++) {
2431 for (plane = 0; plane <= 2; plane++) {
2433 if (inter || plane > 0)
2434 newqr = get_bits1(gb);
2437 if (inter && get_bits1(gb)) {
2441 qtj = (3 * inter + plane - 1) / 3;
2442 plj = (plane + 2) % 3;
2444 s->qr_count[inter][plane] = s->qr_count[qtj][plj];
2445 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj],
2446 sizeof(s->qr_size[0][0]));
2447 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj],
2448 sizeof(s->qr_base[0][0]));
2454 i = get_bits(gb, av_log2(matrices - 1) + 1);
2455 if (i >= matrices) {
2456 av_log(avctx, AV_LOG_ERROR,
2457 "invalid base matrix index\n");
2460 s->qr_base[inter][plane][qri] = i;
2463 i = get_bits(gb, av_log2(63 - qi) + 1) + 1;
2464 s->qr_size[inter][plane][qri++] = i;
2469 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2472 s->qr_count[inter][plane] = qri;
2477 /* Huffman tables */
2478 for (s->hti = 0; s->hti < 80; s->hti++) {
2480 s->huff_code_size = 1;
2481 if (!get_bits1(gb)) {
2483 if (read_huffman_tree(avctx, gb))
2486 if (read_huffman_tree(avctx, gb))
2491 s->theora_tables = 1;
2496 static av_cold int theora_decode_init(AVCodecContext *avctx)
2498 Vp3DecodeContext *s = avctx->priv_data;
2501 const uint8_t *header_start[3];
2506 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
2510 if (!avctx->extradata_size) {
2511 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2515 if (avpriv_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2516 42, header_start, header_len) < 0) {
2517 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2521 for (i = 0; i < 3; i++) {
2522 if (header_len[i] <= 0)
2524 ret = init_get_bits8(&gb, header_start[i], header_len[i]);
2528 ptype = get_bits(&gb, 8);
2530 if (!(ptype & 0x80)) {
2531 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2535 // FIXME: Check for this as well.
2536 skip_bits_long(&gb, 6 * 8); /* "theora" */
2540 if (theora_decode_header(avctx, &gb) < 0)
2544 // FIXME: is this needed? it breaks sometimes
2545 // theora_decode_comments(avctx, gb);
2548 if (theora_decode_tables(avctx, &gb))
2552 av_log(avctx, AV_LOG_ERROR,
2553 "Unknown Theora config packet: %d\n", ptype & ~0x80);
2556 if (ptype != 0x81 && 8 * header_len[i] != get_bits_count(&gb))
2557 av_log(avctx, AV_LOG_WARNING,
2558 "%d bits left in packet %X\n",
2559 8 * header_len[i] - get_bits_count(&gb), ptype);
2560 if (s->theora < 0x030200)
2564 return vp3_decode_init(avctx);
2567 AVCodec ff_theora_decoder = {
2569 .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2570 .type = AVMEDIA_TYPE_VIDEO,
2571 .id = AV_CODEC_ID_THEORA,
2572 .priv_data_size = sizeof(Vp3DecodeContext),
2573 .init = theora_decode_init,
2574 .close = vp3_decode_end,
2575 .decode = vp3_decode_frame,
2576 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DRAW_HORIZ_BAND |
2577 AV_CODEC_CAP_FRAME_THREADS,
2578 .flush = vp3_decode_flush,
2579 .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
2580 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context),
2581 .caps_internal = FF_CODEC_CAP_EXPORTS_CROPPING,
2585 AVCodec ff_vp3_decoder = {
2587 .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2588 .type = AVMEDIA_TYPE_VIDEO,
2589 .id = AV_CODEC_ID_VP3,
2590 .priv_data_size = sizeof(Vp3DecodeContext),
2591 .init = vp3_decode_init,
2592 .close = vp3_decode_end,
2593 .decode = vp3_decode_frame,
2594 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DRAW_HORIZ_BAND |
2595 AV_CODEC_CAP_FRAME_THREADS,
2596 .flush = vp3_decode_flush,
2597 .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
2598 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context),