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
39 #include "bitstream.h"
44 #define FRAGMENT_PIXELS 8
49 * Define one or more of the following compile-time variables to 1 to obtain
50 * elaborate information about certain aspects of the decoding process.
52 * KEYFRAMES_ONLY: set this to 1 to only see keyframes (VP3 slideshow mode)
53 * DEBUG_VP3: high-level decoding flow
54 * DEBUG_INIT: initialization parameters
55 * DEBUG_DEQUANTIZERS: display how the dequanization tables are built
56 * DEBUG_BLOCK_CODING: unpacking the superblock/macroblock/fragment coding
57 * DEBUG_MODES: unpacking the coding modes for individual fragments
58 * DEBUG_VECTORS: display the motion vectors
59 * DEBUG_TOKEN: display exhaustive information about each DCT token
60 * DEBUG_VLC: display the VLCs as they are extracted from the stream
61 * DEBUG_DC_PRED: display the process of reversing DC prediction
62 * DEBUG_IDCT: show every detail of the IDCT process
65 #define KEYFRAMES_ONLY 0
69 #define DEBUG_DEQUANTIZERS 0
70 #define DEBUG_BLOCK_CODING 0
72 #define DEBUG_VECTORS 0
75 #define DEBUG_DC_PRED 0
79 #define debug_vp3(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
81 static inline void debug_vp3(const char *format, ...) { }
85 #define debug_init(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
87 static inline void debug_init(const char *format, ...) { }
90 #if DEBUG_DEQUANTIZERS
91 #define debug_dequantizers(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
93 static inline void debug_dequantizers(const char *format, ...) { }
96 #if DEBUG_BLOCK_CODING
97 #define debug_block_coding(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
99 static inline void debug_block_coding(const char *format, ...) { }
103 #define debug_modes(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
105 static inline void debug_modes(const char *format, ...) { }
109 #define debug_vectors(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
111 static inline void debug_vectors(const char *format, ...) { }
115 #define debug_token(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
117 static inline void debug_token(const char *format, ...) { }
121 #define debug_vlc(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
123 static inline void debug_vlc(const char *format, ...) { }
127 #define debug_dc_pred(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
129 static inline void debug_dc_pred(const char *format, ...) { }
133 #define debug_idct(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
135 static inline void debug_idct(const char *format, ...) { }
138 typedef struct Coeff {
144 //FIXME split things out into their own arrays
145 typedef struct Vp3Fragment {
147 /* address of first pixel taking into account which plane the fragment
148 * lives on as well as the plane stride */
150 /* this is the macroblock that the fragment belongs to */
152 uint8_t coding_method;
158 #define SB_NOT_CODED 0
159 #define SB_PARTIALLY_CODED 1
160 #define SB_FULLY_CODED 2
162 #define MODE_INTER_NO_MV 0
164 #define MODE_INTER_PLUS_MV 2
165 #define MODE_INTER_LAST_MV 3
166 #define MODE_INTER_PRIOR_LAST 4
167 #define MODE_USING_GOLDEN 5
168 #define MODE_GOLDEN_MV 6
169 #define MODE_INTER_FOURMV 7
170 #define CODING_MODE_COUNT 8
172 /* special internal mode */
175 /* There are 6 preset schemes, plus a free-form scheme */
176 static int ModeAlphabet[7][CODING_MODE_COUNT] =
178 /* this is the custom scheme */
179 { 0, 0, 0, 0, 0, 0, 0, 0 },
181 /* scheme 1: Last motion vector dominates */
182 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
183 MODE_INTER_PLUS_MV, MODE_INTER_NO_MV,
184 MODE_INTRA, MODE_USING_GOLDEN,
185 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
188 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
189 MODE_INTER_NO_MV, MODE_INTER_PLUS_MV,
190 MODE_INTRA, MODE_USING_GOLDEN,
191 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
194 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
195 MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
196 MODE_INTRA, MODE_USING_GOLDEN,
197 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
200 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
201 MODE_INTER_NO_MV, MODE_INTER_PRIOR_LAST,
202 MODE_INTRA, MODE_USING_GOLDEN,
203 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
205 /* scheme 5: No motion vector dominates */
206 { MODE_INTER_NO_MV, MODE_INTER_LAST_MV,
207 MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
208 MODE_INTRA, MODE_USING_GOLDEN,
209 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
212 { MODE_INTER_NO_MV, MODE_USING_GOLDEN,
213 MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
214 MODE_INTER_PLUS_MV, MODE_INTRA,
215 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
219 #define MIN_DEQUANT_VAL 2
221 typedef struct Vp3DecodeContext {
222 AVCodecContext *avctx;
223 int theora, theora_tables;
226 AVFrame golden_frame;
228 AVFrame current_frame;
236 int last_quality_index;
238 int superblock_count;
239 int superblock_width;
240 int superblock_height;
241 int y_superblock_width;
242 int y_superblock_height;
243 int c_superblock_width;
244 int c_superblock_height;
245 int u_superblock_start;
246 int v_superblock_start;
247 unsigned char *superblock_coding;
249 int macroblock_count;
250 int macroblock_width;
251 int macroblock_height;
257 Vp3Fragment *all_fragments;
260 int fragment_start[3];
265 uint16_t coded_dc_scale_factor[64];
266 uint32_t coded_ac_scale_factor[64];
267 uint8_t base_matrix[384][64];
268 uint8_t qr_count[2][3];
269 uint8_t qr_size [2][3][64];
270 uint16_t qr_base[2][3][64];
272 /* this is a list of indexes into the all_fragments array indicating
273 * which of the fragments are coded */
274 int *coded_fragment_list;
275 int coded_fragment_list_index;
276 int pixel_addresses_initialized;
284 VLC superblock_run_length_vlc;
285 VLC fragment_run_length_vlc;
287 VLC motion_vector_vlc;
289 /* these arrays need to be on 16-byte boundaries since SSE2 operations
291 DECLARE_ALIGNED_16(int16_t, qmat[2][4][64]); //<qmat[is_inter][plane]
293 /* This table contains superblock_count * 16 entries. Each set of 16
294 * numbers corresponds to the fragment indexes 0..15 of the superblock.
295 * An entry will be -1 to indicate that no entry corresponds to that
297 int *superblock_fragments;
299 /* This table contains superblock_count * 4 entries. Each set of 4
300 * numbers corresponds to the macroblock indexes 0..3 of the superblock.
301 * An entry will be -1 to indicate that no entry corresponds to that
303 int *superblock_macroblocks;
305 /* This table contains macroblock_count * 6 entries. Each set of 6
306 * numbers corresponds to the fragment indexes 0..5 which comprise
307 * the macroblock (4 Y fragments and 2 C fragments). */
308 int *macroblock_fragments;
309 /* This is an array that indicates how a particular macroblock
311 unsigned char *macroblock_coding;
313 int first_coded_y_fragment;
314 int first_coded_c_fragment;
315 int last_coded_y_fragment;
316 int last_coded_c_fragment;
318 uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
319 int8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
326 uint16_t huffman_table[80][32][2];
328 uint32_t filter_limit_values[64];
329 int bounding_values_array[256];
332 /************************************************************************
333 * VP3 specific functions
334 ************************************************************************/
337 * This function sets up all of the various blocks mappings:
338 * superblocks <-> fragments, macroblocks <-> fragments,
339 * superblocks <-> macroblocks
341 * Returns 0 is successful; returns 1 if *anything* went wrong.
343 static int init_block_mapping(Vp3DecodeContext *s)
346 signed int hilbert_walk_mb[4];
348 int current_fragment = 0;
349 int current_width = 0;
350 int current_height = 0;
353 int superblock_row_inc = 0;
355 int mapping_index = 0;
357 int current_macroblock;
360 signed char travel_width[16] = {
367 signed char travel_height[16] = {
374 signed char travel_width_mb[4] = {
378 signed char travel_height_mb[4] = {
382 debug_vp3(" vp3: initialize block mapping tables\n");
384 hilbert_walk_mb[0] = 1;
385 hilbert_walk_mb[1] = s->macroblock_width;
386 hilbert_walk_mb[2] = 1;
387 hilbert_walk_mb[3] = -s->macroblock_width;
389 /* iterate through each superblock (all planes) and map the fragments */
390 for (i = 0; i < s->superblock_count; i++) {
391 debug_init(" superblock %d (u starts @ %d, v starts @ %d)\n",
392 i, s->u_superblock_start, s->v_superblock_start);
394 /* time to re-assign the limits? */
397 /* start of Y superblocks */
398 right_edge = s->fragment_width;
399 bottom_edge = s->fragment_height;
402 superblock_row_inc = 3 * s->fragment_width -
403 (s->y_superblock_width * 4 - s->fragment_width);
405 /* the first operation for this variable is to advance by 1 */
406 current_fragment = -1;
408 } else if (i == s->u_superblock_start) {
410 /* start of U superblocks */
411 right_edge = s->fragment_width / 2;
412 bottom_edge = s->fragment_height / 2;
415 superblock_row_inc = 3 * (s->fragment_width / 2) -
416 (s->c_superblock_width * 4 - s->fragment_width / 2);
418 /* the first operation for this variable is to advance by 1 */
419 current_fragment = s->fragment_start[1] - 1;
421 } else if (i == s->v_superblock_start) {
423 /* start of V superblocks */
424 right_edge = s->fragment_width / 2;
425 bottom_edge = s->fragment_height / 2;
428 superblock_row_inc = 3 * (s->fragment_width / 2) -
429 (s->c_superblock_width * 4 - s->fragment_width / 2);
431 /* the first operation for this variable is to advance by 1 */
432 current_fragment = s->fragment_start[2] - 1;
436 if (current_width >= right_edge - 1) {
437 /* reset width and move to next superblock row */
441 /* fragment is now at the start of a new superblock row */
442 current_fragment += superblock_row_inc;
445 /* iterate through all 16 fragments in a superblock */
446 for (j = 0; j < 16; j++) {
447 current_fragment += travel_width[j] + right_edge * travel_height[j];
448 current_width += travel_width[j];
449 current_height += travel_height[j];
451 /* check if the fragment is in bounds */
452 if ((current_width < right_edge) &&
453 (current_height < bottom_edge)) {
454 s->superblock_fragments[mapping_index] = current_fragment;
455 debug_init(" mapping fragment %d to superblock %d, position %d (%d/%d x %d/%d)\n",
456 s->superblock_fragments[mapping_index], i, j,
457 current_width, right_edge, current_height, bottom_edge);
459 s->superblock_fragments[mapping_index] = -1;
460 debug_init(" superblock %d, position %d has no fragment (%d/%d x %d/%d)\n",
462 current_width, right_edge, current_height, bottom_edge);
469 /* initialize the superblock <-> macroblock mapping; iterate through
470 * all of the Y plane superblocks to build this mapping */
471 right_edge = s->macroblock_width;
472 bottom_edge = s->macroblock_height;
475 superblock_row_inc = s->macroblock_width -
476 (s->y_superblock_width * 2 - s->macroblock_width);
477 hilbert = hilbert_walk_mb;
479 current_macroblock = -1;
480 for (i = 0; i < s->u_superblock_start; i++) {
482 if (current_width >= right_edge - 1) {
483 /* reset width and move to next superblock row */
487 /* macroblock is now at the start of a new superblock row */
488 current_macroblock += superblock_row_inc;
491 /* iterate through each potential macroblock in the superblock */
492 for (j = 0; j < 4; j++) {
493 current_macroblock += hilbert_walk_mb[j];
494 current_width += travel_width_mb[j];
495 current_height += travel_height_mb[j];
497 /* check if the macroblock is in bounds */
498 if ((current_width < right_edge) &&
499 (current_height < bottom_edge)) {
500 s->superblock_macroblocks[mapping_index] = current_macroblock;
501 debug_init(" mapping macroblock %d to superblock %d, position %d (%d/%d x %d/%d)\n",
502 s->superblock_macroblocks[mapping_index], i, j,
503 current_width, right_edge, current_height, bottom_edge);
505 s->superblock_macroblocks[mapping_index] = -1;
506 debug_init(" superblock %d, position %d has no macroblock (%d/%d x %d/%d)\n",
508 current_width, right_edge, current_height, bottom_edge);
515 /* initialize the macroblock <-> fragment mapping */
516 current_fragment = 0;
517 current_macroblock = 0;
519 for (i = 0; i < s->fragment_height; i += 2) {
521 for (j = 0; j < s->fragment_width; j += 2) {
523 debug_init(" macroblock %d contains fragments: ", current_macroblock);
524 s->all_fragments[current_fragment].macroblock = current_macroblock;
525 s->macroblock_fragments[mapping_index++] = current_fragment;
526 debug_init("%d ", current_fragment);
528 if (j + 1 < s->fragment_width) {
529 s->all_fragments[current_fragment + 1].macroblock = current_macroblock;
530 s->macroblock_fragments[mapping_index++] = current_fragment + 1;
531 debug_init("%d ", current_fragment + 1);
533 s->macroblock_fragments[mapping_index++] = -1;
535 if (i + 1 < s->fragment_height) {
536 s->all_fragments[current_fragment + s->fragment_width].macroblock =
538 s->macroblock_fragments[mapping_index++] =
539 current_fragment + s->fragment_width;
540 debug_init("%d ", current_fragment + s->fragment_width);
542 s->macroblock_fragments[mapping_index++] = -1;
544 if ((j + 1 < s->fragment_width) && (i + 1 < s->fragment_height)) {
545 s->all_fragments[current_fragment + s->fragment_width + 1].macroblock =
547 s->macroblock_fragments[mapping_index++] =
548 current_fragment + s->fragment_width + 1;
549 debug_init("%d ", current_fragment + s->fragment_width + 1);
551 s->macroblock_fragments[mapping_index++] = -1;
554 c_fragment = s->fragment_start[1] +
555 (i * s->fragment_width / 4) + (j / 2);
556 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
557 s->macroblock_fragments[mapping_index++] = c_fragment;
558 debug_init("%d ", c_fragment);
560 c_fragment = s->fragment_start[2] +
561 (i * s->fragment_width / 4) + (j / 2);
562 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
563 s->macroblock_fragments[mapping_index++] = c_fragment;
564 debug_init("%d ", c_fragment);
568 if (j + 2 <= s->fragment_width)
569 current_fragment += 2;
572 current_macroblock++;
575 current_fragment += s->fragment_width;
578 return 0; /* successful path out */
582 * This function wipes out all of the fragment data.
584 static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
588 /* zero out all of the fragment information */
589 s->coded_fragment_list_index = 0;
590 for (i = 0; i < s->fragment_count; i++) {
591 s->all_fragments[i].coeff_count = 0;
592 s->all_fragments[i].motion_x = 127;
593 s->all_fragments[i].motion_y = 127;
594 s->all_fragments[i].next_coeff= NULL;
596 s->coeffs[i].coeff=0;
597 s->coeffs[i].next= NULL;
602 * This function sets up the dequantization tables used for a particular
605 static void init_dequantizer(Vp3DecodeContext *s)
607 int ac_scale_factor = s->coded_ac_scale_factor[s->quality_index];
608 int dc_scale_factor = s->coded_dc_scale_factor[s->quality_index];
609 int i, plane, inter, qri, bmi, bmj, qistart;
611 debug_vp3(" vp3: initializing dequantization tables\n");
613 for(inter=0; inter<2; inter++){
614 for(plane=0; plane<3; plane++){
616 for(qri=0; qri<s->qr_count[inter][plane]; qri++){
617 sum+= s->qr_size[inter][plane][qri];
618 if(s->quality_index <= sum)
621 qistart= sum - s->qr_size[inter][plane][qri];
622 bmi= s->qr_base[inter][plane][qri ];
623 bmj= s->qr_base[inter][plane][qri+1];
625 int coeff= ( 2*(sum -s->quality_index)*s->base_matrix[bmi][i]
626 - 2*(qistart-s->quality_index)*s->base_matrix[bmj][i]
627 + s->qr_size[inter][plane][qri])
628 / (2*s->qr_size[inter][plane][qri]);
630 int qmin= 8<<(inter + !i);
631 int qscale= i ? ac_scale_factor : dc_scale_factor;
633 s->qmat[inter][plane][i]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
638 memset(s->qscale_table, (FFMAX(s->qmat[0][0][1], s->qmat[0][1][1])+8)/16, 512); //FIXME finetune
642 * This function initializes the loop filter boundary limits if the frame's
643 * quality index is different from the previous frame's.
645 static void init_loop_filter(Vp3DecodeContext *s)
647 int *bounding_values= s->bounding_values_array+127;
651 filter_limit = s->filter_limit_values[s->quality_index];
653 /* set up the bounding values */
654 memset(s->bounding_values_array, 0, 256 * sizeof(int));
655 for (x = 0; x < filter_limit; x++) {
656 bounding_values[-x - filter_limit] = -filter_limit + x;
657 bounding_values[-x] = -x;
658 bounding_values[x] = x;
659 bounding_values[x + filter_limit] = filter_limit - x;
664 * This function unpacks all of the superblock/macroblock/fragment coding
665 * information from the bitstream.
667 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
670 int current_superblock = 0;
672 int decode_fully_flags = 0;
673 int decode_partial_blocks = 0;
674 int first_c_fragment_seen;
677 int current_fragment;
679 debug_vp3(" vp3: unpacking superblock coding\n");
683 debug_vp3(" keyframe-- all superblocks are fully coded\n");
684 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
688 /* unpack the list of partially-coded superblocks */
690 /* toggle the bit because as soon as the first run length is
691 * fetched the bit will be toggled again */
693 while (current_superblock < s->superblock_count) {
694 if (current_run-- == 0) {
696 current_run = get_vlc2(gb,
697 s->superblock_run_length_vlc.table, 6, 2);
698 if (current_run == 33)
699 current_run += get_bits(gb, 12);
700 debug_block_coding(" setting superblocks %d..%d to %s\n",
702 current_superblock + current_run - 1,
703 (bit) ? "partially coded" : "not coded");
705 /* if any of the superblocks are not partially coded, flag
706 * a boolean to decode the list of fully-coded superblocks */
708 decode_fully_flags = 1;
711 /* make a note of the fact that there are partially coded
713 decode_partial_blocks = 1;
716 s->superblock_coding[current_superblock++] = bit;
719 /* unpack the list of fully coded superblocks if any of the blocks were
720 * not marked as partially coded in the previous step */
721 if (decode_fully_flags) {
723 current_superblock = 0;
726 /* toggle the bit because as soon as the first run length is
727 * fetched the bit will be toggled again */
729 while (current_superblock < s->superblock_count) {
731 /* skip any superblocks already marked as partially coded */
732 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
734 if (current_run-- == 0) {
736 current_run = get_vlc2(gb,
737 s->superblock_run_length_vlc.table, 6, 2);
738 if (current_run == 33)
739 current_run += get_bits(gb, 12);
742 debug_block_coding(" setting superblock %d to %s\n",
744 (bit) ? "fully coded" : "not coded");
745 s->superblock_coding[current_superblock] = 2*bit;
747 current_superblock++;
751 /* if there were partial blocks, initialize bitstream for
752 * unpacking fragment codings */
753 if (decode_partial_blocks) {
757 /* toggle the bit because as soon as the first run length is
758 * fetched the bit will be toggled again */
763 /* figure out which fragments are coded; iterate through each
764 * superblock (all planes) */
765 s->coded_fragment_list_index = 0;
766 s->next_coeff= s->coeffs + s->fragment_count;
767 s->first_coded_y_fragment = s->first_coded_c_fragment = 0;
768 s->last_coded_y_fragment = s->last_coded_c_fragment = -1;
769 first_c_fragment_seen = 0;
770 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
771 for (i = 0; i < s->superblock_count; i++) {
773 /* iterate through all 16 fragments in a superblock */
774 for (j = 0; j < 16; j++) {
776 /* if the fragment is in bounds, check its coding status */
777 current_fragment = s->superblock_fragments[i * 16 + j];
778 if (current_fragment >= s->fragment_count) {
779 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
780 current_fragment, s->fragment_count);
783 if (current_fragment != -1) {
784 if (s->superblock_coding[i] == SB_NOT_CODED) {
786 /* copy all the fragments from the prior frame */
787 s->all_fragments[current_fragment].coding_method =
790 } else if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
792 /* fragment may or may not be coded; this is the case
793 * that cares about the fragment coding runs */
794 if (current_run-- == 0) {
796 current_run = get_vlc2(gb,
797 s->fragment_run_length_vlc.table, 5, 2);
801 /* default mode; actual mode will be decoded in
803 s->all_fragments[current_fragment].coding_method =
805 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
806 s->coded_fragment_list[s->coded_fragment_list_index] =
808 if ((current_fragment >= s->fragment_start[1]) &&
809 (s->last_coded_y_fragment == -1) &&
810 (!first_c_fragment_seen)) {
811 s->first_coded_c_fragment = s->coded_fragment_list_index;
812 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
813 first_c_fragment_seen = 1;
815 s->coded_fragment_list_index++;
816 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
817 debug_block_coding(" superblock %d is partially coded, fragment %d is coded\n",
818 i, current_fragment);
820 /* not coded; copy this fragment from the prior frame */
821 s->all_fragments[current_fragment].coding_method =
823 debug_block_coding(" superblock %d is partially coded, fragment %d is not coded\n",
824 i, current_fragment);
829 /* fragments are fully coded in this superblock; actual
830 * coding will be determined in next step */
831 s->all_fragments[current_fragment].coding_method =
833 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
834 s->coded_fragment_list[s->coded_fragment_list_index] =
836 if ((current_fragment >= s->fragment_start[1]) &&
837 (s->last_coded_y_fragment == -1) &&
838 (!first_c_fragment_seen)) {
839 s->first_coded_c_fragment = s->coded_fragment_list_index;
840 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
841 first_c_fragment_seen = 1;
843 s->coded_fragment_list_index++;
844 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
845 debug_block_coding(" superblock %d is fully coded, fragment %d is coded\n",
846 i, current_fragment);
852 if (!first_c_fragment_seen)
853 /* only Y fragments coded in this frame */
854 s->last_coded_y_fragment = s->coded_fragment_list_index - 1;
856 /* end the list of coded C fragments */
857 s->last_coded_c_fragment = s->coded_fragment_list_index - 1;
859 debug_block_coding(" %d total coded fragments, y: %d -> %d, c: %d -> %d\n",
860 s->coded_fragment_list_index,
861 s->first_coded_y_fragment,
862 s->last_coded_y_fragment,
863 s->first_coded_c_fragment,
864 s->last_coded_c_fragment);
870 * This function unpacks all the coding mode data for individual macroblocks
871 * from the bitstream.
873 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
877 int current_macroblock;
878 int current_fragment;
881 debug_vp3(" vp3: unpacking encoding modes\n");
884 debug_vp3(" keyframe-- all blocks are coded as INTRA\n");
886 for (i = 0; i < s->fragment_count; i++)
887 s->all_fragments[i].coding_method = MODE_INTRA;
891 /* fetch the mode coding scheme for this frame */
892 scheme = get_bits(gb, 3);
893 debug_modes(" using mode alphabet %d\n", scheme);
895 /* is it a custom coding scheme? */
897 debug_modes(" custom mode alphabet ahead:\n");
898 for (i = 0; i < 8; i++)
899 ModeAlphabet[scheme][get_bits(gb, 3)] = i;
902 for (i = 0; i < 8; i++)
903 debug_modes(" mode[%d][%d] = %d\n", scheme, i,
904 ModeAlphabet[scheme][i]);
906 /* iterate through all of the macroblocks that contain 1 or more
908 for (i = 0; i < s->u_superblock_start; i++) {
910 for (j = 0; j < 4; j++) {
911 current_macroblock = s->superblock_macroblocks[i * 4 + j];
912 if ((current_macroblock == -1) ||
913 (s->macroblock_coding[current_macroblock] == MODE_COPY))
915 if (current_macroblock >= s->macroblock_count) {
916 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad macroblock number (%d >= %d)\n",
917 current_macroblock, s->macroblock_count);
921 /* mode 7 means get 3 bits for each coding mode */
923 coding_mode = get_bits(gb, 3);
925 coding_mode = ModeAlphabet[scheme]
926 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
928 s->macroblock_coding[current_macroblock] = coding_mode;
929 for (k = 0; k < 6; k++) {
931 s->macroblock_fragments[current_macroblock * 6 + k];
932 if (current_fragment == -1)
934 if (current_fragment >= s->fragment_count) {
935 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad fragment number (%d >= %d)\n",
936 current_fragment, s->fragment_count);
939 if (s->all_fragments[current_fragment].coding_method !=
941 s->all_fragments[current_fragment].coding_method =
945 debug_modes(" coding method for macroblock starting @ fragment %d = %d\n",
946 s->macroblock_fragments[current_macroblock * 6], coding_mode);
955 * This function unpacks all the motion vectors for the individual
956 * macroblocks from the bitstream.
958 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
964 int last_motion_x = 0;
965 int last_motion_y = 0;
966 int prior_last_motion_x = 0;
967 int prior_last_motion_y = 0;
968 int current_macroblock;
969 int current_fragment;
971 debug_vp3(" vp3: unpacking motion vectors\n");
974 debug_vp3(" keyframe-- there are no motion vectors\n");
978 memset(motion_x, 0, 6 * sizeof(int));
979 memset(motion_y, 0, 6 * sizeof(int));
981 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
982 coding_mode = get_bits1(gb);
983 debug_vectors(" using %s scheme for unpacking motion vectors\n",
984 (coding_mode == 0) ? "VLC" : "fixed-length");
986 /* iterate through all of the macroblocks that contain 1 or more
988 for (i = 0; i < s->u_superblock_start; i++) {
990 for (j = 0; j < 4; j++) {
991 current_macroblock = s->superblock_macroblocks[i * 4 + j];
992 if ((current_macroblock == -1) ||
993 (s->macroblock_coding[current_macroblock] == MODE_COPY))
995 if (current_macroblock >= s->macroblock_count) {
996 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad macroblock number (%d >= %d)\n",
997 current_macroblock, s->macroblock_count);
1001 current_fragment = s->macroblock_fragments[current_macroblock * 6];
1002 if (current_fragment >= s->fragment_count) {
1003 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d\n",
1004 current_fragment, s->fragment_count);
1007 switch (s->macroblock_coding[current_macroblock]) {
1009 case MODE_INTER_PLUS_MV:
1010 case MODE_GOLDEN_MV:
1011 /* all 6 fragments use the same motion vector */
1012 if (coding_mode == 0) {
1013 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1014 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1016 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
1017 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
1020 for (k = 1; k < 6; k++) {
1021 motion_x[k] = motion_x[0];
1022 motion_y[k] = motion_y[0];
1025 /* vector maintenance, only on MODE_INTER_PLUS_MV */
1026 if (s->macroblock_coding[current_macroblock] ==
1027 MODE_INTER_PLUS_MV) {
1028 prior_last_motion_x = last_motion_x;
1029 prior_last_motion_y = last_motion_y;
1030 last_motion_x = motion_x[0];
1031 last_motion_y = motion_y[0];
1035 case MODE_INTER_FOURMV:
1036 /* fetch 4 vectors from the bitstream, one for each
1037 * Y fragment, then average for the C fragment vectors */
1038 motion_x[4] = motion_y[4] = 0;
1039 for (k = 0; k < 4; k++) {
1040 if (coding_mode == 0) {
1041 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1042 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1044 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
1045 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
1047 motion_x[4] += motion_x[k];
1048 motion_y[4] += motion_y[k];
1052 motion_x[4]= RSHIFT(motion_x[4], 2);
1054 motion_y[4]= RSHIFT(motion_y[4], 2);
1056 /* vector maintenance; vector[3] is treated as the
1057 * last vector in this case */
1058 prior_last_motion_x = last_motion_x;
1059 prior_last_motion_y = last_motion_y;
1060 last_motion_x = motion_x[3];
1061 last_motion_y = motion_y[3];
1064 case MODE_INTER_LAST_MV:
1065 /* all 6 fragments use the last motion vector */
1066 motion_x[0] = last_motion_x;
1067 motion_y[0] = last_motion_y;
1068 for (k = 1; k < 6; k++) {
1069 motion_x[k] = motion_x[0];
1070 motion_y[k] = motion_y[0];
1073 /* no vector maintenance (last vector remains the
1077 case MODE_INTER_PRIOR_LAST:
1078 /* all 6 fragments use the motion vector prior to the
1079 * last motion vector */
1080 motion_x[0] = prior_last_motion_x;
1081 motion_y[0] = prior_last_motion_y;
1082 for (k = 1; k < 6; k++) {
1083 motion_x[k] = motion_x[0];
1084 motion_y[k] = motion_y[0];
1087 /* vector maintenance */
1088 prior_last_motion_x = last_motion_x;
1089 prior_last_motion_y = last_motion_y;
1090 last_motion_x = motion_x[0];
1091 last_motion_y = motion_y[0];
1095 /* covers intra, inter without MV, golden without MV */
1096 memset(motion_x, 0, 6 * sizeof(int));
1097 memset(motion_y, 0, 6 * sizeof(int));
1099 /* no vector maintenance */
1103 /* assign the motion vectors to the correct fragments */
1104 debug_vectors(" vectors for macroblock starting @ fragment %d (coding method %d):\n",
1106 s->macroblock_coding[current_macroblock]);
1107 for (k = 0; k < 6; k++) {
1109 s->macroblock_fragments[current_macroblock * 6 + k];
1110 if (current_fragment == -1)
1112 if (current_fragment >= s->fragment_count) {
1113 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d)\n",
1114 current_fragment, s->fragment_count);
1117 s->all_fragments[current_fragment].motion_x = motion_x[k];
1118 s->all_fragments[current_fragment].motion_y = motion_y[k];
1119 debug_vectors(" vector %d: fragment %d = (%d, %d)\n",
1120 k, current_fragment, motion_x[k], motion_y[k]);
1130 * This function is called by unpack_dct_coeffs() to extract the VLCs from
1131 * the bitstream. The VLCs encode tokens which are used to unpack DCT
1132 * data. This function unpacks all the VLCs for either the Y plane or both
1133 * C planes, and is called for DC coefficients or different AC coefficient
1134 * levels (since different coefficient types require different VLC tables.
1136 * This function returns a residual eob run. E.g, if a particular token gave
1137 * instructions to EOB the next 5 fragments and there were only 2 fragments
1138 * left in the current fragment range, 3 would be returned so that it could
1139 * be passed into the next call to this same function.
1141 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
1142 VLC *table, int coeff_index,
1143 int first_fragment, int last_fragment,
1150 Vp3Fragment *fragment;
1151 uint8_t *perm= s->scantable.permutated;
1154 if ((first_fragment >= s->fragment_count) ||
1155 (last_fragment >= s->fragment_count)) {
1157 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vlcs(): bad fragment number (%d -> %d ?)\n",
1158 first_fragment, last_fragment);
1162 for (i = first_fragment; i <= last_fragment; i++) {
1164 fragment = &s->all_fragments[s->coded_fragment_list[i]];
1165 if (fragment->coeff_count > coeff_index)
1169 /* decode a VLC into a token */
1170 token = get_vlc2(gb, table->table, 5, 3);
1171 debug_vlc(" token = %2d, ", token);
1172 /* use the token to get a zero run, a coefficient, and an eob run */
1174 eob_run = eob_run_base[token];
1175 if (eob_run_get_bits[token])
1176 eob_run += get_bits(gb, eob_run_get_bits[token]);
1177 coeff = zero_run = 0;
1179 bits_to_get = coeff_get_bits[token];
1181 coeff = coeff_tables[token][0];
1183 coeff = coeff_tables[token][get_bits(gb, bits_to_get)];
1185 zero_run = zero_run_base[token];
1186 if (zero_run_get_bits[token])
1187 zero_run += get_bits(gb, zero_run_get_bits[token]);
1192 fragment->coeff_count += zero_run;
1193 if (fragment->coeff_count < 64){
1194 fragment->next_coeff->coeff= coeff;
1195 fragment->next_coeff->index= perm[fragment->coeff_count++]; //FIXME perm here already?
1196 fragment->next_coeff->next= s->next_coeff;
1197 s->next_coeff->next=NULL;
1198 fragment->next_coeff= s->next_coeff++;
1200 debug_vlc(" fragment %d coeff = %d\n",
1201 s->coded_fragment_list[i], fragment->next_coeff[coeff_index]);
1203 fragment->coeff_count |= 128;
1204 debug_vlc(" fragment %d eob with %d coefficients\n",
1205 s->coded_fragment_list[i], fragment->coeff_count&127);
1214 * This function unpacks all of the DCT coefficient data from the
1217 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1224 int residual_eob_run = 0;
1226 /* fetch the DC table indexes */
1227 dc_y_table = get_bits(gb, 4);
1228 dc_c_table = get_bits(gb, 4);
1230 /* unpack the Y plane DC coefficients */
1231 debug_vp3(" vp3: unpacking Y plane DC coefficients using table %d\n",
1233 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1234 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1236 /* unpack the C plane DC coefficients */
1237 debug_vp3(" vp3: unpacking C plane DC coefficients using table %d\n",
1239 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1240 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1242 /* fetch the AC table indexes */
1243 ac_y_table = get_bits(gb, 4);
1244 ac_c_table = get_bits(gb, 4);
1246 /* unpack the group 1 AC coefficients (coeffs 1-5) */
1247 for (i = 1; i <= 5; i++) {
1249 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1251 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_y_table], i,
1252 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1254 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1256 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_c_table], i,
1257 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1260 /* unpack the group 2 AC coefficients (coeffs 6-14) */
1261 for (i = 6; i <= 14; i++) {
1263 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1265 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_y_table], i,
1266 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1268 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1270 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_c_table], i,
1271 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1274 /* unpack the group 3 AC coefficients (coeffs 15-27) */
1275 for (i = 15; i <= 27; i++) {
1277 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1279 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_y_table], i,
1280 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1282 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1284 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_c_table], i,
1285 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1288 /* unpack the group 4 AC coefficients (coeffs 28-63) */
1289 for (i = 28; i <= 63; i++) {
1291 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1293 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_y_table], i,
1294 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1296 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1298 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_c_table], i,
1299 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1306 * This function reverses the DC prediction for each coded fragment in
1307 * the frame. Much of this function is adapted directly from the original
1310 #define COMPATIBLE_FRAME(x) \
1311 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1312 #define FRAME_CODED(x) (s->all_fragments[x].coding_method != MODE_COPY)
1313 #define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
1315 static void reverse_dc_prediction(Vp3DecodeContext *s,
1318 int fragment_height)
1327 int i = first_fragment;
1331 /* DC values for the left, up-left, up, and up-right fragments */
1332 int vl, vul, vu, vur;
1334 /* indexes for the left, up-left, up, and up-right fragments */
1338 * The 6 fields mean:
1339 * 0: up-left multiplier
1341 * 2: up-right multiplier
1342 * 3: left multiplier
1344 int predictor_transform[16][4] = {
1346 { 0, 0, 0,128}, // PL
1347 { 0, 0,128, 0}, // PUR
1348 { 0, 0, 53, 75}, // PUR|PL
1349 { 0,128, 0, 0}, // PU
1350 { 0, 64, 0, 64}, // PU|PL
1351 { 0,128, 0, 0}, // PU|PUR
1352 { 0, 0, 53, 75}, // PU|PUR|PL
1353 {128, 0, 0, 0}, // PUL
1354 { 0, 0, 0,128}, // PUL|PL
1355 { 64, 0, 64, 0}, // PUL|PUR
1356 { 0, 0, 53, 75}, // PUL|PUR|PL
1357 { 0,128, 0, 0}, // PUL|PU
1358 {-104,116, 0,116}, // PUL|PU|PL
1359 { 24, 80, 24, 0}, // PUL|PU|PUR
1360 {-104,116, 0,116} // PUL|PU|PUR|PL
1363 /* This table shows which types of blocks can use other blocks for
1364 * prediction. For example, INTRA is the only mode in this table to
1365 * have a frame number of 0. That means INTRA blocks can only predict
1366 * from other INTRA blocks. There are 2 golden frame coding types;
1367 * blocks encoding in these modes can only predict from other blocks
1368 * that were encoded with these 1 of these 2 modes. */
1369 unsigned char compatible_frame[8] = {
1370 1, /* MODE_INTER_NO_MV */
1372 1, /* MODE_INTER_PLUS_MV */
1373 1, /* MODE_INTER_LAST_MV */
1374 1, /* MODE_INTER_PRIOR_MV */
1375 2, /* MODE_USING_GOLDEN */
1376 2, /* MODE_GOLDEN_MV */
1377 1 /* MODE_INTER_FOUR_MV */
1379 int current_frame_type;
1381 /* there is a last DC predictor for each of the 3 frame types */
1386 debug_vp3(" vp3: reversing DC prediction\n");
1388 vul = vu = vur = vl = 0;
1389 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1391 /* for each fragment row... */
1392 for (y = 0; y < fragment_height; y++) {
1394 /* for each fragment in a row... */
1395 for (x = 0; x < fragment_width; x++, i++) {
1397 /* reverse prediction if this block was coded */
1398 if (s->all_fragments[i].coding_method != MODE_COPY) {
1400 current_frame_type =
1401 compatible_frame[s->all_fragments[i].coding_method];
1402 debug_dc_pred(" frag %d: orig DC = %d, ",
1409 if(FRAME_CODED(l) && COMPATIBLE_FRAME(l))
1413 u= i-fragment_width;
1415 if(FRAME_CODED(u) && COMPATIBLE_FRAME(u))
1418 ul= i-fragment_width-1;
1420 if(FRAME_CODED(ul) && COMPATIBLE_FRAME(ul))
1423 if(x + 1 < fragment_width){
1424 ur= i-fragment_width+1;
1426 if(FRAME_CODED(ur) && COMPATIBLE_FRAME(ur))
1431 debug_dc_pred("transform = %d, ", transform);
1433 if (transform == 0) {
1435 /* if there were no fragments to predict from, use last
1437 predicted_dc = last_dc[current_frame_type];
1438 debug_dc_pred("from last DC (%d) = %d\n",
1439 current_frame_type, DC_COEFF(i));
1443 /* apply the appropriate predictor transform */
1445 (predictor_transform[transform][0] * vul) +
1446 (predictor_transform[transform][1] * vu) +
1447 (predictor_transform[transform][2] * vur) +
1448 (predictor_transform[transform][3] * vl);
1450 predicted_dc /= 128;
1452 /* check for outranging on the [ul u l] and
1453 * [ul u ur l] predictors */
1454 if ((transform == 13) || (transform == 15)) {
1455 if (FFABS(predicted_dc - vu) > 128)
1457 else if (FFABS(predicted_dc - vl) > 128)
1459 else if (FFABS(predicted_dc - vul) > 128)
1463 debug_dc_pred("from pred DC = %d\n",
1467 /* at long last, apply the predictor */
1468 if(s->coeffs[i].index){
1469 *s->next_coeff= s->coeffs[i];
1470 s->coeffs[i].index=0;
1471 s->coeffs[i].coeff=0;
1472 s->coeffs[i].next= s->next_coeff++;
1474 s->coeffs[i].coeff += predicted_dc;
1476 last_dc[current_frame_type] = DC_COEFF(i);
1477 if(DC_COEFF(i) && !(s->all_fragments[i].coeff_count&127)){
1478 s->all_fragments[i].coeff_count= 129;
1479 // s->all_fragments[i].next_coeff= s->next_coeff;
1480 s->coeffs[i].next= s->next_coeff;
1481 (s->next_coeff++)->next=NULL;
1489 static void horizontal_filter(unsigned char *first_pixel, int stride,
1490 int *bounding_values);
1491 static void vertical_filter(unsigned char *first_pixel, int stride,
1492 int *bounding_values);
1495 * Perform the final rendering for a particular slice of data.
1496 * The slice number ranges from 0..(macroblock_height - 1).
1498 static void render_slice(Vp3DecodeContext *s, int slice)
1502 int16_t *dequantizer;
1503 DECLARE_ALIGNED_16(DCTELEM, block[64]);
1504 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1505 int motion_halfpel_index;
1506 uint8_t *motion_source;
1508 int current_macroblock_entry = slice * s->macroblock_width * 6;
1510 if (slice >= s->macroblock_height)
1513 for (plane = 0; plane < 3; plane++) {
1514 uint8_t *output_plane = s->current_frame.data [plane];
1515 uint8_t * last_plane = s-> last_frame.data [plane];
1516 uint8_t *golden_plane = s-> golden_frame.data [plane];
1517 int stride = s->current_frame.linesize[plane];
1518 int plane_width = s->width >> !!plane;
1519 int plane_height = s->height >> !!plane;
1520 int y = slice * FRAGMENT_PIXELS << !plane ;
1521 int slice_height = y + (FRAGMENT_PIXELS << !plane);
1522 int i = s->macroblock_fragments[current_macroblock_entry + plane + 3*!!plane];
1524 if (!s->flipped_image) stride = -stride;
1527 if(FFABS(stride) > 2048)
1528 return; //various tables are fixed size
1530 /* for each fragment row in the slice (both of them)... */
1531 for (; y < slice_height; y += 8) {
1533 /* for each fragment in a row... */
1534 for (x = 0; x < plane_width; x += 8, i++) {
1536 if ((i < 0) || (i >= s->fragment_count)) {
1537 av_log(s->avctx, AV_LOG_ERROR, " vp3:render_slice(): bad fragment number (%d)\n", i);
1541 /* transform if this block was coded */
1542 if ((s->all_fragments[i].coding_method != MODE_COPY) &&
1543 !((s->avctx->flags & CODEC_FLAG_GRAY) && plane)) {
1545 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1546 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1547 motion_source= golden_plane;
1549 motion_source= last_plane;
1551 motion_source += s->all_fragments[i].first_pixel;
1552 motion_halfpel_index = 0;
1554 /* sort out the motion vector if this fragment is coded
1555 * using a motion vector method */
1556 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1557 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1559 motion_x = s->all_fragments[i].motion_x;
1560 motion_y = s->all_fragments[i].motion_y;
1562 motion_x= (motion_x>>1) | (motion_x&1);
1563 motion_y= (motion_y>>1) | (motion_y&1);
1566 src_x= (motion_x>>1) + x;
1567 src_y= (motion_y>>1) + y;
1568 if ((motion_x == 127) || (motion_y == 127))
1569 av_log(s->avctx, AV_LOG_ERROR, " help! got invalid motion vector! (%X, %X)\n", motion_x, motion_y);
1571 motion_halfpel_index = motion_x & 0x01;
1572 motion_source += (motion_x >> 1);
1574 motion_halfpel_index |= (motion_y & 0x01) << 1;
1575 motion_source += ((motion_y >> 1) * stride);
1577 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1578 uint8_t *temp= s->edge_emu_buffer;
1579 if(stride<0) temp -= 9*stride;
1580 else temp += 9*stride;
1582 ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1583 motion_source= temp;
1588 /* first, take care of copying a block from either the
1589 * previous or the golden frame */
1590 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1591 /* Note, it is possible to implement all MC cases with
1592 put_no_rnd_pixels_l2 which would look more like the
1593 VP3 source but this would be slower as
1594 put_no_rnd_pixels_tab is better optimzed */
1595 if(motion_halfpel_index != 3){
1596 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1597 output_plane + s->all_fragments[i].first_pixel,
1598 motion_source, stride, 8);
1600 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1601 s->dsp.put_no_rnd_pixels_l2[1](
1602 output_plane + s->all_fragments[i].first_pixel,
1604 motion_source + stride + 1 + d,
1607 dequantizer = s->qmat[1][plane];
1609 dequantizer = s->qmat[0][plane];
1612 /* dequantize the DCT coefficients */
1613 debug_idct("fragment %d, coding mode %d, DC = %d, dequant = %d:\n",
1614 i, s->all_fragments[i].coding_method,
1615 DC_COEFF(i), dequantizer[0]);
1617 if(s->avctx->idct_algo==FF_IDCT_VP3){
1618 Coeff *coeff= s->coeffs + i;
1619 memset(block, 0, sizeof(block));
1621 block[coeff->index]= coeff->coeff * dequantizer[coeff->index];
1625 Coeff *coeff= s->coeffs + i;
1626 memset(block, 0, sizeof(block));
1628 block[coeff->index]= (coeff->coeff * dequantizer[coeff->index] + 2)>>2;
1633 /* invert DCT and place (or add) in final output */
1635 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1636 if(s->avctx->idct_algo!=FF_IDCT_VP3)
1639 output_plane + s->all_fragments[i].first_pixel,
1644 output_plane + s->all_fragments[i].first_pixel,
1649 debug_idct("block after idct_%s():\n",
1650 (s->all_fragments[i].coding_method == MODE_INTRA)?
1652 for (m = 0; m < 8; m++) {
1653 for (n = 0; n < 8; n++) {
1654 debug_idct(" %3d", *(output_plane +
1655 s->all_fragments[i].first_pixel + (m * stride + n)));
1663 /* copy directly from the previous frame */
1664 s->dsp.put_pixels_tab[1][0](
1665 output_plane + s->all_fragments[i].first_pixel,
1666 last_plane + s->all_fragments[i].first_pixel,
1671 /* perform the left edge filter if:
1672 * - the fragment is not on the left column
1673 * - the fragment is coded in this frame
1674 * - the fragment is not coded in this frame but the left
1675 * fragment is coded in this frame (this is done instead
1676 * of a right edge filter when rendering the left fragment
1677 * since this fragment is not available yet) */
1679 ((s->all_fragments[i].coding_method != MODE_COPY) ||
1680 ((s->all_fragments[i].coding_method == MODE_COPY) &&
1681 (s->all_fragments[i - 1].coding_method != MODE_COPY)) )) {
1683 output_plane + s->all_fragments[i].first_pixel + 7*stride,
1684 -stride, s->bounding_values_array + 127);
1687 /* perform the top edge filter if:
1688 * - the fragment is not on the top row
1689 * - the fragment is coded in this frame
1690 * - the fragment is not coded in this frame but the above
1691 * fragment is coded in this frame (this is done instead
1692 * of a bottom edge filter when rendering the above
1693 * fragment since this fragment is not available yet) */
1695 ((s->all_fragments[i].coding_method != MODE_COPY) ||
1696 ((s->all_fragments[i].coding_method == MODE_COPY) &&
1697 (s->all_fragments[i - fragment_width].coding_method != MODE_COPY)) )) {
1699 output_plane + s->all_fragments[i].first_pixel - stride,
1700 -stride, s->bounding_values_array + 127);
1707 /* this looks like a good place for slice dispatch... */
1709 * if (slice == s->macroblock_height - 1)
1710 * dispatch (both last slice & 2nd-to-last slice);
1711 * else if (slice > 0)
1712 * dispatch (slice - 1);
1718 static void horizontal_filter(unsigned char *first_pixel, int stride,
1719 int *bounding_values)
1724 for (end= first_pixel + 8*stride; first_pixel != end; first_pixel += stride) {
1726 (first_pixel[-2] - first_pixel[ 1])
1727 +3*(first_pixel[ 0] - first_pixel[-1]);
1728 filter_value = bounding_values[(filter_value + 4) >> 3];
1729 first_pixel[-1] = av_clip_uint8(first_pixel[-1] + filter_value);
1730 first_pixel[ 0] = av_clip_uint8(first_pixel[ 0] - filter_value);
1734 static void vertical_filter(unsigned char *first_pixel, int stride,
1735 int *bounding_values)
1739 const int nstride= -stride;
1741 for (end= first_pixel + 8; first_pixel < end; first_pixel++) {
1743 (first_pixel[2 * nstride] - first_pixel[ stride])
1744 +3*(first_pixel[0 ] - first_pixel[nstride]);
1745 filter_value = bounding_values[(filter_value + 4) >> 3];
1746 first_pixel[nstride] = av_clip_uint8(first_pixel[nstride] + filter_value);
1747 first_pixel[0] = av_clip_uint8(first_pixel[0] - filter_value);
1751 static void apply_loop_filter(Vp3DecodeContext *s)
1755 int *bounding_values= s->bounding_values_array+127;
1758 int bounding_values_array[256];
1761 /* find the right loop limit value */
1762 for (x = 63; x >= 0; x--) {
1763 if (vp31_ac_scale_factor[x] >= s->quality_index)
1766 filter_limit = vp31_filter_limit_values[s->quality_index];
1768 /* set up the bounding values */
1769 memset(bounding_values_array, 0, 256 * sizeof(int));
1770 for (x = 0; x < filter_limit; x++) {
1771 bounding_values[-x - filter_limit] = -filter_limit + x;
1772 bounding_values[-x] = -x;
1773 bounding_values[x] = x;
1774 bounding_values[x + filter_limit] = filter_limit - x;
1778 for (plane = 0; plane < 3; plane++) {
1779 int width = s->fragment_width >> !!plane;
1780 int height = s->fragment_height >> !!plane;
1781 int fragment = s->fragment_start [plane];
1782 int stride = s->current_frame.linesize[plane];
1783 uint8_t *plane_data = s->current_frame.data [plane];
1784 if (!s->flipped_image) stride = -stride;
1786 for (y = 0; y < height; y++) {
1788 for (x = 0; x < width; x++) {
1790 /* do not perform left edge filter for left columns frags */
1792 (s->all_fragments[fragment].coding_method != MODE_COPY)) {
1794 plane_data + s->all_fragments[fragment].first_pixel,
1795 stride, bounding_values);
1798 /* do not perform top edge filter for top row fragments */
1800 (s->all_fragments[fragment].coding_method != MODE_COPY)) {
1802 plane_data + s->all_fragments[fragment].first_pixel,
1803 stride, bounding_values);
1806 /* do not perform right edge filter for right column
1807 * fragments or if right fragment neighbor is also coded
1808 * in this frame (it will be filtered in next iteration) */
1809 if ((x < width - 1) &&
1810 (s->all_fragments[fragment].coding_method != MODE_COPY) &&
1811 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1813 plane_data + s->all_fragments[fragment + 1].first_pixel,
1814 stride, bounding_values);
1817 /* do not perform bottom edge filter for bottom row
1818 * fragments or if bottom fragment neighbor is also coded
1819 * in this frame (it will be filtered in the next row) */
1820 if ((y < height - 1) &&
1821 (s->all_fragments[fragment].coding_method != MODE_COPY) &&
1822 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1824 plane_data + s->all_fragments[fragment + width].first_pixel,
1825 stride, bounding_values);
1829 STOP_TIMER("loop filter")
1836 * This function computes the first pixel addresses for each fragment.
1837 * This function needs to be invoked after the first frame is allocated
1838 * so that it has access to the plane strides.
1840 static void vp3_calculate_pixel_addresses(Vp3DecodeContext *s)
1845 /* figure out the first pixel addresses for each of the fragments */
1848 for (y = s->fragment_height; y > 0; y--) {
1849 for (x = 0; x < s->fragment_width; x++) {
1850 s->all_fragments[i++].first_pixel =
1851 s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
1852 s->golden_frame.linesize[0] +
1853 x * FRAGMENT_PIXELS;
1854 debug_init(" fragment %d, first pixel @ %d\n",
1855 i-1, s->all_fragments[i-1].first_pixel);
1860 i = s->fragment_start[1];
1861 for (y = s->fragment_height / 2; y > 0; y--) {
1862 for (x = 0; x < s->fragment_width / 2; x++) {
1863 s->all_fragments[i++].first_pixel =
1864 s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
1865 s->golden_frame.linesize[1] +
1866 x * FRAGMENT_PIXELS;
1867 debug_init(" fragment %d, first pixel @ %d\n",
1868 i-1, s->all_fragments[i-1].first_pixel);
1873 i = s->fragment_start[2];
1874 for (y = s->fragment_height / 2; y > 0; y--) {
1875 for (x = 0; x < s->fragment_width / 2; x++) {
1876 s->all_fragments[i++].first_pixel =
1877 s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
1878 s->golden_frame.linesize[2] +
1879 x * FRAGMENT_PIXELS;
1880 debug_init(" fragment %d, first pixel @ %d\n",
1881 i-1, s->all_fragments[i-1].first_pixel);
1886 /* FIXME: this should be merged with the above! */
1887 static void theora_calculate_pixel_addresses(Vp3DecodeContext *s)
1892 /* figure out the first pixel addresses for each of the fragments */
1895 for (y = 1; y <= s->fragment_height; y++) {
1896 for (x = 0; x < s->fragment_width; x++) {
1897 s->all_fragments[i++].first_pixel =
1898 s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
1899 s->golden_frame.linesize[0] +
1900 x * FRAGMENT_PIXELS;
1901 debug_init(" fragment %d, first pixel @ %d\n",
1902 i-1, s->all_fragments[i-1].first_pixel);
1907 i = s->fragment_start[1];
1908 for (y = 1; y <= s->fragment_height / 2; y++) {
1909 for (x = 0; x < s->fragment_width / 2; x++) {
1910 s->all_fragments[i++].first_pixel =
1911 s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
1912 s->golden_frame.linesize[1] +
1913 x * FRAGMENT_PIXELS;
1914 debug_init(" fragment %d, first pixel @ %d\n",
1915 i-1, s->all_fragments[i-1].first_pixel);
1920 i = s->fragment_start[2];
1921 for (y = 1; y <= s->fragment_height / 2; y++) {
1922 for (x = 0; x < s->fragment_width / 2; x++) {
1923 s->all_fragments[i++].first_pixel =
1924 s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
1925 s->golden_frame.linesize[2] +
1926 x * FRAGMENT_PIXELS;
1927 debug_init(" fragment %d, first pixel @ %d\n",
1928 i-1, s->all_fragments[i-1].first_pixel);
1934 * This is the ffmpeg/libavcodec API init function.
1936 static av_cold int vp3_decode_init(AVCodecContext *avctx)
1938 Vp3DecodeContext *s = avctx->priv_data;
1939 int i, inter, plane;
1942 int y_superblock_count;
1943 int c_superblock_count;
1945 if (avctx->codec_tag == MKTAG('V','P','3','0'))
1951 s->width = (avctx->width + 15) & 0xFFFFFFF0;
1952 s->height = (avctx->height + 15) & 0xFFFFFFF0;
1953 avctx->pix_fmt = PIX_FMT_YUV420P;
1954 if(avctx->idct_algo==FF_IDCT_AUTO)
1955 avctx->idct_algo=FF_IDCT_VP3;
1956 dsputil_init(&s->dsp, avctx);
1958 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1960 /* initialize to an impossible value which will force a recalculation
1961 * in the first frame decode */
1962 s->quality_index = -1;
1964 s->y_superblock_width = (s->width + 31) / 32;
1965 s->y_superblock_height = (s->height + 31) / 32;
1966 y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1968 /* work out the dimensions for the C planes */
1969 c_width = s->width / 2;
1970 c_height = s->height / 2;
1971 s->c_superblock_width = (c_width + 31) / 32;
1972 s->c_superblock_height = (c_height + 31) / 32;
1973 c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1975 s->superblock_count = y_superblock_count + (c_superblock_count * 2);
1976 s->u_superblock_start = y_superblock_count;
1977 s->v_superblock_start = s->u_superblock_start + c_superblock_count;
1978 s->superblock_coding = av_malloc(s->superblock_count);
1980 s->macroblock_width = (s->width + 15) / 16;
1981 s->macroblock_height = (s->height + 15) / 16;
1982 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1984 s->fragment_width = s->width / FRAGMENT_PIXELS;
1985 s->fragment_height = s->height / FRAGMENT_PIXELS;
1987 /* fragment count covers all 8x8 blocks for all 3 planes */
1988 s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
1989 s->fragment_start[1] = s->fragment_width * s->fragment_height;
1990 s->fragment_start[2] = s->fragment_width * s->fragment_height * 5 / 4;
1992 debug_init(" Y plane: %d x %d\n", s->width, s->height);
1993 debug_init(" C plane: %d x %d\n", c_width, c_height);
1994 debug_init(" Y superblocks: %d x %d, %d total\n",
1995 s->y_superblock_width, s->y_superblock_height, y_superblock_count);
1996 debug_init(" C superblocks: %d x %d, %d total\n",
1997 s->c_superblock_width, s->c_superblock_height, c_superblock_count);
1998 debug_init(" total superblocks = %d, U starts @ %d, V starts @ %d\n",
1999 s->superblock_count, s->u_superblock_start, s->v_superblock_start);
2000 debug_init(" macroblocks: %d x %d, %d total\n",
2001 s->macroblock_width, s->macroblock_height, s->macroblock_count);
2002 debug_init(" %d fragments, %d x %d, u starts @ %d, v starts @ %d\n",
2006 s->fragment_start[1],
2007 s->fragment_start[2]);
2009 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
2010 s->coeffs = av_malloc(s->fragment_count * sizeof(Coeff) * 65);
2011 s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int));
2012 s->pixel_addresses_initialized = 0;
2014 if (!s->theora_tables)
2016 for (i = 0; i < 64; i++) {
2017 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
2018 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
2019 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
2020 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
2021 s->base_matrix[2][i] = vp31_inter_dequant[i];
2022 s->filter_limit_values[i] = vp31_filter_limit_values[i];
2025 for(inter=0; inter<2; inter++){
2026 for(plane=0; plane<3; plane++){
2027 s->qr_count[inter][plane]= 1;
2028 s->qr_size [inter][plane][0]= 63;
2029 s->qr_base [inter][plane][0]=
2030 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
2034 /* init VLC tables */
2035 for (i = 0; i < 16; i++) {
2038 init_vlc(&s->dc_vlc[i], 5, 32,
2039 &dc_bias[i][0][1], 4, 2,
2040 &dc_bias[i][0][0], 4, 2, 0);
2042 /* group 1 AC histograms */
2043 init_vlc(&s->ac_vlc_1[i], 5, 32,
2044 &ac_bias_0[i][0][1], 4, 2,
2045 &ac_bias_0[i][0][0], 4, 2, 0);
2047 /* group 2 AC histograms */
2048 init_vlc(&s->ac_vlc_2[i], 5, 32,
2049 &ac_bias_1[i][0][1], 4, 2,
2050 &ac_bias_1[i][0][0], 4, 2, 0);
2052 /* group 3 AC histograms */
2053 init_vlc(&s->ac_vlc_3[i], 5, 32,
2054 &ac_bias_2[i][0][1], 4, 2,
2055 &ac_bias_2[i][0][0], 4, 2, 0);
2057 /* group 4 AC histograms */
2058 init_vlc(&s->ac_vlc_4[i], 5, 32,
2059 &ac_bias_3[i][0][1], 4, 2,
2060 &ac_bias_3[i][0][0], 4, 2, 0);
2063 for (i = 0; i < 16; i++) {
2066 init_vlc(&s->dc_vlc[i], 5, 32,
2067 &s->huffman_table[i][0][1], 4, 2,
2068 &s->huffman_table[i][0][0], 4, 2, 0);
2070 /* group 1 AC histograms */
2071 init_vlc(&s->ac_vlc_1[i], 5, 32,
2072 &s->huffman_table[i+16][0][1], 4, 2,
2073 &s->huffman_table[i+16][0][0], 4, 2, 0);
2075 /* group 2 AC histograms */
2076 init_vlc(&s->ac_vlc_2[i], 5, 32,
2077 &s->huffman_table[i+16*2][0][1], 4, 2,
2078 &s->huffman_table[i+16*2][0][0], 4, 2, 0);
2080 /* group 3 AC histograms */
2081 init_vlc(&s->ac_vlc_3[i], 5, 32,
2082 &s->huffman_table[i+16*3][0][1], 4, 2,
2083 &s->huffman_table[i+16*3][0][0], 4, 2, 0);
2085 /* group 4 AC histograms */
2086 init_vlc(&s->ac_vlc_4[i], 5, 32,
2087 &s->huffman_table[i+16*4][0][1], 4, 2,
2088 &s->huffman_table[i+16*4][0][0], 4, 2, 0);
2092 init_vlc(&s->superblock_run_length_vlc, 6, 34,
2093 &superblock_run_length_vlc_table[0][1], 4, 2,
2094 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
2096 init_vlc(&s->fragment_run_length_vlc, 5, 30,
2097 &fragment_run_length_vlc_table[0][1], 4, 2,
2098 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
2100 init_vlc(&s->mode_code_vlc, 3, 8,
2101 &mode_code_vlc_table[0][1], 2, 1,
2102 &mode_code_vlc_table[0][0], 2, 1, 0);
2104 init_vlc(&s->motion_vector_vlc, 6, 63,
2105 &motion_vector_vlc_table[0][1], 2, 1,
2106 &motion_vector_vlc_table[0][0], 2, 1, 0);
2108 /* work out the block mapping tables */
2109 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
2110 s->superblock_macroblocks = av_malloc(s->superblock_count * 4 * sizeof(int));
2111 s->macroblock_fragments = av_malloc(s->macroblock_count * 6 * sizeof(int));
2112 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
2113 init_block_mapping(s);
2115 for (i = 0; i < 3; i++) {
2116 s->current_frame.data[i] = NULL;
2117 s->last_frame.data[i] = NULL;
2118 s->golden_frame.data[i] = NULL;
2125 * This is the ffmpeg/libavcodec API frame decode function.
2127 static int vp3_decode_frame(AVCodecContext *avctx,
2128 void *data, int *data_size,
2129 const uint8_t *buf, int buf_size)
2131 Vp3DecodeContext *s = avctx->priv_data;
2133 static int counter = 0;
2136 init_get_bits(&gb, buf, buf_size * 8);
2138 if (s->theora && get_bits1(&gb))
2140 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
2144 s->keyframe = !get_bits1(&gb);
2147 s->last_quality_index = s->quality_index;
2151 s->qis[s->nqis++]= get_bits(&gb, 6);
2152 } while(s->theora >= 0x030200 && s->nqis<3 && get_bits1(&gb));
2154 s->quality_index= s->qis[0];
2156 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
2157 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
2158 s->keyframe?"key":"", counter, s->quality_index);
2161 if (s->quality_index != s->last_quality_index) {
2162 init_dequantizer(s);
2163 init_loop_filter(s);
2169 skip_bits(&gb, 4); /* width code */
2170 skip_bits(&gb, 4); /* height code */
2173 s->version = get_bits(&gb, 5);
2175 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
2178 if (s->version || s->theora)
2181 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
2182 skip_bits(&gb, 2); /* reserved? */
2185 if (s->last_frame.data[0] == s->golden_frame.data[0]) {
2186 if (s->golden_frame.data[0])
2187 avctx->release_buffer(avctx, &s->golden_frame);
2188 s->last_frame= s->golden_frame; /* ensure that we catch any access to this released frame */
2190 if (s->golden_frame.data[0])
2191 avctx->release_buffer(avctx, &s->golden_frame);
2192 if (s->last_frame.data[0])
2193 avctx->release_buffer(avctx, &s->last_frame);
2196 s->golden_frame.reference = 3;
2197 if(avctx->get_buffer(avctx, &s->golden_frame) < 0) {
2198 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
2202 /* golden frame is also the current frame */
2203 s->current_frame= s->golden_frame;
2205 /* time to figure out pixel addresses? */
2206 if (!s->pixel_addresses_initialized)
2208 if (!s->flipped_image)
2209 vp3_calculate_pixel_addresses(s);
2211 theora_calculate_pixel_addresses(s);
2212 s->pixel_addresses_initialized = 1;
2215 /* allocate a new current frame */
2216 s->current_frame.reference = 3;
2217 if (!s->pixel_addresses_initialized) {
2218 av_log(s->avctx, AV_LOG_ERROR, "vp3: first frame not a keyframe\n");
2221 if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
2222 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
2227 s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
2228 s->current_frame.qstride= 0;
2232 STOP_TIMER("init_frame")}
2237 memcpy(s->current_frame.data[0], s->golden_frame.data[0],
2238 s->current_frame.linesize[0] * s->height);
2239 memcpy(s->current_frame.data[1], s->golden_frame.data[1],
2240 s->current_frame.linesize[1] * s->height / 2);
2241 memcpy(s->current_frame.data[2], s->golden_frame.data[2],
2242 s->current_frame.linesize[2] * s->height / 2);
2248 if (unpack_superblocks(s, &gb)){
2249 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
2252 STOP_TIMER("unpack_superblocks")}
2254 if (unpack_modes(s, &gb)){
2255 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
2258 STOP_TIMER("unpack_modes")}
2260 if (unpack_vectors(s, &gb)){
2261 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2264 STOP_TIMER("unpack_vectors")}
2266 if (unpack_dct_coeffs(s, &gb)){
2267 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
2270 STOP_TIMER("unpack_dct_coeffs")}
2273 reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
2274 if ((avctx->flags & CODEC_FLAG_GRAY) == 0) {
2275 reverse_dc_prediction(s, s->fragment_start[1],
2276 s->fragment_width / 2, s->fragment_height / 2);
2277 reverse_dc_prediction(s, s->fragment_start[2],
2278 s->fragment_width / 2, s->fragment_height / 2);
2280 STOP_TIMER("reverse_dc_prediction")}
2283 for (i = 0; i < s->macroblock_height; i++)
2285 STOP_TIMER("render_fragments")}
2288 apply_loop_filter(s);
2289 STOP_TIMER("apply_loop_filter")}
2294 *data_size=sizeof(AVFrame);
2295 *(AVFrame*)data= s->current_frame;
2297 /* release the last frame, if it is allocated and if it is not the
2299 if ((s->last_frame.data[0]) &&
2300 (s->last_frame.data[0] != s->golden_frame.data[0]))
2301 avctx->release_buffer(avctx, &s->last_frame);
2303 /* shuffle frames (last = current) */
2304 s->last_frame= s->current_frame;
2305 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
2311 * This is the ffmpeg/libavcodec API module cleanup function.
2313 static av_cold int vp3_decode_end(AVCodecContext *avctx)
2315 Vp3DecodeContext *s = avctx->priv_data;
2318 av_free(s->superblock_coding);
2319 av_free(s->all_fragments);
2321 av_free(s->coded_fragment_list);
2322 av_free(s->superblock_fragments);
2323 av_free(s->superblock_macroblocks);
2324 av_free(s->macroblock_fragments);
2325 av_free(s->macroblock_coding);
2327 for (i = 0; i < 16; i++) {
2328 free_vlc(&s->dc_vlc[i]);
2329 free_vlc(&s->ac_vlc_1[i]);
2330 free_vlc(&s->ac_vlc_2[i]);
2331 free_vlc(&s->ac_vlc_3[i]);
2332 free_vlc(&s->ac_vlc_4[i]);
2335 free_vlc(&s->superblock_run_length_vlc);
2336 free_vlc(&s->fragment_run_length_vlc);
2337 free_vlc(&s->mode_code_vlc);
2338 free_vlc(&s->motion_vector_vlc);
2340 /* release all frames */
2341 if (s->golden_frame.data[0] && s->golden_frame.data[0] != s->last_frame.data[0])
2342 avctx->release_buffer(avctx, &s->golden_frame);
2343 if (s->last_frame.data[0])
2344 avctx->release_buffer(avctx, &s->last_frame);
2345 /* no need to release the current_frame since it will always be pointing
2346 * to the same frame as either the golden or last frame */
2351 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2353 Vp3DecodeContext *s = avctx->priv_data;
2355 if (get_bits1(gb)) {
2357 if (s->entries >= 32) { /* overflow */
2358 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2361 token = get_bits(gb, 5);
2362 //av_log(avctx, AV_LOG_DEBUG, "hti %d hbits %x token %d entry : %d size %d\n", s->hti, s->hbits, token, s->entries, s->huff_code_size);
2363 s->huffman_table[s->hti][token][0] = s->hbits;
2364 s->huffman_table[s->hti][token][1] = s->huff_code_size;
2368 if (s->huff_code_size >= 32) {/* overflow */
2369 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2372 s->huff_code_size++;
2374 read_huffman_tree(avctx, gb);
2376 read_huffman_tree(avctx, gb);
2378 s->huff_code_size--;
2383 #ifdef CONFIG_THEORA_DECODER
2384 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2386 Vp3DecodeContext *s = avctx->priv_data;
2387 int visible_width, visible_height;
2389 s->theora = get_bits_long(gb, 24);
2390 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2392 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2393 /* but previous versions have the image flipped relative to vp3 */
2394 if (s->theora < 0x030200)
2396 s->flipped_image = 1;
2397 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2400 s->width = get_bits(gb, 16) << 4;
2401 s->height = get_bits(gb, 16) << 4;
2403 if(avcodec_check_dimensions(avctx, s->width, s->height)){
2404 av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2405 s->width= s->height= 0;
2409 if (s->theora >= 0x030400)
2411 skip_bits(gb, 32); /* total number of superblocks in a frame */
2412 // fixme, the next field is 36bits long
2413 skip_bits(gb, 32); /* total number of blocks in a frame */
2414 skip_bits(gb, 4); /* total number of blocks in a frame */
2415 skip_bits(gb, 32); /* total number of macroblocks in a frame */
2418 visible_width = get_bits_long(gb, 24);
2419 visible_height = get_bits_long(gb, 24);
2421 if (s->theora >= 0x030200) {
2422 skip_bits(gb, 8); /* offset x */
2423 skip_bits(gb, 8); /* offset y */
2426 skip_bits(gb, 32); /* fps numerator */
2427 skip_bits(gb, 32); /* fps denumerator */
2428 skip_bits(gb, 24); /* aspect numerator */
2429 skip_bits(gb, 24); /* aspect denumerator */
2431 if (s->theora < 0x030200)
2432 skip_bits(gb, 5); /* keyframe frequency force */
2433 skip_bits(gb, 8); /* colorspace */
2434 if (s->theora >= 0x030400)
2435 skip_bits(gb, 2); /* pixel format: 420,res,422,444 */
2436 skip_bits(gb, 24); /* bitrate */
2438 skip_bits(gb, 6); /* quality hint */
2440 if (s->theora >= 0x030200)
2442 skip_bits(gb, 5); /* keyframe frequency force */
2444 if (s->theora < 0x030400)
2445 skip_bits(gb, 5); /* spare bits */
2448 // align_get_bits(gb);
2450 if ( visible_width <= s->width && visible_width > s->width-16
2451 && visible_height <= s->height && visible_height > s->height-16)
2452 avcodec_set_dimensions(avctx, visible_width, visible_height);
2454 avcodec_set_dimensions(avctx, s->width, s->height);
2459 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2461 Vp3DecodeContext *s = avctx->priv_data;
2462 int i, n, matrices, inter, plane;
2464 if (s->theora >= 0x030200) {
2465 n = get_bits(gb, 3);
2466 /* loop filter limit values table */
2467 for (i = 0; i < 64; i++)
2468 s->filter_limit_values[i] = get_bits(gb, n);
2471 if (s->theora >= 0x030200)
2472 n = get_bits(gb, 4) + 1;
2475 /* quality threshold table */
2476 for (i = 0; i < 64; i++)
2477 s->coded_ac_scale_factor[i] = get_bits(gb, n);
2479 if (s->theora >= 0x030200)
2480 n = get_bits(gb, 4) + 1;
2483 /* dc scale factor table */
2484 for (i = 0; i < 64; i++)
2485 s->coded_dc_scale_factor[i] = get_bits(gb, n);
2487 if (s->theora >= 0x030200)
2488 matrices = get_bits(gb, 9) + 1;
2493 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2497 for(n=0; n<matrices; n++){
2498 for (i = 0; i < 64; i++)
2499 s->base_matrix[n][i]= get_bits(gb, 8);
2502 for (inter = 0; inter <= 1; inter++) {
2503 for (plane = 0; plane <= 2; plane++) {
2505 if (inter || plane > 0)
2506 newqr = get_bits1(gb);
2509 if(inter && get_bits1(gb)){
2513 qtj= (3*inter + plane - 1) / 3;
2514 plj= (plane + 2) % 3;
2516 s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2517 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2518 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2524 i= get_bits(gb, av_log2(matrices-1)+1);
2526 av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2529 s->qr_base[inter][plane][qri]= i;
2532 i = get_bits(gb, av_log2(63-qi)+1) + 1;
2533 s->qr_size[inter][plane][qri++]= i;
2538 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2541 s->qr_count[inter][plane]= qri;
2546 /* Huffman tables */
2547 for (s->hti = 0; s->hti < 80; s->hti++) {
2549 s->huff_code_size = 1;
2550 if (!get_bits1(gb)) {
2552 read_huffman_tree(avctx, gb);
2554 read_huffman_tree(avctx, gb);
2558 s->theora_tables = 1;
2563 static int theora_decode_init(AVCodecContext *avctx)
2565 Vp3DecodeContext *s = avctx->priv_data;
2568 uint8_t *header_start[3];
2574 if (!avctx->extradata_size)
2576 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2580 if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2581 42, header_start, header_len) < 0) {
2582 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2587 init_get_bits(&gb, header_start[i], header_len[i]);
2589 ptype = get_bits(&gb, 8);
2590 debug_vp3("Theora headerpacket type: %x\n", ptype);
2592 if (!(ptype & 0x80))
2594 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2598 // FIXME: Check for this as well.
2599 skip_bits(&gb, 6*8); /* "theora" */
2604 theora_decode_header(avctx, &gb);
2607 // FIXME: is this needed? it breaks sometimes
2608 // theora_decode_comments(avctx, gb);
2611 theora_decode_tables(avctx, &gb);
2614 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2617 if(8*header_len[i] != get_bits_count(&gb))
2618 av_log(avctx, AV_LOG_ERROR, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2619 if (s->theora < 0x030200)
2623 vp3_decode_init(avctx);
2627 AVCodec theora_decoder = {
2631 sizeof(Vp3DecodeContext),
2638 .long_name = "Theora",
2642 AVCodec vp3_decoder = {
2646 sizeof(Vp3DecodeContext),
2653 .long_name = "On2 VP3",