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
24 * On2 VP3 Video Decoder
26 * VP3 Video Decoder by Mike Melanson (mike at multimedia.cx)
27 * For more information about the VP3 coding process, visit:
28 * http://multimedia.cx/
30 * Theora decoder by Alex Beregszaszi
41 #include "mpegvideo.h"
46 #define FRAGMENT_PIXELS 8
51 * Define one or more of the following compile-time variables to 1 to obtain
52 * elaborate information about certain aspects of the decoding process.
54 * KEYFRAMES_ONLY: set this to 1 to only see keyframes (VP3 slideshow mode)
55 * DEBUG_VP3: high-level decoding flow
56 * DEBUG_INIT: initialization parameters
57 * DEBUG_DEQUANTIZERS: display how the dequanization tables are built
58 * DEBUG_BLOCK_CODING: unpacking the superblock/macroblock/fragment coding
59 * DEBUG_MODES: unpacking the coding modes for individual fragments
60 * DEBUG_VECTORS: display the motion vectors
61 * DEBUG_TOKEN: display exhaustive information about each DCT token
62 * DEBUG_VLC: display the VLCs as they are extracted from the stream
63 * DEBUG_DC_PRED: display the process of reversing DC prediction
64 * DEBUG_IDCT: show every detail of the IDCT process
67 #define KEYFRAMES_ONLY 0
71 #define DEBUG_DEQUANTIZERS 0
72 #define DEBUG_BLOCK_CODING 0
74 #define DEBUG_VECTORS 0
77 #define DEBUG_DC_PRED 0
81 #define debug_vp3(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
83 static inline void debug_vp3(const char *format, ...) { }
87 #define debug_init(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
89 static inline void debug_init(const char *format, ...) { }
92 #if DEBUG_DEQUANTIZERS
93 #define debug_dequantizers(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
95 static inline void debug_dequantizers(const char *format, ...) { }
98 #if DEBUG_BLOCK_CODING
99 #define debug_block_coding(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
101 static inline void debug_block_coding(const char *format, ...) { }
105 #define debug_modes(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
107 static inline void debug_modes(const char *format, ...) { }
111 #define debug_vectors(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
113 static inline void debug_vectors(const char *format, ...) { }
117 #define debug_token(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
119 static inline void debug_token(const char *format, ...) { }
123 #define debug_vlc(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
125 static inline void debug_vlc(const char *format, ...) { }
129 #define debug_dc_pred(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
131 static inline void debug_dc_pred(const char *format, ...) { }
135 #define debug_idct(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
137 static inline void debug_idct(const char *format, ...) { }
140 typedef struct Coeff {
146 //FIXME split things out into their own arrays
147 typedef struct Vp3Fragment {
149 /* address of first pixel taking into account which plane the fragment
150 * lives on as well as the plane stride */
152 /* this is the macroblock that the fragment belongs to */
154 uint8_t coding_method;
160 #define SB_NOT_CODED 0
161 #define SB_PARTIALLY_CODED 1
162 #define SB_FULLY_CODED 2
164 #define MODE_INTER_NO_MV 0
166 #define MODE_INTER_PLUS_MV 2
167 #define MODE_INTER_LAST_MV 3
168 #define MODE_INTER_PRIOR_LAST 4
169 #define MODE_USING_GOLDEN 5
170 #define MODE_GOLDEN_MV 6
171 #define MODE_INTER_FOURMV 7
172 #define CODING_MODE_COUNT 8
174 /* special internal mode */
177 /* There are 6 preset schemes, plus a free-form scheme */
178 static int ModeAlphabet[7][CODING_MODE_COUNT] =
180 /* this is the custom scheme */
181 { 0, 0, 0, 0, 0, 0, 0, 0 },
183 /* scheme 1: Last motion vector dominates */
184 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
185 MODE_INTER_PLUS_MV, MODE_INTER_NO_MV,
186 MODE_INTRA, MODE_USING_GOLDEN,
187 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
190 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
191 MODE_INTER_NO_MV, MODE_INTER_PLUS_MV,
192 MODE_INTRA, MODE_USING_GOLDEN,
193 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
196 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
197 MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
198 MODE_INTRA, MODE_USING_GOLDEN,
199 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
202 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
203 MODE_INTER_NO_MV, MODE_INTER_PRIOR_LAST,
204 MODE_INTRA, MODE_USING_GOLDEN,
205 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
207 /* scheme 5: No motion vector dominates */
208 { MODE_INTER_NO_MV, MODE_INTER_LAST_MV,
209 MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
210 MODE_INTRA, MODE_USING_GOLDEN,
211 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
214 { MODE_INTER_NO_MV, MODE_USING_GOLDEN,
215 MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
216 MODE_INTER_PLUS_MV, MODE_INTRA,
217 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
221 #define MIN_DEQUANT_VAL 2
223 typedef struct Vp3DecodeContext {
224 AVCodecContext *avctx;
225 int theora, theora_tables;
228 AVFrame golden_frame;
230 AVFrame current_frame;
238 int last_quality_index;
240 int superblock_count;
241 int superblock_width;
242 int superblock_height;
243 int y_superblock_width;
244 int y_superblock_height;
245 int c_superblock_width;
246 int c_superblock_height;
247 int u_superblock_start;
248 int v_superblock_start;
249 unsigned char *superblock_coding;
251 int macroblock_count;
252 int macroblock_width;
253 int macroblock_height;
259 Vp3Fragment *all_fragments;
262 int fragment_start[3];
267 uint16_t coded_dc_scale_factor[64];
268 uint32_t coded_ac_scale_factor[64];
269 uint8_t base_matrix[384][64];
270 uint8_t qr_count[2][3];
271 uint8_t qr_size [2][3][64];
272 uint16_t qr_base[2][3][64];
274 /* this is a list of indices into the all_fragments array indicating
275 * which of the fragments are coded */
276 int *coded_fragment_list;
277 int coded_fragment_list_index;
278 int pixel_addresses_inited;
286 VLC superblock_run_length_vlc;
287 VLC fragment_run_length_vlc;
289 VLC motion_vector_vlc;
291 /* these arrays need to be on 16-byte boundaries since SSE2 operations
293 DECLARE_ALIGNED_16(int16_t, qmat[2][4][64]); //<qmat[is_inter][plane]
295 /* This table contains superblock_count * 16 entries. Each set of 16
296 * numbers corresponds to the fragment indices 0..15 of the superblock.
297 * An entry will be -1 to indicate that no entry corresponds to that
299 int *superblock_fragments;
301 /* This table contains superblock_count * 4 entries. Each set of 4
302 * numbers corresponds to the macroblock indices 0..3 of the superblock.
303 * An entry will be -1 to indicate that no entry corresponds to that
305 int *superblock_macroblocks;
307 /* This table contains macroblock_count * 6 entries. Each set of 6
308 * numbers corresponds to the fragment indices 0..5 which comprise
309 * the macroblock (4 Y fragments and 2 C fragments). */
310 int *macroblock_fragments;
311 /* This is an array that indicates how a particular macroblock
313 unsigned char *macroblock_coding;
315 int first_coded_y_fragment;
316 int first_coded_c_fragment;
317 int last_coded_y_fragment;
318 int last_coded_c_fragment;
320 uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
321 int8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
328 uint16_t huffman_table[80][32][2];
330 uint32_t filter_limit_values[64];
331 int bounding_values_array[256];
334 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb);
336 /************************************************************************
337 * VP3 specific functions
338 ************************************************************************/
341 * This function sets up all of the various blocks mappings:
342 * superblocks <-> fragments, macroblocks <-> fragments,
343 * superblocks <-> macroblocks
345 * Returns 0 is successful; returns 1 if *anything* went wrong.
347 static int init_block_mapping(Vp3DecodeContext *s)
350 signed int hilbert_walk_mb[4];
352 int current_fragment = 0;
353 int current_width = 0;
354 int current_height = 0;
357 int superblock_row_inc = 0;
359 int mapping_index = 0;
361 int current_macroblock;
364 signed char travel_width[16] = {
371 signed char travel_height[16] = {
378 signed char travel_width_mb[4] = {
382 signed char travel_height_mb[4] = {
386 debug_vp3(" vp3: initialize block mapping tables\n");
388 hilbert_walk_mb[0] = 1;
389 hilbert_walk_mb[1] = s->macroblock_width;
390 hilbert_walk_mb[2] = 1;
391 hilbert_walk_mb[3] = -s->macroblock_width;
393 /* iterate through each superblock (all planes) and map the fragments */
394 for (i = 0; i < s->superblock_count; i++) {
395 debug_init(" superblock %d (u starts @ %d, v starts @ %d)\n",
396 i, s->u_superblock_start, s->v_superblock_start);
398 /* time to re-assign the limits? */
401 /* start of Y superblocks */
402 right_edge = s->fragment_width;
403 bottom_edge = s->fragment_height;
406 superblock_row_inc = 3 * s->fragment_width -
407 (s->y_superblock_width * 4 - s->fragment_width);
409 /* the first operation for this variable is to advance by 1 */
410 current_fragment = -1;
412 } else if (i == s->u_superblock_start) {
414 /* start of U superblocks */
415 right_edge = s->fragment_width / 2;
416 bottom_edge = s->fragment_height / 2;
419 superblock_row_inc = 3 * (s->fragment_width / 2) -
420 (s->c_superblock_width * 4 - s->fragment_width / 2);
422 /* the first operation for this variable is to advance by 1 */
423 current_fragment = s->fragment_start[1] - 1;
425 } else if (i == s->v_superblock_start) {
427 /* start of V superblocks */
428 right_edge = s->fragment_width / 2;
429 bottom_edge = s->fragment_height / 2;
432 superblock_row_inc = 3 * (s->fragment_width / 2) -
433 (s->c_superblock_width * 4 - s->fragment_width / 2);
435 /* the first operation for this variable is to advance by 1 */
436 current_fragment = s->fragment_start[2] - 1;
440 if (current_width >= right_edge - 1) {
441 /* reset width and move to next superblock row */
445 /* fragment is now at the start of a new superblock row */
446 current_fragment += superblock_row_inc;
449 /* iterate through all 16 fragments in a superblock */
450 for (j = 0; j < 16; j++) {
451 current_fragment += travel_width[j] + right_edge * travel_height[j];
452 current_width += travel_width[j];
453 current_height += travel_height[j];
455 /* check if the fragment is in bounds */
456 if ((current_width < right_edge) &&
457 (current_height < bottom_edge)) {
458 s->superblock_fragments[mapping_index] = current_fragment;
459 debug_init(" mapping fragment %d to superblock %d, position %d (%d/%d x %d/%d)\n",
460 s->superblock_fragments[mapping_index], i, j,
461 current_width, right_edge, current_height, bottom_edge);
463 s->superblock_fragments[mapping_index] = -1;
464 debug_init(" superblock %d, position %d has no fragment (%d/%d x %d/%d)\n",
466 current_width, right_edge, current_height, bottom_edge);
473 /* initialize the superblock <-> macroblock mapping; iterate through
474 * all of the Y plane superblocks to build this mapping */
475 right_edge = s->macroblock_width;
476 bottom_edge = s->macroblock_height;
479 superblock_row_inc = s->macroblock_width -
480 (s->y_superblock_width * 2 - s->macroblock_width);;
481 hilbert = hilbert_walk_mb;
483 current_macroblock = -1;
484 for (i = 0; i < s->u_superblock_start; i++) {
486 if (current_width >= right_edge - 1) {
487 /* reset width and move to next superblock row */
491 /* macroblock is now at the start of a new superblock row */
492 current_macroblock += superblock_row_inc;
495 /* iterate through each potential macroblock in the superblock */
496 for (j = 0; j < 4; j++) {
497 current_macroblock += hilbert_walk_mb[j];
498 current_width += travel_width_mb[j];
499 current_height += travel_height_mb[j];
501 /* check if the macroblock is in bounds */
502 if ((current_width < right_edge) &&
503 (current_height < bottom_edge)) {
504 s->superblock_macroblocks[mapping_index] = current_macroblock;
505 debug_init(" mapping macroblock %d to superblock %d, position %d (%d/%d x %d/%d)\n",
506 s->superblock_macroblocks[mapping_index], i, j,
507 current_width, right_edge, current_height, bottom_edge);
509 s->superblock_macroblocks[mapping_index] = -1;
510 debug_init(" superblock %d, position %d has no macroblock (%d/%d x %d/%d)\n",
512 current_width, right_edge, current_height, bottom_edge);
519 /* initialize the macroblock <-> fragment mapping */
520 current_fragment = 0;
521 current_macroblock = 0;
523 for (i = 0; i < s->fragment_height; i += 2) {
525 for (j = 0; j < s->fragment_width; j += 2) {
527 debug_init(" macroblock %d contains fragments: ", current_macroblock);
528 s->all_fragments[current_fragment].macroblock = current_macroblock;
529 s->macroblock_fragments[mapping_index++] = current_fragment;
530 debug_init("%d ", current_fragment);
532 if (j + 1 < s->fragment_width) {
533 s->all_fragments[current_fragment + 1].macroblock = current_macroblock;
534 s->macroblock_fragments[mapping_index++] = current_fragment + 1;
535 debug_init("%d ", current_fragment + 1);
537 s->macroblock_fragments[mapping_index++] = -1;
539 if (i + 1 < s->fragment_height) {
540 s->all_fragments[current_fragment + s->fragment_width].macroblock =
542 s->macroblock_fragments[mapping_index++] =
543 current_fragment + s->fragment_width;
544 debug_init("%d ", current_fragment + s->fragment_width);
546 s->macroblock_fragments[mapping_index++] = -1;
548 if ((j + 1 < s->fragment_width) && (i + 1 < s->fragment_height)) {
549 s->all_fragments[current_fragment + s->fragment_width + 1].macroblock =
551 s->macroblock_fragments[mapping_index++] =
552 current_fragment + s->fragment_width + 1;
553 debug_init("%d ", current_fragment + s->fragment_width + 1);
555 s->macroblock_fragments[mapping_index++] = -1;
558 c_fragment = s->fragment_start[1] +
559 (i * s->fragment_width / 4) + (j / 2);
560 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
561 s->macroblock_fragments[mapping_index++] = c_fragment;
562 debug_init("%d ", c_fragment);
564 c_fragment = s->fragment_start[2] +
565 (i * s->fragment_width / 4) + (j / 2);
566 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
567 s->macroblock_fragments[mapping_index++] = c_fragment;
568 debug_init("%d ", c_fragment);
572 if (j + 2 <= s->fragment_width)
573 current_fragment += 2;
576 current_macroblock++;
579 current_fragment += s->fragment_width;
582 return 0; /* successful path out */
586 * This function wipes out all of the fragment data.
588 static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
592 /* zero out all of the fragment information */
593 s->coded_fragment_list_index = 0;
594 for (i = 0; i < s->fragment_count; i++) {
595 s->all_fragments[i].coeff_count = 0;
596 s->all_fragments[i].motion_x = 127;
597 s->all_fragments[i].motion_y = 127;
598 s->all_fragments[i].next_coeff= NULL;
600 s->coeffs[i].coeff=0;
601 s->coeffs[i].next= NULL;
606 * This function sets up the dequantization tables used for a particular
609 static void init_dequantizer(Vp3DecodeContext *s)
611 int ac_scale_factor = s->coded_ac_scale_factor[s->quality_index];
612 int dc_scale_factor = s->coded_dc_scale_factor[s->quality_index];
613 int i, plane, inter, qri, bmi, bmj, qistart;
615 debug_vp3(" vp3: initializing dequantization tables\n");
617 for(inter=0; inter<2; inter++){
618 for(plane=0; plane<3; plane++){
620 for(qri=0; qri<s->qr_count[inter][plane]; qri++){
621 sum+= s->qr_size[inter][plane][qri];
622 if(s->quality_index <= sum)
625 qistart= sum - s->qr_size[inter][plane][qri];
626 bmi= s->qr_base[inter][plane][qri ];
627 bmj= s->qr_base[inter][plane][qri+1];
629 int coeff= ( 2*(sum -s->quality_index)*s->base_matrix[bmi][i]
630 - 2*(qistart-s->quality_index)*s->base_matrix[bmj][i]
631 + s->qr_size[inter][plane][qri])
632 / (2*s->qr_size[inter][plane][qri]);
634 int qmin= 8<<(inter + !i);
635 int qscale= i ? ac_scale_factor : dc_scale_factor;
637 s->qmat[inter][plane][i]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
642 memset(s->qscale_table, (FFMAX(s->qmat[0][0][1], s->qmat[0][1][1])+8)/16, 512); //FIXME finetune
646 * This function initializes the loop filter boundary limits if the frame's
647 * quality index is different from the previous frame's.
649 static void init_loop_filter(Vp3DecodeContext *s)
651 int *bounding_values= s->bounding_values_array+127;
655 filter_limit = s->filter_limit_values[s->quality_index];
657 /* set up the bounding values */
658 memset(s->bounding_values_array, 0, 256 * sizeof(int));
659 for (x = 0; x < filter_limit; x++) {
660 bounding_values[-x - filter_limit] = -filter_limit + x;
661 bounding_values[-x] = -x;
662 bounding_values[x] = x;
663 bounding_values[x + filter_limit] = filter_limit - x;
668 * This function unpacks all of the superblock/macroblock/fragment coding
669 * information from the bitstream.
671 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
674 int current_superblock = 0;
676 int decode_fully_flags = 0;
677 int decode_partial_blocks = 0;
678 int first_c_fragment_seen;
681 int current_fragment;
683 debug_vp3(" vp3: unpacking superblock coding\n");
687 debug_vp3(" keyframe-- all superblocks are fully coded\n");
688 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
692 /* unpack the list of partially-coded superblocks */
693 bit = get_bits(gb, 1);
694 /* toggle the bit because as soon as the first run length is
695 * fetched the bit will be toggled again */
697 while (current_superblock < s->superblock_count) {
698 if (current_run-- == 0) {
700 current_run = get_vlc2(gb,
701 s->superblock_run_length_vlc.table, 6, 2);
702 if (current_run == 33)
703 current_run += get_bits(gb, 12);
704 debug_block_coding(" setting superblocks %d..%d to %s\n",
706 current_superblock + current_run - 1,
707 (bit) ? "partially coded" : "not coded");
709 /* if any of the superblocks are not partially coded, flag
710 * a boolean to decode the list of fully-coded superblocks */
712 decode_fully_flags = 1;
715 /* make a note of the fact that there are partially coded
717 decode_partial_blocks = 1;
720 s->superblock_coding[current_superblock++] = bit;
723 /* unpack the list of fully coded superblocks if any of the blocks were
724 * not marked as partially coded in the previous step */
725 if (decode_fully_flags) {
727 current_superblock = 0;
729 bit = get_bits(gb, 1);
730 /* toggle the bit because as soon as the first run length is
731 * fetched the bit will be toggled again */
733 while (current_superblock < s->superblock_count) {
735 /* skip any superblocks already marked as partially coded */
736 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
738 if (current_run-- == 0) {
740 current_run = get_vlc2(gb,
741 s->superblock_run_length_vlc.table, 6, 2);
742 if (current_run == 33)
743 current_run += get_bits(gb, 12);
746 debug_block_coding(" setting superblock %d to %s\n",
748 (bit) ? "fully coded" : "not coded");
749 s->superblock_coding[current_superblock] = 2*bit;
751 current_superblock++;
755 /* if there were partial blocks, initialize bitstream for
756 * unpacking fragment codings */
757 if (decode_partial_blocks) {
760 bit = get_bits(gb, 1);
761 /* toggle the bit because as soon as the first run length is
762 * fetched the bit will be toggled again */
767 /* figure out which fragments are coded; iterate through each
768 * superblock (all planes) */
769 s->coded_fragment_list_index = 0;
770 s->next_coeff= s->coeffs + s->fragment_count;
771 s->first_coded_y_fragment = s->first_coded_c_fragment = 0;
772 s->last_coded_y_fragment = s->last_coded_c_fragment = -1;
773 first_c_fragment_seen = 0;
774 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
775 for (i = 0; i < s->superblock_count; i++) {
777 /* iterate through all 16 fragments in a superblock */
778 for (j = 0; j < 16; j++) {
780 /* if the fragment is in bounds, check its coding status */
781 current_fragment = s->superblock_fragments[i * 16 + j];
782 if (current_fragment >= s->fragment_count) {
783 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
784 current_fragment, s->fragment_count);
787 if (current_fragment != -1) {
788 if (s->superblock_coding[i] == SB_NOT_CODED) {
790 /* copy all the fragments from the prior frame */
791 s->all_fragments[current_fragment].coding_method =
794 } else if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
796 /* fragment may or may not be coded; this is the case
797 * that cares about the fragment coding runs */
798 if (current_run-- == 0) {
800 current_run = get_vlc2(gb,
801 s->fragment_run_length_vlc.table, 5, 2);
805 /* default mode; actual mode will be decoded in
807 s->all_fragments[current_fragment].coding_method =
809 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
810 s->coded_fragment_list[s->coded_fragment_list_index] =
812 if ((current_fragment >= s->fragment_start[1]) &&
813 (s->last_coded_y_fragment == -1) &&
814 (!first_c_fragment_seen)) {
815 s->first_coded_c_fragment = s->coded_fragment_list_index;
816 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
817 first_c_fragment_seen = 1;
819 s->coded_fragment_list_index++;
820 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
821 debug_block_coding(" superblock %d is partially coded, fragment %d is coded\n",
822 i, current_fragment);
824 /* not coded; copy this fragment from the prior frame */
825 s->all_fragments[current_fragment].coding_method =
827 debug_block_coding(" superblock %d is partially coded, fragment %d is not coded\n",
828 i, current_fragment);
833 /* fragments are fully coded in this superblock; actual
834 * coding will be determined in next step */
835 s->all_fragments[current_fragment].coding_method =
837 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
838 s->coded_fragment_list[s->coded_fragment_list_index] =
840 if ((current_fragment >= s->fragment_start[1]) &&
841 (s->last_coded_y_fragment == -1) &&
842 (!first_c_fragment_seen)) {
843 s->first_coded_c_fragment = s->coded_fragment_list_index;
844 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
845 first_c_fragment_seen = 1;
847 s->coded_fragment_list_index++;
848 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
849 debug_block_coding(" superblock %d is fully coded, fragment %d is coded\n",
850 i, current_fragment);
856 if (!first_c_fragment_seen)
857 /* only Y fragments coded in this frame */
858 s->last_coded_y_fragment = s->coded_fragment_list_index - 1;
860 /* end the list of coded C fragments */
861 s->last_coded_c_fragment = s->coded_fragment_list_index - 1;
863 debug_block_coding(" %d total coded fragments, y: %d -> %d, c: %d -> %d\n",
864 s->coded_fragment_list_index,
865 s->first_coded_y_fragment,
866 s->last_coded_y_fragment,
867 s->first_coded_c_fragment,
868 s->last_coded_c_fragment);
874 * This function unpacks all the coding mode data for individual macroblocks
875 * from the bitstream.
877 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
881 int current_macroblock;
882 int current_fragment;
885 debug_vp3(" vp3: unpacking encoding modes\n");
888 debug_vp3(" keyframe-- all blocks are coded as INTRA\n");
890 for (i = 0; i < s->fragment_count; i++)
891 s->all_fragments[i].coding_method = MODE_INTRA;
895 /* fetch the mode coding scheme for this frame */
896 scheme = get_bits(gb, 3);
897 debug_modes(" using mode alphabet %d\n", scheme);
899 /* is it a custom coding scheme? */
901 debug_modes(" custom mode alphabet ahead:\n");
902 for (i = 0; i < 8; i++)
903 ModeAlphabet[scheme][get_bits(gb, 3)] = i;
906 for (i = 0; i < 8; i++)
907 debug_modes(" mode[%d][%d] = %d\n", scheme, i,
908 ModeAlphabet[scheme][i]);
910 /* iterate through all of the macroblocks that contain 1 or more
912 for (i = 0; i < s->u_superblock_start; i++) {
914 for (j = 0; j < 4; j++) {
915 current_macroblock = s->superblock_macroblocks[i * 4 + j];
916 if ((current_macroblock == -1) ||
917 (s->macroblock_coding[current_macroblock] == MODE_COPY))
919 if (current_macroblock >= s->macroblock_count) {
920 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad macroblock number (%d >= %d)\n",
921 current_macroblock, s->macroblock_count);
925 /* mode 7 means get 3 bits for each coding mode */
927 coding_mode = get_bits(gb, 3);
929 coding_mode = ModeAlphabet[scheme]
930 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
932 s->macroblock_coding[current_macroblock] = coding_mode;
933 for (k = 0; k < 6; k++) {
935 s->macroblock_fragments[current_macroblock * 6 + k];
936 if (current_fragment == -1)
938 if (current_fragment >= s->fragment_count) {
939 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad fragment number (%d >= %d)\n",
940 current_fragment, s->fragment_count);
943 if (s->all_fragments[current_fragment].coding_method !=
945 s->all_fragments[current_fragment].coding_method =
949 debug_modes(" coding method for macroblock starting @ fragment %d = %d\n",
950 s->macroblock_fragments[current_macroblock * 6], coding_mode);
959 * This function unpacks all the motion vectors for the individual
960 * macroblocks from the bitstream.
962 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
968 int last_motion_x = 0;
969 int last_motion_y = 0;
970 int prior_last_motion_x = 0;
971 int prior_last_motion_y = 0;
972 int current_macroblock;
973 int current_fragment;
975 debug_vp3(" vp3: unpacking motion vectors\n");
978 debug_vp3(" keyframe-- there are no motion vectors\n");
982 memset(motion_x, 0, 6 * sizeof(int));
983 memset(motion_y, 0, 6 * sizeof(int));
985 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
986 coding_mode = get_bits(gb, 1);
987 debug_vectors(" using %s scheme for unpacking motion vectors\n",
988 (coding_mode == 0) ? "VLC" : "fixed-length");
990 /* iterate through all of the macroblocks that contain 1 or more
992 for (i = 0; i < s->u_superblock_start; i++) {
994 for (j = 0; j < 4; j++) {
995 current_macroblock = s->superblock_macroblocks[i * 4 + j];
996 if ((current_macroblock == -1) ||
997 (s->macroblock_coding[current_macroblock] == MODE_COPY))
999 if (current_macroblock >= s->macroblock_count) {
1000 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad macroblock number (%d >= %d)\n",
1001 current_macroblock, s->macroblock_count);
1005 current_fragment = s->macroblock_fragments[current_macroblock * 6];
1006 if (current_fragment >= s->fragment_count) {
1007 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d\n",
1008 current_fragment, s->fragment_count);
1011 switch (s->macroblock_coding[current_macroblock]) {
1013 case MODE_INTER_PLUS_MV:
1014 case MODE_GOLDEN_MV:
1015 /* all 6 fragments use the same motion vector */
1016 if (coding_mode == 0) {
1017 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1018 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1020 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
1021 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
1024 for (k = 1; k < 6; k++) {
1025 motion_x[k] = motion_x[0];
1026 motion_y[k] = motion_y[0];
1029 /* vector maintenance, only on MODE_INTER_PLUS_MV */
1030 if (s->macroblock_coding[current_macroblock] ==
1031 MODE_INTER_PLUS_MV) {
1032 prior_last_motion_x = last_motion_x;
1033 prior_last_motion_y = last_motion_y;
1034 last_motion_x = motion_x[0];
1035 last_motion_y = motion_y[0];
1039 case MODE_INTER_FOURMV:
1040 /* fetch 4 vectors from the bitstream, one for each
1041 * Y fragment, then average for the C fragment vectors */
1042 motion_x[4] = motion_y[4] = 0;
1043 for (k = 0; k < 4; k++) {
1044 if (coding_mode == 0) {
1045 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1046 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1048 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
1049 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
1051 motion_x[4] += motion_x[k];
1052 motion_y[4] += motion_y[k];
1056 motion_x[4]= RSHIFT(motion_x[4], 2);
1058 motion_y[4]= RSHIFT(motion_y[4], 2);
1060 /* vector maintenance; vector[3] is treated as the
1061 * last vector in this case */
1062 prior_last_motion_x = last_motion_x;
1063 prior_last_motion_y = last_motion_y;
1064 last_motion_x = motion_x[3];
1065 last_motion_y = motion_y[3];
1068 case MODE_INTER_LAST_MV:
1069 /* all 6 fragments use the last motion vector */
1070 motion_x[0] = last_motion_x;
1071 motion_y[0] = last_motion_y;
1072 for (k = 1; k < 6; k++) {
1073 motion_x[k] = motion_x[0];
1074 motion_y[k] = motion_y[0];
1077 /* no vector maintenance (last vector remains the
1081 case MODE_INTER_PRIOR_LAST:
1082 /* all 6 fragments use the motion vector prior to the
1083 * last motion vector */
1084 motion_x[0] = prior_last_motion_x;
1085 motion_y[0] = prior_last_motion_y;
1086 for (k = 1; k < 6; k++) {
1087 motion_x[k] = motion_x[0];
1088 motion_y[k] = motion_y[0];
1091 /* vector maintenance */
1092 prior_last_motion_x = last_motion_x;
1093 prior_last_motion_y = last_motion_y;
1094 last_motion_x = motion_x[0];
1095 last_motion_y = motion_y[0];
1099 /* covers intra, inter without MV, golden without MV */
1100 memset(motion_x, 0, 6 * sizeof(int));
1101 memset(motion_y, 0, 6 * sizeof(int));
1103 /* no vector maintenance */
1107 /* assign the motion vectors to the correct fragments */
1108 debug_vectors(" vectors for macroblock starting @ fragment %d (coding method %d):\n",
1110 s->macroblock_coding[current_macroblock]);
1111 for (k = 0; k < 6; k++) {
1113 s->macroblock_fragments[current_macroblock * 6 + k];
1114 if (current_fragment == -1)
1116 if (current_fragment >= s->fragment_count) {
1117 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d)\n",
1118 current_fragment, s->fragment_count);
1121 s->all_fragments[current_fragment].motion_x = motion_x[k];
1122 s->all_fragments[current_fragment].motion_y = motion_y[k];
1123 debug_vectors(" vector %d: fragment %d = (%d, %d)\n",
1124 k, current_fragment, motion_x[k], motion_y[k]);
1134 * This function is called by unpack_dct_coeffs() to extract the VLCs from
1135 * the bitstream. The VLCs encode tokens which are used to unpack DCT
1136 * data. This function unpacks all the VLCs for either the Y plane or both
1137 * C planes, and is called for DC coefficients or different AC coefficient
1138 * levels (since different coefficient types require different VLC tables.
1140 * This function returns a residual eob run. E.g, if a particular token gave
1141 * instructions to EOB the next 5 fragments and there were only 2 fragments
1142 * left in the current fragment range, 3 would be returned so that it could
1143 * be passed into the next call to this same function.
1145 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
1146 VLC *table, int coeff_index,
1147 int first_fragment, int last_fragment,
1154 Vp3Fragment *fragment;
1155 uint8_t *perm= s->scantable.permutated;
1158 if ((first_fragment >= s->fragment_count) ||
1159 (last_fragment >= s->fragment_count)) {
1161 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vlcs(): bad fragment number (%d -> %d ?)\n",
1162 first_fragment, last_fragment);
1166 for (i = first_fragment; i <= last_fragment; i++) {
1168 fragment = &s->all_fragments[s->coded_fragment_list[i]];
1169 if (fragment->coeff_count > coeff_index)
1173 /* decode a VLC into a token */
1174 token = get_vlc2(gb, table->table, 5, 3);
1175 debug_vlc(" token = %2d, ", token);
1176 /* use the token to get a zero run, a coefficient, and an eob run */
1178 eob_run = eob_run_base[token];
1179 if (eob_run_get_bits[token])
1180 eob_run += get_bits(gb, eob_run_get_bits[token]);
1181 coeff = zero_run = 0;
1183 bits_to_get = coeff_get_bits[token];
1185 coeff = coeff_tables[token][0];
1187 coeff = coeff_tables[token][get_bits(gb, bits_to_get)];
1189 zero_run = zero_run_base[token];
1190 if (zero_run_get_bits[token])
1191 zero_run += get_bits(gb, zero_run_get_bits[token]);
1196 fragment->coeff_count += zero_run;
1197 if (fragment->coeff_count < 64){
1198 fragment->next_coeff->coeff= coeff;
1199 fragment->next_coeff->index= perm[fragment->coeff_count++]; //FIXME perm here already?
1200 fragment->next_coeff->next= s->next_coeff;
1201 s->next_coeff->next=NULL;
1202 fragment->next_coeff= s->next_coeff++;
1204 debug_vlc(" fragment %d coeff = %d\n",
1205 s->coded_fragment_list[i], fragment->next_coeff[coeff_index]);
1207 fragment->coeff_count |= 128;
1208 debug_vlc(" fragment %d eob with %d coefficients\n",
1209 s->coded_fragment_list[i], fragment->coeff_count&127);
1218 * This function unpacks all of the DCT coefficient data from the
1221 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1228 int residual_eob_run = 0;
1230 /* fetch the DC table indices */
1231 dc_y_table = get_bits(gb, 4);
1232 dc_c_table = get_bits(gb, 4);
1234 /* unpack the Y plane DC coefficients */
1235 debug_vp3(" vp3: unpacking Y plane DC coefficients using table %d\n",
1237 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1238 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1240 /* unpack the C plane DC coefficients */
1241 debug_vp3(" vp3: unpacking C plane DC coefficients using table %d\n",
1243 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1244 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1246 /* fetch the AC table indices */
1247 ac_y_table = get_bits(gb, 4);
1248 ac_c_table = get_bits(gb, 4);
1250 /* unpack the group 1 AC coefficients (coeffs 1-5) */
1251 for (i = 1; i <= 5; i++) {
1253 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1255 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_y_table], i,
1256 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1258 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1260 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_c_table], i,
1261 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1264 /* unpack the group 2 AC coefficients (coeffs 6-14) */
1265 for (i = 6; i <= 14; i++) {
1267 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1269 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_y_table], i,
1270 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1272 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1274 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_c_table], i,
1275 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1278 /* unpack the group 3 AC coefficients (coeffs 15-27) */
1279 for (i = 15; i <= 27; i++) {
1281 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1283 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_y_table], i,
1284 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1286 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1288 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_c_table], i,
1289 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1292 /* unpack the group 4 AC coefficients (coeffs 28-63) */
1293 for (i = 28; i <= 63; i++) {
1295 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1297 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_y_table], i,
1298 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1300 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1302 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_c_table], i,
1303 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1310 * This function reverses the DC prediction for each coded fragment in
1311 * the frame. Much of this function is adapted directly from the original
1314 #define COMPATIBLE_FRAME(x) \
1315 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1316 #define FRAME_CODED(x) (s->all_fragments[x].coding_method != MODE_COPY)
1317 #define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
1319 static void reverse_dc_prediction(Vp3DecodeContext *s,
1322 int fragment_height)
1331 int i = first_fragment;
1335 /* DC values for the left, up-left, up, and up-right fragments */
1336 int vl, vul, vu, vur;
1338 /* indices for the left, up-left, up, and up-right fragments */
1342 * The 6 fields mean:
1343 * 0: up-left multiplier
1345 * 2: up-right multiplier
1346 * 3: left multiplier
1348 int predictor_transform[16][4] = {
1350 { 0, 0, 0,128}, // PL
1351 { 0, 0,128, 0}, // PUR
1352 { 0, 0, 53, 75}, // PUR|PL
1353 { 0,128, 0, 0}, // PU
1354 { 0, 64, 0, 64}, // PU|PL
1355 { 0,128, 0, 0}, // PU|PUR
1356 { 0, 0, 53, 75}, // PU|PUR|PL
1357 {128, 0, 0, 0}, // PUL
1358 { 0, 0, 0,128}, // PUL|PL
1359 { 64, 0, 64, 0}, // PUL|PUR
1360 { 0, 0, 53, 75}, // PUL|PUR|PL
1361 { 0,128, 0, 0}, // PUL|PU
1362 {-104,116, 0,116}, // PUL|PU|PL
1363 { 24, 80, 24, 0}, // PUL|PU|PUR
1364 {-104,116, 0,116} // PUL|PU|PUR|PL
1367 /* This table shows which types of blocks can use other blocks for
1368 * prediction. For example, INTRA is the only mode in this table to
1369 * have a frame number of 0. That means INTRA blocks can only predict
1370 * from other INTRA blocks. There are 2 golden frame coding types;
1371 * blocks encoding in these modes can only predict from other blocks
1372 * that were encoded with these 1 of these 2 modes. */
1373 unsigned char compatible_frame[8] = {
1374 1, /* MODE_INTER_NO_MV */
1376 1, /* MODE_INTER_PLUS_MV */
1377 1, /* MODE_INTER_LAST_MV */
1378 1, /* MODE_INTER_PRIOR_MV */
1379 2, /* MODE_USING_GOLDEN */
1380 2, /* MODE_GOLDEN_MV */
1381 1 /* MODE_INTER_FOUR_MV */
1383 int current_frame_type;
1385 /* there is a last DC predictor for each of the 3 frame types */
1390 debug_vp3(" vp3: reversing DC prediction\n");
1392 vul = vu = vur = vl = 0;
1393 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1395 /* for each fragment row... */
1396 for (y = 0; y < fragment_height; y++) {
1398 /* for each fragment in a row... */
1399 for (x = 0; x < fragment_width; x++, i++) {
1401 /* reverse prediction if this block was coded */
1402 if (s->all_fragments[i].coding_method != MODE_COPY) {
1404 current_frame_type =
1405 compatible_frame[s->all_fragments[i].coding_method];
1406 debug_dc_pred(" frag %d: orig DC = %d, ",
1413 if(FRAME_CODED(l) && COMPATIBLE_FRAME(l))
1417 u= i-fragment_width;
1419 if(FRAME_CODED(u) && COMPATIBLE_FRAME(u))
1422 ul= i-fragment_width-1;
1424 if(FRAME_CODED(ul) && COMPATIBLE_FRAME(ul))
1427 if(x + 1 < fragment_width){
1428 ur= i-fragment_width+1;
1430 if(FRAME_CODED(ur) && COMPATIBLE_FRAME(ur))
1435 debug_dc_pred("transform = %d, ", transform);
1437 if (transform == 0) {
1439 /* if there were no fragments to predict from, use last
1441 predicted_dc = last_dc[current_frame_type];
1442 debug_dc_pred("from last DC (%d) = %d\n",
1443 current_frame_type, DC_COEFF(i));
1447 /* apply the appropriate predictor transform */
1449 (predictor_transform[transform][0] * vul) +
1450 (predictor_transform[transform][1] * vu) +
1451 (predictor_transform[transform][2] * vur) +
1452 (predictor_transform[transform][3] * vl);
1454 predicted_dc /= 128;
1456 /* check for outranging on the [ul u l] and
1457 * [ul u ur l] predictors */
1458 if ((transform == 13) || (transform == 15)) {
1459 if (FFABS(predicted_dc - vu) > 128)
1461 else if (FFABS(predicted_dc - vl) > 128)
1463 else if (FFABS(predicted_dc - vul) > 128)
1467 debug_dc_pred("from pred DC = %d\n",
1471 /* at long last, apply the predictor */
1472 if(s->coeffs[i].index){
1473 *s->next_coeff= s->coeffs[i];
1474 s->coeffs[i].index=0;
1475 s->coeffs[i].coeff=0;
1476 s->coeffs[i].next= s->next_coeff++;
1478 s->coeffs[i].coeff += predicted_dc;
1480 last_dc[current_frame_type] = DC_COEFF(i);
1481 if(DC_COEFF(i) && !(s->all_fragments[i].coeff_count&127)){
1482 s->all_fragments[i].coeff_count= 129;
1483 // s->all_fragments[i].next_coeff= s->next_coeff;
1484 s->coeffs[i].next= s->next_coeff;
1485 (s->next_coeff++)->next=NULL;
1493 static void horizontal_filter(unsigned char *first_pixel, int stride,
1494 int *bounding_values);
1495 static void vertical_filter(unsigned char *first_pixel, int stride,
1496 int *bounding_values);
1499 * Perform the final rendering for a particular slice of data.
1500 * The slice number ranges from 0..(macroblock_height - 1).
1502 static void render_slice(Vp3DecodeContext *s, int slice)
1506 int16_t *dequantizer;
1507 DECLARE_ALIGNED_16(DCTELEM, block[64]);
1508 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1509 int motion_halfpel_index;
1510 uint8_t *motion_source;
1512 int current_macroblock_entry = slice * s->macroblock_width * 6;
1514 if (slice >= s->macroblock_height)
1517 for (plane = 0; plane < 3; plane++) {
1518 uint8_t *output_plane = s->current_frame.data [plane];
1519 uint8_t * last_plane = s-> last_frame.data [plane];
1520 uint8_t *golden_plane = s-> golden_frame.data [plane];
1521 int stride = s->current_frame.linesize[plane];
1522 int plane_width = s->width >> !!plane;
1523 int plane_height = s->height >> !!plane;
1524 int y = slice * FRAGMENT_PIXELS << !plane ;
1525 int slice_height = y + (FRAGMENT_PIXELS << !plane);
1526 int i = s->macroblock_fragments[current_macroblock_entry + plane + 3*!!plane];
1528 if (!s->flipped_image) stride = -stride;
1531 if(FFABS(stride) > 2048)
1532 return; //various tables are fixed size
1534 /* for each fragment row in the slice (both of them)... */
1535 for (; y < slice_height; y += 8) {
1537 /* for each fragment in a row... */
1538 for (x = 0; x < plane_width; x += 8, i++) {
1540 if ((i < 0) || (i >= s->fragment_count)) {
1541 av_log(s->avctx, AV_LOG_ERROR, " vp3:render_slice(): bad fragment number (%d)\n", i);
1545 /* transform if this block was coded */
1546 if ((s->all_fragments[i].coding_method != MODE_COPY) &&
1547 !((s->avctx->flags & CODEC_FLAG_GRAY) && plane)) {
1549 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1550 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1551 motion_source= golden_plane;
1553 motion_source= last_plane;
1555 motion_source += s->all_fragments[i].first_pixel;
1556 motion_halfpel_index = 0;
1558 /* sort out the motion vector if this fragment is coded
1559 * using a motion vector method */
1560 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1561 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1563 motion_x = s->all_fragments[i].motion_x;
1564 motion_y = s->all_fragments[i].motion_y;
1566 motion_x= (motion_x>>1) | (motion_x&1);
1567 motion_y= (motion_y>>1) | (motion_y&1);
1570 src_x= (motion_x>>1) + x;
1571 src_y= (motion_y>>1) + y;
1572 if ((motion_x == 127) || (motion_y == 127))
1573 av_log(s->avctx, AV_LOG_ERROR, " help! got invalid motion vector! (%X, %X)\n", motion_x, motion_y);
1575 motion_halfpel_index = motion_x & 0x01;
1576 motion_source += (motion_x >> 1);
1578 motion_halfpel_index |= (motion_y & 0x01) << 1;
1579 motion_source += ((motion_y >> 1) * stride);
1581 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1582 uint8_t *temp= s->edge_emu_buffer;
1583 if(stride<0) temp -= 9*stride;
1584 else temp += 9*stride;
1586 ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1587 motion_source= temp;
1592 /* first, take care of copying a block from either the
1593 * previous or the golden frame */
1594 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1595 /* Note, it is possible to implement all MC cases with
1596 put_no_rnd_pixels_l2 which would look more like the
1597 VP3 source but this would be slower as
1598 put_no_rnd_pixels_tab is better optimzed */
1599 if(motion_halfpel_index != 3){
1600 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1601 output_plane + s->all_fragments[i].first_pixel,
1602 motion_source, stride, 8);
1604 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1605 s->dsp.put_no_rnd_pixels_l2[1](
1606 output_plane + s->all_fragments[i].first_pixel,
1608 motion_source + stride + 1 + d,
1611 dequantizer = s->qmat[1][plane];
1613 dequantizer = s->qmat[0][plane];
1616 /* dequantize the DCT coefficients */
1617 debug_idct("fragment %d, coding mode %d, DC = %d, dequant = %d:\n",
1618 i, s->all_fragments[i].coding_method,
1619 DC_COEFF(i), dequantizer[0]);
1621 if(s->avctx->idct_algo==FF_IDCT_VP3){
1622 Coeff *coeff= s->coeffs + i;
1623 memset(block, 0, sizeof(block));
1625 block[coeff->index]= coeff->coeff * dequantizer[coeff->index];
1629 Coeff *coeff= s->coeffs + i;
1630 memset(block, 0, sizeof(block));
1632 block[coeff->index]= (coeff->coeff * dequantizer[coeff->index] + 2)>>2;
1637 /* invert DCT and place (or add) in final output */
1639 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1640 if(s->avctx->idct_algo!=FF_IDCT_VP3)
1643 output_plane + s->all_fragments[i].first_pixel,
1648 output_plane + s->all_fragments[i].first_pixel,
1653 debug_idct("block after idct_%s():\n",
1654 (s->all_fragments[i].coding_method == MODE_INTRA)?
1656 for (m = 0; m < 8; m++) {
1657 for (n = 0; n < 8; n++) {
1658 debug_idct(" %3d", *(output_plane +
1659 s->all_fragments[i].first_pixel + (m * stride + n)));
1667 /* copy directly from the previous frame */
1668 s->dsp.put_pixels_tab[1][0](
1669 output_plane + s->all_fragments[i].first_pixel,
1670 last_plane + s->all_fragments[i].first_pixel,
1675 /* perform the left edge filter if:
1676 * - the fragment is not on the left column
1677 * - the fragment is coded in this frame
1678 * - the fragment is not coded in this frame but the left
1679 * fragment is coded in this frame (this is done instead
1680 * of a right edge filter when rendering the left fragment
1681 * since this fragment is not available yet) */
1683 ((s->all_fragments[i].coding_method != MODE_COPY) ||
1684 ((s->all_fragments[i].coding_method == MODE_COPY) &&
1685 (s->all_fragments[i - 1].coding_method != MODE_COPY)) )) {
1687 output_plane + s->all_fragments[i].first_pixel + 7*stride,
1688 -stride, s->bounding_values_array + 127);
1691 /* perform the top edge filter if:
1692 * - the fragment is not on the top row
1693 * - the fragment is coded in this frame
1694 * - the fragment is not coded in this frame but the above
1695 * fragment is coded in this frame (this is done instead
1696 * of a bottom edge filter when rendering the above
1697 * fragment since this fragment is not available yet) */
1699 ((s->all_fragments[i].coding_method != MODE_COPY) ||
1700 ((s->all_fragments[i].coding_method == MODE_COPY) &&
1701 (s->all_fragments[i - fragment_width].coding_method != MODE_COPY)) )) {
1703 output_plane + s->all_fragments[i].first_pixel - stride,
1704 -stride, s->bounding_values_array + 127);
1711 /* this looks like a good place for slice dispatch... */
1713 * if (slice == s->macroblock_height - 1)
1714 * dispatch (both last slice & 2nd-to-last slice);
1715 * else if (slice > 0)
1716 * dispatch (slice - 1);
1722 static void horizontal_filter(unsigned char *first_pixel, int stride,
1723 int *bounding_values)
1728 for (end= first_pixel + 8*stride; first_pixel != end; first_pixel += stride) {
1730 (first_pixel[-2] - first_pixel[ 1])
1731 +3*(first_pixel[ 0] - first_pixel[-1]);
1732 filter_value = bounding_values[(filter_value + 4) >> 3];
1733 first_pixel[-1] = av_clip_uint8(first_pixel[-1] + filter_value);
1734 first_pixel[ 0] = av_clip_uint8(first_pixel[ 0] - filter_value);
1738 static void vertical_filter(unsigned char *first_pixel, int stride,
1739 int *bounding_values)
1743 const int nstride= -stride;
1745 for (end= first_pixel + 8; first_pixel < end; first_pixel++) {
1747 (first_pixel[2 * nstride] - first_pixel[ stride])
1748 +3*(first_pixel[0 ] - first_pixel[nstride]);
1749 filter_value = bounding_values[(filter_value + 4) >> 3];
1750 first_pixel[nstride] = av_clip_uint8(first_pixel[nstride] + filter_value);
1751 first_pixel[0] = av_clip_uint8(first_pixel[0] - filter_value);
1755 static void apply_loop_filter(Vp3DecodeContext *s)
1759 int *bounding_values= s->bounding_values_array+127;
1762 int bounding_values_array[256];
1765 /* find the right loop limit value */
1766 for (x = 63; x >= 0; x--) {
1767 if (vp31_ac_scale_factor[x] >= s->quality_index)
1770 filter_limit = vp31_filter_limit_values[s->quality_index];
1772 /* set up the bounding values */
1773 memset(bounding_values_array, 0, 256 * sizeof(int));
1774 for (x = 0; x < filter_limit; x++) {
1775 bounding_values[-x - filter_limit] = -filter_limit + x;
1776 bounding_values[-x] = -x;
1777 bounding_values[x] = x;
1778 bounding_values[x + filter_limit] = filter_limit - x;
1782 for (plane = 0; plane < 3; plane++) {
1783 int width = s->fragment_width >> !!plane;
1784 int height = s->fragment_height >> !!plane;
1785 int fragment = s->fragment_start [plane];
1786 int stride = s->current_frame.linesize[plane];
1787 uint8_t *plane_data = s->current_frame.data [plane];
1788 if (!s->flipped_image) stride = -stride;
1790 for (y = 0; y < height; y++) {
1792 for (x = 0; x < width; x++) {
1794 /* do not perform left edge filter for left columns frags */
1796 (s->all_fragments[fragment].coding_method != MODE_COPY)) {
1798 plane_data + s->all_fragments[fragment].first_pixel,
1799 stride, bounding_values);
1802 /* do not perform top edge filter for top row fragments */
1804 (s->all_fragments[fragment].coding_method != MODE_COPY)) {
1806 plane_data + s->all_fragments[fragment].first_pixel,
1807 stride, bounding_values);
1810 /* do not perform right edge filter for right column
1811 * fragments or if right fragment neighbor is also coded
1812 * in this frame (it will be filtered in next iteration) */
1813 if ((x < width - 1) &&
1814 (s->all_fragments[fragment].coding_method != MODE_COPY) &&
1815 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1817 plane_data + s->all_fragments[fragment + 1].first_pixel,
1818 stride, bounding_values);
1821 /* do not perform bottom edge filter for bottom row
1822 * fragments or if bottom fragment neighbor is also coded
1823 * in this frame (it will be filtered in the next row) */
1824 if ((y < height - 1) &&
1825 (s->all_fragments[fragment].coding_method != MODE_COPY) &&
1826 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1828 plane_data + s->all_fragments[fragment + width].first_pixel,
1829 stride, bounding_values);
1833 STOP_TIMER("loop filter")
1840 * This function computes the first pixel addresses for each fragment.
1841 * This function needs to be invoked after the first frame is allocated
1842 * so that it has access to the plane strides.
1844 static void vp3_calculate_pixel_addresses(Vp3DecodeContext *s)
1849 /* figure out the first pixel addresses for each of the fragments */
1852 for (y = s->fragment_height; y > 0; y--) {
1853 for (x = 0; x < s->fragment_width; x++) {
1854 s->all_fragments[i++].first_pixel =
1855 s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
1856 s->golden_frame.linesize[0] +
1857 x * FRAGMENT_PIXELS;
1858 debug_init(" fragment %d, first pixel @ %d\n",
1859 i-1, s->all_fragments[i-1].first_pixel);
1864 i = s->fragment_start[1];
1865 for (y = s->fragment_height / 2; y > 0; y--) {
1866 for (x = 0; x < s->fragment_width / 2; x++) {
1867 s->all_fragments[i++].first_pixel =
1868 s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
1869 s->golden_frame.linesize[1] +
1870 x * FRAGMENT_PIXELS;
1871 debug_init(" fragment %d, first pixel @ %d\n",
1872 i-1, s->all_fragments[i-1].first_pixel);
1877 i = s->fragment_start[2];
1878 for (y = s->fragment_height / 2; y > 0; y--) {
1879 for (x = 0; x < s->fragment_width / 2; x++) {
1880 s->all_fragments[i++].first_pixel =
1881 s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
1882 s->golden_frame.linesize[2] +
1883 x * FRAGMENT_PIXELS;
1884 debug_init(" fragment %d, first pixel @ %d\n",
1885 i-1, s->all_fragments[i-1].first_pixel);
1890 /* FIXME: this should be merged with the above! */
1891 static void theora_calculate_pixel_addresses(Vp3DecodeContext *s)
1896 /* figure out the first pixel addresses for each of the fragments */
1899 for (y = 1; y <= s->fragment_height; y++) {
1900 for (x = 0; x < s->fragment_width; x++) {
1901 s->all_fragments[i++].first_pixel =
1902 s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
1903 s->golden_frame.linesize[0] +
1904 x * FRAGMENT_PIXELS;
1905 debug_init(" fragment %d, first pixel @ %d\n",
1906 i-1, s->all_fragments[i-1].first_pixel);
1911 i = s->fragment_start[1];
1912 for (y = 1; y <= s->fragment_height / 2; y++) {
1913 for (x = 0; x < s->fragment_width / 2; x++) {
1914 s->all_fragments[i++].first_pixel =
1915 s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
1916 s->golden_frame.linesize[1] +
1917 x * FRAGMENT_PIXELS;
1918 debug_init(" fragment %d, first pixel @ %d\n",
1919 i-1, s->all_fragments[i-1].first_pixel);
1924 i = s->fragment_start[2];
1925 for (y = 1; y <= s->fragment_height / 2; y++) {
1926 for (x = 0; x < s->fragment_width / 2; x++) {
1927 s->all_fragments[i++].first_pixel =
1928 s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
1929 s->golden_frame.linesize[2] +
1930 x * FRAGMENT_PIXELS;
1931 debug_init(" fragment %d, first pixel @ %d\n",
1932 i-1, s->all_fragments[i-1].first_pixel);
1938 * This is the ffmpeg/libavcodec API init function.
1940 static int vp3_decode_init(AVCodecContext *avctx)
1942 Vp3DecodeContext *s = avctx->priv_data;
1943 int i, inter, plane;
1946 int y_superblock_count;
1947 int c_superblock_count;
1949 if (avctx->codec_tag == MKTAG('V','P','3','0'))
1955 s->width = (avctx->width + 15) & 0xFFFFFFF0;
1956 s->height = (avctx->height + 15) & 0xFFFFFFF0;
1957 avctx->pix_fmt = PIX_FMT_YUV420P;
1958 if(avctx->idct_algo==FF_IDCT_AUTO)
1959 avctx->idct_algo=FF_IDCT_VP3;
1960 dsputil_init(&s->dsp, avctx);
1962 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1964 /* initialize to an impossible value which will force a recalculation
1965 * in the first frame decode */
1966 s->quality_index = -1;
1968 s->y_superblock_width = (s->width + 31) / 32;
1969 s->y_superblock_height = (s->height + 31) / 32;
1970 y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1972 /* work out the dimensions for the C planes */
1973 c_width = s->width / 2;
1974 c_height = s->height / 2;
1975 s->c_superblock_width = (c_width + 31) / 32;
1976 s->c_superblock_height = (c_height + 31) / 32;
1977 c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1979 s->superblock_count = y_superblock_count + (c_superblock_count * 2);
1980 s->u_superblock_start = y_superblock_count;
1981 s->v_superblock_start = s->u_superblock_start + c_superblock_count;
1982 s->superblock_coding = av_malloc(s->superblock_count);
1984 s->macroblock_width = (s->width + 15) / 16;
1985 s->macroblock_height = (s->height + 15) / 16;
1986 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1988 s->fragment_width = s->width / FRAGMENT_PIXELS;
1989 s->fragment_height = s->height / FRAGMENT_PIXELS;
1991 /* fragment count covers all 8x8 blocks for all 3 planes */
1992 s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
1993 s->fragment_start[1] = s->fragment_width * s->fragment_height;
1994 s->fragment_start[2] = s->fragment_width * s->fragment_height * 5 / 4;
1996 debug_init(" Y plane: %d x %d\n", s->width, s->height);
1997 debug_init(" C plane: %d x %d\n", c_width, c_height);
1998 debug_init(" Y superblocks: %d x %d, %d total\n",
1999 s->y_superblock_width, s->y_superblock_height, y_superblock_count);
2000 debug_init(" C superblocks: %d x %d, %d total\n",
2001 s->c_superblock_width, s->c_superblock_height, c_superblock_count);
2002 debug_init(" total superblocks = %d, U starts @ %d, V starts @ %d\n",
2003 s->superblock_count, s->u_superblock_start, s->v_superblock_start);
2004 debug_init(" macroblocks: %d x %d, %d total\n",
2005 s->macroblock_width, s->macroblock_height, s->macroblock_count);
2006 debug_init(" %d fragments, %d x %d, u starts @ %d, v starts @ %d\n",
2010 s->fragment_start[1],
2011 s->fragment_start[2]);
2013 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
2014 s->coeffs = av_malloc(s->fragment_count * sizeof(Coeff) * 65);
2015 s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int));
2016 s->pixel_addresses_inited = 0;
2018 if (!s->theora_tables)
2020 for (i = 0; i < 64; i++) {
2021 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
2022 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
2023 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
2024 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
2025 s->base_matrix[2][i] = vp31_inter_dequant[i];
2026 s->filter_limit_values[i] = vp31_filter_limit_values[i];
2029 for(inter=0; inter<2; inter++){
2030 for(plane=0; plane<3; plane++){
2031 s->qr_count[inter][plane]= 1;
2032 s->qr_size [inter][plane][0]= 63;
2033 s->qr_base [inter][plane][0]=
2034 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
2038 /* init VLC tables */
2039 for (i = 0; i < 16; i++) {
2042 init_vlc(&s->dc_vlc[i], 5, 32,
2043 &dc_bias[i][0][1], 4, 2,
2044 &dc_bias[i][0][0], 4, 2, 0);
2046 /* group 1 AC histograms */
2047 init_vlc(&s->ac_vlc_1[i], 5, 32,
2048 &ac_bias_0[i][0][1], 4, 2,
2049 &ac_bias_0[i][0][0], 4, 2, 0);
2051 /* group 2 AC histograms */
2052 init_vlc(&s->ac_vlc_2[i], 5, 32,
2053 &ac_bias_1[i][0][1], 4, 2,
2054 &ac_bias_1[i][0][0], 4, 2, 0);
2056 /* group 3 AC histograms */
2057 init_vlc(&s->ac_vlc_3[i], 5, 32,
2058 &ac_bias_2[i][0][1], 4, 2,
2059 &ac_bias_2[i][0][0], 4, 2, 0);
2061 /* group 4 AC histograms */
2062 init_vlc(&s->ac_vlc_4[i], 5, 32,
2063 &ac_bias_3[i][0][1], 4, 2,
2064 &ac_bias_3[i][0][0], 4, 2, 0);
2067 for (i = 0; i < 16; i++) {
2070 init_vlc(&s->dc_vlc[i], 5, 32,
2071 &s->huffman_table[i][0][1], 4, 2,
2072 &s->huffman_table[i][0][0], 4, 2, 0);
2074 /* group 1 AC histograms */
2075 init_vlc(&s->ac_vlc_1[i], 5, 32,
2076 &s->huffman_table[i+16][0][1], 4, 2,
2077 &s->huffman_table[i+16][0][0], 4, 2, 0);
2079 /* group 2 AC histograms */
2080 init_vlc(&s->ac_vlc_2[i], 5, 32,
2081 &s->huffman_table[i+16*2][0][1], 4, 2,
2082 &s->huffman_table[i+16*2][0][0], 4, 2, 0);
2084 /* group 3 AC histograms */
2085 init_vlc(&s->ac_vlc_3[i], 5, 32,
2086 &s->huffman_table[i+16*3][0][1], 4, 2,
2087 &s->huffman_table[i+16*3][0][0], 4, 2, 0);
2089 /* group 4 AC histograms */
2090 init_vlc(&s->ac_vlc_4[i], 5, 32,
2091 &s->huffman_table[i+16*4][0][1], 4, 2,
2092 &s->huffman_table[i+16*4][0][0], 4, 2, 0);
2096 init_vlc(&s->superblock_run_length_vlc, 6, 34,
2097 &superblock_run_length_vlc_table[0][1], 4, 2,
2098 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
2100 init_vlc(&s->fragment_run_length_vlc, 5, 30,
2101 &fragment_run_length_vlc_table[0][1], 4, 2,
2102 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
2104 init_vlc(&s->mode_code_vlc, 3, 8,
2105 &mode_code_vlc_table[0][1], 2, 1,
2106 &mode_code_vlc_table[0][0], 2, 1, 0);
2108 init_vlc(&s->motion_vector_vlc, 6, 63,
2109 &motion_vector_vlc_table[0][1], 2, 1,
2110 &motion_vector_vlc_table[0][0], 2, 1, 0);
2112 /* work out the block mapping tables */
2113 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
2114 s->superblock_macroblocks = av_malloc(s->superblock_count * 4 * sizeof(int));
2115 s->macroblock_fragments = av_malloc(s->macroblock_count * 6 * sizeof(int));
2116 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
2117 init_block_mapping(s);
2119 for (i = 0; i < 3; i++) {
2120 s->current_frame.data[i] = NULL;
2121 s->last_frame.data[i] = NULL;
2122 s->golden_frame.data[i] = NULL;
2129 * This is the ffmpeg/libavcodec API frame decode function.
2131 static int vp3_decode_frame(AVCodecContext *avctx,
2132 void *data, int *data_size,
2133 uint8_t *buf, int buf_size)
2135 Vp3DecodeContext *s = avctx->priv_data;
2137 static int counter = 0;
2140 init_get_bits(&gb, buf, buf_size * 8);
2142 if (s->theora && get_bits1(&gb))
2145 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
2148 int ptype = get_bits(&gb, 7);
2150 skip_bits(&gb, 6*8); /* "theora" */
2155 theora_decode_comments(avctx, &gb);
2158 theora_decode_tables(avctx, &gb);
2159 init_dequantizer(s);
2162 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype);
2168 s->keyframe = !get_bits1(&gb);
2171 s->last_quality_index = s->quality_index;
2175 s->qis[s->nqis++]= get_bits(&gb, 6);
2176 } while(s->theora >= 0x030200 && s->nqis<3 && get_bits1(&gb));
2178 s->quality_index= s->qis[0];
2180 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
2181 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
2182 s->keyframe?"key":"", counter, s->quality_index);
2185 if (s->quality_index != s->last_quality_index) {
2186 init_dequantizer(s);
2187 init_loop_filter(s);
2193 skip_bits(&gb, 4); /* width code */
2194 skip_bits(&gb, 4); /* height code */
2197 s->version = get_bits(&gb, 5);
2199 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
2202 if (s->version || s->theora)
2205 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
2206 skip_bits(&gb, 2); /* reserved? */
2209 if (s->last_frame.data[0] == s->golden_frame.data[0]) {
2210 if (s->golden_frame.data[0])
2211 avctx->release_buffer(avctx, &s->golden_frame);
2212 s->last_frame= s->golden_frame; /* ensure that we catch any access to this released frame */
2214 if (s->golden_frame.data[0])
2215 avctx->release_buffer(avctx, &s->golden_frame);
2216 if (s->last_frame.data[0])
2217 avctx->release_buffer(avctx, &s->last_frame);
2220 s->golden_frame.reference = 3;
2221 if(avctx->get_buffer(avctx, &s->golden_frame) < 0) {
2222 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
2226 /* golden frame is also the current frame */
2227 s->current_frame= s->golden_frame;
2229 /* time to figure out pixel addresses? */
2230 if (!s->pixel_addresses_inited)
2232 if (!s->flipped_image)
2233 vp3_calculate_pixel_addresses(s);
2235 theora_calculate_pixel_addresses(s);
2236 s->pixel_addresses_inited = 1;
2239 /* allocate a new current frame */
2240 s->current_frame.reference = 3;
2241 if (!s->pixel_addresses_inited) {
2242 av_log(s->avctx, AV_LOG_ERROR, "vp3: first frame not a keyframe\n");
2245 if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
2246 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
2251 s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
2252 s->current_frame.qstride= 0;
2256 STOP_TIMER("init_frame")}
2261 memcpy(s->current_frame.data[0], s->golden_frame.data[0],
2262 s->current_frame.linesize[0] * s->height);
2263 memcpy(s->current_frame.data[1], s->golden_frame.data[1],
2264 s->current_frame.linesize[1] * s->height / 2);
2265 memcpy(s->current_frame.data[2], s->golden_frame.data[2],
2266 s->current_frame.linesize[2] * s->height / 2);
2272 if (unpack_superblocks(s, &gb)){
2273 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
2276 STOP_TIMER("unpack_superblocks")}
2278 if (unpack_modes(s, &gb)){
2279 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
2282 STOP_TIMER("unpack_modes")}
2284 if (unpack_vectors(s, &gb)){
2285 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2288 STOP_TIMER("unpack_vectors")}
2290 if (unpack_dct_coeffs(s, &gb)){
2291 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
2294 STOP_TIMER("unpack_dct_coeffs")}
2297 reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
2298 if ((avctx->flags & CODEC_FLAG_GRAY) == 0) {
2299 reverse_dc_prediction(s, s->fragment_start[1],
2300 s->fragment_width / 2, s->fragment_height / 2);
2301 reverse_dc_prediction(s, s->fragment_start[2],
2302 s->fragment_width / 2, s->fragment_height / 2);
2304 STOP_TIMER("reverse_dc_prediction")}
2307 for (i = 0; i < s->macroblock_height; i++)
2309 STOP_TIMER("render_fragments")}
2312 apply_loop_filter(s);
2313 STOP_TIMER("apply_loop_filter")}
2318 *data_size=sizeof(AVFrame);
2319 *(AVFrame*)data= s->current_frame;
2321 /* release the last frame, if it is allocated and if it is not the
2323 if ((s->last_frame.data[0]) &&
2324 (s->last_frame.data[0] != s->golden_frame.data[0]))
2325 avctx->release_buffer(avctx, &s->last_frame);
2327 /* shuffle frames (last = current) */
2328 s->last_frame= s->current_frame;
2329 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
2335 * This is the ffmpeg/libavcodec API module cleanup function.
2337 static int vp3_decode_end(AVCodecContext *avctx)
2339 Vp3DecodeContext *s = avctx->priv_data;
2341 av_free(s->all_fragments);
2343 av_free(s->coded_fragment_list);
2344 av_free(s->superblock_fragments);
2345 av_free(s->superblock_macroblocks);
2346 av_free(s->macroblock_fragments);
2347 av_free(s->macroblock_coding);
2349 /* release all frames */
2350 if (s->golden_frame.data[0] && s->golden_frame.data[0] != s->last_frame.data[0])
2351 avctx->release_buffer(avctx, &s->golden_frame);
2352 if (s->last_frame.data[0])
2353 avctx->release_buffer(avctx, &s->last_frame);
2354 /* no need to release the current_frame since it will always be pointing
2355 * to the same frame as either the golden or last frame */
2360 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2362 Vp3DecodeContext *s = avctx->priv_data;
2364 if (get_bits(gb, 1)) {
2366 if (s->entries >= 32) { /* overflow */
2367 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2370 token = get_bits(gb, 5);
2371 //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);
2372 s->huffman_table[s->hti][token][0] = s->hbits;
2373 s->huffman_table[s->hti][token][1] = s->huff_code_size;
2377 if (s->huff_code_size >= 32) {/* overflow */
2378 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2381 s->huff_code_size++;
2383 read_huffman_tree(avctx, gb);
2385 read_huffman_tree(avctx, gb);
2387 s->huff_code_size--;
2392 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2394 Vp3DecodeContext *s = avctx->priv_data;
2396 s->theora = get_bits_long(gb, 24);
2397 av_log(avctx, AV_LOG_INFO, "Theora bitstream version %X\n", s->theora);
2399 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2400 /* but previous versions have the image flipped relative to vp3 */
2401 if (s->theora < 0x030200)
2403 s->flipped_image = 1;
2404 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2407 s->width = get_bits(gb, 16) << 4;
2408 s->height = get_bits(gb, 16) << 4;
2410 if(avcodec_check_dimensions(avctx, s->width, s->height)){
2411 av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2412 s->width= s->height= 0;
2416 if (s->theora >= 0x030400)
2418 skip_bits(gb, 32); /* total number of superblocks in a frame */
2419 // fixme, the next field is 36bits long
2420 skip_bits(gb, 32); /* total number of blocks in a frame */
2421 skip_bits(gb, 4); /* total number of blocks in a frame */
2422 skip_bits(gb, 32); /* total number of macroblocks in a frame */
2424 skip_bits(gb, 24); /* frame width */
2425 skip_bits(gb, 24); /* frame height */
2429 skip_bits(gb, 24); /* frame width */
2430 skip_bits(gb, 24); /* frame height */
2433 if (s->theora >= 0x030200) {
2434 skip_bits(gb, 8); /* offset x */
2435 skip_bits(gb, 8); /* offset y */
2438 skip_bits(gb, 32); /* fps numerator */
2439 skip_bits(gb, 32); /* fps denumerator */
2440 skip_bits(gb, 24); /* aspect numerator */
2441 skip_bits(gb, 24); /* aspect denumerator */
2443 if (s->theora < 0x030200)
2444 skip_bits(gb, 5); /* keyframe frequency force */
2445 skip_bits(gb, 8); /* colorspace */
2446 if (s->theora >= 0x030400)
2447 skip_bits(gb, 2); /* pixel format: 420,res,422,444 */
2448 skip_bits(gb, 24); /* bitrate */
2450 skip_bits(gb, 6); /* quality hint */
2452 if (s->theora >= 0x030200)
2454 skip_bits(gb, 5); /* keyframe frequency force */
2456 if (s->theora < 0x030400)
2457 skip_bits(gb, 5); /* spare bits */
2460 // align_get_bits(gb);
2462 avctx->width = s->width;
2463 avctx->height = s->height;
2468 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2470 Vp3DecodeContext *s = avctx->priv_data;
2471 int i, n, matrices, inter, plane;
2473 if (s->theora >= 0x030200) {
2474 n = get_bits(gb, 3);
2475 /* loop filter limit values table */
2476 for (i = 0; i < 64; i++)
2477 s->filter_limit_values[i] = get_bits(gb, n);
2480 if (s->theora >= 0x030200)
2481 n = get_bits(gb, 4) + 1;
2484 /* quality threshold table */
2485 for (i = 0; i < 64; i++)
2486 s->coded_ac_scale_factor[i] = get_bits(gb, n);
2488 if (s->theora >= 0x030200)
2489 n = get_bits(gb, 4) + 1;
2492 /* dc scale factor table */
2493 for (i = 0; i < 64; i++)
2494 s->coded_dc_scale_factor[i] = get_bits(gb, n);
2496 if (s->theora >= 0x030200)
2497 matrices = get_bits(gb, 9) + 1;
2502 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2506 for(n=0; n<matrices; n++){
2507 for (i = 0; i < 64; i++)
2508 s->base_matrix[n][i]= get_bits(gb, 8);
2511 for (inter = 0; inter <= 1; inter++) {
2512 for (plane = 0; plane <= 2; plane++) {
2514 if (inter || plane > 0)
2515 newqr = get_bits(gb, 1);
2518 if(inter && get_bits(gb, 1)){
2522 qtj= (3*inter + plane - 1) / 3;
2523 plj= (plane + 2) % 3;
2525 s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2526 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2527 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2533 i= get_bits(gb, av_log2(matrices-1)+1);
2535 av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2538 s->qr_base[inter][plane][qri]= i;
2541 i = get_bits(gb, av_log2(63-qi)+1) + 1;
2542 s->qr_size[inter][plane][qri++]= i;
2547 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2550 s->qr_count[inter][plane]= qri;
2555 /* Huffman tables */
2556 for (s->hti = 0; s->hti < 80; s->hti++) {
2558 s->huff_code_size = 1;
2559 if (!get_bits(gb, 1)) {
2561 read_huffman_tree(avctx, gb);
2563 read_huffman_tree(avctx, gb);
2567 s->theora_tables = 1;
2572 static int theora_decode_init(AVCodecContext *avctx)
2574 Vp3DecodeContext *s = avctx->priv_data;
2577 uint8_t *header_start[3];
2583 if (!avctx->extradata_size)
2585 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2589 if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2590 42, header_start, header_len) < 0) {
2591 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2596 init_get_bits(&gb, header_start[i], header_len[i]);
2598 ptype = get_bits(&gb, 8);
2599 debug_vp3("Theora headerpacket type: %x\n", ptype);
2601 if (!(ptype & 0x80))
2603 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2607 // FIXME: Check for this as well.
2608 skip_bits(&gb, 6*8); /* "theora" */
2613 theora_decode_header(avctx, &gb);
2616 // FIXME: is this needed? it breaks sometimes
2617 // theora_decode_comments(avctx, gb);
2620 theora_decode_tables(avctx, &gb);
2623 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2626 if(8*header_len[i] != get_bits_count(&gb))
2627 av_log(avctx, AV_LOG_ERROR, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2628 if (s->theora < 0x030200)
2632 vp3_decode_init(avctx);
2636 AVCodec vp3_decoder = {
2640 sizeof(Vp3DecodeContext),
2649 #ifdef CONFIG_THEORA_DECODER
2650 AVCodec theora_decoder = {
2654 sizeof(Vp3DecodeContext),