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 /************************************************************************
335 * VP3 specific functions
336 ************************************************************************/
339 * This function sets up all of the various blocks mappings:
340 * superblocks <-> fragments, macroblocks <-> fragments,
341 * superblocks <-> macroblocks
343 * Returns 0 is successful; returns 1 if *anything* went wrong.
345 static int init_block_mapping(Vp3DecodeContext *s)
348 signed int hilbert_walk_mb[4];
350 int current_fragment = 0;
351 int current_width = 0;
352 int current_height = 0;
355 int superblock_row_inc = 0;
357 int mapping_index = 0;
359 int current_macroblock;
362 signed char travel_width[16] = {
369 signed char travel_height[16] = {
376 signed char travel_width_mb[4] = {
380 signed char travel_height_mb[4] = {
384 debug_vp3(" vp3: initialize block mapping tables\n");
386 hilbert_walk_mb[0] = 1;
387 hilbert_walk_mb[1] = s->macroblock_width;
388 hilbert_walk_mb[2] = 1;
389 hilbert_walk_mb[3] = -s->macroblock_width;
391 /* iterate through each superblock (all planes) and map the fragments */
392 for (i = 0; i < s->superblock_count; i++) {
393 debug_init(" superblock %d (u starts @ %d, v starts @ %d)\n",
394 i, s->u_superblock_start, s->v_superblock_start);
396 /* time to re-assign the limits? */
399 /* start of Y superblocks */
400 right_edge = s->fragment_width;
401 bottom_edge = s->fragment_height;
404 superblock_row_inc = 3 * s->fragment_width -
405 (s->y_superblock_width * 4 - s->fragment_width);
407 /* the first operation for this variable is to advance by 1 */
408 current_fragment = -1;
410 } else if (i == s->u_superblock_start) {
412 /* start of U superblocks */
413 right_edge = s->fragment_width / 2;
414 bottom_edge = s->fragment_height / 2;
417 superblock_row_inc = 3 * (s->fragment_width / 2) -
418 (s->c_superblock_width * 4 - s->fragment_width / 2);
420 /* the first operation for this variable is to advance by 1 */
421 current_fragment = s->fragment_start[1] - 1;
423 } else if (i == s->v_superblock_start) {
425 /* start of V superblocks */
426 right_edge = s->fragment_width / 2;
427 bottom_edge = s->fragment_height / 2;
430 superblock_row_inc = 3 * (s->fragment_width / 2) -
431 (s->c_superblock_width * 4 - s->fragment_width / 2);
433 /* the first operation for this variable is to advance by 1 */
434 current_fragment = s->fragment_start[2] - 1;
438 if (current_width >= right_edge - 1) {
439 /* reset width and move to next superblock row */
443 /* fragment is now at the start of a new superblock row */
444 current_fragment += superblock_row_inc;
447 /* iterate through all 16 fragments in a superblock */
448 for (j = 0; j < 16; j++) {
449 current_fragment += travel_width[j] + right_edge * travel_height[j];
450 current_width += travel_width[j];
451 current_height += travel_height[j];
453 /* check if the fragment is in bounds */
454 if ((current_width < right_edge) &&
455 (current_height < bottom_edge)) {
456 s->superblock_fragments[mapping_index] = current_fragment;
457 debug_init(" mapping fragment %d to superblock %d, position %d (%d/%d x %d/%d)\n",
458 s->superblock_fragments[mapping_index], i, j,
459 current_width, right_edge, current_height, bottom_edge);
461 s->superblock_fragments[mapping_index] = -1;
462 debug_init(" superblock %d, position %d has no fragment (%d/%d x %d/%d)\n",
464 current_width, right_edge, current_height, bottom_edge);
471 /* initialize the superblock <-> macroblock mapping; iterate through
472 * all of the Y plane superblocks to build this mapping */
473 right_edge = s->macroblock_width;
474 bottom_edge = s->macroblock_height;
477 superblock_row_inc = s->macroblock_width -
478 (s->y_superblock_width * 2 - s->macroblock_width);;
479 hilbert = hilbert_walk_mb;
481 current_macroblock = -1;
482 for (i = 0; i < s->u_superblock_start; i++) {
484 if (current_width >= right_edge - 1) {
485 /* reset width and move to next superblock row */
489 /* macroblock is now at the start of a new superblock row */
490 current_macroblock += superblock_row_inc;
493 /* iterate through each potential macroblock in the superblock */
494 for (j = 0; j < 4; j++) {
495 current_macroblock += hilbert_walk_mb[j];
496 current_width += travel_width_mb[j];
497 current_height += travel_height_mb[j];
499 /* check if the macroblock is in bounds */
500 if ((current_width < right_edge) &&
501 (current_height < bottom_edge)) {
502 s->superblock_macroblocks[mapping_index] = current_macroblock;
503 debug_init(" mapping macroblock %d to superblock %d, position %d (%d/%d x %d/%d)\n",
504 s->superblock_macroblocks[mapping_index], i, j,
505 current_width, right_edge, current_height, bottom_edge);
507 s->superblock_macroblocks[mapping_index] = -1;
508 debug_init(" superblock %d, position %d has no macroblock (%d/%d x %d/%d)\n",
510 current_width, right_edge, current_height, bottom_edge);
517 /* initialize the macroblock <-> fragment mapping */
518 current_fragment = 0;
519 current_macroblock = 0;
521 for (i = 0; i < s->fragment_height; i += 2) {
523 for (j = 0; j < s->fragment_width; j += 2) {
525 debug_init(" macroblock %d contains fragments: ", current_macroblock);
526 s->all_fragments[current_fragment].macroblock = current_macroblock;
527 s->macroblock_fragments[mapping_index++] = current_fragment;
528 debug_init("%d ", current_fragment);
530 if (j + 1 < s->fragment_width) {
531 s->all_fragments[current_fragment + 1].macroblock = current_macroblock;
532 s->macroblock_fragments[mapping_index++] = current_fragment + 1;
533 debug_init("%d ", current_fragment + 1);
535 s->macroblock_fragments[mapping_index++] = -1;
537 if (i + 1 < s->fragment_height) {
538 s->all_fragments[current_fragment + s->fragment_width].macroblock =
540 s->macroblock_fragments[mapping_index++] =
541 current_fragment + s->fragment_width;
542 debug_init("%d ", current_fragment + s->fragment_width);
544 s->macroblock_fragments[mapping_index++] = -1;
546 if ((j + 1 < s->fragment_width) && (i + 1 < s->fragment_height)) {
547 s->all_fragments[current_fragment + s->fragment_width + 1].macroblock =
549 s->macroblock_fragments[mapping_index++] =
550 current_fragment + s->fragment_width + 1;
551 debug_init("%d ", current_fragment + s->fragment_width + 1);
553 s->macroblock_fragments[mapping_index++] = -1;
556 c_fragment = s->fragment_start[1] +
557 (i * s->fragment_width / 4) + (j / 2);
558 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
559 s->macroblock_fragments[mapping_index++] = c_fragment;
560 debug_init("%d ", c_fragment);
562 c_fragment = s->fragment_start[2] +
563 (i * s->fragment_width / 4) + (j / 2);
564 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
565 s->macroblock_fragments[mapping_index++] = c_fragment;
566 debug_init("%d ", c_fragment);
570 if (j + 2 <= s->fragment_width)
571 current_fragment += 2;
574 current_macroblock++;
577 current_fragment += s->fragment_width;
580 return 0; /* successful path out */
584 * This function wipes out all of the fragment data.
586 static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
590 /* zero out all of the fragment information */
591 s->coded_fragment_list_index = 0;
592 for (i = 0; i < s->fragment_count; i++) {
593 s->all_fragments[i].coeff_count = 0;
594 s->all_fragments[i].motion_x = 127;
595 s->all_fragments[i].motion_y = 127;
596 s->all_fragments[i].next_coeff= NULL;
598 s->coeffs[i].coeff=0;
599 s->coeffs[i].next= NULL;
604 * This function sets up the dequantization tables used for a particular
607 static void init_dequantizer(Vp3DecodeContext *s)
609 int ac_scale_factor = s->coded_ac_scale_factor[s->quality_index];
610 int dc_scale_factor = s->coded_dc_scale_factor[s->quality_index];
611 int i, plane, inter, qri, bmi, bmj, qistart;
613 debug_vp3(" vp3: initializing dequantization tables\n");
615 for(inter=0; inter<2; inter++){
616 for(plane=0; plane<3; plane++){
618 for(qri=0; qri<s->qr_count[inter][plane]; qri++){
619 sum+= s->qr_size[inter][plane][qri];
620 if(s->quality_index <= sum)
623 qistart= sum - s->qr_size[inter][plane][qri];
624 bmi= s->qr_base[inter][plane][qri ];
625 bmj= s->qr_base[inter][plane][qri+1];
627 int coeff= ( 2*(sum -s->quality_index)*s->base_matrix[bmi][i]
628 - 2*(qistart-s->quality_index)*s->base_matrix[bmj][i]
629 + s->qr_size[inter][plane][qri])
630 / (2*s->qr_size[inter][plane][qri]);
632 int qmin= 8<<(inter + !i);
633 int qscale= i ? ac_scale_factor : dc_scale_factor;
635 s->qmat[inter][plane][i]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
640 memset(s->qscale_table, (FFMAX(s->qmat[0][0][1], s->qmat[0][1][1])+8)/16, 512); //FIXME finetune
644 * This function initializes the loop filter boundary limits if the frame's
645 * quality index is different from the previous frame's.
647 static void init_loop_filter(Vp3DecodeContext *s)
649 int *bounding_values= s->bounding_values_array+127;
653 filter_limit = s->filter_limit_values[s->quality_index];
655 /* set up the bounding values */
656 memset(s->bounding_values_array, 0, 256 * sizeof(int));
657 for (x = 0; x < filter_limit; x++) {
658 bounding_values[-x - filter_limit] = -filter_limit + x;
659 bounding_values[-x] = -x;
660 bounding_values[x] = x;
661 bounding_values[x + filter_limit] = filter_limit - x;
666 * This function unpacks all of the superblock/macroblock/fragment coding
667 * information from the bitstream.
669 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
672 int current_superblock = 0;
674 int decode_fully_flags = 0;
675 int decode_partial_blocks = 0;
676 int first_c_fragment_seen;
679 int current_fragment;
681 debug_vp3(" vp3: unpacking superblock coding\n");
685 debug_vp3(" keyframe-- all superblocks are fully coded\n");
686 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
690 /* unpack the list of partially-coded superblocks */
691 bit = get_bits(gb, 1);
692 /* toggle the bit because as soon as the first run length is
693 * fetched the bit will be toggled again */
695 while (current_superblock < s->superblock_count) {
696 if (current_run-- == 0) {
698 current_run = get_vlc2(gb,
699 s->superblock_run_length_vlc.table, 6, 2);
700 if (current_run == 33)
701 current_run += get_bits(gb, 12);
702 debug_block_coding(" setting superblocks %d..%d to %s\n",
704 current_superblock + current_run - 1,
705 (bit) ? "partially coded" : "not coded");
707 /* if any of the superblocks are not partially coded, flag
708 * a boolean to decode the list of fully-coded superblocks */
710 decode_fully_flags = 1;
713 /* make a note of the fact that there are partially coded
715 decode_partial_blocks = 1;
718 s->superblock_coding[current_superblock++] = bit;
721 /* unpack the list of fully coded superblocks if any of the blocks were
722 * not marked as partially coded in the previous step */
723 if (decode_fully_flags) {
725 current_superblock = 0;
727 bit = get_bits(gb, 1);
728 /* toggle the bit because as soon as the first run length is
729 * fetched the bit will be toggled again */
731 while (current_superblock < s->superblock_count) {
733 /* skip any superblocks already marked as partially coded */
734 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
736 if (current_run-- == 0) {
738 current_run = get_vlc2(gb,
739 s->superblock_run_length_vlc.table, 6, 2);
740 if (current_run == 33)
741 current_run += get_bits(gb, 12);
744 debug_block_coding(" setting superblock %d to %s\n",
746 (bit) ? "fully coded" : "not coded");
747 s->superblock_coding[current_superblock] = 2*bit;
749 current_superblock++;
753 /* if there were partial blocks, initialize bitstream for
754 * unpacking fragment codings */
755 if (decode_partial_blocks) {
758 bit = get_bits(gb, 1);
759 /* toggle the bit because as soon as the first run length is
760 * fetched the bit will be toggled again */
765 /* figure out which fragments are coded; iterate through each
766 * superblock (all planes) */
767 s->coded_fragment_list_index = 0;
768 s->next_coeff= s->coeffs + s->fragment_count;
769 s->first_coded_y_fragment = s->first_coded_c_fragment = 0;
770 s->last_coded_y_fragment = s->last_coded_c_fragment = -1;
771 first_c_fragment_seen = 0;
772 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
773 for (i = 0; i < s->superblock_count; i++) {
775 /* iterate through all 16 fragments in a superblock */
776 for (j = 0; j < 16; j++) {
778 /* if the fragment is in bounds, check its coding status */
779 current_fragment = s->superblock_fragments[i * 16 + j];
780 if (current_fragment >= s->fragment_count) {
781 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
782 current_fragment, s->fragment_count);
785 if (current_fragment != -1) {
786 if (s->superblock_coding[i] == SB_NOT_CODED) {
788 /* copy all the fragments from the prior frame */
789 s->all_fragments[current_fragment].coding_method =
792 } else if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
794 /* fragment may or may not be coded; this is the case
795 * that cares about the fragment coding runs */
796 if (current_run-- == 0) {
798 current_run = get_vlc2(gb,
799 s->fragment_run_length_vlc.table, 5, 2);
803 /* default mode; actual mode will be decoded in
805 s->all_fragments[current_fragment].coding_method =
807 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
808 s->coded_fragment_list[s->coded_fragment_list_index] =
810 if ((current_fragment >= s->fragment_start[1]) &&
811 (s->last_coded_y_fragment == -1) &&
812 (!first_c_fragment_seen)) {
813 s->first_coded_c_fragment = s->coded_fragment_list_index;
814 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
815 first_c_fragment_seen = 1;
817 s->coded_fragment_list_index++;
818 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
819 debug_block_coding(" superblock %d is partially coded, fragment %d is coded\n",
820 i, current_fragment);
822 /* not coded; copy this fragment from the prior frame */
823 s->all_fragments[current_fragment].coding_method =
825 debug_block_coding(" superblock %d is partially coded, fragment %d is not coded\n",
826 i, current_fragment);
831 /* fragments are fully coded in this superblock; actual
832 * coding will be determined in next step */
833 s->all_fragments[current_fragment].coding_method =
835 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
836 s->coded_fragment_list[s->coded_fragment_list_index] =
838 if ((current_fragment >= s->fragment_start[1]) &&
839 (s->last_coded_y_fragment == -1) &&
840 (!first_c_fragment_seen)) {
841 s->first_coded_c_fragment = s->coded_fragment_list_index;
842 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
843 first_c_fragment_seen = 1;
845 s->coded_fragment_list_index++;
846 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
847 debug_block_coding(" superblock %d is fully coded, fragment %d is coded\n",
848 i, current_fragment);
854 if (!first_c_fragment_seen)
855 /* only Y fragments coded in this frame */
856 s->last_coded_y_fragment = s->coded_fragment_list_index - 1;
858 /* end the list of coded C fragments */
859 s->last_coded_c_fragment = s->coded_fragment_list_index - 1;
861 debug_block_coding(" %d total coded fragments, y: %d -> %d, c: %d -> %d\n",
862 s->coded_fragment_list_index,
863 s->first_coded_y_fragment,
864 s->last_coded_y_fragment,
865 s->first_coded_c_fragment,
866 s->last_coded_c_fragment);
872 * This function unpacks all the coding mode data for individual macroblocks
873 * from the bitstream.
875 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
879 int current_macroblock;
880 int current_fragment;
883 debug_vp3(" vp3: unpacking encoding modes\n");
886 debug_vp3(" keyframe-- all blocks are coded as INTRA\n");
888 for (i = 0; i < s->fragment_count; i++)
889 s->all_fragments[i].coding_method = MODE_INTRA;
893 /* fetch the mode coding scheme for this frame */
894 scheme = get_bits(gb, 3);
895 debug_modes(" using mode alphabet %d\n", scheme);
897 /* is it a custom coding scheme? */
899 debug_modes(" custom mode alphabet ahead:\n");
900 for (i = 0; i < 8; i++)
901 ModeAlphabet[scheme][get_bits(gb, 3)] = i;
904 for (i = 0; i < 8; i++)
905 debug_modes(" mode[%d][%d] = %d\n", scheme, i,
906 ModeAlphabet[scheme][i]);
908 /* iterate through all of the macroblocks that contain 1 or more
910 for (i = 0; i < s->u_superblock_start; i++) {
912 for (j = 0; j < 4; j++) {
913 current_macroblock = s->superblock_macroblocks[i * 4 + j];
914 if ((current_macroblock == -1) ||
915 (s->macroblock_coding[current_macroblock] == MODE_COPY))
917 if (current_macroblock >= s->macroblock_count) {
918 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad macroblock number (%d >= %d)\n",
919 current_macroblock, s->macroblock_count);
923 /* mode 7 means get 3 bits for each coding mode */
925 coding_mode = get_bits(gb, 3);
927 coding_mode = ModeAlphabet[scheme]
928 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
930 s->macroblock_coding[current_macroblock] = coding_mode;
931 for (k = 0; k < 6; k++) {
933 s->macroblock_fragments[current_macroblock * 6 + k];
934 if (current_fragment == -1)
936 if (current_fragment >= s->fragment_count) {
937 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad fragment number (%d >= %d)\n",
938 current_fragment, s->fragment_count);
941 if (s->all_fragments[current_fragment].coding_method !=
943 s->all_fragments[current_fragment].coding_method =
947 debug_modes(" coding method for macroblock starting @ fragment %d = %d\n",
948 s->macroblock_fragments[current_macroblock * 6], coding_mode);
957 * This function unpacks all the motion vectors for the individual
958 * macroblocks from the bitstream.
960 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
966 int last_motion_x = 0;
967 int last_motion_y = 0;
968 int prior_last_motion_x = 0;
969 int prior_last_motion_y = 0;
970 int current_macroblock;
971 int current_fragment;
973 debug_vp3(" vp3: unpacking motion vectors\n");
976 debug_vp3(" keyframe-- there are no motion vectors\n");
980 memset(motion_x, 0, 6 * sizeof(int));
981 memset(motion_y, 0, 6 * sizeof(int));
983 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
984 coding_mode = get_bits(gb, 1);
985 debug_vectors(" using %s scheme for unpacking motion vectors\n",
986 (coding_mode == 0) ? "VLC" : "fixed-length");
988 /* iterate through all of the macroblocks that contain 1 or more
990 for (i = 0; i < s->u_superblock_start; i++) {
992 for (j = 0; j < 4; j++) {
993 current_macroblock = s->superblock_macroblocks[i * 4 + j];
994 if ((current_macroblock == -1) ||
995 (s->macroblock_coding[current_macroblock] == MODE_COPY))
997 if (current_macroblock >= s->macroblock_count) {
998 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad macroblock number (%d >= %d)\n",
999 current_macroblock, s->macroblock_count);
1003 current_fragment = s->macroblock_fragments[current_macroblock * 6];
1004 if (current_fragment >= s->fragment_count) {
1005 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d\n",
1006 current_fragment, s->fragment_count);
1009 switch (s->macroblock_coding[current_macroblock]) {
1011 case MODE_INTER_PLUS_MV:
1012 case MODE_GOLDEN_MV:
1013 /* all 6 fragments use the same motion vector */
1014 if (coding_mode == 0) {
1015 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1016 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1018 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
1019 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
1022 for (k = 1; k < 6; k++) {
1023 motion_x[k] = motion_x[0];
1024 motion_y[k] = motion_y[0];
1027 /* vector maintenance, only on MODE_INTER_PLUS_MV */
1028 if (s->macroblock_coding[current_macroblock] ==
1029 MODE_INTER_PLUS_MV) {
1030 prior_last_motion_x = last_motion_x;
1031 prior_last_motion_y = last_motion_y;
1032 last_motion_x = motion_x[0];
1033 last_motion_y = motion_y[0];
1037 case MODE_INTER_FOURMV:
1038 /* fetch 4 vectors from the bitstream, one for each
1039 * Y fragment, then average for the C fragment vectors */
1040 motion_x[4] = motion_y[4] = 0;
1041 for (k = 0; k < 4; k++) {
1042 if (coding_mode == 0) {
1043 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1044 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1046 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
1047 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
1049 motion_x[4] += motion_x[k];
1050 motion_y[4] += motion_y[k];
1054 motion_x[4]= RSHIFT(motion_x[4], 2);
1056 motion_y[4]= RSHIFT(motion_y[4], 2);
1058 /* vector maintenance; vector[3] is treated as the
1059 * last vector in this case */
1060 prior_last_motion_x = last_motion_x;
1061 prior_last_motion_y = last_motion_y;
1062 last_motion_x = motion_x[3];
1063 last_motion_y = motion_y[3];
1066 case MODE_INTER_LAST_MV:
1067 /* all 6 fragments use the last motion vector */
1068 motion_x[0] = last_motion_x;
1069 motion_y[0] = last_motion_y;
1070 for (k = 1; k < 6; k++) {
1071 motion_x[k] = motion_x[0];
1072 motion_y[k] = motion_y[0];
1075 /* no vector maintenance (last vector remains the
1079 case MODE_INTER_PRIOR_LAST:
1080 /* all 6 fragments use the motion vector prior to the
1081 * last motion vector */
1082 motion_x[0] = prior_last_motion_x;
1083 motion_y[0] = prior_last_motion_y;
1084 for (k = 1; k < 6; k++) {
1085 motion_x[k] = motion_x[0];
1086 motion_y[k] = motion_y[0];
1089 /* vector maintenance */
1090 prior_last_motion_x = last_motion_x;
1091 prior_last_motion_y = last_motion_y;
1092 last_motion_x = motion_x[0];
1093 last_motion_y = motion_y[0];
1097 /* covers intra, inter without MV, golden without MV */
1098 memset(motion_x, 0, 6 * sizeof(int));
1099 memset(motion_y, 0, 6 * sizeof(int));
1101 /* no vector maintenance */
1105 /* assign the motion vectors to the correct fragments */
1106 debug_vectors(" vectors for macroblock starting @ fragment %d (coding method %d):\n",
1108 s->macroblock_coding[current_macroblock]);
1109 for (k = 0; k < 6; k++) {
1111 s->macroblock_fragments[current_macroblock * 6 + k];
1112 if (current_fragment == -1)
1114 if (current_fragment >= s->fragment_count) {
1115 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d)\n",
1116 current_fragment, s->fragment_count);
1119 s->all_fragments[current_fragment].motion_x = motion_x[k];
1120 s->all_fragments[current_fragment].motion_y = motion_y[k];
1121 debug_vectors(" vector %d: fragment %d = (%d, %d)\n",
1122 k, current_fragment, motion_x[k], motion_y[k]);
1132 * This function is called by unpack_dct_coeffs() to extract the VLCs from
1133 * the bitstream. The VLCs encode tokens which are used to unpack DCT
1134 * data. This function unpacks all the VLCs for either the Y plane or both
1135 * C planes, and is called for DC coefficients or different AC coefficient
1136 * levels (since different coefficient types require different VLC tables.
1138 * This function returns a residual eob run. E.g, if a particular token gave
1139 * instructions to EOB the next 5 fragments and there were only 2 fragments
1140 * left in the current fragment range, 3 would be returned so that it could
1141 * be passed into the next call to this same function.
1143 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
1144 VLC *table, int coeff_index,
1145 int first_fragment, int last_fragment,
1152 Vp3Fragment *fragment;
1153 uint8_t *perm= s->scantable.permutated;
1156 if ((first_fragment >= s->fragment_count) ||
1157 (last_fragment >= s->fragment_count)) {
1159 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vlcs(): bad fragment number (%d -> %d ?)\n",
1160 first_fragment, last_fragment);
1164 for (i = first_fragment; i <= last_fragment; i++) {
1166 fragment = &s->all_fragments[s->coded_fragment_list[i]];
1167 if (fragment->coeff_count > coeff_index)
1171 /* decode a VLC into a token */
1172 token = get_vlc2(gb, table->table, 5, 3);
1173 debug_vlc(" token = %2d, ", token);
1174 /* use the token to get a zero run, a coefficient, and an eob run */
1176 eob_run = eob_run_base[token];
1177 if (eob_run_get_bits[token])
1178 eob_run += get_bits(gb, eob_run_get_bits[token]);
1179 coeff = zero_run = 0;
1181 bits_to_get = coeff_get_bits[token];
1183 coeff = coeff_tables[token][0];
1185 coeff = coeff_tables[token][get_bits(gb, bits_to_get)];
1187 zero_run = zero_run_base[token];
1188 if (zero_run_get_bits[token])
1189 zero_run += get_bits(gb, zero_run_get_bits[token]);
1194 fragment->coeff_count += zero_run;
1195 if (fragment->coeff_count < 64){
1196 fragment->next_coeff->coeff= coeff;
1197 fragment->next_coeff->index= perm[fragment->coeff_count++]; //FIXME perm here already?
1198 fragment->next_coeff->next= s->next_coeff;
1199 s->next_coeff->next=NULL;
1200 fragment->next_coeff= s->next_coeff++;
1202 debug_vlc(" fragment %d coeff = %d\n",
1203 s->coded_fragment_list[i], fragment->next_coeff[coeff_index]);
1205 fragment->coeff_count |= 128;
1206 debug_vlc(" fragment %d eob with %d coefficients\n",
1207 s->coded_fragment_list[i], fragment->coeff_count&127);
1216 * This function unpacks all of the DCT coefficient data from the
1219 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1226 int residual_eob_run = 0;
1228 /* fetch the DC table indices */
1229 dc_y_table = get_bits(gb, 4);
1230 dc_c_table = get_bits(gb, 4);
1232 /* unpack the Y plane DC coefficients */
1233 debug_vp3(" vp3: unpacking Y plane DC coefficients using table %d\n",
1235 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1236 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1238 /* unpack the C plane DC coefficients */
1239 debug_vp3(" vp3: unpacking C plane DC coefficients using table %d\n",
1241 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1242 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1244 /* fetch the AC table indices */
1245 ac_y_table = get_bits(gb, 4);
1246 ac_c_table = get_bits(gb, 4);
1248 /* unpack the group 1 AC coefficients (coeffs 1-5) */
1249 for (i = 1; i <= 5; i++) {
1251 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1253 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_y_table], i,
1254 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1256 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1258 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_c_table], i,
1259 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1262 /* unpack the group 2 AC coefficients (coeffs 6-14) */
1263 for (i = 6; i <= 14; i++) {
1265 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1267 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_y_table], i,
1268 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1270 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1272 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_c_table], i,
1273 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1276 /* unpack the group 3 AC coefficients (coeffs 15-27) */
1277 for (i = 15; i <= 27; i++) {
1279 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1281 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_y_table], i,
1282 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1284 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1286 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_c_table], i,
1287 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1290 /* unpack the group 4 AC coefficients (coeffs 28-63) */
1291 for (i = 28; i <= 63; i++) {
1293 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1295 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_y_table], i,
1296 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1298 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1300 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_c_table], i,
1301 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1308 * This function reverses the DC prediction for each coded fragment in
1309 * the frame. Much of this function is adapted directly from the original
1312 #define COMPATIBLE_FRAME(x) \
1313 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1314 #define FRAME_CODED(x) (s->all_fragments[x].coding_method != MODE_COPY)
1315 #define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
1317 static void reverse_dc_prediction(Vp3DecodeContext *s,
1320 int fragment_height)
1329 int i = first_fragment;
1333 /* DC values for the left, up-left, up, and up-right fragments */
1334 int vl, vul, vu, vur;
1336 /* indices for the left, up-left, up, and up-right fragments */
1340 * The 6 fields mean:
1341 * 0: up-left multiplier
1343 * 2: up-right multiplier
1344 * 3: left multiplier
1346 int predictor_transform[16][4] = {
1348 { 0, 0, 0,128}, // PL
1349 { 0, 0,128, 0}, // PUR
1350 { 0, 0, 53, 75}, // PUR|PL
1351 { 0,128, 0, 0}, // PU
1352 { 0, 64, 0, 64}, // PU|PL
1353 { 0,128, 0, 0}, // PU|PUR
1354 { 0, 0, 53, 75}, // PU|PUR|PL
1355 {128, 0, 0, 0}, // PUL
1356 { 0, 0, 0,128}, // PUL|PL
1357 { 64, 0, 64, 0}, // PUL|PUR
1358 { 0, 0, 53, 75}, // PUL|PUR|PL
1359 { 0,128, 0, 0}, // PUL|PU
1360 {-104,116, 0,116}, // PUL|PU|PL
1361 { 24, 80, 24, 0}, // PUL|PU|PUR
1362 {-104,116, 0,116} // PUL|PU|PUR|PL
1365 /* This table shows which types of blocks can use other blocks for
1366 * prediction. For example, INTRA is the only mode in this table to
1367 * have a frame number of 0. That means INTRA blocks can only predict
1368 * from other INTRA blocks. There are 2 golden frame coding types;
1369 * blocks encoding in these modes can only predict from other blocks
1370 * that were encoded with these 1 of these 2 modes. */
1371 unsigned char compatible_frame[8] = {
1372 1, /* MODE_INTER_NO_MV */
1374 1, /* MODE_INTER_PLUS_MV */
1375 1, /* MODE_INTER_LAST_MV */
1376 1, /* MODE_INTER_PRIOR_MV */
1377 2, /* MODE_USING_GOLDEN */
1378 2, /* MODE_GOLDEN_MV */
1379 1 /* MODE_INTER_FOUR_MV */
1381 int current_frame_type;
1383 /* there is a last DC predictor for each of the 3 frame types */
1388 debug_vp3(" vp3: reversing DC prediction\n");
1390 vul = vu = vur = vl = 0;
1391 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1393 /* for each fragment row... */
1394 for (y = 0; y < fragment_height; y++) {
1396 /* for each fragment in a row... */
1397 for (x = 0; x < fragment_width; x++, i++) {
1399 /* reverse prediction if this block was coded */
1400 if (s->all_fragments[i].coding_method != MODE_COPY) {
1402 current_frame_type =
1403 compatible_frame[s->all_fragments[i].coding_method];
1404 debug_dc_pred(" frag %d: orig DC = %d, ",
1411 if(FRAME_CODED(l) && COMPATIBLE_FRAME(l))
1415 u= i-fragment_width;
1417 if(FRAME_CODED(u) && COMPATIBLE_FRAME(u))
1420 ul= i-fragment_width-1;
1422 if(FRAME_CODED(ul) && COMPATIBLE_FRAME(ul))
1425 if(x + 1 < fragment_width){
1426 ur= i-fragment_width+1;
1428 if(FRAME_CODED(ur) && COMPATIBLE_FRAME(ur))
1433 debug_dc_pred("transform = %d, ", transform);
1435 if (transform == 0) {
1437 /* if there were no fragments to predict from, use last
1439 predicted_dc = last_dc[current_frame_type];
1440 debug_dc_pred("from last DC (%d) = %d\n",
1441 current_frame_type, DC_COEFF(i));
1445 /* apply the appropriate predictor transform */
1447 (predictor_transform[transform][0] * vul) +
1448 (predictor_transform[transform][1] * vu) +
1449 (predictor_transform[transform][2] * vur) +
1450 (predictor_transform[transform][3] * vl);
1452 predicted_dc /= 128;
1454 /* check for outranging on the [ul u l] and
1455 * [ul u ur l] predictors */
1456 if ((transform == 13) || (transform == 15)) {
1457 if (FFABS(predicted_dc - vu) > 128)
1459 else if (FFABS(predicted_dc - vl) > 128)
1461 else if (FFABS(predicted_dc - vul) > 128)
1465 debug_dc_pred("from pred DC = %d\n",
1469 /* at long last, apply the predictor */
1470 if(s->coeffs[i].index){
1471 *s->next_coeff= s->coeffs[i];
1472 s->coeffs[i].index=0;
1473 s->coeffs[i].coeff=0;
1474 s->coeffs[i].next= s->next_coeff++;
1476 s->coeffs[i].coeff += predicted_dc;
1478 last_dc[current_frame_type] = DC_COEFF(i);
1479 if(DC_COEFF(i) && !(s->all_fragments[i].coeff_count&127)){
1480 s->all_fragments[i].coeff_count= 129;
1481 // s->all_fragments[i].next_coeff= s->next_coeff;
1482 s->coeffs[i].next= s->next_coeff;
1483 (s->next_coeff++)->next=NULL;
1491 static void horizontal_filter(unsigned char *first_pixel, int stride,
1492 int *bounding_values);
1493 static void vertical_filter(unsigned char *first_pixel, int stride,
1494 int *bounding_values);
1497 * Perform the final rendering for a particular slice of data.
1498 * The slice number ranges from 0..(macroblock_height - 1).
1500 static void render_slice(Vp3DecodeContext *s, int slice)
1504 int16_t *dequantizer;
1505 DECLARE_ALIGNED_16(DCTELEM, block[64]);
1506 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1507 int motion_halfpel_index;
1508 uint8_t *motion_source;
1510 int current_macroblock_entry = slice * s->macroblock_width * 6;
1512 if (slice >= s->macroblock_height)
1515 for (plane = 0; plane < 3; plane++) {
1516 uint8_t *output_plane = s->current_frame.data [plane];
1517 uint8_t * last_plane = s-> last_frame.data [plane];
1518 uint8_t *golden_plane = s-> golden_frame.data [plane];
1519 int stride = s->current_frame.linesize[plane];
1520 int plane_width = s->width >> !!plane;
1521 int plane_height = s->height >> !!plane;
1522 int y = slice * FRAGMENT_PIXELS << !plane ;
1523 int slice_height = y + (FRAGMENT_PIXELS << !plane);
1524 int i = s->macroblock_fragments[current_macroblock_entry + plane + 3*!!plane];
1526 if (!s->flipped_image) stride = -stride;
1529 if(FFABS(stride) > 2048)
1530 return; //various tables are fixed size
1532 /* for each fragment row in the slice (both of them)... */
1533 for (; y < slice_height; y += 8) {
1535 /* for each fragment in a row... */
1536 for (x = 0; x < plane_width; x += 8, i++) {
1538 if ((i < 0) || (i >= s->fragment_count)) {
1539 av_log(s->avctx, AV_LOG_ERROR, " vp3:render_slice(): bad fragment number (%d)\n", i);
1543 /* transform if this block was coded */
1544 if ((s->all_fragments[i].coding_method != MODE_COPY) &&
1545 !((s->avctx->flags & CODEC_FLAG_GRAY) && plane)) {
1547 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1548 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1549 motion_source= golden_plane;
1551 motion_source= last_plane;
1553 motion_source += s->all_fragments[i].first_pixel;
1554 motion_halfpel_index = 0;
1556 /* sort out the motion vector if this fragment is coded
1557 * using a motion vector method */
1558 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1559 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1561 motion_x = s->all_fragments[i].motion_x;
1562 motion_y = s->all_fragments[i].motion_y;
1564 motion_x= (motion_x>>1) | (motion_x&1);
1565 motion_y= (motion_y>>1) | (motion_y&1);
1568 src_x= (motion_x>>1) + x;
1569 src_y= (motion_y>>1) + y;
1570 if ((motion_x == 127) || (motion_y == 127))
1571 av_log(s->avctx, AV_LOG_ERROR, " help! got invalid motion vector! (%X, %X)\n", motion_x, motion_y);
1573 motion_halfpel_index = motion_x & 0x01;
1574 motion_source += (motion_x >> 1);
1576 motion_halfpel_index |= (motion_y & 0x01) << 1;
1577 motion_source += ((motion_y >> 1) * stride);
1579 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1580 uint8_t *temp= s->edge_emu_buffer;
1581 if(stride<0) temp -= 9*stride;
1582 else temp += 9*stride;
1584 ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1585 motion_source= temp;
1590 /* first, take care of copying a block from either the
1591 * previous or the golden frame */
1592 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1593 /* Note, it is possible to implement all MC cases with
1594 put_no_rnd_pixels_l2 which would look more like the
1595 VP3 source but this would be slower as
1596 put_no_rnd_pixels_tab is better optimzed */
1597 if(motion_halfpel_index != 3){
1598 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1599 output_plane + s->all_fragments[i].first_pixel,
1600 motion_source, stride, 8);
1602 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1603 s->dsp.put_no_rnd_pixels_l2[1](
1604 output_plane + s->all_fragments[i].first_pixel,
1606 motion_source + stride + 1 + d,
1609 dequantizer = s->qmat[1][plane];
1611 dequantizer = s->qmat[0][plane];
1614 /* dequantize the DCT coefficients */
1615 debug_idct("fragment %d, coding mode %d, DC = %d, dequant = %d:\n",
1616 i, s->all_fragments[i].coding_method,
1617 DC_COEFF(i), dequantizer[0]);
1619 if(s->avctx->idct_algo==FF_IDCT_VP3){
1620 Coeff *coeff= s->coeffs + i;
1621 memset(block, 0, sizeof(block));
1623 block[coeff->index]= coeff->coeff * dequantizer[coeff->index];
1627 Coeff *coeff= s->coeffs + i;
1628 memset(block, 0, sizeof(block));
1630 block[coeff->index]= (coeff->coeff * dequantizer[coeff->index] + 2)>>2;
1635 /* invert DCT and place (or add) in final output */
1637 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1638 if(s->avctx->idct_algo!=FF_IDCT_VP3)
1641 output_plane + s->all_fragments[i].first_pixel,
1646 output_plane + s->all_fragments[i].first_pixel,
1651 debug_idct("block after idct_%s():\n",
1652 (s->all_fragments[i].coding_method == MODE_INTRA)?
1654 for (m = 0; m < 8; m++) {
1655 for (n = 0; n < 8; n++) {
1656 debug_idct(" %3d", *(output_plane +
1657 s->all_fragments[i].first_pixel + (m * stride + n)));
1665 /* copy directly from the previous frame */
1666 s->dsp.put_pixels_tab[1][0](
1667 output_plane + s->all_fragments[i].first_pixel,
1668 last_plane + s->all_fragments[i].first_pixel,
1673 /* perform the left edge filter if:
1674 * - the fragment is not on the left column
1675 * - the fragment is coded in this frame
1676 * - the fragment is not coded in this frame but the left
1677 * fragment is coded in this frame (this is done instead
1678 * of a right edge filter when rendering the left fragment
1679 * since this fragment is not available yet) */
1681 ((s->all_fragments[i].coding_method != MODE_COPY) ||
1682 ((s->all_fragments[i].coding_method == MODE_COPY) &&
1683 (s->all_fragments[i - 1].coding_method != MODE_COPY)) )) {
1685 output_plane + s->all_fragments[i].first_pixel + 7*stride,
1686 -stride, s->bounding_values_array + 127);
1689 /* perform the top edge filter if:
1690 * - the fragment is not on the top row
1691 * - the fragment is coded in this frame
1692 * - the fragment is not coded in this frame but the above
1693 * fragment is coded in this frame (this is done instead
1694 * of a bottom edge filter when rendering the above
1695 * fragment since this fragment is not available yet) */
1697 ((s->all_fragments[i].coding_method != MODE_COPY) ||
1698 ((s->all_fragments[i].coding_method == MODE_COPY) &&
1699 (s->all_fragments[i - fragment_width].coding_method != MODE_COPY)) )) {
1701 output_plane + s->all_fragments[i].first_pixel - stride,
1702 -stride, s->bounding_values_array + 127);
1709 /* this looks like a good place for slice dispatch... */
1711 * if (slice == s->macroblock_height - 1)
1712 * dispatch (both last slice & 2nd-to-last slice);
1713 * else if (slice > 0)
1714 * dispatch (slice - 1);
1720 static void horizontal_filter(unsigned char *first_pixel, int stride,
1721 int *bounding_values)
1726 for (end= first_pixel + 8*stride; first_pixel != end; first_pixel += stride) {
1728 (first_pixel[-2] - first_pixel[ 1])
1729 +3*(first_pixel[ 0] - first_pixel[-1]);
1730 filter_value = bounding_values[(filter_value + 4) >> 3];
1731 first_pixel[-1] = av_clip_uint8(first_pixel[-1] + filter_value);
1732 first_pixel[ 0] = av_clip_uint8(first_pixel[ 0] - filter_value);
1736 static void vertical_filter(unsigned char *first_pixel, int stride,
1737 int *bounding_values)
1741 const int nstride= -stride;
1743 for (end= first_pixel + 8; first_pixel < end; first_pixel++) {
1745 (first_pixel[2 * nstride] - first_pixel[ stride])
1746 +3*(first_pixel[0 ] - first_pixel[nstride]);
1747 filter_value = bounding_values[(filter_value + 4) >> 3];
1748 first_pixel[nstride] = av_clip_uint8(first_pixel[nstride] + filter_value);
1749 first_pixel[0] = av_clip_uint8(first_pixel[0] - filter_value);
1753 static void apply_loop_filter(Vp3DecodeContext *s)
1757 int *bounding_values= s->bounding_values_array+127;
1760 int bounding_values_array[256];
1763 /* find the right loop limit value */
1764 for (x = 63; x >= 0; x--) {
1765 if (vp31_ac_scale_factor[x] >= s->quality_index)
1768 filter_limit = vp31_filter_limit_values[s->quality_index];
1770 /* set up the bounding values */
1771 memset(bounding_values_array, 0, 256 * sizeof(int));
1772 for (x = 0; x < filter_limit; x++) {
1773 bounding_values[-x - filter_limit] = -filter_limit + x;
1774 bounding_values[-x] = -x;
1775 bounding_values[x] = x;
1776 bounding_values[x + filter_limit] = filter_limit - x;
1780 for (plane = 0; plane < 3; plane++) {
1781 int width = s->fragment_width >> !!plane;
1782 int height = s->fragment_height >> !!plane;
1783 int fragment = s->fragment_start [plane];
1784 int stride = s->current_frame.linesize[plane];
1785 uint8_t *plane_data = s->current_frame.data [plane];
1786 if (!s->flipped_image) stride = -stride;
1788 for (y = 0; y < height; y++) {
1790 for (x = 0; x < width; x++) {
1792 /* do not perform left edge filter for left columns frags */
1794 (s->all_fragments[fragment].coding_method != MODE_COPY)) {
1796 plane_data + s->all_fragments[fragment].first_pixel,
1797 stride, bounding_values);
1800 /* do not perform top edge filter for top row fragments */
1802 (s->all_fragments[fragment].coding_method != MODE_COPY)) {
1804 plane_data + s->all_fragments[fragment].first_pixel,
1805 stride, bounding_values);
1808 /* do not perform right edge filter for right column
1809 * fragments or if right fragment neighbor is also coded
1810 * in this frame (it will be filtered in next iteration) */
1811 if ((x < width - 1) &&
1812 (s->all_fragments[fragment].coding_method != MODE_COPY) &&
1813 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1815 plane_data + s->all_fragments[fragment + 1].first_pixel,
1816 stride, bounding_values);
1819 /* do not perform bottom edge filter for bottom row
1820 * fragments or if bottom fragment neighbor is also coded
1821 * in this frame (it will be filtered in the next row) */
1822 if ((y < height - 1) &&
1823 (s->all_fragments[fragment].coding_method != MODE_COPY) &&
1824 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1826 plane_data + s->all_fragments[fragment + width].first_pixel,
1827 stride, bounding_values);
1831 STOP_TIMER("loop filter")
1838 * This function computes the first pixel addresses for each fragment.
1839 * This function needs to be invoked after the first frame is allocated
1840 * so that it has access to the plane strides.
1842 static void vp3_calculate_pixel_addresses(Vp3DecodeContext *s)
1847 /* figure out the first pixel addresses for each of the fragments */
1850 for (y = s->fragment_height; y > 0; y--) {
1851 for (x = 0; x < s->fragment_width; x++) {
1852 s->all_fragments[i++].first_pixel =
1853 s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
1854 s->golden_frame.linesize[0] +
1855 x * FRAGMENT_PIXELS;
1856 debug_init(" fragment %d, first pixel @ %d\n",
1857 i-1, s->all_fragments[i-1].first_pixel);
1862 i = s->fragment_start[1];
1863 for (y = s->fragment_height / 2; y > 0; y--) {
1864 for (x = 0; x < s->fragment_width / 2; x++) {
1865 s->all_fragments[i++].first_pixel =
1866 s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
1867 s->golden_frame.linesize[1] +
1868 x * FRAGMENT_PIXELS;
1869 debug_init(" fragment %d, first pixel @ %d\n",
1870 i-1, s->all_fragments[i-1].first_pixel);
1875 i = s->fragment_start[2];
1876 for (y = s->fragment_height / 2; y > 0; y--) {
1877 for (x = 0; x < s->fragment_width / 2; x++) {
1878 s->all_fragments[i++].first_pixel =
1879 s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
1880 s->golden_frame.linesize[2] +
1881 x * FRAGMENT_PIXELS;
1882 debug_init(" fragment %d, first pixel @ %d\n",
1883 i-1, s->all_fragments[i-1].first_pixel);
1888 /* FIXME: this should be merged with the above! */
1889 static void theora_calculate_pixel_addresses(Vp3DecodeContext *s)
1894 /* figure out the first pixel addresses for each of the fragments */
1897 for (y = 1; y <= s->fragment_height; y++) {
1898 for (x = 0; x < s->fragment_width; x++) {
1899 s->all_fragments[i++].first_pixel =
1900 s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
1901 s->golden_frame.linesize[0] +
1902 x * FRAGMENT_PIXELS;
1903 debug_init(" fragment %d, first pixel @ %d\n",
1904 i-1, s->all_fragments[i-1].first_pixel);
1909 i = s->fragment_start[1];
1910 for (y = 1; y <= s->fragment_height / 2; y++) {
1911 for (x = 0; x < s->fragment_width / 2; x++) {
1912 s->all_fragments[i++].first_pixel =
1913 s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
1914 s->golden_frame.linesize[1] +
1915 x * FRAGMENT_PIXELS;
1916 debug_init(" fragment %d, first pixel @ %d\n",
1917 i-1, s->all_fragments[i-1].first_pixel);
1922 i = s->fragment_start[2];
1923 for (y = 1; y <= s->fragment_height / 2; y++) {
1924 for (x = 0; x < s->fragment_width / 2; x++) {
1925 s->all_fragments[i++].first_pixel =
1926 s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
1927 s->golden_frame.linesize[2] +
1928 x * FRAGMENT_PIXELS;
1929 debug_init(" fragment %d, first pixel @ %d\n",
1930 i-1, s->all_fragments[i-1].first_pixel);
1936 * This is the ffmpeg/libavcodec API init function.
1938 static int vp3_decode_init(AVCodecContext *avctx)
1940 Vp3DecodeContext *s = avctx->priv_data;
1941 int i, inter, plane;
1944 int y_superblock_count;
1945 int c_superblock_count;
1947 if (avctx->codec_tag == MKTAG('V','P','3','0'))
1953 s->width = (avctx->width + 15) & 0xFFFFFFF0;
1954 s->height = (avctx->height + 15) & 0xFFFFFFF0;
1955 avctx->pix_fmt = PIX_FMT_YUV420P;
1956 if(avctx->idct_algo==FF_IDCT_AUTO)
1957 avctx->idct_algo=FF_IDCT_VP3;
1958 dsputil_init(&s->dsp, avctx);
1960 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1962 /* initialize to an impossible value which will force a recalculation
1963 * in the first frame decode */
1964 s->quality_index = -1;
1966 s->y_superblock_width = (s->width + 31) / 32;
1967 s->y_superblock_height = (s->height + 31) / 32;
1968 y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1970 /* work out the dimensions for the C planes */
1971 c_width = s->width / 2;
1972 c_height = s->height / 2;
1973 s->c_superblock_width = (c_width + 31) / 32;
1974 s->c_superblock_height = (c_height + 31) / 32;
1975 c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1977 s->superblock_count = y_superblock_count + (c_superblock_count * 2);
1978 s->u_superblock_start = y_superblock_count;
1979 s->v_superblock_start = s->u_superblock_start + c_superblock_count;
1980 s->superblock_coding = av_malloc(s->superblock_count);
1982 s->macroblock_width = (s->width + 15) / 16;
1983 s->macroblock_height = (s->height + 15) / 16;
1984 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1986 s->fragment_width = s->width / FRAGMENT_PIXELS;
1987 s->fragment_height = s->height / FRAGMENT_PIXELS;
1989 /* fragment count covers all 8x8 blocks for all 3 planes */
1990 s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
1991 s->fragment_start[1] = s->fragment_width * s->fragment_height;
1992 s->fragment_start[2] = s->fragment_width * s->fragment_height * 5 / 4;
1994 debug_init(" Y plane: %d x %d\n", s->width, s->height);
1995 debug_init(" C plane: %d x %d\n", c_width, c_height);
1996 debug_init(" Y superblocks: %d x %d, %d total\n",
1997 s->y_superblock_width, s->y_superblock_height, y_superblock_count);
1998 debug_init(" C superblocks: %d x %d, %d total\n",
1999 s->c_superblock_width, s->c_superblock_height, c_superblock_count);
2000 debug_init(" total superblocks = %d, U starts @ %d, V starts @ %d\n",
2001 s->superblock_count, s->u_superblock_start, s->v_superblock_start);
2002 debug_init(" macroblocks: %d x %d, %d total\n",
2003 s->macroblock_width, s->macroblock_height, s->macroblock_count);
2004 debug_init(" %d fragments, %d x %d, u starts @ %d, v starts @ %d\n",
2008 s->fragment_start[1],
2009 s->fragment_start[2]);
2011 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
2012 s->coeffs = av_malloc(s->fragment_count * sizeof(Coeff) * 65);
2013 s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int));
2014 s->pixel_addresses_inited = 0;
2016 if (!s->theora_tables)
2018 for (i = 0; i < 64; i++) {
2019 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
2020 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
2021 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
2022 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
2023 s->base_matrix[2][i] = vp31_inter_dequant[i];
2024 s->filter_limit_values[i] = vp31_filter_limit_values[i];
2027 for(inter=0; inter<2; inter++){
2028 for(plane=0; plane<3; plane++){
2029 s->qr_count[inter][plane]= 1;
2030 s->qr_size [inter][plane][0]= 63;
2031 s->qr_base [inter][plane][0]=
2032 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
2036 /* init VLC tables */
2037 for (i = 0; i < 16; i++) {
2040 init_vlc(&s->dc_vlc[i], 5, 32,
2041 &dc_bias[i][0][1], 4, 2,
2042 &dc_bias[i][0][0], 4, 2, 0);
2044 /* group 1 AC histograms */
2045 init_vlc(&s->ac_vlc_1[i], 5, 32,
2046 &ac_bias_0[i][0][1], 4, 2,
2047 &ac_bias_0[i][0][0], 4, 2, 0);
2049 /* group 2 AC histograms */
2050 init_vlc(&s->ac_vlc_2[i], 5, 32,
2051 &ac_bias_1[i][0][1], 4, 2,
2052 &ac_bias_1[i][0][0], 4, 2, 0);
2054 /* group 3 AC histograms */
2055 init_vlc(&s->ac_vlc_3[i], 5, 32,
2056 &ac_bias_2[i][0][1], 4, 2,
2057 &ac_bias_2[i][0][0], 4, 2, 0);
2059 /* group 4 AC histograms */
2060 init_vlc(&s->ac_vlc_4[i], 5, 32,
2061 &ac_bias_3[i][0][1], 4, 2,
2062 &ac_bias_3[i][0][0], 4, 2, 0);
2065 for (i = 0; i < 16; i++) {
2068 init_vlc(&s->dc_vlc[i], 5, 32,
2069 &s->huffman_table[i][0][1], 4, 2,
2070 &s->huffman_table[i][0][0], 4, 2, 0);
2072 /* group 1 AC histograms */
2073 init_vlc(&s->ac_vlc_1[i], 5, 32,
2074 &s->huffman_table[i+16][0][1], 4, 2,
2075 &s->huffman_table[i+16][0][0], 4, 2, 0);
2077 /* group 2 AC histograms */
2078 init_vlc(&s->ac_vlc_2[i], 5, 32,
2079 &s->huffman_table[i+16*2][0][1], 4, 2,
2080 &s->huffman_table[i+16*2][0][0], 4, 2, 0);
2082 /* group 3 AC histograms */
2083 init_vlc(&s->ac_vlc_3[i], 5, 32,
2084 &s->huffman_table[i+16*3][0][1], 4, 2,
2085 &s->huffman_table[i+16*3][0][0], 4, 2, 0);
2087 /* group 4 AC histograms */
2088 init_vlc(&s->ac_vlc_4[i], 5, 32,
2089 &s->huffman_table[i+16*4][0][1], 4, 2,
2090 &s->huffman_table[i+16*4][0][0], 4, 2, 0);
2094 init_vlc(&s->superblock_run_length_vlc, 6, 34,
2095 &superblock_run_length_vlc_table[0][1], 4, 2,
2096 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
2098 init_vlc(&s->fragment_run_length_vlc, 5, 30,
2099 &fragment_run_length_vlc_table[0][1], 4, 2,
2100 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
2102 init_vlc(&s->mode_code_vlc, 3, 8,
2103 &mode_code_vlc_table[0][1], 2, 1,
2104 &mode_code_vlc_table[0][0], 2, 1, 0);
2106 init_vlc(&s->motion_vector_vlc, 6, 63,
2107 &motion_vector_vlc_table[0][1], 2, 1,
2108 &motion_vector_vlc_table[0][0], 2, 1, 0);
2110 /* work out the block mapping tables */
2111 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
2112 s->superblock_macroblocks = av_malloc(s->superblock_count * 4 * sizeof(int));
2113 s->macroblock_fragments = av_malloc(s->macroblock_count * 6 * sizeof(int));
2114 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
2115 init_block_mapping(s);
2117 for (i = 0; i < 3; i++) {
2118 s->current_frame.data[i] = NULL;
2119 s->last_frame.data[i] = NULL;
2120 s->golden_frame.data[i] = NULL;
2127 * This is the ffmpeg/libavcodec API frame decode function.
2129 static int vp3_decode_frame(AVCodecContext *avctx,
2130 void *data, int *data_size,
2131 uint8_t *buf, int buf_size)
2133 Vp3DecodeContext *s = avctx->priv_data;
2135 static int counter = 0;
2138 init_get_bits(&gb, buf, buf_size * 8);
2140 if (s->theora && get_bits1(&gb))
2142 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
2146 s->keyframe = !get_bits1(&gb);
2149 s->last_quality_index = s->quality_index;
2153 s->qis[s->nqis++]= get_bits(&gb, 6);
2154 } while(s->theora >= 0x030200 && s->nqis<3 && get_bits1(&gb));
2156 s->quality_index= s->qis[0];
2158 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
2159 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
2160 s->keyframe?"key":"", counter, s->quality_index);
2163 if (s->quality_index != s->last_quality_index) {
2164 init_dequantizer(s);
2165 init_loop_filter(s);
2171 skip_bits(&gb, 4); /* width code */
2172 skip_bits(&gb, 4); /* height code */
2175 s->version = get_bits(&gb, 5);
2177 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
2180 if (s->version || s->theora)
2183 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
2184 skip_bits(&gb, 2); /* reserved? */
2187 if (s->last_frame.data[0] == s->golden_frame.data[0]) {
2188 if (s->golden_frame.data[0])
2189 avctx->release_buffer(avctx, &s->golden_frame);
2190 s->last_frame= s->golden_frame; /* ensure that we catch any access to this released frame */
2192 if (s->golden_frame.data[0])
2193 avctx->release_buffer(avctx, &s->golden_frame);
2194 if (s->last_frame.data[0])
2195 avctx->release_buffer(avctx, &s->last_frame);
2198 s->golden_frame.reference = 3;
2199 if(avctx->get_buffer(avctx, &s->golden_frame) < 0) {
2200 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
2204 /* golden frame is also the current frame */
2205 s->current_frame= s->golden_frame;
2207 /* time to figure out pixel addresses? */
2208 if (!s->pixel_addresses_inited)
2210 if (!s->flipped_image)
2211 vp3_calculate_pixel_addresses(s);
2213 theora_calculate_pixel_addresses(s);
2214 s->pixel_addresses_inited = 1;
2217 /* allocate a new current frame */
2218 s->current_frame.reference = 3;
2219 if (!s->pixel_addresses_inited) {
2220 av_log(s->avctx, AV_LOG_ERROR, "vp3: first frame not a keyframe\n");
2223 if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
2224 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
2229 s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
2230 s->current_frame.qstride= 0;
2234 STOP_TIMER("init_frame")}
2239 memcpy(s->current_frame.data[0], s->golden_frame.data[0],
2240 s->current_frame.linesize[0] * s->height);
2241 memcpy(s->current_frame.data[1], s->golden_frame.data[1],
2242 s->current_frame.linesize[1] * s->height / 2);
2243 memcpy(s->current_frame.data[2], s->golden_frame.data[2],
2244 s->current_frame.linesize[2] * s->height / 2);
2250 if (unpack_superblocks(s, &gb)){
2251 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
2254 STOP_TIMER("unpack_superblocks")}
2256 if (unpack_modes(s, &gb)){
2257 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
2260 STOP_TIMER("unpack_modes")}
2262 if (unpack_vectors(s, &gb)){
2263 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2266 STOP_TIMER("unpack_vectors")}
2268 if (unpack_dct_coeffs(s, &gb)){
2269 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
2272 STOP_TIMER("unpack_dct_coeffs")}
2275 reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
2276 if ((avctx->flags & CODEC_FLAG_GRAY) == 0) {
2277 reverse_dc_prediction(s, s->fragment_start[1],
2278 s->fragment_width / 2, s->fragment_height / 2);
2279 reverse_dc_prediction(s, s->fragment_start[2],
2280 s->fragment_width / 2, s->fragment_height / 2);
2282 STOP_TIMER("reverse_dc_prediction")}
2285 for (i = 0; i < s->macroblock_height; i++)
2287 STOP_TIMER("render_fragments")}
2290 apply_loop_filter(s);
2291 STOP_TIMER("apply_loop_filter")}
2296 *data_size=sizeof(AVFrame);
2297 *(AVFrame*)data= s->current_frame;
2299 /* release the last frame, if it is allocated and if it is not the
2301 if ((s->last_frame.data[0]) &&
2302 (s->last_frame.data[0] != s->golden_frame.data[0]))
2303 avctx->release_buffer(avctx, &s->last_frame);
2305 /* shuffle frames (last = current) */
2306 s->last_frame= s->current_frame;
2307 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
2313 * This is the ffmpeg/libavcodec API module cleanup function.
2315 static int vp3_decode_end(AVCodecContext *avctx)
2317 Vp3DecodeContext *s = avctx->priv_data;
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 /* release all frames */
2328 if (s->golden_frame.data[0] && s->golden_frame.data[0] != s->last_frame.data[0])
2329 avctx->release_buffer(avctx, &s->golden_frame);
2330 if (s->last_frame.data[0])
2331 avctx->release_buffer(avctx, &s->last_frame);
2332 /* no need to release the current_frame since it will always be pointing
2333 * to the same frame as either the golden or last frame */
2338 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2340 Vp3DecodeContext *s = avctx->priv_data;
2342 if (get_bits(gb, 1)) {
2344 if (s->entries >= 32) { /* overflow */
2345 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2348 token = get_bits(gb, 5);
2349 //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);
2350 s->huffman_table[s->hti][token][0] = s->hbits;
2351 s->huffman_table[s->hti][token][1] = s->huff_code_size;
2355 if (s->huff_code_size >= 32) {/* overflow */
2356 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2359 s->huff_code_size++;
2361 read_huffman_tree(avctx, gb);
2363 read_huffman_tree(avctx, gb);
2365 s->huff_code_size--;
2370 #ifdef CONFIG_THEORA_DECODER
2371 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2373 Vp3DecodeContext *s = avctx->priv_data;
2375 s->theora = get_bits_long(gb, 24);
2376 av_log(avctx, AV_LOG_INFO, "Theora bitstream version %X\n", s->theora);
2378 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2379 /* but previous versions have the image flipped relative to vp3 */
2380 if (s->theora < 0x030200)
2382 s->flipped_image = 1;
2383 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2386 s->width = get_bits(gb, 16) << 4;
2387 s->height = get_bits(gb, 16) << 4;
2389 if(avcodec_check_dimensions(avctx, s->width, s->height)){
2390 av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2391 s->width= s->height= 0;
2395 if (s->theora >= 0x030400)
2397 skip_bits(gb, 32); /* total number of superblocks in a frame */
2398 // fixme, the next field is 36bits long
2399 skip_bits(gb, 32); /* total number of blocks in a frame */
2400 skip_bits(gb, 4); /* total number of blocks in a frame */
2401 skip_bits(gb, 32); /* total number of macroblocks in a frame */
2403 skip_bits(gb, 24); /* frame width */
2404 skip_bits(gb, 24); /* frame height */
2408 skip_bits(gb, 24); /* frame width */
2409 skip_bits(gb, 24); /* frame height */
2412 if (s->theora >= 0x030200) {
2413 skip_bits(gb, 8); /* offset x */
2414 skip_bits(gb, 8); /* offset y */
2417 skip_bits(gb, 32); /* fps numerator */
2418 skip_bits(gb, 32); /* fps denumerator */
2419 skip_bits(gb, 24); /* aspect numerator */
2420 skip_bits(gb, 24); /* aspect denumerator */
2422 if (s->theora < 0x030200)
2423 skip_bits(gb, 5); /* keyframe frequency force */
2424 skip_bits(gb, 8); /* colorspace */
2425 if (s->theora >= 0x030400)
2426 skip_bits(gb, 2); /* pixel format: 420,res,422,444 */
2427 skip_bits(gb, 24); /* bitrate */
2429 skip_bits(gb, 6); /* quality hint */
2431 if (s->theora >= 0x030200)
2433 skip_bits(gb, 5); /* keyframe frequency force */
2435 if (s->theora < 0x030400)
2436 skip_bits(gb, 5); /* spare bits */
2439 // align_get_bits(gb);
2441 avctx->width = s->width;
2442 avctx->height = s->height;
2447 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2449 Vp3DecodeContext *s = avctx->priv_data;
2450 int i, n, matrices, inter, plane;
2452 if (s->theora >= 0x030200) {
2453 n = get_bits(gb, 3);
2454 /* loop filter limit values table */
2455 for (i = 0; i < 64; i++)
2456 s->filter_limit_values[i] = get_bits(gb, n);
2459 if (s->theora >= 0x030200)
2460 n = get_bits(gb, 4) + 1;
2463 /* quality threshold table */
2464 for (i = 0; i < 64; i++)
2465 s->coded_ac_scale_factor[i] = get_bits(gb, n);
2467 if (s->theora >= 0x030200)
2468 n = get_bits(gb, 4) + 1;
2471 /* dc scale factor table */
2472 for (i = 0; i < 64; i++)
2473 s->coded_dc_scale_factor[i] = get_bits(gb, n);
2475 if (s->theora >= 0x030200)
2476 matrices = get_bits(gb, 9) + 1;
2481 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2485 for(n=0; n<matrices; n++){
2486 for (i = 0; i < 64; i++)
2487 s->base_matrix[n][i]= get_bits(gb, 8);
2490 for (inter = 0; inter <= 1; inter++) {
2491 for (plane = 0; plane <= 2; plane++) {
2493 if (inter || plane > 0)
2494 newqr = get_bits(gb, 1);
2497 if(inter && get_bits(gb, 1)){
2501 qtj= (3*inter + plane - 1) / 3;
2502 plj= (plane + 2) % 3;
2504 s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2505 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2506 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2512 i= get_bits(gb, av_log2(matrices-1)+1);
2514 av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2517 s->qr_base[inter][plane][qri]= i;
2520 i = get_bits(gb, av_log2(63-qi)+1) + 1;
2521 s->qr_size[inter][plane][qri++]= i;
2526 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2529 s->qr_count[inter][plane]= qri;
2534 /* Huffman tables */
2535 for (s->hti = 0; s->hti < 80; s->hti++) {
2537 s->huff_code_size = 1;
2538 if (!get_bits(gb, 1)) {
2540 read_huffman_tree(avctx, gb);
2542 read_huffman_tree(avctx, gb);
2546 s->theora_tables = 1;
2551 static int theora_decode_init(AVCodecContext *avctx)
2553 Vp3DecodeContext *s = avctx->priv_data;
2556 uint8_t *header_start[3];
2562 if (!avctx->extradata_size)
2564 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2568 if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2569 42, header_start, header_len) < 0) {
2570 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2575 init_get_bits(&gb, header_start[i], header_len[i]);
2577 ptype = get_bits(&gb, 8);
2578 debug_vp3("Theora headerpacket type: %x\n", ptype);
2580 if (!(ptype & 0x80))
2582 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2586 // FIXME: Check for this as well.
2587 skip_bits(&gb, 6*8); /* "theora" */
2592 theora_decode_header(avctx, &gb);
2595 // FIXME: is this needed? it breaks sometimes
2596 // theora_decode_comments(avctx, gb);
2599 theora_decode_tables(avctx, &gb);
2602 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2605 if(8*header_len[i] != get_bits_count(&gb))
2606 av_log(avctx, AV_LOG_ERROR, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2607 if (s->theora < 0x030200)
2611 vp3_decode_init(avctx);
2615 AVCodec theora_decoder = {
2619 sizeof(Vp3DecodeContext),
2629 AVCodec vp3_decoder = {
2633 sizeof(Vp3DecodeContext),