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"
45 #define FRAGMENT_PIXELS 8
50 * Define one or more of the following compile-time variables to 1 to obtain
51 * elaborate information about certain aspects of the decoding process.
53 * KEYFRAMES_ONLY: set this to 1 to only see keyframes (VP3 slideshow mode)
54 * DEBUG_VP3: high-level decoding flow
55 * DEBUG_INIT: initialization parameters
56 * DEBUG_DEQUANTIZERS: display how the dequanization tables are built
57 * DEBUG_BLOCK_CODING: unpacking the superblock/macroblock/fragment coding
58 * DEBUG_MODES: unpacking the coding modes for individual fragments
59 * DEBUG_VECTORS: display the motion vectors
60 * DEBUG_TOKEN: display exhaustive information about each DCT token
61 * DEBUG_VLC: display the VLCs as they are extracted from the stream
62 * DEBUG_DC_PRED: display the process of reversing DC prediction
63 * DEBUG_IDCT: show every detail of the IDCT process
66 #define KEYFRAMES_ONLY 0
70 #define DEBUG_DEQUANTIZERS 0
71 #define DEBUG_BLOCK_CODING 0
73 #define DEBUG_VECTORS 0
76 #define DEBUG_DC_PRED 0
80 #define debug_vp3(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
82 static inline void debug_vp3(const char *format, ...) { }
86 #define debug_init(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
88 static inline void debug_init(const char *format, ...) { }
91 #if DEBUG_DEQUANTIZERS
92 #define debug_dequantizers(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
94 static inline void debug_dequantizers(const char *format, ...) { }
97 #if DEBUG_BLOCK_CODING
98 #define debug_block_coding(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
100 static inline void debug_block_coding(const char *format, ...) { }
104 #define debug_modes(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
106 static inline void debug_modes(const char *format, ...) { }
110 #define debug_vectors(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
112 static inline void debug_vectors(const char *format, ...) { }
116 #define debug_token(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
118 static inline void debug_token(const char *format, ...) { }
122 #define debug_vlc(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
124 static inline void debug_vlc(const char *format, ...) { }
128 #define debug_dc_pred(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
130 static inline void debug_dc_pred(const char *format, ...) { }
134 #define debug_idct(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
136 static inline void debug_idct(const char *format, ...) { }
139 typedef struct Coeff {
145 //FIXME split things out into their own arrays
146 typedef struct Vp3Fragment {
148 /* address of first pixel taking into account which plane the fragment
149 * lives on as well as the plane stride */
151 /* this is the macroblock that the fragment belongs to */
153 uint8_t coding_method;
159 #define SB_NOT_CODED 0
160 #define SB_PARTIALLY_CODED 1
161 #define SB_FULLY_CODED 2
163 #define MODE_INTER_NO_MV 0
165 #define MODE_INTER_PLUS_MV 2
166 #define MODE_INTER_LAST_MV 3
167 #define MODE_INTER_PRIOR_LAST 4
168 #define MODE_USING_GOLDEN 5
169 #define MODE_GOLDEN_MV 6
170 #define MODE_INTER_FOURMV 7
171 #define CODING_MODE_COUNT 8
173 /* special internal mode */
176 /* There are 6 preset schemes, plus a free-form scheme */
177 static int ModeAlphabet[7][CODING_MODE_COUNT] =
179 /* this is the custom scheme */
180 { 0, 0, 0, 0, 0, 0, 0, 0 },
182 /* scheme 1: Last motion vector dominates */
183 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
184 MODE_INTER_PLUS_MV, MODE_INTER_NO_MV,
185 MODE_INTRA, MODE_USING_GOLDEN,
186 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
189 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
190 MODE_INTER_NO_MV, MODE_INTER_PLUS_MV,
191 MODE_INTRA, MODE_USING_GOLDEN,
192 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
195 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
196 MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
197 MODE_INTRA, MODE_USING_GOLDEN,
198 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
201 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
202 MODE_INTER_NO_MV, MODE_INTER_PRIOR_LAST,
203 MODE_INTRA, MODE_USING_GOLDEN,
204 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
206 /* scheme 5: No motion vector dominates */
207 { MODE_INTER_NO_MV, MODE_INTER_LAST_MV,
208 MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
209 MODE_INTRA, MODE_USING_GOLDEN,
210 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
213 { MODE_INTER_NO_MV, MODE_USING_GOLDEN,
214 MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
215 MODE_INTER_PLUS_MV, MODE_INTRA,
216 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
220 #define MIN_DEQUANT_VAL 2
222 typedef struct Vp3DecodeContext {
223 AVCodecContext *avctx;
224 int theora, theora_tables;
227 AVFrame golden_frame;
229 AVFrame current_frame;
237 int last_quality_index;
239 int superblock_count;
240 int superblock_width;
241 int superblock_height;
242 int y_superblock_width;
243 int y_superblock_height;
244 int c_superblock_width;
245 int c_superblock_height;
246 int u_superblock_start;
247 int v_superblock_start;
248 unsigned char *superblock_coding;
250 int macroblock_count;
251 int macroblock_width;
252 int macroblock_height;
258 Vp3Fragment *all_fragments;
261 int fragment_start[3];
266 uint16_t coded_dc_scale_factor[64];
267 uint32_t coded_ac_scale_factor[64];
268 uint8_t base_matrix[384][64];
269 uint8_t qr_count[2][3];
270 uint8_t qr_size [2][3][64];
271 uint16_t qr_base[2][3][64];
273 /* this is a list of indices into the all_fragments array indicating
274 * which of the fragments are coded */
275 int *coded_fragment_list;
276 int coded_fragment_list_index;
277 int pixel_addresses_inited;
285 VLC superblock_run_length_vlc;
286 VLC fragment_run_length_vlc;
288 VLC motion_vector_vlc;
290 /* these arrays need to be on 16-byte boundaries since SSE2 operations
292 DECLARE_ALIGNED_16(int16_t, qmat[2][4][64]); //<qmat[is_inter][plane]
294 /* This table contains superblock_count * 16 entries. Each set of 16
295 * numbers corresponds to the fragment indices 0..15 of the superblock.
296 * An entry will be -1 to indicate that no entry corresponds to that
298 int *superblock_fragments;
300 /* This table contains superblock_count * 4 entries. Each set of 4
301 * numbers corresponds to the macroblock indices 0..3 of the superblock.
302 * An entry will be -1 to indicate that no entry corresponds to that
304 int *superblock_macroblocks;
306 /* This table contains macroblock_count * 6 entries. Each set of 6
307 * numbers corresponds to the fragment indices 0..5 which comprise
308 * the macroblock (4 Y fragments and 2 C fragments). */
309 int *macroblock_fragments;
310 /* This is an array that indicates how a particular macroblock
312 unsigned char *macroblock_coding;
314 int first_coded_y_fragment;
315 int first_coded_c_fragment;
316 int last_coded_y_fragment;
317 int last_coded_c_fragment;
319 uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
320 int8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
327 uint16_t huffman_table[80][32][2];
329 uint32_t filter_limit_values[64];
330 int bounding_values_array[256];
333 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb);
335 /************************************************************************
336 * VP3 specific functions
337 ************************************************************************/
340 * This function sets up all of the various blocks mappings:
341 * superblocks <-> fragments, macroblocks <-> fragments,
342 * superblocks <-> macroblocks
344 * Returns 0 is successful; returns 1 if *anything* went wrong.
346 static int init_block_mapping(Vp3DecodeContext *s)
349 signed int hilbert_walk_mb[4];
351 int current_fragment = 0;
352 int current_width = 0;
353 int current_height = 0;
356 int superblock_row_inc = 0;
358 int mapping_index = 0;
360 int current_macroblock;
363 signed char travel_width[16] = {
370 signed char travel_height[16] = {
377 signed char travel_width_mb[4] = {
381 signed char travel_height_mb[4] = {
385 debug_vp3(" vp3: initialize block mapping tables\n");
387 hilbert_walk_mb[0] = 1;
388 hilbert_walk_mb[1] = s->macroblock_width;
389 hilbert_walk_mb[2] = 1;
390 hilbert_walk_mb[3] = -s->macroblock_width;
392 /* iterate through each superblock (all planes) and map the fragments */
393 for (i = 0; i < s->superblock_count; i++) {
394 debug_init(" superblock %d (u starts @ %d, v starts @ %d)\n",
395 i, s->u_superblock_start, s->v_superblock_start);
397 /* time to re-assign the limits? */
400 /* start of Y superblocks */
401 right_edge = s->fragment_width;
402 bottom_edge = s->fragment_height;
405 superblock_row_inc = 3 * s->fragment_width -
406 (s->y_superblock_width * 4 - s->fragment_width);
408 /* the first operation for this variable is to advance by 1 */
409 current_fragment = -1;
411 } else if (i == s->u_superblock_start) {
413 /* start of U superblocks */
414 right_edge = s->fragment_width / 2;
415 bottom_edge = s->fragment_height / 2;
418 superblock_row_inc = 3 * (s->fragment_width / 2) -
419 (s->c_superblock_width * 4 - s->fragment_width / 2);
421 /* the first operation for this variable is to advance by 1 */
422 current_fragment = s->fragment_start[1] - 1;
424 } else if (i == s->v_superblock_start) {
426 /* start of V superblocks */
427 right_edge = s->fragment_width / 2;
428 bottom_edge = s->fragment_height / 2;
431 superblock_row_inc = 3 * (s->fragment_width / 2) -
432 (s->c_superblock_width * 4 - s->fragment_width / 2);
434 /* the first operation for this variable is to advance by 1 */
435 current_fragment = s->fragment_start[2] - 1;
439 if (current_width >= right_edge - 1) {
440 /* reset width and move to next superblock row */
444 /* fragment is now at the start of a new superblock row */
445 current_fragment += superblock_row_inc;
448 /* iterate through all 16 fragments in a superblock */
449 for (j = 0; j < 16; j++) {
450 current_fragment += travel_width[j] + right_edge * travel_height[j];
451 current_width += travel_width[j];
452 current_height += travel_height[j];
454 /* check if the fragment is in bounds */
455 if ((current_width < right_edge) &&
456 (current_height < bottom_edge)) {
457 s->superblock_fragments[mapping_index] = current_fragment;
458 debug_init(" mapping fragment %d to superblock %d, position %d (%d/%d x %d/%d)\n",
459 s->superblock_fragments[mapping_index], i, j,
460 current_width, right_edge, current_height, bottom_edge);
462 s->superblock_fragments[mapping_index] = -1;
463 debug_init(" superblock %d, position %d has no fragment (%d/%d x %d/%d)\n",
465 current_width, right_edge, current_height, bottom_edge);
472 /* initialize the superblock <-> macroblock mapping; iterate through
473 * all of the Y plane superblocks to build this mapping */
474 right_edge = s->macroblock_width;
475 bottom_edge = s->macroblock_height;
478 superblock_row_inc = s->macroblock_width -
479 (s->y_superblock_width * 2 - s->macroblock_width);;
480 hilbert = hilbert_walk_mb;
482 current_macroblock = -1;
483 for (i = 0; i < s->u_superblock_start; i++) {
485 if (current_width >= right_edge - 1) {
486 /* reset width and move to next superblock row */
490 /* macroblock is now at the start of a new superblock row */
491 current_macroblock += superblock_row_inc;
494 /* iterate through each potential macroblock in the superblock */
495 for (j = 0; j < 4; j++) {
496 current_macroblock += hilbert_walk_mb[j];
497 current_width += travel_width_mb[j];
498 current_height += travel_height_mb[j];
500 /* check if the macroblock is in bounds */
501 if ((current_width < right_edge) &&
502 (current_height < bottom_edge)) {
503 s->superblock_macroblocks[mapping_index] = current_macroblock;
504 debug_init(" mapping macroblock %d to superblock %d, position %d (%d/%d x %d/%d)\n",
505 s->superblock_macroblocks[mapping_index], i, j,
506 current_width, right_edge, current_height, bottom_edge);
508 s->superblock_macroblocks[mapping_index] = -1;
509 debug_init(" superblock %d, position %d has no macroblock (%d/%d x %d/%d)\n",
511 current_width, right_edge, current_height, bottom_edge);
518 /* initialize the macroblock <-> fragment mapping */
519 current_fragment = 0;
520 current_macroblock = 0;
522 for (i = 0; i < s->fragment_height; i += 2) {
524 for (j = 0; j < s->fragment_width; j += 2) {
526 debug_init(" macroblock %d contains fragments: ", current_macroblock);
527 s->all_fragments[current_fragment].macroblock = current_macroblock;
528 s->macroblock_fragments[mapping_index++] = current_fragment;
529 debug_init("%d ", current_fragment);
531 if (j + 1 < s->fragment_width) {
532 s->all_fragments[current_fragment + 1].macroblock = current_macroblock;
533 s->macroblock_fragments[mapping_index++] = current_fragment + 1;
534 debug_init("%d ", current_fragment + 1);
536 s->macroblock_fragments[mapping_index++] = -1;
538 if (i + 1 < s->fragment_height) {
539 s->all_fragments[current_fragment + s->fragment_width].macroblock =
541 s->macroblock_fragments[mapping_index++] =
542 current_fragment + s->fragment_width;
543 debug_init("%d ", current_fragment + s->fragment_width);
545 s->macroblock_fragments[mapping_index++] = -1;
547 if ((j + 1 < s->fragment_width) && (i + 1 < s->fragment_height)) {
548 s->all_fragments[current_fragment + s->fragment_width + 1].macroblock =
550 s->macroblock_fragments[mapping_index++] =
551 current_fragment + s->fragment_width + 1;
552 debug_init("%d ", current_fragment + s->fragment_width + 1);
554 s->macroblock_fragments[mapping_index++] = -1;
557 c_fragment = s->fragment_start[1] +
558 (i * s->fragment_width / 4) + (j / 2);
559 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
560 s->macroblock_fragments[mapping_index++] = c_fragment;
561 debug_init("%d ", c_fragment);
563 c_fragment = s->fragment_start[2] +
564 (i * s->fragment_width / 4) + (j / 2);
565 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
566 s->macroblock_fragments[mapping_index++] = c_fragment;
567 debug_init("%d ", c_fragment);
571 if (j + 2 <= s->fragment_width)
572 current_fragment += 2;
575 current_macroblock++;
578 current_fragment += s->fragment_width;
581 return 0; /* successful path out */
585 * This function wipes out all of the fragment data.
587 static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
591 /* zero out all of the fragment information */
592 s->coded_fragment_list_index = 0;
593 for (i = 0; i < s->fragment_count; i++) {
594 s->all_fragments[i].coeff_count = 0;
595 s->all_fragments[i].motion_x = 127;
596 s->all_fragments[i].motion_y = 127;
597 s->all_fragments[i].next_coeff= NULL;
599 s->coeffs[i].coeff=0;
600 s->coeffs[i].next= NULL;
605 * This function sets up the dequantization tables used for a particular
608 static void init_dequantizer(Vp3DecodeContext *s)
610 int ac_scale_factor = s->coded_ac_scale_factor[s->quality_index];
611 int dc_scale_factor = s->coded_dc_scale_factor[s->quality_index];
612 int i, plane, inter, qri, bmi, bmj, qistart;
614 debug_vp3(" vp3: initializing dequantization tables\n");
616 for(inter=0; inter<2; inter++){
617 for(plane=0; plane<3; plane++){
619 for(qri=0; qri<s->qr_count[inter][plane]; qri++){
620 sum+= s->qr_size[inter][plane][qri];
621 if(s->quality_index <= sum)
624 qistart= sum - s->qr_size[inter][plane][qri];
625 bmi= s->qr_base[inter][plane][qri ];
626 bmj= s->qr_base[inter][plane][qri+1];
628 int coeff= ( 2*(sum -s->quality_index)*s->base_matrix[bmi][i]
629 - 2*(qistart-s->quality_index)*s->base_matrix[bmj][i]
630 + s->qr_size[inter][plane][qri])
631 / (2*s->qr_size[inter][plane][qri]);
633 int qmin= 8<<(inter + !i);
634 int qscale= i ? ac_scale_factor : dc_scale_factor;
636 s->qmat[inter][plane][i]= clip((qscale * coeff)/100 * 4, qmin, 4096);
641 memset(s->qscale_table, (FFMAX(s->qmat[0][0][1], s->qmat[0][1][1])+8)/16, 512); //FIXME finetune
645 * This function initializes the loop filter boundary limits if the frame's
646 * quality index is different from the previous frame's.
648 static void init_loop_filter(Vp3DecodeContext *s)
650 int *bounding_values= s->bounding_values_array+127;
654 filter_limit = s->filter_limit_values[s->quality_index];
656 /* set up the bounding values */
657 memset(s->bounding_values_array, 0, 256 * sizeof(int));
658 for (x = 0; x < filter_limit; x++) {
659 bounding_values[-x - filter_limit] = -filter_limit + x;
660 bounding_values[-x] = -x;
661 bounding_values[x] = x;
662 bounding_values[x + filter_limit] = filter_limit - x;
667 * This function unpacks all of the superblock/macroblock/fragment coding
668 * information from the bitstream.
670 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
673 int current_superblock = 0;
675 int decode_fully_flags = 0;
676 int decode_partial_blocks = 0;
677 int first_c_fragment_seen;
680 int current_fragment;
682 debug_vp3(" vp3: unpacking superblock coding\n");
686 debug_vp3(" keyframe-- all superblocks are fully coded\n");
687 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
691 /* unpack the list of partially-coded superblocks */
692 bit = get_bits(gb, 1);
693 /* toggle the bit because as soon as the first run length is
694 * fetched the bit will be toggled again */
696 while (current_superblock < s->superblock_count) {
697 if (current_run-- == 0) {
699 current_run = get_vlc2(gb,
700 s->superblock_run_length_vlc.table, 6, 2);
701 if (current_run == 33)
702 current_run += get_bits(gb, 12);
703 debug_block_coding(" setting superblocks %d..%d to %s\n",
705 current_superblock + current_run - 1,
706 (bit) ? "partially coded" : "not coded");
708 /* if any of the superblocks are not partially coded, flag
709 * a boolean to decode the list of fully-coded superblocks */
711 decode_fully_flags = 1;
714 /* make a note of the fact that there are partially coded
716 decode_partial_blocks = 1;
719 s->superblock_coding[current_superblock++] = bit;
722 /* unpack the list of fully coded superblocks if any of the blocks were
723 * not marked as partially coded in the previous step */
724 if (decode_fully_flags) {
726 current_superblock = 0;
728 bit = get_bits(gb, 1);
729 /* toggle the bit because as soon as the first run length is
730 * fetched the bit will be toggled again */
732 while (current_superblock < s->superblock_count) {
734 /* skip any superblocks already marked as partially coded */
735 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
737 if (current_run-- == 0) {
739 current_run = get_vlc2(gb,
740 s->superblock_run_length_vlc.table, 6, 2);
741 if (current_run == 33)
742 current_run += get_bits(gb, 12);
745 debug_block_coding(" setting superblock %d to %s\n",
747 (bit) ? "fully coded" : "not coded");
748 s->superblock_coding[current_superblock] = 2*bit;
750 current_superblock++;
754 /* if there were partial blocks, initialize bitstream for
755 * unpacking fragment codings */
756 if (decode_partial_blocks) {
759 bit = get_bits(gb, 1);
760 /* toggle the bit because as soon as the first run length is
761 * fetched the bit will be toggled again */
766 /* figure out which fragments are coded; iterate through each
767 * superblock (all planes) */
768 s->coded_fragment_list_index = 0;
769 s->next_coeff= s->coeffs + s->fragment_count;
770 s->first_coded_y_fragment = s->first_coded_c_fragment = 0;
771 s->last_coded_y_fragment = s->last_coded_c_fragment = -1;
772 first_c_fragment_seen = 0;
773 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
774 for (i = 0; i < s->superblock_count; i++) {
776 /* iterate through all 16 fragments in a superblock */
777 for (j = 0; j < 16; j++) {
779 /* if the fragment is in bounds, check its coding status */
780 current_fragment = s->superblock_fragments[i * 16 + j];
781 if (current_fragment >= s->fragment_count) {
782 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
783 current_fragment, s->fragment_count);
786 if (current_fragment != -1) {
787 if (s->superblock_coding[i] == SB_NOT_CODED) {
789 /* copy all the fragments from the prior frame */
790 s->all_fragments[current_fragment].coding_method =
793 } else if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
795 /* fragment may or may not be coded; this is the case
796 * that cares about the fragment coding runs */
797 if (current_run-- == 0) {
799 current_run = get_vlc2(gb,
800 s->fragment_run_length_vlc.table, 5, 2);
804 /* default mode; actual mode will be decoded in
806 s->all_fragments[current_fragment].coding_method =
808 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
809 s->coded_fragment_list[s->coded_fragment_list_index] =
811 if ((current_fragment >= s->fragment_start[1]) &&
812 (s->last_coded_y_fragment == -1) &&
813 (!first_c_fragment_seen)) {
814 s->first_coded_c_fragment = s->coded_fragment_list_index;
815 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
816 first_c_fragment_seen = 1;
818 s->coded_fragment_list_index++;
819 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
820 debug_block_coding(" superblock %d is partially coded, fragment %d is coded\n",
821 i, current_fragment);
823 /* not coded; copy this fragment from the prior frame */
824 s->all_fragments[current_fragment].coding_method =
826 debug_block_coding(" superblock %d is partially coded, fragment %d is not coded\n",
827 i, current_fragment);
832 /* fragments are fully coded in this superblock; actual
833 * coding will be determined in next step */
834 s->all_fragments[current_fragment].coding_method =
836 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
837 s->coded_fragment_list[s->coded_fragment_list_index] =
839 if ((current_fragment >= s->fragment_start[1]) &&
840 (s->last_coded_y_fragment == -1) &&
841 (!first_c_fragment_seen)) {
842 s->first_coded_c_fragment = s->coded_fragment_list_index;
843 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
844 first_c_fragment_seen = 1;
846 s->coded_fragment_list_index++;
847 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
848 debug_block_coding(" superblock %d is fully coded, fragment %d is coded\n",
849 i, current_fragment);
855 if (!first_c_fragment_seen)
856 /* only Y fragments coded in this frame */
857 s->last_coded_y_fragment = s->coded_fragment_list_index - 1;
859 /* end the list of coded C fragments */
860 s->last_coded_c_fragment = s->coded_fragment_list_index - 1;
862 debug_block_coding(" %d total coded fragments, y: %d -> %d, c: %d -> %d\n",
863 s->coded_fragment_list_index,
864 s->first_coded_y_fragment,
865 s->last_coded_y_fragment,
866 s->first_coded_c_fragment,
867 s->last_coded_c_fragment);
873 * This function unpacks all the coding mode data for individual macroblocks
874 * from the bitstream.
876 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
880 int current_macroblock;
881 int current_fragment;
884 debug_vp3(" vp3: unpacking encoding modes\n");
887 debug_vp3(" keyframe-- all blocks are coded as INTRA\n");
889 for (i = 0; i < s->fragment_count; i++)
890 s->all_fragments[i].coding_method = MODE_INTRA;
894 /* fetch the mode coding scheme for this frame */
895 scheme = get_bits(gb, 3);
896 debug_modes(" using mode alphabet %d\n", scheme);
898 /* is it a custom coding scheme? */
900 debug_modes(" custom mode alphabet ahead:\n");
901 for (i = 0; i < 8; i++)
902 ModeAlphabet[scheme][get_bits(gb, 3)] = i;
905 for (i = 0; i < 8; i++)
906 debug_modes(" mode[%d][%d] = %d\n", scheme, i,
907 ModeAlphabet[scheme][i]);
909 /* iterate through all of the macroblocks that contain 1 or more
911 for (i = 0; i < s->u_superblock_start; i++) {
913 for (j = 0; j < 4; j++) {
914 current_macroblock = s->superblock_macroblocks[i * 4 + j];
915 if ((current_macroblock == -1) ||
916 (s->macroblock_coding[current_macroblock] == MODE_COPY))
918 if (current_macroblock >= s->macroblock_count) {
919 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad macroblock number (%d >= %d)\n",
920 current_macroblock, s->macroblock_count);
924 /* mode 7 means get 3 bits for each coding mode */
926 coding_mode = get_bits(gb, 3);
928 coding_mode = ModeAlphabet[scheme]
929 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
931 s->macroblock_coding[current_macroblock] = coding_mode;
932 for (k = 0; k < 6; k++) {
934 s->macroblock_fragments[current_macroblock * 6 + k];
935 if (current_fragment == -1)
937 if (current_fragment >= s->fragment_count) {
938 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad fragment number (%d >= %d)\n",
939 current_fragment, s->fragment_count);
942 if (s->all_fragments[current_fragment].coding_method !=
944 s->all_fragments[current_fragment].coding_method =
948 debug_modes(" coding method for macroblock starting @ fragment %d = %d\n",
949 s->macroblock_fragments[current_macroblock * 6], coding_mode);
958 * This function unpacks all the motion vectors for the individual
959 * macroblocks from the bitstream.
961 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
967 int last_motion_x = 0;
968 int last_motion_y = 0;
969 int prior_last_motion_x = 0;
970 int prior_last_motion_y = 0;
971 int current_macroblock;
972 int current_fragment;
974 debug_vp3(" vp3: unpacking motion vectors\n");
977 debug_vp3(" keyframe-- there are no motion vectors\n");
981 memset(motion_x, 0, 6 * sizeof(int));
982 memset(motion_y, 0, 6 * sizeof(int));
984 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
985 coding_mode = get_bits(gb, 1);
986 debug_vectors(" using %s scheme for unpacking motion vectors\n",
987 (coding_mode == 0) ? "VLC" : "fixed-length");
989 /* iterate through all of the macroblocks that contain 1 or more
991 for (i = 0; i < s->u_superblock_start; i++) {
993 for (j = 0; j < 4; j++) {
994 current_macroblock = s->superblock_macroblocks[i * 4 + j];
995 if ((current_macroblock == -1) ||
996 (s->macroblock_coding[current_macroblock] == MODE_COPY))
998 if (current_macroblock >= s->macroblock_count) {
999 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad macroblock number (%d >= %d)\n",
1000 current_macroblock, s->macroblock_count);
1004 current_fragment = s->macroblock_fragments[current_macroblock * 6];
1005 if (current_fragment >= s->fragment_count) {
1006 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d\n",
1007 current_fragment, s->fragment_count);
1010 switch (s->macroblock_coding[current_macroblock]) {
1012 case MODE_INTER_PLUS_MV:
1013 case MODE_GOLDEN_MV:
1014 /* all 6 fragments use the same motion vector */
1015 if (coding_mode == 0) {
1016 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1017 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1019 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
1020 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
1023 for (k = 1; k < 6; k++) {
1024 motion_x[k] = motion_x[0];
1025 motion_y[k] = motion_y[0];
1028 /* vector maintenance, only on MODE_INTER_PLUS_MV */
1029 if (s->macroblock_coding[current_macroblock] ==
1030 MODE_INTER_PLUS_MV) {
1031 prior_last_motion_x = last_motion_x;
1032 prior_last_motion_y = last_motion_y;
1033 last_motion_x = motion_x[0];
1034 last_motion_y = motion_y[0];
1038 case MODE_INTER_FOURMV:
1039 /* fetch 4 vectors from the bitstream, one for each
1040 * Y fragment, then average for the C fragment vectors */
1041 motion_x[4] = motion_y[4] = 0;
1042 for (k = 0; k < 4; k++) {
1043 if (coding_mode == 0) {
1044 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1045 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1047 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
1048 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
1050 motion_x[4] += motion_x[k];
1051 motion_y[4] += motion_y[k];
1055 motion_x[4]= RSHIFT(motion_x[4], 2);
1057 motion_y[4]= RSHIFT(motion_y[4], 2);
1059 /* vector maintenance; vector[3] is treated as the
1060 * last vector in this case */
1061 prior_last_motion_x = last_motion_x;
1062 prior_last_motion_y = last_motion_y;
1063 last_motion_x = motion_x[3];
1064 last_motion_y = motion_y[3];
1067 case MODE_INTER_LAST_MV:
1068 /* all 6 fragments use the last motion vector */
1069 motion_x[0] = last_motion_x;
1070 motion_y[0] = last_motion_y;
1071 for (k = 1; k < 6; k++) {
1072 motion_x[k] = motion_x[0];
1073 motion_y[k] = motion_y[0];
1076 /* no vector maintenance (last vector remains the
1080 case MODE_INTER_PRIOR_LAST:
1081 /* all 6 fragments use the motion vector prior to the
1082 * last motion vector */
1083 motion_x[0] = prior_last_motion_x;
1084 motion_y[0] = prior_last_motion_y;
1085 for (k = 1; k < 6; k++) {
1086 motion_x[k] = motion_x[0];
1087 motion_y[k] = motion_y[0];
1090 /* vector maintenance */
1091 prior_last_motion_x = last_motion_x;
1092 prior_last_motion_y = last_motion_y;
1093 last_motion_x = motion_x[0];
1094 last_motion_y = motion_y[0];
1098 /* covers intra, inter without MV, golden without MV */
1099 memset(motion_x, 0, 6 * sizeof(int));
1100 memset(motion_y, 0, 6 * sizeof(int));
1102 /* no vector maintenance */
1106 /* assign the motion vectors to the correct fragments */
1107 debug_vectors(" vectors for macroblock starting @ fragment %d (coding method %d):\n",
1109 s->macroblock_coding[current_macroblock]);
1110 for (k = 0; k < 6; k++) {
1112 s->macroblock_fragments[current_macroblock * 6 + k];
1113 if (current_fragment == -1)
1115 if (current_fragment >= s->fragment_count) {
1116 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d)\n",
1117 current_fragment, s->fragment_count);
1120 s->all_fragments[current_fragment].motion_x = motion_x[k];
1121 s->all_fragments[current_fragment].motion_y = motion_y[k];
1122 debug_vectors(" vector %d: fragment %d = (%d, %d)\n",
1123 k, current_fragment, motion_x[k], motion_y[k]);
1133 * This function is called by unpack_dct_coeffs() to extract the VLCs from
1134 * the bitstream. The VLCs encode tokens which are used to unpack DCT
1135 * data. This function unpacks all the VLCs for either the Y plane or both
1136 * C planes, and is called for DC coefficients or different AC coefficient
1137 * levels (since different coefficient types require different VLC tables.
1139 * This function returns a residual eob run. E.g, if a particular token gave
1140 * instructions to EOB the next 5 fragments and there were only 2 fragments
1141 * left in the current fragment range, 3 would be returned so that it could
1142 * be passed into the next call to this same function.
1144 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
1145 VLC *table, int coeff_index,
1146 int first_fragment, int last_fragment,
1153 Vp3Fragment *fragment;
1154 uint8_t *perm= s->scantable.permutated;
1157 if ((first_fragment >= s->fragment_count) ||
1158 (last_fragment >= s->fragment_count)) {
1160 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vlcs(): bad fragment number (%d -> %d ?)\n",
1161 first_fragment, last_fragment);
1165 for (i = first_fragment; i <= last_fragment; i++) {
1167 fragment = &s->all_fragments[s->coded_fragment_list[i]];
1168 if (fragment->coeff_count > coeff_index)
1172 /* decode a VLC into a token */
1173 token = get_vlc2(gb, table->table, 5, 3);
1174 debug_vlc(" token = %2d, ", token);
1175 /* use the token to get a zero run, a coefficient, and an eob run */
1177 eob_run = eob_run_base[token];
1178 if (eob_run_get_bits[token])
1179 eob_run += get_bits(gb, eob_run_get_bits[token]);
1180 coeff = zero_run = 0;
1182 bits_to_get = coeff_get_bits[token];
1184 coeff = coeff_tables[token][0];
1186 coeff = coeff_tables[token][get_bits(gb, bits_to_get)];
1188 zero_run = zero_run_base[token];
1189 if (zero_run_get_bits[token])
1190 zero_run += get_bits(gb, zero_run_get_bits[token]);
1195 fragment->coeff_count += zero_run;
1196 if (fragment->coeff_count < 64){
1197 fragment->next_coeff->coeff= coeff;
1198 fragment->next_coeff->index= perm[fragment->coeff_count++]; //FIXME perm here already?
1199 fragment->next_coeff->next= s->next_coeff;
1200 s->next_coeff->next=NULL;
1201 fragment->next_coeff= s->next_coeff++;
1203 debug_vlc(" fragment %d coeff = %d\n",
1204 s->coded_fragment_list[i], fragment->next_coeff[coeff_index]);
1206 fragment->coeff_count |= 128;
1207 debug_vlc(" fragment %d eob with %d coefficients\n",
1208 s->coded_fragment_list[i], fragment->coeff_count&127);
1217 * This function unpacks all of the DCT coefficient data from the
1220 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1227 int residual_eob_run = 0;
1229 /* fetch the DC table indices */
1230 dc_y_table = get_bits(gb, 4);
1231 dc_c_table = get_bits(gb, 4);
1233 /* unpack the Y plane DC coefficients */
1234 debug_vp3(" vp3: unpacking Y plane DC coefficients using table %d\n",
1236 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1237 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1239 /* unpack the C plane DC coefficients */
1240 debug_vp3(" vp3: unpacking C plane DC coefficients using table %d\n",
1242 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1243 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1245 /* fetch the AC table indices */
1246 ac_y_table = get_bits(gb, 4);
1247 ac_c_table = get_bits(gb, 4);
1249 /* unpack the group 1 AC coefficients (coeffs 1-5) */
1250 for (i = 1; i <= 5; i++) {
1252 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1254 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_y_table], i,
1255 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1257 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1259 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_c_table], i,
1260 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1263 /* unpack the group 2 AC coefficients (coeffs 6-14) */
1264 for (i = 6; i <= 14; i++) {
1266 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1268 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_y_table], i,
1269 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1271 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1273 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_c_table], i,
1274 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1277 /* unpack the group 3 AC coefficients (coeffs 15-27) */
1278 for (i = 15; i <= 27; i++) {
1280 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1282 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_y_table], i,
1283 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1285 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1287 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_c_table], i,
1288 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1291 /* unpack the group 4 AC coefficients (coeffs 28-63) */
1292 for (i = 28; i <= 63; i++) {
1294 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1296 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_y_table], i,
1297 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1299 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1301 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_c_table], i,
1302 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1309 * This function reverses the DC prediction for each coded fragment in
1310 * the frame. Much of this function is adapted directly from the original
1313 #define COMPATIBLE_FRAME(x) \
1314 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1315 #define FRAME_CODED(x) (s->all_fragments[x].coding_method != MODE_COPY)
1316 #define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
1318 static void reverse_dc_prediction(Vp3DecodeContext *s,
1321 int fragment_height)
1330 int i = first_fragment;
1334 /* DC values for the left, up-left, up, and up-right fragments */
1335 int vl, vul, vu, vur;
1337 /* indices for the left, up-left, up, and up-right fragments */
1341 * The 6 fields mean:
1342 * 0: up-left multiplier
1344 * 2: up-right multiplier
1345 * 3: left multiplier
1347 int predictor_transform[16][4] = {
1349 { 0, 0, 0,128}, // PL
1350 { 0, 0,128, 0}, // PUR
1351 { 0, 0, 53, 75}, // PUR|PL
1352 { 0,128, 0, 0}, // PU
1353 { 0, 64, 0, 64}, // PU|PL
1354 { 0,128, 0, 0}, // PU|PUR
1355 { 0, 0, 53, 75}, // PU|PUR|PL
1356 {128, 0, 0, 0}, // PUL
1357 { 0, 0, 0,128}, // PUL|PL
1358 { 64, 0, 64, 0}, // PUL|PUR
1359 { 0, 0, 53, 75}, // PUL|PUR|PL
1360 { 0,128, 0, 0}, // PUL|PU
1361 {-104,116, 0,116}, // PUL|PU|PL
1362 { 24, 80, 24, 0}, // PUL|PU|PUR
1363 {-104,116, 0,116} // PUL|PU|PUR|PL
1366 /* This table shows which types of blocks can use other blocks for
1367 * prediction. For example, INTRA is the only mode in this table to
1368 * have a frame number of 0. That means INTRA blocks can only predict
1369 * from other INTRA blocks. There are 2 golden frame coding types;
1370 * blocks encoding in these modes can only predict from other blocks
1371 * that were encoded with these 1 of these 2 modes. */
1372 unsigned char compatible_frame[8] = {
1373 1, /* MODE_INTER_NO_MV */
1375 1, /* MODE_INTER_PLUS_MV */
1376 1, /* MODE_INTER_LAST_MV */
1377 1, /* MODE_INTER_PRIOR_MV */
1378 2, /* MODE_USING_GOLDEN */
1379 2, /* MODE_GOLDEN_MV */
1380 1 /* MODE_INTER_FOUR_MV */
1382 int current_frame_type;
1384 /* there is a last DC predictor for each of the 3 frame types */
1389 debug_vp3(" vp3: reversing DC prediction\n");
1391 vul = vu = vur = vl = 0;
1392 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1394 /* for each fragment row... */
1395 for (y = 0; y < fragment_height; y++) {
1397 /* for each fragment in a row... */
1398 for (x = 0; x < fragment_width; x++, i++) {
1400 /* reverse prediction if this block was coded */
1401 if (s->all_fragments[i].coding_method != MODE_COPY) {
1403 current_frame_type =
1404 compatible_frame[s->all_fragments[i].coding_method];
1405 debug_dc_pred(" frag %d: orig DC = %d, ",
1412 if(FRAME_CODED(l) && COMPATIBLE_FRAME(l))
1416 u= i-fragment_width;
1418 if(FRAME_CODED(u) && COMPATIBLE_FRAME(u))
1421 ul= i-fragment_width-1;
1423 if(FRAME_CODED(ul) && COMPATIBLE_FRAME(ul))
1426 if(x + 1 < fragment_width){
1427 ur= i-fragment_width+1;
1429 if(FRAME_CODED(ur) && COMPATIBLE_FRAME(ur))
1434 debug_dc_pred("transform = %d, ", transform);
1436 if (transform == 0) {
1438 /* if there were no fragments to predict from, use last
1440 predicted_dc = last_dc[current_frame_type];
1441 debug_dc_pred("from last DC (%d) = %d\n",
1442 current_frame_type, DC_COEFF(i));
1446 /* apply the appropriate predictor transform */
1448 (predictor_transform[transform][0] * vul) +
1449 (predictor_transform[transform][1] * vu) +
1450 (predictor_transform[transform][2] * vur) +
1451 (predictor_transform[transform][3] * vl);
1453 predicted_dc /= 128;
1455 /* check for outranging on the [ul u l] and
1456 * [ul u ur l] predictors */
1457 if ((transform == 13) || (transform == 15)) {
1458 if (FFABS(predicted_dc - vu) > 128)
1460 else if (FFABS(predicted_dc - vl) > 128)
1462 else if (FFABS(predicted_dc - vul) > 128)
1466 debug_dc_pred("from pred DC = %d\n",
1470 /* at long last, apply the predictor */
1471 if(s->coeffs[i].index){
1472 *s->next_coeff= s->coeffs[i];
1473 s->coeffs[i].index=0;
1474 s->coeffs[i].coeff=0;
1475 s->coeffs[i].next= s->next_coeff++;
1477 s->coeffs[i].coeff += predicted_dc;
1479 last_dc[current_frame_type] = DC_COEFF(i);
1480 if(DC_COEFF(i) && !(s->all_fragments[i].coeff_count&127)){
1481 s->all_fragments[i].coeff_count= 129;
1482 // s->all_fragments[i].next_coeff= s->next_coeff;
1483 s->coeffs[i].next= s->next_coeff;
1484 (s->next_coeff++)->next=NULL;
1492 static void horizontal_filter(unsigned char *first_pixel, int stride,
1493 int *bounding_values);
1494 static void vertical_filter(unsigned char *first_pixel, int stride,
1495 int *bounding_values);
1498 * Perform the final rendering for a particular slice of data.
1499 * The slice number ranges from 0..(macroblock_height - 1).
1501 static void render_slice(Vp3DecodeContext *s, int slice)
1505 int16_t *dequantizer;
1506 DECLARE_ALIGNED_16(DCTELEM, block[64]);
1507 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1508 int motion_halfpel_index;
1509 uint8_t *motion_source;
1511 int current_macroblock_entry = slice * s->macroblock_width * 6;
1513 if (slice >= s->macroblock_height)
1516 for (plane = 0; plane < 3; plane++) {
1517 uint8_t *output_plane = s->current_frame.data [plane];
1518 uint8_t * last_plane = s-> last_frame.data [plane];
1519 uint8_t *golden_plane = s-> golden_frame.data [plane];
1520 int stride = s->current_frame.linesize[plane];
1521 int plane_width = s->width >> !!plane;
1522 int plane_height = s->height >> !!plane;
1523 int y = slice * FRAGMENT_PIXELS << !plane ;
1524 int slice_height = y + (FRAGMENT_PIXELS << !plane);
1525 int i = s->macroblock_fragments[current_macroblock_entry + plane + 3*!!plane];
1527 if (!s->flipped_image) stride = -stride;
1530 if(FFABS(stride) > 2048)
1531 return; //various tables are fixed size
1533 /* for each fragment row in the slice (both of them)... */
1534 for (; y < slice_height; y += 8) {
1536 /* for each fragment in a row... */
1537 for (x = 0; x < plane_width; x += 8, i++) {
1539 if ((i < 0) || (i >= s->fragment_count)) {
1540 av_log(s->avctx, AV_LOG_ERROR, " vp3:render_slice(): bad fragment number (%d)\n", i);
1544 /* transform if this block was coded */
1545 if ((s->all_fragments[i].coding_method != MODE_COPY) &&
1546 !((s->avctx->flags & CODEC_FLAG_GRAY) && plane)) {
1548 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1549 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1550 motion_source= golden_plane;
1552 motion_source= last_plane;
1554 motion_source += s->all_fragments[i].first_pixel;
1555 motion_halfpel_index = 0;
1557 /* sort out the motion vector if this fragment is coded
1558 * using a motion vector method */
1559 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1560 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1562 motion_x = s->all_fragments[i].motion_x;
1563 motion_y = s->all_fragments[i].motion_y;
1565 motion_x= (motion_x>>1) | (motion_x&1);
1566 motion_y= (motion_y>>1) | (motion_y&1);
1569 src_x= (motion_x>>1) + x;
1570 src_y= (motion_y>>1) + y;
1571 if ((motion_x == 127) || (motion_y == 127))
1572 av_log(s->avctx, AV_LOG_ERROR, " help! got invalid motion vector! (%X, %X)\n", motion_x, motion_y);
1574 motion_halfpel_index = motion_x & 0x01;
1575 motion_source += (motion_x >> 1);
1577 motion_halfpel_index |= (motion_y & 0x01) << 1;
1578 motion_source += ((motion_y >> 1) * stride);
1580 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1581 uint8_t *temp= s->edge_emu_buffer;
1582 if(stride<0) temp -= 9*stride;
1583 else temp += 9*stride;
1585 ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1586 motion_source= temp;
1591 /* first, take care of copying a block from either the
1592 * previous or the golden frame */
1593 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1594 /* Note, it is possible to implement all MC cases with
1595 put_no_rnd_pixels_l2 which would look more like the
1596 VP3 source but this would be slower as
1597 put_no_rnd_pixels_tab is better optimzed */
1598 if(motion_halfpel_index != 3){
1599 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1600 output_plane + s->all_fragments[i].first_pixel,
1601 motion_source, stride, 8);
1603 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1604 s->dsp.put_no_rnd_pixels_l2[1](
1605 output_plane + s->all_fragments[i].first_pixel,
1607 motion_source + stride + 1 + d,
1610 dequantizer = s->qmat[1][plane];
1612 dequantizer = s->qmat[0][plane];
1615 /* dequantize the DCT coefficients */
1616 debug_idct("fragment %d, coding mode %d, DC = %d, dequant = %d:\n",
1617 i, s->all_fragments[i].coding_method,
1618 DC_COEFF(i), dequantizer[0]);
1620 if(s->avctx->idct_algo==FF_IDCT_VP3){
1621 Coeff *coeff= s->coeffs + i;
1622 memset(block, 0, sizeof(block));
1624 block[coeff->index]= coeff->coeff * dequantizer[coeff->index];
1628 Coeff *coeff= s->coeffs + i;
1629 memset(block, 0, sizeof(block));
1631 block[coeff->index]= (coeff->coeff * dequantizer[coeff->index] + 2)>>2;
1636 /* invert DCT and place (or add) in final output */
1638 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1639 if(s->avctx->idct_algo!=FF_IDCT_VP3)
1642 output_plane + s->all_fragments[i].first_pixel,
1647 output_plane + s->all_fragments[i].first_pixel,
1652 debug_idct("block after idct_%s():\n",
1653 (s->all_fragments[i].coding_method == MODE_INTRA)?
1655 for (m = 0; m < 8; m++) {
1656 for (n = 0; n < 8; n++) {
1657 debug_idct(" %3d", *(output_plane +
1658 s->all_fragments[i].first_pixel + (m * stride + n)));
1666 /* copy directly from the previous frame */
1667 s->dsp.put_pixels_tab[1][0](
1668 output_plane + s->all_fragments[i].first_pixel,
1669 last_plane + s->all_fragments[i].first_pixel,
1674 /* perform the left edge filter if:
1675 * - the fragment is not on the left column
1676 * - the fragment is coded in this frame
1677 * - the fragment is not coded in this frame but the left
1678 * fragment is coded in this frame (this is done instead
1679 * of a right edge filter when rendering the left fragment
1680 * since this fragment is not available yet) */
1682 ((s->all_fragments[i].coding_method != MODE_COPY) ||
1683 ((s->all_fragments[i].coding_method == MODE_COPY) &&
1684 (s->all_fragments[i - 1].coding_method != MODE_COPY)) )) {
1686 output_plane + s->all_fragments[i].first_pixel + 7*stride,
1687 -stride, s->bounding_values_array + 127);
1690 /* perform the top edge filter if:
1691 * - the fragment is not on the top row
1692 * - the fragment is coded in this frame
1693 * - the fragment is not coded in this frame but the above
1694 * fragment is coded in this frame (this is done instead
1695 * of a bottom edge filter when rendering the above
1696 * fragment since this fragment is not available yet) */
1698 ((s->all_fragments[i].coding_method != MODE_COPY) ||
1699 ((s->all_fragments[i].coding_method == MODE_COPY) &&
1700 (s->all_fragments[i - fragment_width].coding_method != MODE_COPY)) )) {
1702 output_plane + s->all_fragments[i].first_pixel - stride,
1703 -stride, s->bounding_values_array + 127);
1710 /* this looks like a good place for slice dispatch... */
1712 * if (slice == s->macroblock_height - 1)
1713 * dispatch (both last slice & 2nd-to-last slice);
1714 * else if (slice > 0)
1715 * dispatch (slice - 1);
1721 static void horizontal_filter(unsigned char *first_pixel, int stride,
1722 int *bounding_values)
1727 for (end= first_pixel + 8*stride; first_pixel != end; first_pixel += stride) {
1729 (first_pixel[-2] - first_pixel[ 1])
1730 +3*(first_pixel[ 0] - first_pixel[-1]);
1731 filter_value = bounding_values[(filter_value + 4) >> 3];
1732 first_pixel[-1] = clip_uint8(first_pixel[-1] + filter_value);
1733 first_pixel[ 0] = clip_uint8(first_pixel[ 0] - filter_value);
1737 static void vertical_filter(unsigned char *first_pixel, int stride,
1738 int *bounding_values)
1742 const int nstride= -stride;
1744 for (end= first_pixel + 8; first_pixel < end; first_pixel++) {
1746 (first_pixel[2 * nstride] - first_pixel[ stride])
1747 +3*(first_pixel[0 ] - first_pixel[nstride]);
1748 filter_value = bounding_values[(filter_value + 4) >> 3];
1749 first_pixel[nstride] = clip_uint8(first_pixel[nstride] + filter_value);
1750 first_pixel[0] = clip_uint8(first_pixel[0] - filter_value);
1754 static void apply_loop_filter(Vp3DecodeContext *s)
1758 int *bounding_values= s->bounding_values_array+127;
1761 int bounding_values_array[256];
1764 /* find the right loop limit value */
1765 for (x = 63; x >= 0; x--) {
1766 if (vp31_ac_scale_factor[x] >= s->quality_index)
1769 filter_limit = vp31_filter_limit_values[s->quality_index];
1771 /* set up the bounding values */
1772 memset(bounding_values_array, 0, 256 * sizeof(int));
1773 for (x = 0; x < filter_limit; x++) {
1774 bounding_values[-x - filter_limit] = -filter_limit + x;
1775 bounding_values[-x] = -x;
1776 bounding_values[x] = x;
1777 bounding_values[x + filter_limit] = filter_limit - x;
1781 for (plane = 0; plane < 3; plane++) {
1782 int width = s->fragment_width >> !!plane;
1783 int height = s->fragment_height >> !!plane;
1784 int fragment = s->fragment_start [plane];
1785 int stride = s->current_frame.linesize[plane];
1786 uint8_t *plane_data = s->current_frame.data [plane];
1787 if (!s->flipped_image) stride = -stride;
1789 for (y = 0; y < height; y++) {
1791 for (x = 0; x < width; x++) {
1793 /* do not perform left edge filter for left columns frags */
1795 (s->all_fragments[fragment].coding_method != MODE_COPY)) {
1797 plane_data + s->all_fragments[fragment].first_pixel,
1798 stride, bounding_values);
1801 /* do not perform top edge filter for top row fragments */
1803 (s->all_fragments[fragment].coding_method != MODE_COPY)) {
1805 plane_data + s->all_fragments[fragment].first_pixel,
1806 stride, bounding_values);
1809 /* do not perform right edge filter for right column
1810 * fragments or if right fragment neighbor is also coded
1811 * in this frame (it will be filtered in next iteration) */
1812 if ((x < width - 1) &&
1813 (s->all_fragments[fragment].coding_method != MODE_COPY) &&
1814 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1816 plane_data + s->all_fragments[fragment + 1].first_pixel,
1817 stride, bounding_values);
1820 /* do not perform bottom edge filter for bottom row
1821 * fragments or if bottom fragment neighbor is also coded
1822 * in this frame (it will be filtered in the next row) */
1823 if ((y < height - 1) &&
1824 (s->all_fragments[fragment].coding_method != MODE_COPY) &&
1825 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1827 plane_data + s->all_fragments[fragment + width].first_pixel,
1828 stride, bounding_values);
1832 STOP_TIMER("loop filter")
1839 * This function computes the first pixel addresses for each fragment.
1840 * This function needs to be invoked after the first frame is allocated
1841 * so that it has access to the plane strides.
1843 static void vp3_calculate_pixel_addresses(Vp3DecodeContext *s)
1848 /* figure out the first pixel addresses for each of the fragments */
1851 for (y = s->fragment_height; y > 0; y--) {
1852 for (x = 0; x < s->fragment_width; x++) {
1853 s->all_fragments[i++].first_pixel =
1854 s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
1855 s->golden_frame.linesize[0] +
1856 x * FRAGMENT_PIXELS;
1857 debug_init(" fragment %d, first pixel @ %d\n",
1858 i-1, s->all_fragments[i-1].first_pixel);
1863 i = s->fragment_start[1];
1864 for (y = s->fragment_height / 2; y > 0; y--) {
1865 for (x = 0; x < s->fragment_width / 2; x++) {
1866 s->all_fragments[i++].first_pixel =
1867 s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
1868 s->golden_frame.linesize[1] +
1869 x * FRAGMENT_PIXELS;
1870 debug_init(" fragment %d, first pixel @ %d\n",
1871 i-1, s->all_fragments[i-1].first_pixel);
1876 i = s->fragment_start[2];
1877 for (y = s->fragment_height / 2; y > 0; y--) {
1878 for (x = 0; x < s->fragment_width / 2; x++) {
1879 s->all_fragments[i++].first_pixel =
1880 s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
1881 s->golden_frame.linesize[2] +
1882 x * FRAGMENT_PIXELS;
1883 debug_init(" fragment %d, first pixel @ %d\n",
1884 i-1, s->all_fragments[i-1].first_pixel);
1889 /* FIXME: this should be merged with the above! */
1890 static void theora_calculate_pixel_addresses(Vp3DecodeContext *s)
1895 /* figure out the first pixel addresses for each of the fragments */
1898 for (y = 1; y <= s->fragment_height; y++) {
1899 for (x = 0; x < s->fragment_width; x++) {
1900 s->all_fragments[i++].first_pixel =
1901 s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
1902 s->golden_frame.linesize[0] +
1903 x * FRAGMENT_PIXELS;
1904 debug_init(" fragment %d, first pixel @ %d\n",
1905 i-1, s->all_fragments[i-1].first_pixel);
1910 i = s->fragment_start[1];
1911 for (y = 1; y <= s->fragment_height / 2; y++) {
1912 for (x = 0; x < s->fragment_width / 2; x++) {
1913 s->all_fragments[i++].first_pixel =
1914 s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
1915 s->golden_frame.linesize[1] +
1916 x * FRAGMENT_PIXELS;
1917 debug_init(" fragment %d, first pixel @ %d\n",
1918 i-1, s->all_fragments[i-1].first_pixel);
1923 i = s->fragment_start[2];
1924 for (y = 1; y <= s->fragment_height / 2; y++) {
1925 for (x = 0; x < s->fragment_width / 2; x++) {
1926 s->all_fragments[i++].first_pixel =
1927 s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
1928 s->golden_frame.linesize[2] +
1929 x * FRAGMENT_PIXELS;
1930 debug_init(" fragment %d, first pixel @ %d\n",
1931 i-1, s->all_fragments[i-1].first_pixel);
1937 * This is the ffmpeg/libavcodec API init function.
1939 static int vp3_decode_init(AVCodecContext *avctx)
1941 Vp3DecodeContext *s = avctx->priv_data;
1942 int i, inter, plane;
1945 int y_superblock_count;
1946 int c_superblock_count;
1948 if (avctx->codec_tag == MKTAG('V','P','3','0'))
1954 s->width = (avctx->width + 15) & 0xFFFFFFF0;
1955 s->height = (avctx->height + 15) & 0xFFFFFFF0;
1956 avctx->pix_fmt = PIX_FMT_YUV420P;
1957 avctx->has_b_frames = 0;
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 *p= avctx->extradata;
2582 if (!avctx->extradata_size)
2584 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2589 op_bytes = *(p++)<<8;
2592 init_get_bits(&gb, p, op_bytes);
2595 ptype = get_bits(&gb, 8);
2596 debug_vp3("Theora headerpacket type: %x\n", ptype);
2598 if (!(ptype & 0x80))
2600 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2604 // FIXME: check for this aswell
2605 skip_bits(&gb, 6*8); /* "theora" */
2610 theora_decode_header(avctx, &gb);
2613 // FIXME: is this needed? it breaks sometimes
2614 // theora_decode_comments(avctx, gb);
2617 theora_decode_tables(avctx, &gb);
2620 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2623 if(8*op_bytes != get_bits_count(&gb))
2624 av_log(avctx, AV_LOG_ERROR, "%d bits left in packet %X\n", 8*op_bytes - get_bits_count(&gb), ptype);
2625 if (s->theora < 0x030200)
2629 vp3_decode_init(avctx);
2633 AVCodec vp3_decoder = {
2637 sizeof(Vp3DecodeContext),
2646 #ifndef CONFIG_LIBTHEORA
2647 AVCodec theora_decoder = {
2651 sizeof(Vp3DecodeContext),