2 * Copyright (C) 2003-2004 the ffmpeg project
4 * This library is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU Lesser General Public
6 * License as published by the Free Software Foundation; either
7 * version 2 of the License, or (at your option) any later version.
9 * This library is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * Lesser General Public License for more details.
14 * You should have received a copy of the GNU Lesser General Public
15 * License along with this library; if not, write to the Free Software
16 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 * On2 VP3 Video Decoder
24 * VP3 Video Decoder by Mike Melanson (mike at multimedia.cx)
25 * For more information about the VP3 coding process, visit:
26 * http://multimedia.cx/
28 * Theora decoder by Alex Beregszaszi
39 #include "mpegvideo.h"
43 #define FRAGMENT_PIXELS 8
48 * Define one or more of the following compile-time variables to 1 to obtain
49 * elaborate information about certain aspects of the decoding process.
51 * KEYFRAMES_ONLY: set this to 1 to only see keyframes (VP3 slideshow mode)
52 * DEBUG_VP3: high-level decoding flow
53 * DEBUG_INIT: initialization parameters
54 * DEBUG_DEQUANTIZERS: display how the dequanization tables are built
55 * DEBUG_BLOCK_CODING: unpacking the superblock/macroblock/fragment coding
56 * DEBUG_MODES: unpacking the coding modes for individual fragments
57 * DEBUG_VECTORS: display the motion vectors
58 * DEBUG_TOKEN: display exhaustive information about each DCT token
59 * DEBUG_VLC: display the VLCs as they are extracted from the stream
60 * DEBUG_DC_PRED: display the process of reversing DC prediction
61 * DEBUG_IDCT: show every detail of the IDCT process
64 #define KEYFRAMES_ONLY 0
68 #define DEBUG_DEQUANTIZERS 0
69 #define DEBUG_BLOCK_CODING 0
71 #define DEBUG_VECTORS 0
74 #define DEBUG_DC_PRED 0
78 #define debug_vp3(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
80 static inline void debug_vp3(const char *format, ...) { }
84 #define debug_init(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
86 static inline void debug_init(const char *format, ...) { }
89 #if DEBUG_DEQUANTIZERS
90 #define debug_dequantizers(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
92 static inline void debug_dequantizers(const char *format, ...) { }
95 #if DEBUG_BLOCK_CODING
96 #define debug_block_coding(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
98 static inline void debug_block_coding(const char *format, ...) { }
102 #define debug_modes(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
104 static inline void debug_modes(const char *format, ...) { }
108 #define debug_vectors(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
110 static inline void debug_vectors(const char *format, ...) { }
114 #define debug_token(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
116 static inline void debug_token(const char *format, ...) { }
120 #define debug_vlc(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
122 static inline void debug_vlc(const char *format, ...) { }
126 #define debug_dc_pred(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
128 static inline void debug_dc_pred(const char *format, ...) { }
132 #define debug_idct(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
134 static inline void debug_idct(const char *format, ...) { }
137 typedef struct Coeff {
143 //FIXME split things out into their own arrays
144 typedef struct Vp3Fragment {
146 /* address of first pixel taking into account which plane the fragment
147 * lives on as well as the plane stride */
149 /* this is the macroblock that the fragment belongs to */
151 uint8_t coding_method;
157 #define SB_NOT_CODED 0
158 #define SB_PARTIALLY_CODED 1
159 #define SB_FULLY_CODED 2
161 #define MODE_INTER_NO_MV 0
163 #define MODE_INTER_PLUS_MV 2
164 #define MODE_INTER_LAST_MV 3
165 #define MODE_INTER_PRIOR_LAST 4
166 #define MODE_USING_GOLDEN 5
167 #define MODE_GOLDEN_MV 6
168 #define MODE_INTER_FOURMV 7
169 #define CODING_MODE_COUNT 8
171 /* special internal mode */
174 /* There are 6 preset schemes, plus a free-form scheme */
175 static int ModeAlphabet[7][CODING_MODE_COUNT] =
177 /* this is the custom scheme */
178 { 0, 0, 0, 0, 0, 0, 0, 0 },
180 /* scheme 1: Last motion vector dominates */
181 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
182 MODE_INTER_PLUS_MV, MODE_INTER_NO_MV,
183 MODE_INTRA, MODE_USING_GOLDEN,
184 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
187 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
188 MODE_INTER_NO_MV, MODE_INTER_PLUS_MV,
189 MODE_INTRA, MODE_USING_GOLDEN,
190 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
193 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
194 MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
195 MODE_INTRA, MODE_USING_GOLDEN,
196 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
199 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
200 MODE_INTER_NO_MV, MODE_INTER_PRIOR_LAST,
201 MODE_INTRA, MODE_USING_GOLDEN,
202 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
204 /* scheme 5: No motion vector dominates */
205 { MODE_INTER_NO_MV, MODE_INTER_LAST_MV,
206 MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
207 MODE_INTRA, MODE_USING_GOLDEN,
208 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
211 { MODE_INTER_NO_MV, MODE_USING_GOLDEN,
212 MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
213 MODE_INTER_PLUS_MV, MODE_INTRA,
214 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
218 #define MIN_DEQUANT_VAL 2
220 typedef struct Vp3DecodeContext {
221 AVCodecContext *avctx;
222 int theora, theora_tables;
225 AVFrame golden_frame;
227 AVFrame current_frame;
233 int last_quality_index;
235 int superblock_count;
236 int superblock_width;
237 int superblock_height;
238 int y_superblock_width;
239 int y_superblock_height;
240 int c_superblock_width;
241 int c_superblock_height;
242 int u_superblock_start;
243 int v_superblock_start;
244 unsigned char *superblock_coding;
246 int macroblock_count;
247 int macroblock_width;
248 int macroblock_height;
254 Vp3Fragment *all_fragments;
257 int u_fragment_start;
258 int v_fragment_start;
263 uint16_t coded_dc_scale_factor[64];
264 uint32_t coded_ac_scale_factor[64];
265 uint8_t base_matrix[384][64];
266 uint8_t qr_count[2][3];
267 uint8_t qr_size [2][3][64];
268 uint16_t qr_base[2][3][64];
270 /* this is a list of indices into the all_fragments array indicating
271 * which of the fragments are coded */
272 int *coded_fragment_list;
273 int coded_fragment_list_index;
274 int pixel_addresses_inited;
282 VLC superblock_run_length_vlc;
283 VLC fragment_run_length_vlc;
285 VLC motion_vector_vlc;
287 /* these arrays need to be on 16-byte boundaries since SSE2 operations
289 DECLARE_ALIGNED_16(int16_t, qmat[2][4][64]); //<qmat[is_inter][plane]
291 /* This table contains superblock_count * 16 entries. Each set of 16
292 * numbers corresponds to the fragment indices 0..15 of the superblock.
293 * An entry will be -1 to indicate that no entry corresponds to that
295 int *superblock_fragments;
297 /* This table contains superblock_count * 4 entries. Each set of 4
298 * numbers corresponds to the macroblock indices 0..3 of the superblock.
299 * An entry will be -1 to indicate that no entry corresponds to that
301 int *superblock_macroblocks;
303 /* This table contains macroblock_count * 6 entries. Each set of 6
304 * numbers corresponds to the fragment indices 0..5 which comprise
305 * the macroblock (4 Y fragments and 2 C fragments). */
306 int *macroblock_fragments;
307 /* This is an array that indicates how a particular macroblock
309 unsigned char *macroblock_coding;
311 int first_coded_y_fragment;
312 int first_coded_c_fragment;
313 int last_coded_y_fragment;
314 int last_coded_c_fragment;
316 uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
317 uint8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
324 uint16_t huffman_table[80][32][2];
326 uint32_t filter_limit_values[64];
327 int bounding_values_array[256];
330 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb);
332 /************************************************************************
333 * VP3 specific functions
334 ************************************************************************/
337 * This function sets up all of the various blocks mappings:
338 * superblocks <-> fragments, macroblocks <-> fragments,
339 * superblocks <-> macroblocks
341 * Returns 0 is successful; returns 1 if *anything* went wrong.
343 static int init_block_mapping(Vp3DecodeContext *s)
346 signed int hilbert_walk_y[16];
347 signed int hilbert_walk_c[16];
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 /* figure out hilbert pattern per these frame dimensions */
387 hilbert_walk_y[0] = 1;
388 hilbert_walk_y[1] = 1;
389 hilbert_walk_y[2] = s->fragment_width;
390 hilbert_walk_y[3] = -1;
391 hilbert_walk_y[4] = s->fragment_width;
392 hilbert_walk_y[5] = s->fragment_width;
393 hilbert_walk_y[6] = 1;
394 hilbert_walk_y[7] = -s->fragment_width;
395 hilbert_walk_y[8] = 1;
396 hilbert_walk_y[9] = s->fragment_width;
397 hilbert_walk_y[10] = 1;
398 hilbert_walk_y[11] = -s->fragment_width;
399 hilbert_walk_y[12] = -s->fragment_width;
400 hilbert_walk_y[13] = -1;
401 hilbert_walk_y[14] = -s->fragment_width;
402 hilbert_walk_y[15] = 1;
404 hilbert_walk_c[0] = 1;
405 hilbert_walk_c[1] = 1;
406 hilbert_walk_c[2] = s->fragment_width / 2;
407 hilbert_walk_c[3] = -1;
408 hilbert_walk_c[4] = s->fragment_width / 2;
409 hilbert_walk_c[5] = s->fragment_width / 2;
410 hilbert_walk_c[6] = 1;
411 hilbert_walk_c[7] = -s->fragment_width / 2;
412 hilbert_walk_c[8] = 1;
413 hilbert_walk_c[9] = s->fragment_width / 2;
414 hilbert_walk_c[10] = 1;
415 hilbert_walk_c[11] = -s->fragment_width / 2;
416 hilbert_walk_c[12] = -s->fragment_width / 2;
417 hilbert_walk_c[13] = -1;
418 hilbert_walk_c[14] = -s->fragment_width / 2;
419 hilbert_walk_c[15] = 1;
421 hilbert_walk_mb[0] = 1;
422 hilbert_walk_mb[1] = s->macroblock_width;
423 hilbert_walk_mb[2] = 1;
424 hilbert_walk_mb[3] = -s->macroblock_width;
426 /* iterate through each superblock (all planes) and map the fragments */
427 for (i = 0; i < s->superblock_count; i++) {
428 debug_init(" superblock %d (u starts @ %d, v starts @ %d)\n",
429 i, s->u_superblock_start, s->v_superblock_start);
431 /* time to re-assign the limits? */
434 /* start of Y superblocks */
435 right_edge = s->fragment_width;
436 bottom_edge = s->fragment_height;
439 superblock_row_inc = 3 * s->fragment_width -
440 (s->y_superblock_width * 4 - s->fragment_width);
441 hilbert = hilbert_walk_y;
443 /* the first operation for this variable is to advance by 1 */
444 current_fragment = -1;
446 } else if (i == s->u_superblock_start) {
448 /* start of U superblocks */
449 right_edge = s->fragment_width / 2;
450 bottom_edge = s->fragment_height / 2;
453 superblock_row_inc = 3 * (s->fragment_width / 2) -
454 (s->c_superblock_width * 4 - s->fragment_width / 2);
455 hilbert = hilbert_walk_c;
457 /* the first operation for this variable is to advance by 1 */
458 current_fragment = s->u_fragment_start - 1;
460 } else if (i == s->v_superblock_start) {
462 /* start of V superblocks */
463 right_edge = s->fragment_width / 2;
464 bottom_edge = s->fragment_height / 2;
467 superblock_row_inc = 3 * (s->fragment_width / 2) -
468 (s->c_superblock_width * 4 - s->fragment_width / 2);
469 hilbert = hilbert_walk_c;
471 /* the first operation for this variable is to advance by 1 */
472 current_fragment = s->v_fragment_start - 1;
476 if (current_width >= right_edge - 1) {
477 /* reset width and move to next superblock row */
481 /* fragment is now at the start of a new superblock row */
482 current_fragment += superblock_row_inc;
485 /* iterate through all 16 fragments in a superblock */
486 for (j = 0; j < 16; j++) {
487 current_fragment += hilbert[j];
488 current_width += travel_width[j];
489 current_height += travel_height[j];
491 /* check if the fragment is in bounds */
492 if ((current_width < right_edge) &&
493 (current_height < bottom_edge)) {
494 s->superblock_fragments[mapping_index] = current_fragment;
495 debug_init(" mapping fragment %d to superblock %d, position %d (%d/%d x %d/%d)\n",
496 s->superblock_fragments[mapping_index], i, j,
497 current_width, right_edge, current_height, bottom_edge);
499 s->superblock_fragments[mapping_index] = -1;
500 debug_init(" superblock %d, position %d has no fragment (%d/%d x %d/%d)\n",
502 current_width, right_edge, current_height, bottom_edge);
509 /* initialize the superblock <-> macroblock mapping; iterate through
510 * all of the Y plane superblocks to build this mapping */
511 right_edge = s->macroblock_width;
512 bottom_edge = s->macroblock_height;
515 superblock_row_inc = s->macroblock_width -
516 (s->y_superblock_width * 2 - s->macroblock_width);;
517 hilbert = hilbert_walk_mb;
519 current_macroblock = -1;
520 for (i = 0; i < s->u_superblock_start; i++) {
522 if (current_width >= right_edge - 1) {
523 /* reset width and move to next superblock row */
527 /* macroblock is now at the start of a new superblock row */
528 current_macroblock += superblock_row_inc;
531 /* iterate through each potential macroblock in the superblock */
532 for (j = 0; j < 4; j++) {
533 current_macroblock += hilbert_walk_mb[j];
534 current_width += travel_width_mb[j];
535 current_height += travel_height_mb[j];
537 /* check if the macroblock is in bounds */
538 if ((current_width < right_edge) &&
539 (current_height < bottom_edge)) {
540 s->superblock_macroblocks[mapping_index] = current_macroblock;
541 debug_init(" mapping macroblock %d to superblock %d, position %d (%d/%d x %d/%d)\n",
542 s->superblock_macroblocks[mapping_index], i, j,
543 current_width, right_edge, current_height, bottom_edge);
545 s->superblock_macroblocks[mapping_index] = -1;
546 debug_init(" superblock %d, position %d has no macroblock (%d/%d x %d/%d)\n",
548 current_width, right_edge, current_height, bottom_edge);
555 /* initialize the macroblock <-> fragment mapping */
556 current_fragment = 0;
557 current_macroblock = 0;
559 for (i = 0; i < s->fragment_height; i += 2) {
561 for (j = 0; j < s->fragment_width; j += 2) {
563 debug_init(" macroblock %d contains fragments: ", current_macroblock);
564 s->all_fragments[current_fragment].macroblock = current_macroblock;
565 s->macroblock_fragments[mapping_index++] = current_fragment;
566 debug_init("%d ", current_fragment);
568 if (j + 1 < s->fragment_width) {
569 s->all_fragments[current_fragment + 1].macroblock = current_macroblock;
570 s->macroblock_fragments[mapping_index++] = current_fragment + 1;
571 debug_init("%d ", current_fragment + 1);
573 s->macroblock_fragments[mapping_index++] = -1;
575 if (i + 1 < s->fragment_height) {
576 s->all_fragments[current_fragment + s->fragment_width].macroblock =
578 s->macroblock_fragments[mapping_index++] =
579 current_fragment + s->fragment_width;
580 debug_init("%d ", current_fragment + s->fragment_width);
582 s->macroblock_fragments[mapping_index++] = -1;
584 if ((j + 1 < s->fragment_width) && (i + 1 < s->fragment_height)) {
585 s->all_fragments[current_fragment + s->fragment_width + 1].macroblock =
587 s->macroblock_fragments[mapping_index++] =
588 current_fragment + s->fragment_width + 1;
589 debug_init("%d ", current_fragment + s->fragment_width + 1);
591 s->macroblock_fragments[mapping_index++] = -1;
594 c_fragment = s->u_fragment_start +
595 (i * s->fragment_width / 4) + (j / 2);
596 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
597 s->macroblock_fragments[mapping_index++] = c_fragment;
598 debug_init("%d ", c_fragment);
600 c_fragment = s->v_fragment_start +
601 (i * s->fragment_width / 4) + (j / 2);
602 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
603 s->macroblock_fragments[mapping_index++] = c_fragment;
604 debug_init("%d ", c_fragment);
608 if (j + 2 <= s->fragment_width)
609 current_fragment += 2;
612 current_macroblock++;
615 current_fragment += s->fragment_width;
618 return 0; /* successful path out */
622 * This function wipes out all of the fragment data.
624 static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
628 /* zero out all of the fragment information */
629 s->coded_fragment_list_index = 0;
630 for (i = 0; i < s->fragment_count; i++) {
631 s->all_fragments[i].coeff_count = 0;
632 s->all_fragments[i].motion_x = 127;
633 s->all_fragments[i].motion_y = 127;
634 s->all_fragments[i].next_coeff= NULL;
636 s->coeffs[i].coeff=0;
637 s->coeffs[i].next= NULL;
642 * This function sets up the dequantization tables used for a particular
645 static void init_dequantizer(Vp3DecodeContext *s)
647 int ac_scale_factor = s->coded_ac_scale_factor[s->quality_index];
648 int dc_scale_factor = s->coded_dc_scale_factor[s->quality_index];
649 int i, j, plane, inter, qri, bmi, bmj, qistart;
651 debug_vp3(" vp3: initializing dequantization tables\n");
653 for(inter=0; inter<2; inter++){
654 for(plane=0; plane<3; plane++){
656 for(qri=0; qri<s->qr_count[inter][plane]; qri++){
657 sum+= s->qr_size[inter][plane][qri];
658 if(s->quality_index <= sum)
661 qistart= sum - s->qr_size[inter][plane][qri];
662 bmi= s->qr_base[inter][plane][qri ];
663 bmj= s->qr_base[inter][plane][qri+1];
665 int coeff= ( 2*(sum -s->quality_index)*s->base_matrix[bmi][i]
666 - 2*(qistart-s->quality_index)*s->base_matrix[bmj][i]
667 + s->qr_size[inter][plane][qri])
668 / (2*s->qr_size[inter][plane][qri]);
670 int qmin= 8<<(inter + !plane);
671 int qscale= i ? ac_scale_factor : dc_scale_factor;
673 s->qmat[inter][plane][i]= clip((qscale * coeff)/100 * 4, qmin, 4096);
678 memset(s->qscale_table, (FFMAX(s->qmat[0][0][1], s->qmat[0][1][1])+8)/16, 512); //FIXME finetune
682 * This function initializes the loop filter boundary limits if the frame's
683 * quality index is different from the previous frame's.
685 static void init_loop_filter(Vp3DecodeContext *s)
687 int *bounding_values= s->bounding_values_array+127;
691 filter_limit = s->filter_limit_values[s->quality_index];
693 /* set up the bounding values */
694 memset(s->bounding_values_array, 0, 256 * sizeof(int));
695 for (x = 0; x < filter_limit; x++) {
696 bounding_values[-x - filter_limit] = -filter_limit + x;
697 bounding_values[-x] = -x;
698 bounding_values[x] = x;
699 bounding_values[x + filter_limit] = filter_limit - x;
704 * This function unpacks all of the superblock/macroblock/fragment coding
705 * information from the bitstream.
707 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
710 int current_superblock = 0;
712 int decode_fully_flags = 0;
713 int decode_partial_blocks = 0;
714 int first_c_fragment_seen;
717 int current_fragment;
719 debug_vp3(" vp3: unpacking superblock coding\n");
723 debug_vp3(" keyframe-- all superblocks are fully coded\n");
724 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
728 /* unpack the list of partially-coded superblocks */
729 bit = get_bits(gb, 1);
730 /* toggle the bit because as soon as the first run length is
731 * fetched the bit will be toggled again */
733 while (current_superblock < s->superblock_count) {
734 if (current_run-- == 0) {
736 current_run = get_vlc2(gb,
737 s->superblock_run_length_vlc.table, 6, 2);
738 if (current_run == 33)
739 current_run += get_bits(gb, 12);
740 debug_block_coding(" setting superblocks %d..%d to %s\n",
742 current_superblock + current_run - 1,
743 (bit) ? "partially coded" : "not coded");
745 /* if any of the superblocks are not partially coded, flag
746 * a boolean to decode the list of fully-coded superblocks */
748 decode_fully_flags = 1;
751 /* make a note of the fact that there are partially coded
753 decode_partial_blocks = 1;
756 s->superblock_coding[current_superblock++] = bit;
759 /* unpack the list of fully coded superblocks if any of the blocks were
760 * not marked as partially coded in the previous step */
761 if (decode_fully_flags) {
763 current_superblock = 0;
765 bit = get_bits(gb, 1);
766 /* toggle the bit because as soon as the first run length is
767 * fetched the bit will be toggled again */
769 while (current_superblock < s->superblock_count) {
771 /* skip any superblocks already marked as partially coded */
772 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
774 if (current_run-- == 0) {
776 current_run = get_vlc2(gb,
777 s->superblock_run_length_vlc.table, 6, 2);
778 if (current_run == 33)
779 current_run += get_bits(gb, 12);
782 debug_block_coding(" setting superblock %d to %s\n",
784 (bit) ? "fully coded" : "not coded");
785 s->superblock_coding[current_superblock] = 2*bit;
787 current_superblock++;
791 /* if there were partial blocks, initialize bitstream for
792 * unpacking fragment codings */
793 if (decode_partial_blocks) {
796 bit = get_bits(gb, 1);
797 /* toggle the bit because as soon as the first run length is
798 * fetched the bit will be toggled again */
803 /* figure out which fragments are coded; iterate through each
804 * superblock (all planes) */
805 s->coded_fragment_list_index = 0;
806 s->next_coeff= s->coeffs + s->fragment_count;
807 s->first_coded_y_fragment = s->first_coded_c_fragment = 0;
808 s->last_coded_y_fragment = s->last_coded_c_fragment = -1;
809 first_c_fragment_seen = 0;
810 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
811 for (i = 0; i < s->superblock_count; i++) {
813 /* iterate through all 16 fragments in a superblock */
814 for (j = 0; j < 16; j++) {
816 /* if the fragment is in bounds, check its coding status */
817 current_fragment = s->superblock_fragments[i * 16 + j];
818 if (current_fragment >= s->fragment_count) {
819 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
820 current_fragment, s->fragment_count);
823 if (current_fragment != -1) {
824 if (s->superblock_coding[i] == SB_NOT_CODED) {
826 /* copy all the fragments from the prior frame */
827 s->all_fragments[current_fragment].coding_method =
830 } else if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
832 /* fragment may or may not be coded; this is the case
833 * that cares about the fragment coding runs */
834 if (current_run-- == 0) {
836 current_run = get_vlc2(gb,
837 s->fragment_run_length_vlc.table, 5, 2);
841 /* default mode; actual mode will be decoded in
843 s->all_fragments[current_fragment].coding_method =
845 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
846 s->coded_fragment_list[s->coded_fragment_list_index] =
848 if ((current_fragment >= s->u_fragment_start) &&
849 (s->last_coded_y_fragment == -1) &&
850 (!first_c_fragment_seen)) {
851 s->first_coded_c_fragment = s->coded_fragment_list_index;
852 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
853 first_c_fragment_seen = 1;
855 s->coded_fragment_list_index++;
856 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
857 debug_block_coding(" superblock %d is partially coded, fragment %d is coded\n",
858 i, current_fragment);
860 /* not coded; copy this fragment from the prior frame */
861 s->all_fragments[current_fragment].coding_method =
863 debug_block_coding(" superblock %d is partially coded, fragment %d is not coded\n",
864 i, current_fragment);
869 /* fragments are fully coded in this superblock; actual
870 * coding will be determined in next step */
871 s->all_fragments[current_fragment].coding_method =
873 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
874 s->coded_fragment_list[s->coded_fragment_list_index] =
876 if ((current_fragment >= s->u_fragment_start) &&
877 (s->last_coded_y_fragment == -1) &&
878 (!first_c_fragment_seen)) {
879 s->first_coded_c_fragment = s->coded_fragment_list_index;
880 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
881 first_c_fragment_seen = 1;
883 s->coded_fragment_list_index++;
884 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
885 debug_block_coding(" superblock %d is fully coded, fragment %d is coded\n",
886 i, current_fragment);
892 if (!first_c_fragment_seen)
893 /* only Y fragments coded in this frame */
894 s->last_coded_y_fragment = s->coded_fragment_list_index - 1;
896 /* end the list of coded C fragments */
897 s->last_coded_c_fragment = s->coded_fragment_list_index - 1;
899 debug_block_coding(" %d total coded fragments, y: %d -> %d, c: %d -> %d\n",
900 s->coded_fragment_list_index,
901 s->first_coded_y_fragment,
902 s->last_coded_y_fragment,
903 s->first_coded_c_fragment,
904 s->last_coded_c_fragment);
910 * This function unpacks all the coding mode data for individual macroblocks
911 * from the bitstream.
913 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
917 int current_macroblock;
918 int current_fragment;
921 debug_vp3(" vp3: unpacking encoding modes\n");
924 debug_vp3(" keyframe-- all blocks are coded as INTRA\n");
926 for (i = 0; i < s->fragment_count; i++)
927 s->all_fragments[i].coding_method = MODE_INTRA;
931 /* fetch the mode coding scheme for this frame */
932 scheme = get_bits(gb, 3);
933 debug_modes(" using mode alphabet %d\n", scheme);
935 /* is it a custom coding scheme? */
937 debug_modes(" custom mode alphabet ahead:\n");
938 for (i = 0; i < 8; i++)
939 ModeAlphabet[scheme][get_bits(gb, 3)] = i;
942 for (i = 0; i < 8; i++)
943 debug_modes(" mode[%d][%d] = %d\n", scheme, i,
944 ModeAlphabet[scheme][i]);
946 /* iterate through all of the macroblocks that contain 1 or more
948 for (i = 0; i < s->u_superblock_start; i++) {
950 for (j = 0; j < 4; j++) {
951 current_macroblock = s->superblock_macroblocks[i * 4 + j];
952 if ((current_macroblock == -1) ||
953 (s->macroblock_coding[current_macroblock] == MODE_COPY))
955 if (current_macroblock >= s->macroblock_count) {
956 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad macroblock number (%d >= %d)\n",
957 current_macroblock, s->macroblock_count);
961 /* mode 7 means get 3 bits for each coding mode */
963 coding_mode = get_bits(gb, 3);
965 coding_mode = ModeAlphabet[scheme]
966 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
968 s->macroblock_coding[current_macroblock] = coding_mode;
969 for (k = 0; k < 6; k++) {
971 s->macroblock_fragments[current_macroblock * 6 + k];
972 if (current_fragment == -1)
974 if (current_fragment >= s->fragment_count) {
975 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad fragment number (%d >= %d)\n",
976 current_fragment, s->fragment_count);
979 if (s->all_fragments[current_fragment].coding_method !=
981 s->all_fragments[current_fragment].coding_method =
985 debug_modes(" coding method for macroblock starting @ fragment %d = %d\n",
986 s->macroblock_fragments[current_macroblock * 6], coding_mode);
995 * This function unpacks all the motion vectors for the individual
996 * macroblocks from the bitstream.
998 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
1004 int last_motion_x = 0;
1005 int last_motion_y = 0;
1006 int prior_last_motion_x = 0;
1007 int prior_last_motion_y = 0;
1008 int current_macroblock;
1009 int current_fragment;
1011 debug_vp3(" vp3: unpacking motion vectors\n");
1014 debug_vp3(" keyframe-- there are no motion vectors\n");
1018 memset(motion_x, 0, 6 * sizeof(int));
1019 memset(motion_y, 0, 6 * sizeof(int));
1021 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
1022 coding_mode = get_bits(gb, 1);
1023 debug_vectors(" using %s scheme for unpacking motion vectors\n",
1024 (coding_mode == 0) ? "VLC" : "fixed-length");
1026 /* iterate through all of the macroblocks that contain 1 or more
1027 * coded fragments */
1028 for (i = 0; i < s->u_superblock_start; i++) {
1030 for (j = 0; j < 4; j++) {
1031 current_macroblock = s->superblock_macroblocks[i * 4 + j];
1032 if ((current_macroblock == -1) ||
1033 (s->macroblock_coding[current_macroblock] == MODE_COPY))
1035 if (current_macroblock >= s->macroblock_count) {
1036 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad macroblock number (%d >= %d)\n",
1037 current_macroblock, s->macroblock_count);
1041 current_fragment = s->macroblock_fragments[current_macroblock * 6];
1042 if (current_fragment >= s->fragment_count) {
1043 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d\n",
1044 current_fragment, s->fragment_count);
1047 switch (s->macroblock_coding[current_macroblock]) {
1049 case MODE_INTER_PLUS_MV:
1050 case MODE_GOLDEN_MV:
1051 /* all 6 fragments use the same motion vector */
1052 if (coding_mode == 0) {
1053 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1054 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1056 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
1057 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
1060 for (k = 1; k < 6; k++) {
1061 motion_x[k] = motion_x[0];
1062 motion_y[k] = motion_y[0];
1065 /* vector maintenance, only on MODE_INTER_PLUS_MV */
1066 if (s->macroblock_coding[current_macroblock] ==
1067 MODE_INTER_PLUS_MV) {
1068 prior_last_motion_x = last_motion_x;
1069 prior_last_motion_y = last_motion_y;
1070 last_motion_x = motion_x[0];
1071 last_motion_y = motion_y[0];
1075 case MODE_INTER_FOURMV:
1076 /* fetch 4 vectors from the bitstream, one for each
1077 * Y fragment, then average for the C fragment vectors */
1078 motion_x[4] = motion_y[4] = 0;
1079 for (k = 0; k < 4; k++) {
1080 if (coding_mode == 0) {
1081 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1082 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1084 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
1085 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
1087 motion_x[4] += motion_x[k];
1088 motion_y[4] += motion_y[k];
1092 motion_x[4]= RSHIFT(motion_x[4], 2);
1094 motion_y[4]= RSHIFT(motion_y[4], 2);
1096 /* vector maintenance; vector[3] is treated as the
1097 * last vector in this case */
1098 prior_last_motion_x = last_motion_x;
1099 prior_last_motion_y = last_motion_y;
1100 last_motion_x = motion_x[3];
1101 last_motion_y = motion_y[3];
1104 case MODE_INTER_LAST_MV:
1105 /* all 6 fragments use the last motion vector */
1106 motion_x[0] = last_motion_x;
1107 motion_y[0] = last_motion_y;
1108 for (k = 1; k < 6; k++) {
1109 motion_x[k] = motion_x[0];
1110 motion_y[k] = motion_y[0];
1113 /* no vector maintenance (last vector remains the
1117 case MODE_INTER_PRIOR_LAST:
1118 /* all 6 fragments use the motion vector prior to the
1119 * last motion vector */
1120 motion_x[0] = prior_last_motion_x;
1121 motion_y[0] = prior_last_motion_y;
1122 for (k = 1; k < 6; k++) {
1123 motion_x[k] = motion_x[0];
1124 motion_y[k] = motion_y[0];
1127 /* vector maintenance */
1128 prior_last_motion_x = last_motion_x;
1129 prior_last_motion_y = last_motion_y;
1130 last_motion_x = motion_x[0];
1131 last_motion_y = motion_y[0];
1135 /* covers intra, inter without MV, golden without MV */
1136 memset(motion_x, 0, 6 * sizeof(int));
1137 memset(motion_y, 0, 6 * sizeof(int));
1139 /* no vector maintenance */
1143 /* assign the motion vectors to the correct fragments */
1144 debug_vectors(" vectors for macroblock starting @ fragment %d (coding method %d):\n",
1146 s->macroblock_coding[current_macroblock]);
1147 for (k = 0; k < 6; k++) {
1149 s->macroblock_fragments[current_macroblock * 6 + k];
1150 if (current_fragment == -1)
1152 if (current_fragment >= s->fragment_count) {
1153 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d)\n",
1154 current_fragment, s->fragment_count);
1157 s->all_fragments[current_fragment].motion_x = motion_x[k];
1158 s->all_fragments[current_fragment].motion_y = motion_y[k];
1159 debug_vectors(" vector %d: fragment %d = (%d, %d)\n",
1160 k, current_fragment, motion_x[k], motion_y[k]);
1170 * This function is called by unpack_dct_coeffs() to extract the VLCs from
1171 * the bitstream. The VLCs encode tokens which are used to unpack DCT
1172 * data. This function unpacks all the VLCs for either the Y plane or both
1173 * C planes, and is called for DC coefficients or different AC coefficient
1174 * levels (since different coefficient types require different VLC tables.
1176 * This function returns a residual eob run. E.g, if a particular token gave
1177 * instructions to EOB the next 5 fragments and there were only 2 fragments
1178 * left in the current fragment range, 3 would be returned so that it could
1179 * be passed into the next call to this same function.
1181 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
1182 VLC *table, int coeff_index,
1183 int first_fragment, int last_fragment,
1190 Vp3Fragment *fragment;
1191 uint8_t *perm= s->scantable.permutated;
1194 if ((first_fragment >= s->fragment_count) ||
1195 (last_fragment >= s->fragment_count)) {
1197 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vlcs(): bad fragment number (%d -> %d ?)\n",
1198 first_fragment, last_fragment);
1202 for (i = first_fragment; i <= last_fragment; i++) {
1204 fragment = &s->all_fragments[s->coded_fragment_list[i]];
1205 if (fragment->coeff_count > coeff_index)
1209 /* decode a VLC into a token */
1210 token = get_vlc2(gb, table->table, 5, 3);
1211 debug_vlc(" token = %2d, ", token);
1212 /* use the token to get a zero run, a coefficient, and an eob run */
1214 eob_run = eob_run_base[token];
1215 if (eob_run_get_bits[token])
1216 eob_run += get_bits(gb, eob_run_get_bits[token]);
1217 coeff = zero_run = 0;
1219 bits_to_get = coeff_get_bits[token];
1221 coeff = coeff_tables[token][0];
1223 coeff = coeff_tables[token][get_bits(gb, bits_to_get)];
1225 zero_run = zero_run_base[token];
1226 if (zero_run_get_bits[token])
1227 zero_run += get_bits(gb, zero_run_get_bits[token]);
1232 fragment->coeff_count += zero_run;
1233 if (fragment->coeff_count < 64){
1234 fragment->next_coeff->coeff= coeff;
1235 fragment->next_coeff->index= perm[fragment->coeff_count++]; //FIXME perm here already?
1236 fragment->next_coeff->next= s->next_coeff;
1237 s->next_coeff->next=NULL;
1238 fragment->next_coeff= s->next_coeff++;
1240 debug_vlc(" fragment %d coeff = %d\n",
1241 s->coded_fragment_list[i], fragment->next_coeff[coeff_index]);
1243 fragment->coeff_count |= 128;
1244 debug_vlc(" fragment %d eob with %d coefficients\n",
1245 s->coded_fragment_list[i], fragment->coeff_count&127);
1254 * This function unpacks all of the DCT coefficient data from the
1257 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1264 int residual_eob_run = 0;
1266 /* fetch the DC table indices */
1267 dc_y_table = get_bits(gb, 4);
1268 dc_c_table = get_bits(gb, 4);
1270 /* unpack the Y plane DC coefficients */
1271 debug_vp3(" vp3: unpacking Y plane DC coefficients using table %d\n",
1273 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1274 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1276 /* unpack the C plane DC coefficients */
1277 debug_vp3(" vp3: unpacking C plane DC coefficients using table %d\n",
1279 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1280 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1282 /* fetch the AC table indices */
1283 ac_y_table = get_bits(gb, 4);
1284 ac_c_table = get_bits(gb, 4);
1286 /* unpack the group 1 AC coefficients (coeffs 1-5) */
1287 for (i = 1; i <= 5; i++) {
1289 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1291 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_y_table], i,
1292 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1294 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1296 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_c_table], i,
1297 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1300 /* unpack the group 2 AC coefficients (coeffs 6-14) */
1301 for (i = 6; i <= 14; i++) {
1303 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1305 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_y_table], i,
1306 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1308 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1310 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_c_table], i,
1311 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1314 /* unpack the group 3 AC coefficients (coeffs 15-27) */
1315 for (i = 15; i <= 27; i++) {
1317 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1319 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_y_table], i,
1320 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1322 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1324 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_c_table], i,
1325 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1328 /* unpack the group 4 AC coefficients (coeffs 28-63) */
1329 for (i = 28; i <= 63; i++) {
1331 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1333 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_y_table], i,
1334 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1336 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1338 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_c_table], i,
1339 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1346 * This function reverses the DC prediction for each coded fragment in
1347 * the frame. Much of this function is adapted directly from the original
1350 #define COMPATIBLE_FRAME(x) \
1351 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1352 #define FRAME_CODED(x) (s->all_fragments[x].coding_method != MODE_COPY)
1353 #define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
1354 static inline int iabs (int x) { return ((x < 0) ? -x : x); }
1356 static void reverse_dc_prediction(Vp3DecodeContext *s,
1359 int fragment_height)
1368 int i = first_fragment;
1371 * Fragment prediction groups:
1379 * Note: Groups 5 and 7 do not exist as it would mean that the
1380 * fragment's x coordinate is both 0 and (width - 1) at the same time.
1382 int predictor_group;
1385 /* validity flags for the left, up-left, up, and up-right fragments */
1386 int fl, ful, fu, fur;
1388 /* DC values for the left, up-left, up, and up-right fragments */
1389 int vl, vul, vu, vur;
1391 /* indices for the left, up-left, up, and up-right fragments */
1395 * The 6 fields mean:
1396 * 0: up-left multiplier
1398 * 2: up-right multiplier
1399 * 3: left multiplier
1401 * 5: right bit shift divisor (e.g., 7 means >>=7, a.k.a. div by 128)
1403 int predictor_transform[16][6] = {
1404 { 0, 0, 0, 0, 0, 0 },
1405 { 0, 0, 0, 1, 0, 0 }, // PL
1406 { 0, 0, 1, 0, 0, 0 }, // PUR
1407 { 0, 0, 53, 75, 127, 7 }, // PUR|PL
1408 { 0, 1, 0, 0, 0, 0 }, // PU
1409 { 0, 1, 0, 1, 1, 1 }, // PU|PL
1410 { 0, 1, 0, 0, 0, 0 }, // PU|PUR
1411 { 0, 0, 53, 75, 127, 7 }, // PU|PUR|PL
1412 { 1, 0, 0, 0, 0, 0 }, // PUL
1413 { 0, 0, 0, 1, 0, 0 }, // PUL|PL
1414 { 1, 0, 1, 0, 1, 1 }, // PUL|PUR
1415 { 0, 0, 53, 75, 127, 7 }, // PUL|PUR|PL
1416 { 0, 1, 0, 0, 0, 0 }, // PUL|PU
1417 {-26, 29, 0, 29, 31, 5 }, // PUL|PU|PL
1418 { 3, 10, 3, 0, 15, 4 }, // PUL|PU|PUR
1419 {-26, 29, 0, 29, 31, 5 } // PUL|PU|PUR|PL
1422 /* This table shows which types of blocks can use other blocks for
1423 * prediction. For example, INTRA is the only mode in this table to
1424 * have a frame number of 0. That means INTRA blocks can only predict
1425 * from other INTRA blocks. There are 2 golden frame coding types;
1426 * blocks encoding in these modes can only predict from other blocks
1427 * that were encoded with these 1 of these 2 modes. */
1428 unsigned char compatible_frame[8] = {
1429 1, /* MODE_INTER_NO_MV */
1431 1, /* MODE_INTER_PLUS_MV */
1432 1, /* MODE_INTER_LAST_MV */
1433 1, /* MODE_INTER_PRIOR_MV */
1434 2, /* MODE_USING_GOLDEN */
1435 2, /* MODE_GOLDEN_MV */
1436 1 /* MODE_INTER_FOUR_MV */
1438 int current_frame_type;
1440 /* there is a last DC predictor for each of the 3 frame types */
1445 debug_vp3(" vp3: reversing DC prediction\n");
1447 vul = vu = vur = vl = 0;
1448 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1450 /* for each fragment row... */
1451 for (y = 0; y < fragment_height; y++) {
1453 /* for each fragment in a row... */
1454 for (x = 0; x < fragment_width; x++, i++) {
1456 /* reverse prediction if this block was coded */
1457 if (s->all_fragments[i].coding_method != MODE_COPY) {
1459 current_frame_type =
1460 compatible_frame[s->all_fragments[i].coding_method];
1461 predictor_group = (x == 0) + ((y == 0) << 1) +
1462 ((x + 1 == fragment_width) << 2);
1463 debug_dc_pred(" frag %d: group %d, orig DC = %d, ",
1464 i, predictor_group, DC_COEFF(i));
1466 switch (predictor_group) {
1469 /* main body of fragments; consider all 4 possible
1470 * fragments for prediction */
1472 /* calculate the indices of the predicting fragments */
1473 ul = i - fragment_width - 1;
1474 u = i - fragment_width;
1475 ur = i - fragment_width + 1;
1478 /* fetch the DC values for the predicting fragments */
1484 /* figure out which fragments are valid */
1485 ful = FRAME_CODED(ul) && COMPATIBLE_FRAME(ul);
1486 fu = FRAME_CODED(u) && COMPATIBLE_FRAME(u);
1487 fur = FRAME_CODED(ur) && COMPATIBLE_FRAME(ur);
1488 fl = FRAME_CODED(l) && COMPATIBLE_FRAME(l);
1490 /* decide which predictor transform to use */
1491 transform = (fl*PL) | (fu*PU) | (ful*PUL) | (fur*PUR);
1496 /* left column of fragments, not including top corner;
1497 * only consider up and up-right fragments */
1499 /* calculate the indices of the predicting fragments */
1500 u = i - fragment_width;
1501 ur = i - fragment_width + 1;
1503 /* fetch the DC values for the predicting fragments */
1507 /* figure out which fragments are valid */
1508 fur = FRAME_CODED(ur) && COMPATIBLE_FRAME(ur);
1509 fu = FRAME_CODED(u) && COMPATIBLE_FRAME(u);
1511 /* decide which predictor transform to use */
1512 transform = (fu*PU) | (fur*PUR);
1518 /* top row of fragments, not including top-left frag;
1519 * only consider the left fragment for prediction */
1521 /* calculate the indices of the predicting fragments */
1524 /* fetch the DC values for the predicting fragments */
1527 /* figure out which fragments are valid */
1528 fl = FRAME_CODED(l) && COMPATIBLE_FRAME(l);
1530 /* decide which predictor transform to use */
1531 transform = (fl*PL);
1536 /* top-left fragment */
1538 /* nothing to predict from in this case */
1544 /* right column of fragments, not including top corner;
1545 * consider up-left, up, and left fragments for
1548 /* calculate the indices of the predicting fragments */
1549 ul = i - fragment_width - 1;
1550 u = i - fragment_width;
1553 /* fetch the DC values for the predicting fragments */
1558 /* figure out which fragments are valid */
1559 ful = FRAME_CODED(ul) && COMPATIBLE_FRAME(ul);
1560 fu = FRAME_CODED(u) && COMPATIBLE_FRAME(u);
1561 fl = FRAME_CODED(l) && COMPATIBLE_FRAME(l);
1563 /* decide which predictor transform to use */
1564 transform = (fl*PL) | (fu*PU) | (ful*PUL);
1570 debug_dc_pred("transform = %d, ", transform);
1572 if (transform == 0) {
1574 /* if there were no fragments to predict from, use last
1576 predicted_dc = last_dc[current_frame_type];
1577 debug_dc_pred("from last DC (%d) = %d\n",
1578 current_frame_type, DC_COEFF(i));
1582 /* apply the appropriate predictor transform */
1584 (predictor_transform[transform][0] * vul) +
1585 (predictor_transform[transform][1] * vu) +
1586 (predictor_transform[transform][2] * vur) +
1587 (predictor_transform[transform][3] * vl);
1589 /* if there is a shift value in the transform, add
1590 * the sign bit before the shift */
1591 if (predictor_transform[transform][5] != 0) {
1592 predicted_dc += ((predicted_dc >> 15) &
1593 predictor_transform[transform][4]);
1594 predicted_dc >>= predictor_transform[transform][5];
1597 /* check for outranging on the [ul u l] and
1598 * [ul u ur l] predictors */
1599 if ((transform == 13) || (transform == 15)) {
1600 if (iabs(predicted_dc - vu) > 128)
1602 else if (iabs(predicted_dc - vl) > 128)
1604 else if (iabs(predicted_dc - vul) > 128)
1608 debug_dc_pred("from pred DC = %d\n",
1612 /* at long last, apply the predictor */
1613 if(s->coeffs[i].index){
1614 *s->next_coeff= s->coeffs[i];
1615 s->coeffs[i].index=0;
1616 s->coeffs[i].coeff=0;
1617 s->coeffs[i].next= s->next_coeff++;
1619 s->coeffs[i].coeff += predicted_dc;
1621 last_dc[current_frame_type] = DC_COEFF(i);
1622 if(DC_COEFF(i) && !(s->all_fragments[i].coeff_count&127)){
1623 s->all_fragments[i].coeff_count= 129;
1624 // s->all_fragments[i].next_coeff= s->next_coeff;
1625 s->coeffs[i].next= s->next_coeff;
1626 (s->next_coeff++)->next=NULL;
1634 static void horizontal_filter(unsigned char *first_pixel, int stride,
1635 int *bounding_values);
1636 static void vertical_filter(unsigned char *first_pixel, int stride,
1637 int *bounding_values);
1640 * Perform the final rendering for a particular slice of data.
1641 * The slice number ranges from 0..(macroblock_height - 1).
1643 static void render_slice(Vp3DecodeContext *s, int slice)
1647 int i; /* indicates current fragment */
1648 int16_t *dequantizer;
1649 DECLARE_ALIGNED_16(DCTELEM, block[64]);
1650 unsigned char *output_plane;
1651 unsigned char *last_plane;
1652 unsigned char *golden_plane;
1654 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1655 int upper_motion_limit, lower_motion_limit;
1656 int motion_halfpel_index;
1657 uint8_t *motion_source;
1662 int current_macroblock_entry = slice * s->macroblock_width * 6;
1665 if (slice >= s->macroblock_height)
1668 for (plane = 0; plane < 3; plane++) {
1670 /* set up plane-specific parameters */
1672 output_plane = s->current_frame.data[0];
1673 last_plane = s->last_frame.data[0];
1674 golden_plane = s->golden_frame.data[0];
1675 stride = s->current_frame.linesize[0];
1676 if (!s->flipped_image) stride = -stride;
1677 upper_motion_limit = 7 * s->current_frame.linesize[0];
1678 lower_motion_limit = s->height * s->current_frame.linesize[0] + s->width - 8;
1679 y = slice * FRAGMENT_PIXELS * 2;
1680 plane_width = s->width;
1681 plane_height = s->height;
1682 slice_height = y + FRAGMENT_PIXELS * 2;
1683 i = s->macroblock_fragments[current_macroblock_entry + 0];
1684 } else if (plane == 1) {
1685 output_plane = s->current_frame.data[1];
1686 last_plane = s->last_frame.data[1];
1687 golden_plane = s->golden_frame.data[1];
1688 stride = s->current_frame.linesize[1];
1689 if (!s->flipped_image) stride = -stride;
1690 upper_motion_limit = 7 * s->current_frame.linesize[1];
1691 lower_motion_limit = (s->height / 2) * s->current_frame.linesize[1] + (s->width / 2) - 8;
1692 y = slice * FRAGMENT_PIXELS;
1693 plane_width = s->width / 2;
1694 plane_height = s->height / 2;
1695 slice_height = y + FRAGMENT_PIXELS;
1696 i = s->macroblock_fragments[current_macroblock_entry + 4];
1698 output_plane = s->current_frame.data[2];
1699 last_plane = s->last_frame.data[2];
1700 golden_plane = s->golden_frame.data[2];
1701 stride = s->current_frame.linesize[2];
1702 if (!s->flipped_image) stride = -stride;
1703 upper_motion_limit = 7 * s->current_frame.linesize[2];
1704 lower_motion_limit = (s->height / 2) * s->current_frame.linesize[2] + (s->width / 2) - 8;
1705 y = slice * FRAGMENT_PIXELS;
1706 plane_width = s->width / 2;
1707 plane_height = s->height / 2;
1708 slice_height = y + FRAGMENT_PIXELS;
1709 i = s->macroblock_fragments[current_macroblock_entry + 5];
1711 fragment_width = plane_width / FRAGMENT_PIXELS;
1713 if(ABS(stride) > 2048)
1714 return; //various tables are fixed size
1716 /* for each fragment row in the slice (both of them)... */
1717 for (; y < slice_height; y += 8) {
1719 /* for each fragment in a row... */
1720 for (x = 0; x < plane_width; x += 8, i++) {
1722 if ((i < 0) || (i >= s->fragment_count)) {
1723 av_log(s->avctx, AV_LOG_ERROR, " vp3:render_slice(): bad fragment number (%d)\n", i);
1727 /* transform if this block was coded */
1728 if ((s->all_fragments[i].coding_method != MODE_COPY) &&
1729 !((s->avctx->flags & CODEC_FLAG_GRAY) && plane)) {
1731 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1732 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1733 motion_source= golden_plane;
1735 motion_source= last_plane;
1737 motion_source += s->all_fragments[i].first_pixel;
1738 motion_halfpel_index = 0;
1740 /* sort out the motion vector if this fragment is coded
1741 * using a motion vector method */
1742 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1743 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1745 motion_x = s->all_fragments[i].motion_x;
1746 motion_y = s->all_fragments[i].motion_y;
1748 motion_x= (motion_x>>1) | (motion_x&1);
1749 motion_y= (motion_y>>1) | (motion_y&1);
1752 src_x= (motion_x>>1) + x;
1753 src_y= (motion_y>>1) + y;
1754 if ((motion_x == 127) || (motion_y == 127))
1755 av_log(s->avctx, AV_LOG_ERROR, " help! got invalid motion vector! (%X, %X)\n", motion_x, motion_y);
1757 motion_halfpel_index = motion_x & 0x01;
1758 motion_source += (motion_x >> 1);
1760 motion_halfpel_index |= (motion_y & 0x01) << 1;
1761 motion_source += ((motion_y >> 1) * stride);
1763 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1764 uint8_t *temp= s->edge_emu_buffer;
1765 if(stride<0) temp -= 9*stride;
1766 else temp += 9*stride;
1768 ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1769 motion_source= temp;
1774 /* first, take care of copying a block from either the
1775 * previous or the golden frame */
1776 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1777 /* Note, it is possible to implement all MC cases with
1778 put_no_rnd_pixels_l2 which would look more like the
1779 VP3 source but this would be slower as
1780 put_no_rnd_pixels_tab is better optimzed */
1781 if(motion_halfpel_index != 3){
1782 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1783 output_plane + s->all_fragments[i].first_pixel,
1784 motion_source, stride, 8);
1786 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1787 s->dsp.put_no_rnd_pixels_l2[1](
1788 output_plane + s->all_fragments[i].first_pixel,
1790 motion_source + stride + 1 + d,
1793 dequantizer = s->qmat[1][plane];
1795 dequantizer = s->qmat[0][plane];
1798 /* dequantize the DCT coefficients */
1799 debug_idct("fragment %d, coding mode %d, DC = %d, dequant = %d:\n",
1800 i, s->all_fragments[i].coding_method,
1801 DC_COEFF(i), dequantizer[0]);
1803 if(s->avctx->idct_algo==FF_IDCT_VP3){
1804 Coeff *coeff= s->coeffs + i;
1805 memset(block, 0, sizeof(block));
1807 block[coeff->index]= coeff->coeff * dequantizer[coeff->index];
1811 Coeff *coeff= s->coeffs + i;
1812 memset(block, 0, sizeof(block));
1814 block[coeff->index]= (coeff->coeff * dequantizer[coeff->index] + 2)>>2;
1819 /* invert DCT and place (or add) in final output */
1821 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1822 if(s->avctx->idct_algo!=FF_IDCT_VP3)
1825 output_plane + s->all_fragments[i].first_pixel,
1830 output_plane + s->all_fragments[i].first_pixel,
1835 debug_idct("block after idct_%s():\n",
1836 (s->all_fragments[i].coding_method == MODE_INTRA)?
1838 for (m = 0; m < 8; m++) {
1839 for (n = 0; n < 8; n++) {
1840 debug_idct(" %3d", *(output_plane +
1841 s->all_fragments[i].first_pixel + (m * stride + n)));
1849 /* copy directly from the previous frame */
1850 s->dsp.put_pixels_tab[1][0](
1851 output_plane + s->all_fragments[i].first_pixel,
1852 last_plane + s->all_fragments[i].first_pixel,
1857 /* perform the left edge filter if:
1858 * - the fragment is not on the left column
1859 * - the fragment is coded in this frame
1860 * - the fragment is not coded in this frame but the left
1861 * fragment is coded in this frame (this is done instead
1862 * of a right edge filter when rendering the left fragment
1863 * since this fragment is not available yet) */
1865 ((s->all_fragments[i].coding_method != MODE_COPY) ||
1866 ((s->all_fragments[i].coding_method == MODE_COPY) &&
1867 (s->all_fragments[i - 1].coding_method != MODE_COPY)) )) {
1869 output_plane + s->all_fragments[i].first_pixel + 7*stride,
1870 -stride, bounding_values);
1873 /* perform the top edge filter if:
1874 * - the fragment is not on the top row
1875 * - the fragment is coded in this frame
1876 * - the fragment is not coded in this frame but the above
1877 * fragment is coded in this frame (this is done instead
1878 * of a bottom edge filter when rendering the above
1879 * fragment since this fragment is not available yet) */
1881 ((s->all_fragments[i].coding_method != MODE_COPY) ||
1882 ((s->all_fragments[i].coding_method == MODE_COPY) &&
1883 (s->all_fragments[i - fragment_width].coding_method != MODE_COPY)) )) {
1885 output_plane + s->all_fragments[i].first_pixel - stride,
1886 -stride, bounding_values);
1893 /* this looks like a good place for slice dispatch... */
1895 * if (slice == s->macroblock_height - 1)
1896 * dispatch (both last slice & 2nd-to-last slice);
1897 * else if (slice > 0)
1898 * dispatch (slice - 1);
1904 static void horizontal_filter(unsigned char *first_pixel, int stride,
1905 int *bounding_values)
1910 for (end= first_pixel + 8*stride; first_pixel < end; first_pixel += stride) {
1912 (first_pixel[-2] - first_pixel[ 1])
1913 +3*(first_pixel[ 0] - first_pixel[-1]);
1914 filter_value = bounding_values[(filter_value + 4) >> 3];
1915 first_pixel[-1] = clip_uint8(first_pixel[-1] + filter_value);
1916 first_pixel[ 0] = clip_uint8(first_pixel[ 0] - filter_value);
1920 static void vertical_filter(unsigned char *first_pixel, int stride,
1921 int *bounding_values)
1925 const int nstride= -stride;
1927 for (end= first_pixel + 8; first_pixel < end; first_pixel++) {
1929 (first_pixel[2 * nstride] - first_pixel[ stride])
1930 +3*(first_pixel[0 ] - first_pixel[nstride]);
1931 filter_value = bounding_values[(filter_value + 4) >> 3];
1932 first_pixel[nstride] = clip_uint8(first_pixel[nstride] + filter_value);
1933 first_pixel[0] = clip_uint8(first_pixel[0] - filter_value);
1937 static void apply_loop_filter(Vp3DecodeContext *s)
1943 unsigned char *plane_data;
1944 int *bounding_values= s->bounding_values_array+127;
1947 int bounding_values_array[256];
1950 /* find the right loop limit value */
1951 for (x = 63; x >= 0; x--) {
1952 if (vp31_ac_scale_factor[x] >= s->quality_index)
1955 filter_limit = vp31_filter_limit_values[s->quality_index];
1957 /* set up the bounding values */
1958 memset(bounding_values_array, 0, 256 * sizeof(int));
1959 for (x = 0; x < filter_limit; x++) {
1960 bounding_values[-x - filter_limit] = -filter_limit + x;
1961 bounding_values[-x] = -x;
1962 bounding_values[x] = x;
1963 bounding_values[x + filter_limit] = filter_limit - x;
1967 for (plane = 0; plane < 3; plane++) {
1970 /* Y plane parameters */
1972 width = s->fragment_width;
1973 height = s->fragment_height;
1974 stride = s->current_frame.linesize[0];
1975 plane_data = s->current_frame.data[0];
1976 } else if (plane == 1) {
1977 /* U plane parameters */
1978 fragment = s->u_fragment_start;
1979 width = s->fragment_width / 2;
1980 height = s->fragment_height / 2;
1981 stride = s->current_frame.linesize[1];
1982 plane_data = s->current_frame.data[1];
1984 /* V plane parameters */
1985 fragment = s->v_fragment_start;
1986 width = s->fragment_width / 2;
1987 height = s->fragment_height / 2;
1988 stride = s->current_frame.linesize[2];
1989 plane_data = s->current_frame.data[2];
1992 for (y = 0; y < height; y++) {
1994 for (x = 0; x < width; x++) {
1996 /* do not perform left edge filter for left columns frags */
1998 (s->all_fragments[fragment].coding_method != MODE_COPY)) {
2000 plane_data + s->all_fragments[fragment].first_pixel - 7*stride,
2001 stride, bounding_values);
2004 /* do not perform top edge filter for top row fragments */
2006 (s->all_fragments[fragment].coding_method != MODE_COPY)) {
2008 plane_data + s->all_fragments[fragment].first_pixel + stride,
2009 stride, bounding_values);
2012 /* do not perform right edge filter for right column
2013 * fragments or if right fragment neighbor is also coded
2014 * in this frame (it will be filtered in next iteration) */
2015 if ((x < width - 1) &&
2016 (s->all_fragments[fragment].coding_method != MODE_COPY) &&
2017 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
2019 plane_data + s->all_fragments[fragment + 1].first_pixel - 7*stride,
2020 stride, bounding_values);
2023 /* do not perform bottom edge filter for bottom row
2024 * fragments or if bottom fragment neighbor is also coded
2025 * in this frame (it will be filtered in the next row) */
2026 if ((y < height - 1) &&
2027 (s->all_fragments[fragment].coding_method != MODE_COPY) &&
2028 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
2030 plane_data + s->all_fragments[fragment + width].first_pixel + stride,
2031 stride, bounding_values);
2035 STOP_TIMER("loop filter")
2042 * This function computes the first pixel addresses for each fragment.
2043 * This function needs to be invoked after the first frame is allocated
2044 * so that it has access to the plane strides.
2046 static void vp3_calculate_pixel_addresses(Vp3DecodeContext *s)
2051 /* figure out the first pixel addresses for each of the fragments */
2054 for (y = s->fragment_height; y > 0; y--) {
2055 for (x = 0; x < s->fragment_width; x++) {
2056 s->all_fragments[i++].first_pixel =
2057 s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
2058 s->golden_frame.linesize[0] +
2059 x * FRAGMENT_PIXELS;
2060 debug_init(" fragment %d, first pixel @ %d\n",
2061 i-1, s->all_fragments[i-1].first_pixel);
2066 i = s->u_fragment_start;
2067 for (y = s->fragment_height / 2; y > 0; y--) {
2068 for (x = 0; x < s->fragment_width / 2; x++) {
2069 s->all_fragments[i++].first_pixel =
2070 s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
2071 s->golden_frame.linesize[1] +
2072 x * FRAGMENT_PIXELS;
2073 debug_init(" fragment %d, first pixel @ %d\n",
2074 i-1, s->all_fragments[i-1].first_pixel);
2079 i = s->v_fragment_start;
2080 for (y = s->fragment_height / 2; y > 0; y--) {
2081 for (x = 0; x < s->fragment_width / 2; x++) {
2082 s->all_fragments[i++].first_pixel =
2083 s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
2084 s->golden_frame.linesize[2] +
2085 x * FRAGMENT_PIXELS;
2086 debug_init(" fragment %d, first pixel @ %d\n",
2087 i-1, s->all_fragments[i-1].first_pixel);
2092 /* FIXME: this should be merged with the above! */
2093 static void theora_calculate_pixel_addresses(Vp3DecodeContext *s)
2098 /* figure out the first pixel addresses for each of the fragments */
2101 for (y = 1; y <= s->fragment_height; y++) {
2102 for (x = 0; x < s->fragment_width; x++) {
2103 s->all_fragments[i++].first_pixel =
2104 s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
2105 s->golden_frame.linesize[0] +
2106 x * FRAGMENT_PIXELS;
2107 debug_init(" fragment %d, first pixel @ %d\n",
2108 i-1, s->all_fragments[i-1].first_pixel);
2113 i = s->u_fragment_start;
2114 for (y = 1; y <= s->fragment_height / 2; y++) {
2115 for (x = 0; x < s->fragment_width / 2; x++) {
2116 s->all_fragments[i++].first_pixel =
2117 s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
2118 s->golden_frame.linesize[1] +
2119 x * FRAGMENT_PIXELS;
2120 debug_init(" fragment %d, first pixel @ %d\n",
2121 i-1, s->all_fragments[i-1].first_pixel);
2126 i = s->v_fragment_start;
2127 for (y = 1; y <= s->fragment_height / 2; y++) {
2128 for (x = 0; x < s->fragment_width / 2; x++) {
2129 s->all_fragments[i++].first_pixel =
2130 s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
2131 s->golden_frame.linesize[2] +
2132 x * FRAGMENT_PIXELS;
2133 debug_init(" fragment %d, first pixel @ %d\n",
2134 i-1, s->all_fragments[i-1].first_pixel);
2140 * This is the ffmpeg/libavcodec API init function.
2142 static int vp3_decode_init(AVCodecContext *avctx)
2144 Vp3DecodeContext *s = avctx->priv_data;
2145 int i, inter, plane;
2148 int y_superblock_count;
2149 int c_superblock_count;
2151 if (avctx->codec_tag == MKTAG('V','P','3','0'))
2157 s->width = (avctx->width + 15) & 0xFFFFFFF0;
2158 s->height = (avctx->height + 15) & 0xFFFFFFF0;
2159 avctx->pix_fmt = PIX_FMT_YUV420P;
2160 avctx->has_b_frames = 0;
2161 if(avctx->idct_algo==FF_IDCT_AUTO)
2162 avctx->idct_algo=FF_IDCT_VP3;
2163 dsputil_init(&s->dsp, avctx);
2165 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
2167 /* initialize to an impossible value which will force a recalculation
2168 * in the first frame decode */
2169 s->quality_index = -1;
2171 s->y_superblock_width = (s->width + 31) / 32;
2172 s->y_superblock_height = (s->height + 31) / 32;
2173 y_superblock_count = s->y_superblock_width * s->y_superblock_height;
2175 /* work out the dimensions for the C planes */
2176 c_width = s->width / 2;
2177 c_height = s->height / 2;
2178 s->c_superblock_width = (c_width + 31) / 32;
2179 s->c_superblock_height = (c_height + 31) / 32;
2180 c_superblock_count = s->c_superblock_width * s->c_superblock_height;
2182 s->superblock_count = y_superblock_count + (c_superblock_count * 2);
2183 s->u_superblock_start = y_superblock_count;
2184 s->v_superblock_start = s->u_superblock_start + c_superblock_count;
2185 s->superblock_coding = av_malloc(s->superblock_count);
2187 s->macroblock_width = (s->width + 15) / 16;
2188 s->macroblock_height = (s->height + 15) / 16;
2189 s->macroblock_count = s->macroblock_width * s->macroblock_height;
2191 s->fragment_width = s->width / FRAGMENT_PIXELS;
2192 s->fragment_height = s->height / FRAGMENT_PIXELS;
2194 /* fragment count covers all 8x8 blocks for all 3 planes */
2195 s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
2196 s->u_fragment_start = s->fragment_width * s->fragment_height;
2197 s->v_fragment_start = s->fragment_width * s->fragment_height * 5 / 4;
2199 debug_init(" Y plane: %d x %d\n", s->width, s->height);
2200 debug_init(" C plane: %d x %d\n", c_width, c_height);
2201 debug_init(" Y superblocks: %d x %d, %d total\n",
2202 s->y_superblock_width, s->y_superblock_height, y_superblock_count);
2203 debug_init(" C superblocks: %d x %d, %d total\n",
2204 s->c_superblock_width, s->c_superblock_height, c_superblock_count);
2205 debug_init(" total superblocks = %d, U starts @ %d, V starts @ %d\n",
2206 s->superblock_count, s->u_superblock_start, s->v_superblock_start);
2207 debug_init(" macroblocks: %d x %d, %d total\n",
2208 s->macroblock_width, s->macroblock_height, s->macroblock_count);
2209 debug_init(" %d fragments, %d x %d, u starts @ %d, v starts @ %d\n",
2213 s->u_fragment_start,
2214 s->v_fragment_start);
2216 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
2217 s->coeffs = av_malloc(s->fragment_count * sizeof(Coeff) * 65);
2218 s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int));
2219 s->pixel_addresses_inited = 0;
2221 if (!s->theora_tables)
2223 for (i = 0; i < 64; i++)
2224 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
2225 for (i = 0; i < 64; i++)
2226 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
2227 for (i = 0; i < 64; i++)
2228 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
2229 for (i = 0; i < 64; i++)
2230 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
2231 for (i = 0; i < 64; i++)
2232 s->base_matrix[2][i] = vp31_inter_dequant[i];
2233 for (i = 0; i < 64; i++)
2234 s->filter_limit_values[i] = vp31_filter_limit_values[i];
2236 for(inter=0; inter<2; inter++){
2237 for(plane=0; plane<3; plane++){
2238 s->qr_count[inter][plane]= 1;
2239 s->qr_size [inter][plane][0]= 63;
2240 s->qr_base [inter][plane][0]=
2241 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
2245 /* init VLC tables */
2246 for (i = 0; i < 16; i++) {
2249 init_vlc(&s->dc_vlc[i], 5, 32,
2250 &dc_bias[i][0][1], 4, 2,
2251 &dc_bias[i][0][0], 4, 2, 0);
2253 /* group 1 AC histograms */
2254 init_vlc(&s->ac_vlc_1[i], 5, 32,
2255 &ac_bias_0[i][0][1], 4, 2,
2256 &ac_bias_0[i][0][0], 4, 2, 0);
2258 /* group 2 AC histograms */
2259 init_vlc(&s->ac_vlc_2[i], 5, 32,
2260 &ac_bias_1[i][0][1], 4, 2,
2261 &ac_bias_1[i][0][0], 4, 2, 0);
2263 /* group 3 AC histograms */
2264 init_vlc(&s->ac_vlc_3[i], 5, 32,
2265 &ac_bias_2[i][0][1], 4, 2,
2266 &ac_bias_2[i][0][0], 4, 2, 0);
2268 /* group 4 AC histograms */
2269 init_vlc(&s->ac_vlc_4[i], 5, 32,
2270 &ac_bias_3[i][0][1], 4, 2,
2271 &ac_bias_3[i][0][0], 4, 2, 0);
2274 for (i = 0; i < 16; i++) {
2277 init_vlc(&s->dc_vlc[i], 5, 32,
2278 &s->huffman_table[i][0][1], 4, 2,
2279 &s->huffman_table[i][0][0], 4, 2, 0);
2281 /* group 1 AC histograms */
2282 init_vlc(&s->ac_vlc_1[i], 5, 32,
2283 &s->huffman_table[i+16][0][1], 4, 2,
2284 &s->huffman_table[i+16][0][0], 4, 2, 0);
2286 /* group 2 AC histograms */
2287 init_vlc(&s->ac_vlc_2[i], 5, 32,
2288 &s->huffman_table[i+16*2][0][1], 4, 2,
2289 &s->huffman_table[i+16*2][0][0], 4, 2, 0);
2291 /* group 3 AC histograms */
2292 init_vlc(&s->ac_vlc_3[i], 5, 32,
2293 &s->huffman_table[i+16*3][0][1], 4, 2,
2294 &s->huffman_table[i+16*3][0][0], 4, 2, 0);
2296 /* group 4 AC histograms */
2297 init_vlc(&s->ac_vlc_4[i], 5, 32,
2298 &s->huffman_table[i+16*4][0][1], 4, 2,
2299 &s->huffman_table[i+16*4][0][0], 4, 2, 0);
2303 init_vlc(&s->superblock_run_length_vlc, 6, 34,
2304 &superblock_run_length_vlc_table[0][1], 4, 2,
2305 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
2307 init_vlc(&s->fragment_run_length_vlc, 5, 30,
2308 &fragment_run_length_vlc_table[0][1], 4, 2,
2309 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
2311 init_vlc(&s->mode_code_vlc, 3, 8,
2312 &mode_code_vlc_table[0][1], 2, 1,
2313 &mode_code_vlc_table[0][0], 2, 1, 0);
2315 init_vlc(&s->motion_vector_vlc, 6, 63,
2316 &motion_vector_vlc_table[0][1], 2, 1,
2317 &motion_vector_vlc_table[0][0], 2, 1, 0);
2319 /* work out the block mapping tables */
2320 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
2321 s->superblock_macroblocks = av_malloc(s->superblock_count * 4 * sizeof(int));
2322 s->macroblock_fragments = av_malloc(s->macroblock_count * 6 * sizeof(int));
2323 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
2324 init_block_mapping(s);
2326 for (i = 0; i < 3; i++) {
2327 s->current_frame.data[i] = NULL;
2328 s->last_frame.data[i] = NULL;
2329 s->golden_frame.data[i] = NULL;
2336 * This is the ffmpeg/libavcodec API frame decode function.
2338 static int vp3_decode_frame(AVCodecContext *avctx,
2339 void *data, int *data_size,
2340 uint8_t *buf, int buf_size)
2342 Vp3DecodeContext *s = avctx->priv_data;
2344 static int counter = 0;
2347 init_get_bits(&gb, buf, buf_size * 8);
2349 if (s->theora && get_bits1(&gb))
2352 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
2355 int ptype = get_bits(&gb, 7);
2357 skip_bits(&gb, 6*8); /* "theora" */
2362 theora_decode_comments(avctx, &gb);
2365 theora_decode_tables(avctx, &gb);
2366 init_dequantizer(s);
2369 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype);
2375 s->keyframe = !get_bits1(&gb);
2378 s->last_quality_index = s->quality_index;
2379 s->quality_index = get_bits(&gb, 6);
2380 if (s->theora >= 0x030200)
2383 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
2384 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
2385 s->keyframe?"key":"", counter, s->quality_index);
2388 if (s->quality_index != s->last_quality_index) {
2389 init_dequantizer(s);
2390 init_loop_filter(s);
2396 skip_bits(&gb, 4); /* width code */
2397 skip_bits(&gb, 4); /* height code */
2400 s->version = get_bits(&gb, 5);
2402 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
2405 if (s->version || s->theora)
2408 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
2409 skip_bits(&gb, 2); /* reserved? */
2412 if (s->last_frame.data[0] == s->golden_frame.data[0]) {
2413 if (s->golden_frame.data[0])
2414 avctx->release_buffer(avctx, &s->golden_frame);
2415 s->last_frame= s->golden_frame; /* ensure that we catch any access to this released frame */
2417 if (s->golden_frame.data[0])
2418 avctx->release_buffer(avctx, &s->golden_frame);
2419 if (s->last_frame.data[0])
2420 avctx->release_buffer(avctx, &s->last_frame);
2423 s->golden_frame.reference = 3;
2424 if(avctx->get_buffer(avctx, &s->golden_frame) < 0) {
2425 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
2429 /* golden frame is also the current frame */
2430 s->current_frame= s->golden_frame;
2432 /* time to figure out pixel addresses? */
2433 if (!s->pixel_addresses_inited)
2435 if (!s->flipped_image)
2436 vp3_calculate_pixel_addresses(s);
2438 theora_calculate_pixel_addresses(s);
2441 /* allocate a new current frame */
2442 s->current_frame.reference = 3;
2443 if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
2444 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
2449 s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
2450 s->current_frame.qstride= 0;
2454 STOP_TIMER("init_frame")}
2459 memcpy(s->current_frame.data[0], s->golden_frame.data[0],
2460 s->current_frame.linesize[0] * s->height);
2461 memcpy(s->current_frame.data[1], s->golden_frame.data[1],
2462 s->current_frame.linesize[1] * s->height / 2);
2463 memcpy(s->current_frame.data[2], s->golden_frame.data[2],
2464 s->current_frame.linesize[2] * s->height / 2);
2470 if (unpack_superblocks(s, &gb)){
2471 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
2474 STOP_TIMER("unpack_superblocks")}
2476 if (unpack_modes(s, &gb)){
2477 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
2480 STOP_TIMER("unpack_modes")}
2482 if (unpack_vectors(s, &gb)){
2483 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2486 STOP_TIMER("unpack_vectors")}
2488 if (unpack_dct_coeffs(s, &gb)){
2489 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
2492 STOP_TIMER("unpack_dct_coeffs")}
2495 reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
2496 if ((avctx->flags & CODEC_FLAG_GRAY) == 0) {
2497 reverse_dc_prediction(s, s->u_fragment_start,
2498 s->fragment_width / 2, s->fragment_height / 2);
2499 reverse_dc_prediction(s, s->v_fragment_start,
2500 s->fragment_width / 2, s->fragment_height / 2);
2502 STOP_TIMER("reverse_dc_prediction")}
2505 for (i = 0; i < s->macroblock_height; i++)
2507 STOP_TIMER("render_fragments")}
2510 apply_loop_filter(s);
2511 STOP_TIMER("apply_loop_filter")}
2516 *data_size=sizeof(AVFrame);
2517 *(AVFrame*)data= s->current_frame;
2519 /* release the last frame, if it is allocated and if it is not the
2521 if ((s->last_frame.data[0]) &&
2522 (s->last_frame.data[0] != s->golden_frame.data[0]))
2523 avctx->release_buffer(avctx, &s->last_frame);
2525 /* shuffle frames (last = current) */
2526 s->last_frame= s->current_frame;
2527 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
2533 * This is the ffmpeg/libavcodec API module cleanup function.
2535 static int vp3_decode_end(AVCodecContext *avctx)
2537 Vp3DecodeContext *s = avctx->priv_data;
2539 av_free(s->all_fragments);
2541 av_free(s->coded_fragment_list);
2542 av_free(s->superblock_fragments);
2543 av_free(s->superblock_macroblocks);
2544 av_free(s->macroblock_fragments);
2545 av_free(s->macroblock_coding);
2547 /* release all frames */
2548 if (s->golden_frame.data[0] && s->golden_frame.data[0] != s->last_frame.data[0])
2549 avctx->release_buffer(avctx, &s->golden_frame);
2550 if (s->last_frame.data[0])
2551 avctx->release_buffer(avctx, &s->last_frame);
2552 /* no need to release the current_frame since it will always be pointing
2553 * to the same frame as either the golden or last frame */
2558 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2560 Vp3DecodeContext *s = avctx->priv_data;
2562 if (get_bits(gb, 1)) {
2564 if (s->entries >= 32) { /* overflow */
2565 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2568 token = get_bits(gb, 5);
2569 //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);
2570 s->huffman_table[s->hti][token][0] = s->hbits;
2571 s->huffman_table[s->hti][token][1] = s->huff_code_size;
2575 if (s->huff_code_size >= 32) {/* overflow */
2576 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2579 s->huff_code_size++;
2581 read_huffman_tree(avctx, gb);
2583 read_huffman_tree(avctx, gb);
2585 s->huff_code_size--;
2590 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2592 Vp3DecodeContext *s = avctx->priv_data;
2594 s->theora = get_bits_long(gb, 24);
2595 av_log(avctx, AV_LOG_INFO, "Theora bitstream version %X\n", s->theora);
2597 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2598 /* but previous versions have the image flipped relative to vp3 */
2599 if (s->theora < 0x030200)
2601 s->flipped_image = 1;
2602 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2605 s->width = get_bits(gb, 16) << 4;
2606 s->height = get_bits(gb, 16) << 4;
2608 if(avcodec_check_dimensions(avctx, s->width, s->height)){
2609 av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2610 s->width= s->height= 0;
2614 if (s->theora >= 0x030400)
2616 skip_bits(gb, 32); /* total number of superblocks in a frame */
2617 // fixme, the next field is 36bits long
2618 skip_bits(gb, 32); /* total number of blocks in a frame */
2619 skip_bits(gb, 4); /* total number of blocks in a frame */
2620 skip_bits(gb, 32); /* total number of macroblocks in a frame */
2622 skip_bits(gb, 24); /* frame width */
2623 skip_bits(gb, 24); /* frame height */
2627 skip_bits(gb, 24); /* frame width */
2628 skip_bits(gb, 24); /* frame height */
2631 skip_bits(gb, 8); /* offset x */
2632 skip_bits(gb, 8); /* offset y */
2634 skip_bits(gb, 32); /* fps numerator */
2635 skip_bits(gb, 32); /* fps denumerator */
2636 skip_bits(gb, 24); /* aspect numerator */
2637 skip_bits(gb, 24); /* aspect denumerator */
2639 if (s->theora < 0x030200)
2640 skip_bits(gb, 5); /* keyframe frequency force */
2641 skip_bits(gb, 8); /* colorspace */
2642 if (s->theora >= 0x030400)
2643 skip_bits(gb, 2); /* pixel format: 420,res,422,444 */
2644 skip_bits(gb, 24); /* bitrate */
2646 skip_bits(gb, 6); /* quality hint */
2648 if (s->theora >= 0x030200)
2650 skip_bits(gb, 5); /* keyframe frequency force */
2652 if (s->theora < 0x030400)
2653 skip_bits(gb, 5); /* spare bits */
2656 // align_get_bits(gb);
2658 avctx->width = s->width;
2659 avctx->height = s->height;
2664 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2666 Vp3DecodeContext *s = avctx->priv_data;
2667 int i, n, matrices, inter, plane;
2669 if (s->theora >= 0x030200) {
2670 n = get_bits(gb, 3);
2671 /* loop filter limit values table */
2672 for (i = 0; i < 64; i++)
2673 s->filter_limit_values[i] = get_bits(gb, n);
2676 if (s->theora >= 0x030200)
2677 n = get_bits(gb, 4) + 1;
2680 /* quality threshold table */
2681 for (i = 0; i < 64; i++)
2682 s->coded_ac_scale_factor[i] = get_bits(gb, n);
2684 if (s->theora >= 0x030200)
2685 n = get_bits(gb, 4) + 1;
2688 /* dc scale factor table */
2689 for (i = 0; i < 64; i++)
2690 s->coded_dc_scale_factor[i] = get_bits(gb, n);
2692 if (s->theora >= 0x030200)
2693 matrices = get_bits(gb, 9) + 1;
2698 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2702 for(n=0; n<matrices; n++){
2703 for (i = 0; i < 64; i++)
2704 s->base_matrix[n][i]= get_bits(gb, 8);
2707 for (inter = 0; inter <= 1; inter++) {
2708 for (plane = 0; plane <= 2; plane++) {
2710 if (inter || plane > 0)
2711 newqr = get_bits(gb, 1);
2714 if(inter && get_bits(gb, 1)){
2718 qtj= (3*inter + plane - 1) / 3;
2719 plj= (plane + 2) % 3;
2721 s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2722 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2723 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2729 i= get_bits(gb, av_log2(matrices-1)+1);
2731 av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2734 s->qr_base[inter][plane][qri]= i;
2737 i = get_bits(gb, av_log2(63-qi)+1) + 1;
2738 s->qr_size[inter][plane][qri++]= i;
2743 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2746 s->qr_count[inter][plane]= qri;
2751 /* Huffman tables */
2752 for (s->hti = 0; s->hti < 80; s->hti++) {
2754 s->huff_code_size = 1;
2755 if (!get_bits(gb, 1)) {
2757 read_huffman_tree(avctx, gb);
2759 read_huffman_tree(avctx, gb);
2763 s->theora_tables = 1;
2768 static int theora_decode_init(AVCodecContext *avctx)
2770 Vp3DecodeContext *s = avctx->priv_data;
2773 uint8_t *p= avctx->extradata;
2778 if (!avctx->extradata_size)
2780 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2785 op_bytes = *(p++)<<8;
2788 init_get_bits(&gb, p, op_bytes);
2791 ptype = get_bits(&gb, 8);
2792 debug_vp3("Theora headerpacket type: %x\n", ptype);
2794 if (!(ptype & 0x80))
2796 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2800 // FIXME: check for this aswell
2801 skip_bits(&gb, 6*8); /* "theora" */
2806 theora_decode_header(avctx, &gb);
2809 // FIXME: is this needed? it breaks sometimes
2810 // theora_decode_comments(avctx, gb);
2813 theora_decode_tables(avctx, &gb);
2816 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2819 if(8*op_bytes != get_bits_count(&gb))
2820 av_log(avctx, AV_LOG_ERROR, "%d bits left in packet %X\n", 8*op_bytes - get_bits_count(&gb), ptype);
2823 vp3_decode_init(avctx);
2827 AVCodec vp3_decoder = {
2831 sizeof(Vp3DecodeContext),
2840 #ifndef CONFIG_LIBTHEORA
2841 AVCodec theora_decoder = {
2845 sizeof(Vp3DecodeContext),