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;
235 int last_quality_index;
237 int superblock_count;
238 int superblock_width;
239 int superblock_height;
240 int y_superblock_width;
241 int y_superblock_height;
242 int c_superblock_width;
243 int c_superblock_height;
244 int u_superblock_start;
245 int v_superblock_start;
246 unsigned char *superblock_coding;
248 int macroblock_count;
249 int macroblock_width;
250 int macroblock_height;
256 Vp3Fragment *all_fragments;
259 int fragment_start[3];
264 uint16_t coded_dc_scale_factor[64];
265 uint32_t coded_ac_scale_factor[64];
266 uint8_t base_matrix[384][64];
267 uint8_t qr_count[2][3];
268 uint8_t qr_size [2][3][64];
269 uint16_t qr_base[2][3][64];
271 /* this is a list of indices into the all_fragments array indicating
272 * which of the fragments are coded */
273 int *coded_fragment_list;
274 int coded_fragment_list_index;
275 int pixel_addresses_inited;
283 VLC superblock_run_length_vlc;
284 VLC fragment_run_length_vlc;
286 VLC motion_vector_vlc;
288 /* these arrays need to be on 16-byte boundaries since SSE2 operations
290 DECLARE_ALIGNED_16(int16_t, qmat[2][4][64]); //<qmat[is_inter][plane]
292 /* This table contains superblock_count * 16 entries. Each set of 16
293 * numbers corresponds to the fragment indices 0..15 of the superblock.
294 * An entry will be -1 to indicate that no entry corresponds to that
296 int *superblock_fragments;
298 /* This table contains superblock_count * 4 entries. Each set of 4
299 * numbers corresponds to the macroblock indices 0..3 of the superblock.
300 * An entry will be -1 to indicate that no entry corresponds to that
302 int *superblock_macroblocks;
304 /* This table contains macroblock_count * 6 entries. Each set of 6
305 * numbers corresponds to the fragment indices 0..5 which comprise
306 * the macroblock (4 Y fragments and 2 C fragments). */
307 int *macroblock_fragments;
308 /* This is an array that indicates how a particular macroblock
310 unsigned char *macroblock_coding;
312 int first_coded_y_fragment;
313 int first_coded_c_fragment;
314 int last_coded_y_fragment;
315 int last_coded_c_fragment;
317 uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
318 uint8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
325 uint16_t huffman_table[80][32][2];
327 uint32_t filter_limit_values[64];
328 int bounding_values_array[256];
331 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb);
333 /************************************************************************
334 * VP3 specific functions
335 ************************************************************************/
338 * This function sets up all of the various blocks mappings:
339 * superblocks <-> fragments, macroblocks <-> fragments,
340 * superblocks <-> macroblocks
342 * Returns 0 is successful; returns 1 if *anything* went wrong.
344 static int init_block_mapping(Vp3DecodeContext *s)
347 signed int hilbert_walk_y[16];
348 signed int hilbert_walk_c[16];
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 /* figure out hilbert pattern per these frame dimensions */
388 hilbert_walk_y[0] = 1;
389 hilbert_walk_y[1] = 1;
390 hilbert_walk_y[2] = s->fragment_width;
391 hilbert_walk_y[3] = -1;
392 hilbert_walk_y[4] = s->fragment_width;
393 hilbert_walk_y[5] = s->fragment_width;
394 hilbert_walk_y[6] = 1;
395 hilbert_walk_y[7] = -s->fragment_width;
396 hilbert_walk_y[8] = 1;
397 hilbert_walk_y[9] = s->fragment_width;
398 hilbert_walk_y[10] = 1;
399 hilbert_walk_y[11] = -s->fragment_width;
400 hilbert_walk_y[12] = -s->fragment_width;
401 hilbert_walk_y[13] = -1;
402 hilbert_walk_y[14] = -s->fragment_width;
403 hilbert_walk_y[15] = 1;
405 hilbert_walk_c[0] = 1;
406 hilbert_walk_c[1] = 1;
407 hilbert_walk_c[2] = s->fragment_width / 2;
408 hilbert_walk_c[3] = -1;
409 hilbert_walk_c[4] = s->fragment_width / 2;
410 hilbert_walk_c[5] = s->fragment_width / 2;
411 hilbert_walk_c[6] = 1;
412 hilbert_walk_c[7] = -s->fragment_width / 2;
413 hilbert_walk_c[8] = 1;
414 hilbert_walk_c[9] = s->fragment_width / 2;
415 hilbert_walk_c[10] = 1;
416 hilbert_walk_c[11] = -s->fragment_width / 2;
417 hilbert_walk_c[12] = -s->fragment_width / 2;
418 hilbert_walk_c[13] = -1;
419 hilbert_walk_c[14] = -s->fragment_width / 2;
420 hilbert_walk_c[15] = 1;
422 hilbert_walk_mb[0] = 1;
423 hilbert_walk_mb[1] = s->macroblock_width;
424 hilbert_walk_mb[2] = 1;
425 hilbert_walk_mb[3] = -s->macroblock_width;
427 /* iterate through each superblock (all planes) and map the fragments */
428 for (i = 0; i < s->superblock_count; i++) {
429 debug_init(" superblock %d (u starts @ %d, v starts @ %d)\n",
430 i, s->u_superblock_start, s->v_superblock_start);
432 /* time to re-assign the limits? */
435 /* start of Y superblocks */
436 right_edge = s->fragment_width;
437 bottom_edge = s->fragment_height;
440 superblock_row_inc = 3 * s->fragment_width -
441 (s->y_superblock_width * 4 - s->fragment_width);
442 hilbert = hilbert_walk_y;
444 /* the first operation for this variable is to advance by 1 */
445 current_fragment = -1;
447 } else if (i == s->u_superblock_start) {
449 /* start of U superblocks */
450 right_edge = s->fragment_width / 2;
451 bottom_edge = s->fragment_height / 2;
454 superblock_row_inc = 3 * (s->fragment_width / 2) -
455 (s->c_superblock_width * 4 - s->fragment_width / 2);
456 hilbert = hilbert_walk_c;
458 /* the first operation for this variable is to advance by 1 */
459 current_fragment = s->fragment_start[1] - 1;
461 } else if (i == s->v_superblock_start) {
463 /* start of V superblocks */
464 right_edge = s->fragment_width / 2;
465 bottom_edge = s->fragment_height / 2;
468 superblock_row_inc = 3 * (s->fragment_width / 2) -
469 (s->c_superblock_width * 4 - s->fragment_width / 2);
470 hilbert = hilbert_walk_c;
472 /* the first operation for this variable is to advance by 1 */
473 current_fragment = s->fragment_start[2] - 1;
477 if (current_width >= right_edge - 1) {
478 /* reset width and move to next superblock row */
482 /* fragment is now at the start of a new superblock row */
483 current_fragment += superblock_row_inc;
486 /* iterate through all 16 fragments in a superblock */
487 for (j = 0; j < 16; j++) {
488 current_fragment += hilbert[j];
489 current_width += travel_width[j];
490 current_height += travel_height[j];
492 /* check if the fragment is in bounds */
493 if ((current_width < right_edge) &&
494 (current_height < bottom_edge)) {
495 s->superblock_fragments[mapping_index] = current_fragment;
496 debug_init(" mapping fragment %d to superblock %d, position %d (%d/%d x %d/%d)\n",
497 s->superblock_fragments[mapping_index], i, j,
498 current_width, right_edge, current_height, bottom_edge);
500 s->superblock_fragments[mapping_index] = -1;
501 debug_init(" superblock %d, position %d has no fragment (%d/%d x %d/%d)\n",
503 current_width, right_edge, current_height, bottom_edge);
510 /* initialize the superblock <-> macroblock mapping; iterate through
511 * all of the Y plane superblocks to build this mapping */
512 right_edge = s->macroblock_width;
513 bottom_edge = s->macroblock_height;
516 superblock_row_inc = s->macroblock_width -
517 (s->y_superblock_width * 2 - s->macroblock_width);;
518 hilbert = hilbert_walk_mb;
520 current_macroblock = -1;
521 for (i = 0; i < s->u_superblock_start; i++) {
523 if (current_width >= right_edge - 1) {
524 /* reset width and move to next superblock row */
528 /* macroblock is now at the start of a new superblock row */
529 current_macroblock += superblock_row_inc;
532 /* iterate through each potential macroblock in the superblock */
533 for (j = 0; j < 4; j++) {
534 current_macroblock += hilbert_walk_mb[j];
535 current_width += travel_width_mb[j];
536 current_height += travel_height_mb[j];
538 /* check if the macroblock is in bounds */
539 if ((current_width < right_edge) &&
540 (current_height < bottom_edge)) {
541 s->superblock_macroblocks[mapping_index] = current_macroblock;
542 debug_init(" mapping macroblock %d to superblock %d, position %d (%d/%d x %d/%d)\n",
543 s->superblock_macroblocks[mapping_index], i, j,
544 current_width, right_edge, current_height, bottom_edge);
546 s->superblock_macroblocks[mapping_index] = -1;
547 debug_init(" superblock %d, position %d has no macroblock (%d/%d x %d/%d)\n",
549 current_width, right_edge, current_height, bottom_edge);
556 /* initialize the macroblock <-> fragment mapping */
557 current_fragment = 0;
558 current_macroblock = 0;
560 for (i = 0; i < s->fragment_height; i += 2) {
562 for (j = 0; j < s->fragment_width; j += 2) {
564 debug_init(" macroblock %d contains fragments: ", current_macroblock);
565 s->all_fragments[current_fragment].macroblock = current_macroblock;
566 s->macroblock_fragments[mapping_index++] = current_fragment;
567 debug_init("%d ", current_fragment);
569 if (j + 1 < s->fragment_width) {
570 s->all_fragments[current_fragment + 1].macroblock = current_macroblock;
571 s->macroblock_fragments[mapping_index++] = current_fragment + 1;
572 debug_init("%d ", current_fragment + 1);
574 s->macroblock_fragments[mapping_index++] = -1;
576 if (i + 1 < s->fragment_height) {
577 s->all_fragments[current_fragment + s->fragment_width].macroblock =
579 s->macroblock_fragments[mapping_index++] =
580 current_fragment + s->fragment_width;
581 debug_init("%d ", current_fragment + s->fragment_width);
583 s->macroblock_fragments[mapping_index++] = -1;
585 if ((j + 1 < s->fragment_width) && (i + 1 < s->fragment_height)) {
586 s->all_fragments[current_fragment + s->fragment_width + 1].macroblock =
588 s->macroblock_fragments[mapping_index++] =
589 current_fragment + s->fragment_width + 1;
590 debug_init("%d ", current_fragment + s->fragment_width + 1);
592 s->macroblock_fragments[mapping_index++] = -1;
595 c_fragment = s->fragment_start[1] +
596 (i * s->fragment_width / 4) + (j / 2);
597 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
598 s->macroblock_fragments[mapping_index++] = c_fragment;
599 debug_init("%d ", c_fragment);
601 c_fragment = s->fragment_start[2] +
602 (i * s->fragment_width / 4) + (j / 2);
603 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
604 s->macroblock_fragments[mapping_index++] = c_fragment;
605 debug_init("%d ", c_fragment);
609 if (j + 2 <= s->fragment_width)
610 current_fragment += 2;
613 current_macroblock++;
616 current_fragment += s->fragment_width;
619 return 0; /* successful path out */
623 * This function wipes out all of the fragment data.
625 static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
629 /* zero out all of the fragment information */
630 s->coded_fragment_list_index = 0;
631 for (i = 0; i < s->fragment_count; i++) {
632 s->all_fragments[i].coeff_count = 0;
633 s->all_fragments[i].motion_x = 127;
634 s->all_fragments[i].motion_y = 127;
635 s->all_fragments[i].next_coeff= NULL;
637 s->coeffs[i].coeff=0;
638 s->coeffs[i].next= NULL;
643 * This function sets up the dequantization tables used for a particular
646 static void init_dequantizer(Vp3DecodeContext *s)
648 int ac_scale_factor = s->coded_ac_scale_factor[s->quality_index];
649 int dc_scale_factor = s->coded_dc_scale_factor[s->quality_index];
650 int i, j, plane, inter, qri, bmi, bmj, qistart;
652 debug_vp3(" vp3: initializing dequantization tables\n");
654 for(inter=0; inter<2; inter++){
655 for(plane=0; plane<3; plane++){
657 for(qri=0; qri<s->qr_count[inter][plane]; qri++){
658 sum+= s->qr_size[inter][plane][qri];
659 if(s->quality_index <= sum)
662 qistart= sum - s->qr_size[inter][plane][qri];
663 bmi= s->qr_base[inter][plane][qri ];
664 bmj= s->qr_base[inter][plane][qri+1];
666 int coeff= ( 2*(sum -s->quality_index)*s->base_matrix[bmi][i]
667 - 2*(qistart-s->quality_index)*s->base_matrix[bmj][i]
668 + s->qr_size[inter][plane][qri])
669 / (2*s->qr_size[inter][plane][qri]);
671 int qmin= 8<<(inter + !i);
672 int qscale= i ? ac_scale_factor : dc_scale_factor;
674 s->qmat[inter][plane][i]= clip((qscale * coeff)/100 * 4, qmin, 4096);
679 memset(s->qscale_table, (FFMAX(s->qmat[0][0][1], s->qmat[0][1][1])+8)/16, 512); //FIXME finetune
683 * This function initializes the loop filter boundary limits if the frame's
684 * quality index is different from the previous frame's.
686 static void init_loop_filter(Vp3DecodeContext *s)
688 int *bounding_values= s->bounding_values_array+127;
692 filter_limit = s->filter_limit_values[s->quality_index];
694 /* set up the bounding values */
695 memset(s->bounding_values_array, 0, 256 * sizeof(int));
696 for (x = 0; x < filter_limit; x++) {
697 bounding_values[-x - filter_limit] = -filter_limit + x;
698 bounding_values[-x] = -x;
699 bounding_values[x] = x;
700 bounding_values[x + filter_limit] = filter_limit - x;
705 * This function unpacks all of the superblock/macroblock/fragment coding
706 * information from the bitstream.
708 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
711 int current_superblock = 0;
713 int decode_fully_flags = 0;
714 int decode_partial_blocks = 0;
715 int first_c_fragment_seen;
718 int current_fragment;
720 debug_vp3(" vp3: unpacking superblock coding\n");
724 debug_vp3(" keyframe-- all superblocks are fully coded\n");
725 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
729 /* unpack the list of partially-coded superblocks */
730 bit = get_bits(gb, 1);
731 /* toggle the bit because as soon as the first run length is
732 * fetched the bit will be toggled again */
734 while (current_superblock < s->superblock_count) {
735 if (current_run-- == 0) {
737 current_run = get_vlc2(gb,
738 s->superblock_run_length_vlc.table, 6, 2);
739 if (current_run == 33)
740 current_run += get_bits(gb, 12);
741 debug_block_coding(" setting superblocks %d..%d to %s\n",
743 current_superblock + current_run - 1,
744 (bit) ? "partially coded" : "not coded");
746 /* if any of the superblocks are not partially coded, flag
747 * a boolean to decode the list of fully-coded superblocks */
749 decode_fully_flags = 1;
752 /* make a note of the fact that there are partially coded
754 decode_partial_blocks = 1;
757 s->superblock_coding[current_superblock++] = bit;
760 /* unpack the list of fully coded superblocks if any of the blocks were
761 * not marked as partially coded in the previous step */
762 if (decode_fully_flags) {
764 current_superblock = 0;
766 bit = get_bits(gb, 1);
767 /* toggle the bit because as soon as the first run length is
768 * fetched the bit will be toggled again */
770 while (current_superblock < s->superblock_count) {
772 /* skip any superblocks already marked as partially coded */
773 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
775 if (current_run-- == 0) {
777 current_run = get_vlc2(gb,
778 s->superblock_run_length_vlc.table, 6, 2);
779 if (current_run == 33)
780 current_run += get_bits(gb, 12);
783 debug_block_coding(" setting superblock %d to %s\n",
785 (bit) ? "fully coded" : "not coded");
786 s->superblock_coding[current_superblock] = 2*bit;
788 current_superblock++;
792 /* if there were partial blocks, initialize bitstream for
793 * unpacking fragment codings */
794 if (decode_partial_blocks) {
797 bit = get_bits(gb, 1);
798 /* toggle the bit because as soon as the first run length is
799 * fetched the bit will be toggled again */
804 /* figure out which fragments are coded; iterate through each
805 * superblock (all planes) */
806 s->coded_fragment_list_index = 0;
807 s->next_coeff= s->coeffs + s->fragment_count;
808 s->first_coded_y_fragment = s->first_coded_c_fragment = 0;
809 s->last_coded_y_fragment = s->last_coded_c_fragment = -1;
810 first_c_fragment_seen = 0;
811 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
812 for (i = 0; i < s->superblock_count; i++) {
814 /* iterate through all 16 fragments in a superblock */
815 for (j = 0; j < 16; j++) {
817 /* if the fragment is in bounds, check its coding status */
818 current_fragment = s->superblock_fragments[i * 16 + j];
819 if (current_fragment >= s->fragment_count) {
820 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
821 current_fragment, s->fragment_count);
824 if (current_fragment != -1) {
825 if (s->superblock_coding[i] == SB_NOT_CODED) {
827 /* copy all the fragments from the prior frame */
828 s->all_fragments[current_fragment].coding_method =
831 } else if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
833 /* fragment may or may not be coded; this is the case
834 * that cares about the fragment coding runs */
835 if (current_run-- == 0) {
837 current_run = get_vlc2(gb,
838 s->fragment_run_length_vlc.table, 5, 2);
842 /* default mode; actual mode will be decoded in
844 s->all_fragments[current_fragment].coding_method =
846 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
847 s->coded_fragment_list[s->coded_fragment_list_index] =
849 if ((current_fragment >= s->fragment_start[1]) &&
850 (s->last_coded_y_fragment == -1) &&
851 (!first_c_fragment_seen)) {
852 s->first_coded_c_fragment = s->coded_fragment_list_index;
853 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
854 first_c_fragment_seen = 1;
856 s->coded_fragment_list_index++;
857 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
858 debug_block_coding(" superblock %d is partially coded, fragment %d is coded\n",
859 i, current_fragment);
861 /* not coded; copy this fragment from the prior frame */
862 s->all_fragments[current_fragment].coding_method =
864 debug_block_coding(" superblock %d is partially coded, fragment %d is not coded\n",
865 i, current_fragment);
870 /* fragments are fully coded in this superblock; actual
871 * coding will be determined in next step */
872 s->all_fragments[current_fragment].coding_method =
874 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
875 s->coded_fragment_list[s->coded_fragment_list_index] =
877 if ((current_fragment >= s->fragment_start[1]) &&
878 (s->last_coded_y_fragment == -1) &&
879 (!first_c_fragment_seen)) {
880 s->first_coded_c_fragment = s->coded_fragment_list_index;
881 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
882 first_c_fragment_seen = 1;
884 s->coded_fragment_list_index++;
885 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
886 debug_block_coding(" superblock %d is fully coded, fragment %d is coded\n",
887 i, current_fragment);
893 if (!first_c_fragment_seen)
894 /* only Y fragments coded in this frame */
895 s->last_coded_y_fragment = s->coded_fragment_list_index - 1;
897 /* end the list of coded C fragments */
898 s->last_coded_c_fragment = s->coded_fragment_list_index - 1;
900 debug_block_coding(" %d total coded fragments, y: %d -> %d, c: %d -> %d\n",
901 s->coded_fragment_list_index,
902 s->first_coded_y_fragment,
903 s->last_coded_y_fragment,
904 s->first_coded_c_fragment,
905 s->last_coded_c_fragment);
911 * This function unpacks all the coding mode data for individual macroblocks
912 * from the bitstream.
914 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
918 int current_macroblock;
919 int current_fragment;
922 debug_vp3(" vp3: unpacking encoding modes\n");
925 debug_vp3(" keyframe-- all blocks are coded as INTRA\n");
927 for (i = 0; i < s->fragment_count; i++)
928 s->all_fragments[i].coding_method = MODE_INTRA;
932 /* fetch the mode coding scheme for this frame */
933 scheme = get_bits(gb, 3);
934 debug_modes(" using mode alphabet %d\n", scheme);
936 /* is it a custom coding scheme? */
938 debug_modes(" custom mode alphabet ahead:\n");
939 for (i = 0; i < 8; i++)
940 ModeAlphabet[scheme][get_bits(gb, 3)] = i;
943 for (i = 0; i < 8; i++)
944 debug_modes(" mode[%d][%d] = %d\n", scheme, i,
945 ModeAlphabet[scheme][i]);
947 /* iterate through all of the macroblocks that contain 1 or more
949 for (i = 0; i < s->u_superblock_start; i++) {
951 for (j = 0; j < 4; j++) {
952 current_macroblock = s->superblock_macroblocks[i * 4 + j];
953 if ((current_macroblock == -1) ||
954 (s->macroblock_coding[current_macroblock] == MODE_COPY))
956 if (current_macroblock >= s->macroblock_count) {
957 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad macroblock number (%d >= %d)\n",
958 current_macroblock, s->macroblock_count);
962 /* mode 7 means get 3 bits for each coding mode */
964 coding_mode = get_bits(gb, 3);
966 coding_mode = ModeAlphabet[scheme]
967 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
969 s->macroblock_coding[current_macroblock] = coding_mode;
970 for (k = 0; k < 6; k++) {
972 s->macroblock_fragments[current_macroblock * 6 + k];
973 if (current_fragment == -1)
975 if (current_fragment >= s->fragment_count) {
976 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad fragment number (%d >= %d)\n",
977 current_fragment, s->fragment_count);
980 if (s->all_fragments[current_fragment].coding_method !=
982 s->all_fragments[current_fragment].coding_method =
986 debug_modes(" coding method for macroblock starting @ fragment %d = %d\n",
987 s->macroblock_fragments[current_macroblock * 6], coding_mode);
996 * This function unpacks all the motion vectors for the individual
997 * macroblocks from the bitstream.
999 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
1005 int last_motion_x = 0;
1006 int last_motion_y = 0;
1007 int prior_last_motion_x = 0;
1008 int prior_last_motion_y = 0;
1009 int current_macroblock;
1010 int current_fragment;
1012 debug_vp3(" vp3: unpacking motion vectors\n");
1015 debug_vp3(" keyframe-- there are no motion vectors\n");
1019 memset(motion_x, 0, 6 * sizeof(int));
1020 memset(motion_y, 0, 6 * sizeof(int));
1022 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
1023 coding_mode = get_bits(gb, 1);
1024 debug_vectors(" using %s scheme for unpacking motion vectors\n",
1025 (coding_mode == 0) ? "VLC" : "fixed-length");
1027 /* iterate through all of the macroblocks that contain 1 or more
1028 * coded fragments */
1029 for (i = 0; i < s->u_superblock_start; i++) {
1031 for (j = 0; j < 4; j++) {
1032 current_macroblock = s->superblock_macroblocks[i * 4 + j];
1033 if ((current_macroblock == -1) ||
1034 (s->macroblock_coding[current_macroblock] == MODE_COPY))
1036 if (current_macroblock >= s->macroblock_count) {
1037 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad macroblock number (%d >= %d)\n",
1038 current_macroblock, s->macroblock_count);
1042 current_fragment = s->macroblock_fragments[current_macroblock * 6];
1043 if (current_fragment >= s->fragment_count) {
1044 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d\n",
1045 current_fragment, s->fragment_count);
1048 switch (s->macroblock_coding[current_macroblock]) {
1050 case MODE_INTER_PLUS_MV:
1051 case MODE_GOLDEN_MV:
1052 /* all 6 fragments use the same motion vector */
1053 if (coding_mode == 0) {
1054 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1055 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1057 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
1058 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
1061 for (k = 1; k < 6; k++) {
1062 motion_x[k] = motion_x[0];
1063 motion_y[k] = motion_y[0];
1066 /* vector maintenance, only on MODE_INTER_PLUS_MV */
1067 if (s->macroblock_coding[current_macroblock] ==
1068 MODE_INTER_PLUS_MV) {
1069 prior_last_motion_x = last_motion_x;
1070 prior_last_motion_y = last_motion_y;
1071 last_motion_x = motion_x[0];
1072 last_motion_y = motion_y[0];
1076 case MODE_INTER_FOURMV:
1077 /* fetch 4 vectors from the bitstream, one for each
1078 * Y fragment, then average for the C fragment vectors */
1079 motion_x[4] = motion_y[4] = 0;
1080 for (k = 0; k < 4; k++) {
1081 if (coding_mode == 0) {
1082 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1083 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1085 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
1086 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
1088 motion_x[4] += motion_x[k];
1089 motion_y[4] += motion_y[k];
1093 motion_x[4]= RSHIFT(motion_x[4], 2);
1095 motion_y[4]= RSHIFT(motion_y[4], 2);
1097 /* vector maintenance; vector[3] is treated as the
1098 * last vector in this case */
1099 prior_last_motion_x = last_motion_x;
1100 prior_last_motion_y = last_motion_y;
1101 last_motion_x = motion_x[3];
1102 last_motion_y = motion_y[3];
1105 case MODE_INTER_LAST_MV:
1106 /* all 6 fragments use the last motion vector */
1107 motion_x[0] = last_motion_x;
1108 motion_y[0] = last_motion_y;
1109 for (k = 1; k < 6; k++) {
1110 motion_x[k] = motion_x[0];
1111 motion_y[k] = motion_y[0];
1114 /* no vector maintenance (last vector remains the
1118 case MODE_INTER_PRIOR_LAST:
1119 /* all 6 fragments use the motion vector prior to the
1120 * last motion vector */
1121 motion_x[0] = prior_last_motion_x;
1122 motion_y[0] = prior_last_motion_y;
1123 for (k = 1; k < 6; k++) {
1124 motion_x[k] = motion_x[0];
1125 motion_y[k] = motion_y[0];
1128 /* vector maintenance */
1129 prior_last_motion_x = last_motion_x;
1130 prior_last_motion_y = last_motion_y;
1131 last_motion_x = motion_x[0];
1132 last_motion_y = motion_y[0];
1136 /* covers intra, inter without MV, golden without MV */
1137 memset(motion_x, 0, 6 * sizeof(int));
1138 memset(motion_y, 0, 6 * sizeof(int));
1140 /* no vector maintenance */
1144 /* assign the motion vectors to the correct fragments */
1145 debug_vectors(" vectors for macroblock starting @ fragment %d (coding method %d):\n",
1147 s->macroblock_coding[current_macroblock]);
1148 for (k = 0; k < 6; k++) {
1150 s->macroblock_fragments[current_macroblock * 6 + k];
1151 if (current_fragment == -1)
1153 if (current_fragment >= s->fragment_count) {
1154 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d)\n",
1155 current_fragment, s->fragment_count);
1158 s->all_fragments[current_fragment].motion_x = motion_x[k];
1159 s->all_fragments[current_fragment].motion_y = motion_y[k];
1160 debug_vectors(" vector %d: fragment %d = (%d, %d)\n",
1161 k, current_fragment, motion_x[k], motion_y[k]);
1171 * This function is called by unpack_dct_coeffs() to extract the VLCs from
1172 * the bitstream. The VLCs encode tokens which are used to unpack DCT
1173 * data. This function unpacks all the VLCs for either the Y plane or both
1174 * C planes, and is called for DC coefficients or different AC coefficient
1175 * levels (since different coefficient types require different VLC tables.
1177 * This function returns a residual eob run. E.g, if a particular token gave
1178 * instructions to EOB the next 5 fragments and there were only 2 fragments
1179 * left in the current fragment range, 3 would be returned so that it could
1180 * be passed into the next call to this same function.
1182 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
1183 VLC *table, int coeff_index,
1184 int first_fragment, int last_fragment,
1191 Vp3Fragment *fragment;
1192 uint8_t *perm= s->scantable.permutated;
1195 if ((first_fragment >= s->fragment_count) ||
1196 (last_fragment >= s->fragment_count)) {
1198 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vlcs(): bad fragment number (%d -> %d ?)\n",
1199 first_fragment, last_fragment);
1203 for (i = first_fragment; i <= last_fragment; i++) {
1205 fragment = &s->all_fragments[s->coded_fragment_list[i]];
1206 if (fragment->coeff_count > coeff_index)
1210 /* decode a VLC into a token */
1211 token = get_vlc2(gb, table->table, 5, 3);
1212 debug_vlc(" token = %2d, ", token);
1213 /* use the token to get a zero run, a coefficient, and an eob run */
1215 eob_run = eob_run_base[token];
1216 if (eob_run_get_bits[token])
1217 eob_run += get_bits(gb, eob_run_get_bits[token]);
1218 coeff = zero_run = 0;
1220 bits_to_get = coeff_get_bits[token];
1222 coeff = coeff_tables[token][0];
1224 coeff = coeff_tables[token][get_bits(gb, bits_to_get)];
1226 zero_run = zero_run_base[token];
1227 if (zero_run_get_bits[token])
1228 zero_run += get_bits(gb, zero_run_get_bits[token]);
1233 fragment->coeff_count += zero_run;
1234 if (fragment->coeff_count < 64){
1235 fragment->next_coeff->coeff= coeff;
1236 fragment->next_coeff->index= perm[fragment->coeff_count++]; //FIXME perm here already?
1237 fragment->next_coeff->next= s->next_coeff;
1238 s->next_coeff->next=NULL;
1239 fragment->next_coeff= s->next_coeff++;
1241 debug_vlc(" fragment %d coeff = %d\n",
1242 s->coded_fragment_list[i], fragment->next_coeff[coeff_index]);
1244 fragment->coeff_count |= 128;
1245 debug_vlc(" fragment %d eob with %d coefficients\n",
1246 s->coded_fragment_list[i], fragment->coeff_count&127);
1255 * This function unpacks all of the DCT coefficient data from the
1258 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1265 int residual_eob_run = 0;
1267 /* fetch the DC table indices */
1268 dc_y_table = get_bits(gb, 4);
1269 dc_c_table = get_bits(gb, 4);
1271 /* unpack the Y plane DC coefficients */
1272 debug_vp3(" vp3: unpacking Y plane DC coefficients using table %d\n",
1274 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1275 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1277 /* unpack the C plane DC coefficients */
1278 debug_vp3(" vp3: unpacking C plane DC coefficients using table %d\n",
1280 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1281 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1283 /* fetch the AC table indices */
1284 ac_y_table = get_bits(gb, 4);
1285 ac_c_table = get_bits(gb, 4);
1287 /* unpack the group 1 AC coefficients (coeffs 1-5) */
1288 for (i = 1; i <= 5; i++) {
1290 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1292 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_y_table], i,
1293 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1295 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1297 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_c_table], i,
1298 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1301 /* unpack the group 2 AC coefficients (coeffs 6-14) */
1302 for (i = 6; i <= 14; i++) {
1304 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1306 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_y_table], i,
1307 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1309 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1311 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_c_table], i,
1312 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1315 /* unpack the group 3 AC coefficients (coeffs 15-27) */
1316 for (i = 15; i <= 27; i++) {
1318 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1320 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_y_table], i,
1321 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1323 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1325 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_c_table], i,
1326 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1329 /* unpack the group 4 AC coefficients (coeffs 28-63) */
1330 for (i = 28; i <= 63; i++) {
1332 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1334 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_y_table], i,
1335 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1337 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1339 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_c_table], i,
1340 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1347 * This function reverses the DC prediction for each coded fragment in
1348 * the frame. Much of this function is adapted directly from the original
1351 #define COMPATIBLE_FRAME(x) \
1352 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1353 #define FRAME_CODED(x) (s->all_fragments[x].coding_method != MODE_COPY)
1354 #define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
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 (ABS(predicted_dc - vu) > 128)
1602 else if (ABS(predicted_dc - vl) > 128)
1604 else if (ABS(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 int16_t *dequantizer;
1648 DECLARE_ALIGNED_16(DCTELEM, block[64]);
1649 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1650 int motion_halfpel_index;
1651 uint8_t *motion_source;
1653 int current_macroblock_entry = slice * s->macroblock_width * 6;
1655 if (slice >= s->macroblock_height)
1658 for (plane = 0; plane < 3; plane++) {
1659 uint8_t *output_plane = s->current_frame.data [plane];
1660 uint8_t * last_plane = s-> last_frame.data [plane];
1661 uint8_t *golden_plane = s-> golden_frame.data [plane];
1662 int stride = s->current_frame.linesize[plane];
1663 int plane_width = s->width >> !!plane;
1664 int plane_height = s->height >> !!plane;
1665 int y = slice * FRAGMENT_PIXELS << !plane ;
1666 int slice_height = y + (FRAGMENT_PIXELS << !plane);
1667 int i = s->macroblock_fragments[current_macroblock_entry + plane + 3*!!plane];
1669 if (!s->flipped_image) stride = -stride;
1672 if(ABS(stride) > 2048)
1673 return; //various tables are fixed size
1675 /* for each fragment row in the slice (both of them)... */
1676 for (; y < slice_height; y += 8) {
1678 /* for each fragment in a row... */
1679 for (x = 0; x < plane_width; x += 8, i++) {
1681 if ((i < 0) || (i >= s->fragment_count)) {
1682 av_log(s->avctx, AV_LOG_ERROR, " vp3:render_slice(): bad fragment number (%d)\n", i);
1686 /* transform if this block was coded */
1687 if ((s->all_fragments[i].coding_method != MODE_COPY) &&
1688 !((s->avctx->flags & CODEC_FLAG_GRAY) && plane)) {
1690 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1691 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1692 motion_source= golden_plane;
1694 motion_source= last_plane;
1696 motion_source += s->all_fragments[i].first_pixel;
1697 motion_halfpel_index = 0;
1699 /* sort out the motion vector if this fragment is coded
1700 * using a motion vector method */
1701 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1702 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1704 motion_x = s->all_fragments[i].motion_x;
1705 motion_y = s->all_fragments[i].motion_y;
1707 motion_x= (motion_x>>1) | (motion_x&1);
1708 motion_y= (motion_y>>1) | (motion_y&1);
1711 src_x= (motion_x>>1) + x;
1712 src_y= (motion_y>>1) + y;
1713 if ((motion_x == 127) || (motion_y == 127))
1714 av_log(s->avctx, AV_LOG_ERROR, " help! got invalid motion vector! (%X, %X)\n", motion_x, motion_y);
1716 motion_halfpel_index = motion_x & 0x01;
1717 motion_source += (motion_x >> 1);
1719 motion_halfpel_index |= (motion_y & 0x01) << 1;
1720 motion_source += ((motion_y >> 1) * stride);
1722 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1723 uint8_t *temp= s->edge_emu_buffer;
1724 if(stride<0) temp -= 9*stride;
1725 else temp += 9*stride;
1727 ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1728 motion_source= temp;
1733 /* first, take care of copying a block from either the
1734 * previous or the golden frame */
1735 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1736 /* Note, it is possible to implement all MC cases with
1737 put_no_rnd_pixels_l2 which would look more like the
1738 VP3 source but this would be slower as
1739 put_no_rnd_pixels_tab is better optimzed */
1740 if(motion_halfpel_index != 3){
1741 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1742 output_plane + s->all_fragments[i].first_pixel,
1743 motion_source, stride, 8);
1745 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1746 s->dsp.put_no_rnd_pixels_l2[1](
1747 output_plane + s->all_fragments[i].first_pixel,
1749 motion_source + stride + 1 + d,
1752 dequantizer = s->qmat[1][plane];
1754 dequantizer = s->qmat[0][plane];
1757 /* dequantize the DCT coefficients */
1758 debug_idct("fragment %d, coding mode %d, DC = %d, dequant = %d:\n",
1759 i, s->all_fragments[i].coding_method,
1760 DC_COEFF(i), dequantizer[0]);
1762 if(s->avctx->idct_algo==FF_IDCT_VP3){
1763 Coeff *coeff= s->coeffs + i;
1764 memset(block, 0, sizeof(block));
1766 block[coeff->index]= coeff->coeff * dequantizer[coeff->index];
1770 Coeff *coeff= s->coeffs + i;
1771 memset(block, 0, sizeof(block));
1773 block[coeff->index]= (coeff->coeff * dequantizer[coeff->index] + 2)>>2;
1778 /* invert DCT and place (or add) in final output */
1780 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1781 if(s->avctx->idct_algo!=FF_IDCT_VP3)
1784 output_plane + s->all_fragments[i].first_pixel,
1789 output_plane + s->all_fragments[i].first_pixel,
1794 debug_idct("block after idct_%s():\n",
1795 (s->all_fragments[i].coding_method == MODE_INTRA)?
1797 for (m = 0; m < 8; m++) {
1798 for (n = 0; n < 8; n++) {
1799 debug_idct(" %3d", *(output_plane +
1800 s->all_fragments[i].first_pixel + (m * stride + n)));
1808 /* copy directly from the previous frame */
1809 s->dsp.put_pixels_tab[1][0](
1810 output_plane + s->all_fragments[i].first_pixel,
1811 last_plane + s->all_fragments[i].first_pixel,
1816 /* perform the left edge filter if:
1817 * - the fragment is not on the left column
1818 * - the fragment is coded in this frame
1819 * - the fragment is not coded in this frame but the left
1820 * fragment is coded in this frame (this is done instead
1821 * of a right edge filter when rendering the left fragment
1822 * since this fragment is not available yet) */
1824 ((s->all_fragments[i].coding_method != MODE_COPY) ||
1825 ((s->all_fragments[i].coding_method == MODE_COPY) &&
1826 (s->all_fragments[i - 1].coding_method != MODE_COPY)) )) {
1828 output_plane + s->all_fragments[i].first_pixel + 7*stride,
1829 -stride, s->bounding_values_array + 127);
1832 /* perform the top edge filter if:
1833 * - the fragment is not on the top row
1834 * - the fragment is coded in this frame
1835 * - the fragment is not coded in this frame but the above
1836 * fragment is coded in this frame (this is done instead
1837 * of a bottom edge filter when rendering the above
1838 * fragment since this fragment is not available yet) */
1840 ((s->all_fragments[i].coding_method != MODE_COPY) ||
1841 ((s->all_fragments[i].coding_method == MODE_COPY) &&
1842 (s->all_fragments[i - fragment_width].coding_method != MODE_COPY)) )) {
1844 output_plane + s->all_fragments[i].first_pixel - stride,
1845 -stride, s->bounding_values_array + 127);
1852 /* this looks like a good place for slice dispatch... */
1854 * if (slice == s->macroblock_height - 1)
1855 * dispatch (both last slice & 2nd-to-last slice);
1856 * else if (slice > 0)
1857 * dispatch (slice - 1);
1863 static void horizontal_filter(unsigned char *first_pixel, int stride,
1864 int *bounding_values)
1869 for (end= first_pixel + 8*stride; first_pixel != end; first_pixel += stride) {
1871 (first_pixel[-2] - first_pixel[ 1])
1872 +3*(first_pixel[ 0] - first_pixel[-1]);
1873 filter_value = bounding_values[(filter_value + 4) >> 3];
1874 first_pixel[-1] = clip_uint8(first_pixel[-1] + filter_value);
1875 first_pixel[ 0] = clip_uint8(first_pixel[ 0] - filter_value);
1879 static void vertical_filter(unsigned char *first_pixel, int stride,
1880 int *bounding_values)
1884 const int nstride= -stride;
1886 for (end= first_pixel + 8; first_pixel < end; first_pixel++) {
1888 (first_pixel[2 * nstride] - first_pixel[ stride])
1889 +3*(first_pixel[0 ] - first_pixel[nstride]);
1890 filter_value = bounding_values[(filter_value + 4) >> 3];
1891 first_pixel[nstride] = clip_uint8(first_pixel[nstride] + filter_value);
1892 first_pixel[0] = clip_uint8(first_pixel[0] - filter_value);
1896 static void apply_loop_filter(Vp3DecodeContext *s)
1900 int *bounding_values= s->bounding_values_array+127;
1903 int bounding_values_array[256];
1906 /* find the right loop limit value */
1907 for (x = 63; x >= 0; x--) {
1908 if (vp31_ac_scale_factor[x] >= s->quality_index)
1911 filter_limit = vp31_filter_limit_values[s->quality_index];
1913 /* set up the bounding values */
1914 memset(bounding_values_array, 0, 256 * sizeof(int));
1915 for (x = 0; x < filter_limit; x++) {
1916 bounding_values[-x - filter_limit] = -filter_limit + x;
1917 bounding_values[-x] = -x;
1918 bounding_values[x] = x;
1919 bounding_values[x + filter_limit] = filter_limit - x;
1923 for (plane = 0; plane < 3; plane++) {
1924 int width = s->fragment_width >> !!plane;
1925 int height = s->fragment_height >> !!plane;
1926 int fragment = s->fragment_start [plane];
1927 int stride = s->current_frame.linesize[plane];
1928 uint8_t *plane_data = s->current_frame.data [plane];
1929 if (!s->flipped_image) stride = -stride;
1931 for (y = 0; y < height; y++) {
1933 for (x = 0; x < width; x++) {
1935 /* do not perform left edge filter for left columns frags */
1937 (s->all_fragments[fragment].coding_method != MODE_COPY)) {
1939 plane_data + s->all_fragments[fragment].first_pixel,
1940 stride, bounding_values);
1943 /* do not perform top edge filter for top row fragments */
1945 (s->all_fragments[fragment].coding_method != MODE_COPY)) {
1947 plane_data + s->all_fragments[fragment].first_pixel,
1948 stride, bounding_values);
1951 /* do not perform right edge filter for right column
1952 * fragments or if right fragment neighbor is also coded
1953 * in this frame (it will be filtered in next iteration) */
1954 if ((x < width - 1) &&
1955 (s->all_fragments[fragment].coding_method != MODE_COPY) &&
1956 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1958 plane_data + s->all_fragments[fragment + 1].first_pixel,
1959 stride, bounding_values);
1962 /* do not perform bottom edge filter for bottom row
1963 * fragments or if bottom fragment neighbor is also coded
1964 * in this frame (it will be filtered in the next row) */
1965 if ((y < height - 1) &&
1966 (s->all_fragments[fragment].coding_method != MODE_COPY) &&
1967 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1969 plane_data + s->all_fragments[fragment + width].first_pixel,
1970 stride, bounding_values);
1974 STOP_TIMER("loop filter")
1981 * This function computes the first pixel addresses for each fragment.
1982 * This function needs to be invoked after the first frame is allocated
1983 * so that it has access to the plane strides.
1985 static void vp3_calculate_pixel_addresses(Vp3DecodeContext *s)
1990 /* figure out the first pixel addresses for each of the fragments */
1993 for (y = s->fragment_height; y > 0; y--) {
1994 for (x = 0; x < s->fragment_width; x++) {
1995 s->all_fragments[i++].first_pixel =
1996 s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
1997 s->golden_frame.linesize[0] +
1998 x * FRAGMENT_PIXELS;
1999 debug_init(" fragment %d, first pixel @ %d\n",
2000 i-1, s->all_fragments[i-1].first_pixel);
2005 i = s->fragment_start[1];
2006 for (y = s->fragment_height / 2; y > 0; y--) {
2007 for (x = 0; x < s->fragment_width / 2; x++) {
2008 s->all_fragments[i++].first_pixel =
2009 s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
2010 s->golden_frame.linesize[1] +
2011 x * FRAGMENT_PIXELS;
2012 debug_init(" fragment %d, first pixel @ %d\n",
2013 i-1, s->all_fragments[i-1].first_pixel);
2018 i = s->fragment_start[2];
2019 for (y = s->fragment_height / 2; y > 0; y--) {
2020 for (x = 0; x < s->fragment_width / 2; x++) {
2021 s->all_fragments[i++].first_pixel =
2022 s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
2023 s->golden_frame.linesize[2] +
2024 x * FRAGMENT_PIXELS;
2025 debug_init(" fragment %d, first pixel @ %d\n",
2026 i-1, s->all_fragments[i-1].first_pixel);
2031 /* FIXME: this should be merged with the above! */
2032 static void theora_calculate_pixel_addresses(Vp3DecodeContext *s)
2037 /* figure out the first pixel addresses for each of the fragments */
2040 for (y = 1; y <= s->fragment_height; y++) {
2041 for (x = 0; x < s->fragment_width; x++) {
2042 s->all_fragments[i++].first_pixel =
2043 s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
2044 s->golden_frame.linesize[0] +
2045 x * FRAGMENT_PIXELS;
2046 debug_init(" fragment %d, first pixel @ %d\n",
2047 i-1, s->all_fragments[i-1].first_pixel);
2052 i = s->fragment_start[1];
2053 for (y = 1; y <= s->fragment_height / 2; y++) {
2054 for (x = 0; x < s->fragment_width / 2; x++) {
2055 s->all_fragments[i++].first_pixel =
2056 s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
2057 s->golden_frame.linesize[1] +
2058 x * FRAGMENT_PIXELS;
2059 debug_init(" fragment %d, first pixel @ %d\n",
2060 i-1, s->all_fragments[i-1].first_pixel);
2065 i = s->fragment_start[2];
2066 for (y = 1; y <= s->fragment_height / 2; y++) {
2067 for (x = 0; x < s->fragment_width / 2; x++) {
2068 s->all_fragments[i++].first_pixel =
2069 s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
2070 s->golden_frame.linesize[2] +
2071 x * FRAGMENT_PIXELS;
2072 debug_init(" fragment %d, first pixel @ %d\n",
2073 i-1, s->all_fragments[i-1].first_pixel);
2079 * This is the ffmpeg/libavcodec API init function.
2081 static int vp3_decode_init(AVCodecContext *avctx)
2083 Vp3DecodeContext *s = avctx->priv_data;
2084 int i, inter, plane;
2087 int y_superblock_count;
2088 int c_superblock_count;
2090 if (avctx->codec_tag == MKTAG('V','P','3','0'))
2096 s->width = (avctx->width + 15) & 0xFFFFFFF0;
2097 s->height = (avctx->height + 15) & 0xFFFFFFF0;
2098 avctx->pix_fmt = PIX_FMT_YUV420P;
2099 avctx->has_b_frames = 0;
2100 if(avctx->idct_algo==FF_IDCT_AUTO)
2101 avctx->idct_algo=FF_IDCT_VP3;
2102 dsputil_init(&s->dsp, avctx);
2104 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
2106 /* initialize to an impossible value which will force a recalculation
2107 * in the first frame decode */
2108 s->quality_index = -1;
2110 s->y_superblock_width = (s->width + 31) / 32;
2111 s->y_superblock_height = (s->height + 31) / 32;
2112 y_superblock_count = s->y_superblock_width * s->y_superblock_height;
2114 /* work out the dimensions for the C planes */
2115 c_width = s->width / 2;
2116 c_height = s->height / 2;
2117 s->c_superblock_width = (c_width + 31) / 32;
2118 s->c_superblock_height = (c_height + 31) / 32;
2119 c_superblock_count = s->c_superblock_width * s->c_superblock_height;
2121 s->superblock_count = y_superblock_count + (c_superblock_count * 2);
2122 s->u_superblock_start = y_superblock_count;
2123 s->v_superblock_start = s->u_superblock_start + c_superblock_count;
2124 s->superblock_coding = av_malloc(s->superblock_count);
2126 s->macroblock_width = (s->width + 15) / 16;
2127 s->macroblock_height = (s->height + 15) / 16;
2128 s->macroblock_count = s->macroblock_width * s->macroblock_height;
2130 s->fragment_width = s->width / FRAGMENT_PIXELS;
2131 s->fragment_height = s->height / FRAGMENT_PIXELS;
2133 /* fragment count covers all 8x8 blocks for all 3 planes */
2134 s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
2135 s->fragment_start[1] = s->fragment_width * s->fragment_height;
2136 s->fragment_start[2] = s->fragment_width * s->fragment_height * 5 / 4;
2138 debug_init(" Y plane: %d x %d\n", s->width, s->height);
2139 debug_init(" C plane: %d x %d\n", c_width, c_height);
2140 debug_init(" Y superblocks: %d x %d, %d total\n",
2141 s->y_superblock_width, s->y_superblock_height, y_superblock_count);
2142 debug_init(" C superblocks: %d x %d, %d total\n",
2143 s->c_superblock_width, s->c_superblock_height, c_superblock_count);
2144 debug_init(" total superblocks = %d, U starts @ %d, V starts @ %d\n",
2145 s->superblock_count, s->u_superblock_start, s->v_superblock_start);
2146 debug_init(" macroblocks: %d x %d, %d total\n",
2147 s->macroblock_width, s->macroblock_height, s->macroblock_count);
2148 debug_init(" %d fragments, %d x %d, u starts @ %d, v starts @ %d\n",
2152 s->fragment_start[1],
2153 s->fragment_start[2]);
2155 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
2156 s->coeffs = av_malloc(s->fragment_count * sizeof(Coeff) * 65);
2157 s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int));
2158 s->pixel_addresses_inited = 0;
2160 if (!s->theora_tables)
2162 for (i = 0; i < 64; i++)
2163 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
2164 for (i = 0; i < 64; i++)
2165 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
2166 for (i = 0; i < 64; i++)
2167 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
2168 for (i = 0; i < 64; i++)
2169 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
2170 for (i = 0; i < 64; i++)
2171 s->base_matrix[2][i] = vp31_inter_dequant[i];
2172 for (i = 0; i < 64; i++)
2173 s->filter_limit_values[i] = vp31_filter_limit_values[i];
2175 for(inter=0; inter<2; inter++){
2176 for(plane=0; plane<3; plane++){
2177 s->qr_count[inter][plane]= 1;
2178 s->qr_size [inter][plane][0]= 63;
2179 s->qr_base [inter][plane][0]=
2180 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
2184 /* init VLC tables */
2185 for (i = 0; i < 16; i++) {
2188 init_vlc(&s->dc_vlc[i], 5, 32,
2189 &dc_bias[i][0][1], 4, 2,
2190 &dc_bias[i][0][0], 4, 2, 0);
2192 /* group 1 AC histograms */
2193 init_vlc(&s->ac_vlc_1[i], 5, 32,
2194 &ac_bias_0[i][0][1], 4, 2,
2195 &ac_bias_0[i][0][0], 4, 2, 0);
2197 /* group 2 AC histograms */
2198 init_vlc(&s->ac_vlc_2[i], 5, 32,
2199 &ac_bias_1[i][0][1], 4, 2,
2200 &ac_bias_1[i][0][0], 4, 2, 0);
2202 /* group 3 AC histograms */
2203 init_vlc(&s->ac_vlc_3[i], 5, 32,
2204 &ac_bias_2[i][0][1], 4, 2,
2205 &ac_bias_2[i][0][0], 4, 2, 0);
2207 /* group 4 AC histograms */
2208 init_vlc(&s->ac_vlc_4[i], 5, 32,
2209 &ac_bias_3[i][0][1], 4, 2,
2210 &ac_bias_3[i][0][0], 4, 2, 0);
2213 for (i = 0; i < 16; i++) {
2216 init_vlc(&s->dc_vlc[i], 5, 32,
2217 &s->huffman_table[i][0][1], 4, 2,
2218 &s->huffman_table[i][0][0], 4, 2, 0);
2220 /* group 1 AC histograms */
2221 init_vlc(&s->ac_vlc_1[i], 5, 32,
2222 &s->huffman_table[i+16][0][1], 4, 2,
2223 &s->huffman_table[i+16][0][0], 4, 2, 0);
2225 /* group 2 AC histograms */
2226 init_vlc(&s->ac_vlc_2[i], 5, 32,
2227 &s->huffman_table[i+16*2][0][1], 4, 2,
2228 &s->huffman_table[i+16*2][0][0], 4, 2, 0);
2230 /* group 3 AC histograms */
2231 init_vlc(&s->ac_vlc_3[i], 5, 32,
2232 &s->huffman_table[i+16*3][0][1], 4, 2,
2233 &s->huffman_table[i+16*3][0][0], 4, 2, 0);
2235 /* group 4 AC histograms */
2236 init_vlc(&s->ac_vlc_4[i], 5, 32,
2237 &s->huffman_table[i+16*4][0][1], 4, 2,
2238 &s->huffman_table[i+16*4][0][0], 4, 2, 0);
2242 init_vlc(&s->superblock_run_length_vlc, 6, 34,
2243 &superblock_run_length_vlc_table[0][1], 4, 2,
2244 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
2246 init_vlc(&s->fragment_run_length_vlc, 5, 30,
2247 &fragment_run_length_vlc_table[0][1], 4, 2,
2248 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
2250 init_vlc(&s->mode_code_vlc, 3, 8,
2251 &mode_code_vlc_table[0][1], 2, 1,
2252 &mode_code_vlc_table[0][0], 2, 1, 0);
2254 init_vlc(&s->motion_vector_vlc, 6, 63,
2255 &motion_vector_vlc_table[0][1], 2, 1,
2256 &motion_vector_vlc_table[0][0], 2, 1, 0);
2258 /* work out the block mapping tables */
2259 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
2260 s->superblock_macroblocks = av_malloc(s->superblock_count * 4 * sizeof(int));
2261 s->macroblock_fragments = av_malloc(s->macroblock_count * 6 * sizeof(int));
2262 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
2263 init_block_mapping(s);
2265 for (i = 0; i < 3; i++) {
2266 s->current_frame.data[i] = NULL;
2267 s->last_frame.data[i] = NULL;
2268 s->golden_frame.data[i] = NULL;
2275 * This is the ffmpeg/libavcodec API frame decode function.
2277 static int vp3_decode_frame(AVCodecContext *avctx,
2278 void *data, int *data_size,
2279 uint8_t *buf, int buf_size)
2281 Vp3DecodeContext *s = avctx->priv_data;
2283 static int counter = 0;
2286 init_get_bits(&gb, buf, buf_size * 8);
2288 if (s->theora && get_bits1(&gb))
2291 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
2294 int ptype = get_bits(&gb, 7);
2296 skip_bits(&gb, 6*8); /* "theora" */
2301 theora_decode_comments(avctx, &gb);
2304 theora_decode_tables(avctx, &gb);
2305 init_dequantizer(s);
2308 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype);
2314 s->keyframe = !get_bits1(&gb);
2317 s->last_quality_index = s->quality_index;
2321 s->qis[s->nqis++]= get_bits(&gb, 6);
2322 } while(s->theora >= 0x030200 && s->nqis<3 && get_bits1(&gb));
2324 s->quality_index= s->qis[0];
2326 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
2327 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
2328 s->keyframe?"key":"", counter, s->quality_index);
2331 if (s->quality_index != s->last_quality_index) {
2332 init_dequantizer(s);
2333 init_loop_filter(s);
2339 skip_bits(&gb, 4); /* width code */
2340 skip_bits(&gb, 4); /* height code */
2343 s->version = get_bits(&gb, 5);
2345 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
2348 if (s->version || s->theora)
2351 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
2352 skip_bits(&gb, 2); /* reserved? */
2355 if (s->last_frame.data[0] == s->golden_frame.data[0]) {
2356 if (s->golden_frame.data[0])
2357 avctx->release_buffer(avctx, &s->golden_frame);
2358 s->last_frame= s->golden_frame; /* ensure that we catch any access to this released frame */
2360 if (s->golden_frame.data[0])
2361 avctx->release_buffer(avctx, &s->golden_frame);
2362 if (s->last_frame.data[0])
2363 avctx->release_buffer(avctx, &s->last_frame);
2366 s->golden_frame.reference = 3;
2367 if(avctx->get_buffer(avctx, &s->golden_frame) < 0) {
2368 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
2372 /* golden frame is also the current frame */
2373 s->current_frame= s->golden_frame;
2375 /* time to figure out pixel addresses? */
2376 if (!s->pixel_addresses_inited)
2378 if (!s->flipped_image)
2379 vp3_calculate_pixel_addresses(s);
2381 theora_calculate_pixel_addresses(s);
2384 /* allocate a new current frame */
2385 s->current_frame.reference = 3;
2386 if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
2387 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
2392 s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
2393 s->current_frame.qstride= 0;
2397 STOP_TIMER("init_frame")}
2402 memcpy(s->current_frame.data[0], s->golden_frame.data[0],
2403 s->current_frame.linesize[0] * s->height);
2404 memcpy(s->current_frame.data[1], s->golden_frame.data[1],
2405 s->current_frame.linesize[1] * s->height / 2);
2406 memcpy(s->current_frame.data[2], s->golden_frame.data[2],
2407 s->current_frame.linesize[2] * s->height / 2);
2413 if (unpack_superblocks(s, &gb)){
2414 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
2417 STOP_TIMER("unpack_superblocks")}
2419 if (unpack_modes(s, &gb)){
2420 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
2423 STOP_TIMER("unpack_modes")}
2425 if (unpack_vectors(s, &gb)){
2426 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2429 STOP_TIMER("unpack_vectors")}
2431 if (unpack_dct_coeffs(s, &gb)){
2432 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
2435 STOP_TIMER("unpack_dct_coeffs")}
2438 reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
2439 if ((avctx->flags & CODEC_FLAG_GRAY) == 0) {
2440 reverse_dc_prediction(s, s->fragment_start[1],
2441 s->fragment_width / 2, s->fragment_height / 2);
2442 reverse_dc_prediction(s, s->fragment_start[2],
2443 s->fragment_width / 2, s->fragment_height / 2);
2445 STOP_TIMER("reverse_dc_prediction")}
2448 for (i = 0; i < s->macroblock_height; i++)
2450 STOP_TIMER("render_fragments")}
2453 apply_loop_filter(s);
2454 STOP_TIMER("apply_loop_filter")}
2459 *data_size=sizeof(AVFrame);
2460 *(AVFrame*)data= s->current_frame;
2462 /* release the last frame, if it is allocated and if it is not the
2464 if ((s->last_frame.data[0]) &&
2465 (s->last_frame.data[0] != s->golden_frame.data[0]))
2466 avctx->release_buffer(avctx, &s->last_frame);
2468 /* shuffle frames (last = current) */
2469 s->last_frame= s->current_frame;
2470 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
2476 * This is the ffmpeg/libavcodec API module cleanup function.
2478 static int vp3_decode_end(AVCodecContext *avctx)
2480 Vp3DecodeContext *s = avctx->priv_data;
2482 av_free(s->all_fragments);
2484 av_free(s->coded_fragment_list);
2485 av_free(s->superblock_fragments);
2486 av_free(s->superblock_macroblocks);
2487 av_free(s->macroblock_fragments);
2488 av_free(s->macroblock_coding);
2490 /* release all frames */
2491 if (s->golden_frame.data[0] && s->golden_frame.data[0] != s->last_frame.data[0])
2492 avctx->release_buffer(avctx, &s->golden_frame);
2493 if (s->last_frame.data[0])
2494 avctx->release_buffer(avctx, &s->last_frame);
2495 /* no need to release the current_frame since it will always be pointing
2496 * to the same frame as either the golden or last frame */
2501 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2503 Vp3DecodeContext *s = avctx->priv_data;
2505 if (get_bits(gb, 1)) {
2507 if (s->entries >= 32) { /* overflow */
2508 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2511 token = get_bits(gb, 5);
2512 //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);
2513 s->huffman_table[s->hti][token][0] = s->hbits;
2514 s->huffman_table[s->hti][token][1] = s->huff_code_size;
2518 if (s->huff_code_size >= 32) {/* overflow */
2519 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2522 s->huff_code_size++;
2524 read_huffman_tree(avctx, gb);
2526 read_huffman_tree(avctx, gb);
2528 s->huff_code_size--;
2533 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2535 Vp3DecodeContext *s = avctx->priv_data;
2537 s->theora = get_bits_long(gb, 24);
2538 av_log(avctx, AV_LOG_INFO, "Theora bitstream version %X\n", s->theora);
2540 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2541 /* but previous versions have the image flipped relative to vp3 */
2542 if (s->theora < 0x030200)
2544 s->flipped_image = 1;
2545 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2548 s->width = get_bits(gb, 16) << 4;
2549 s->height = get_bits(gb, 16) << 4;
2551 if(avcodec_check_dimensions(avctx, s->width, s->height)){
2552 av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2553 s->width= s->height= 0;
2557 if (s->theora >= 0x030400)
2559 skip_bits(gb, 32); /* total number of superblocks in a frame */
2560 // fixme, the next field is 36bits long
2561 skip_bits(gb, 32); /* total number of blocks in a frame */
2562 skip_bits(gb, 4); /* total number of blocks in a frame */
2563 skip_bits(gb, 32); /* total number of macroblocks in a frame */
2565 skip_bits(gb, 24); /* frame width */
2566 skip_bits(gb, 24); /* frame height */
2570 skip_bits(gb, 24); /* frame width */
2571 skip_bits(gb, 24); /* frame height */
2574 if (s->theora >= 0x030200) {
2575 skip_bits(gb, 8); /* offset x */
2576 skip_bits(gb, 8); /* offset y */
2579 skip_bits(gb, 32); /* fps numerator */
2580 skip_bits(gb, 32); /* fps denumerator */
2581 skip_bits(gb, 24); /* aspect numerator */
2582 skip_bits(gb, 24); /* aspect denumerator */
2584 if (s->theora < 0x030200)
2585 skip_bits(gb, 5); /* keyframe frequency force */
2586 skip_bits(gb, 8); /* colorspace */
2587 if (s->theora >= 0x030400)
2588 skip_bits(gb, 2); /* pixel format: 420,res,422,444 */
2589 skip_bits(gb, 24); /* bitrate */
2591 skip_bits(gb, 6); /* quality hint */
2593 if (s->theora >= 0x030200)
2595 skip_bits(gb, 5); /* keyframe frequency force */
2597 if (s->theora < 0x030400)
2598 skip_bits(gb, 5); /* spare bits */
2601 // align_get_bits(gb);
2603 avctx->width = s->width;
2604 avctx->height = s->height;
2609 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2611 Vp3DecodeContext *s = avctx->priv_data;
2612 int i, n, matrices, inter, plane;
2614 if (s->theora >= 0x030200) {
2615 n = get_bits(gb, 3);
2616 /* loop filter limit values table */
2617 for (i = 0; i < 64; i++)
2618 s->filter_limit_values[i] = get_bits(gb, n);
2621 if (s->theora >= 0x030200)
2622 n = get_bits(gb, 4) + 1;
2625 /* quality threshold table */
2626 for (i = 0; i < 64; i++)
2627 s->coded_ac_scale_factor[i] = get_bits(gb, n);
2629 if (s->theora >= 0x030200)
2630 n = get_bits(gb, 4) + 1;
2633 /* dc scale factor table */
2634 for (i = 0; i < 64; i++)
2635 s->coded_dc_scale_factor[i] = get_bits(gb, n);
2637 if (s->theora >= 0x030200)
2638 matrices = get_bits(gb, 9) + 1;
2643 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2647 for(n=0; n<matrices; n++){
2648 for (i = 0; i < 64; i++)
2649 s->base_matrix[n][i]= get_bits(gb, 8);
2652 for (inter = 0; inter <= 1; inter++) {
2653 for (plane = 0; plane <= 2; plane++) {
2655 if (inter || plane > 0)
2656 newqr = get_bits(gb, 1);
2659 if(inter && get_bits(gb, 1)){
2663 qtj= (3*inter + plane - 1) / 3;
2664 plj= (plane + 2) % 3;
2666 s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2667 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2668 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2674 i= get_bits(gb, av_log2(matrices-1)+1);
2676 av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2679 s->qr_base[inter][plane][qri]= i;
2682 i = get_bits(gb, av_log2(63-qi)+1) + 1;
2683 s->qr_size[inter][plane][qri++]= i;
2688 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2691 s->qr_count[inter][plane]= qri;
2696 /* Huffman tables */
2697 for (s->hti = 0; s->hti < 80; s->hti++) {
2699 s->huff_code_size = 1;
2700 if (!get_bits(gb, 1)) {
2702 read_huffman_tree(avctx, gb);
2704 read_huffman_tree(avctx, gb);
2708 s->theora_tables = 1;
2713 static int theora_decode_init(AVCodecContext *avctx)
2715 Vp3DecodeContext *s = avctx->priv_data;
2718 uint8_t *p= avctx->extradata;
2723 if (!avctx->extradata_size)
2725 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2730 op_bytes = *(p++)<<8;
2733 init_get_bits(&gb, p, op_bytes);
2736 ptype = get_bits(&gb, 8);
2737 debug_vp3("Theora headerpacket type: %x\n", ptype);
2739 if (!(ptype & 0x80))
2741 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2745 // FIXME: check for this aswell
2746 skip_bits(&gb, 6*8); /* "theora" */
2751 theora_decode_header(avctx, &gb);
2754 // FIXME: is this needed? it breaks sometimes
2755 // theora_decode_comments(avctx, gb);
2758 theora_decode_tables(avctx, &gb);
2761 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2764 if(8*op_bytes != get_bits_count(&gb))
2765 av_log(avctx, AV_LOG_ERROR, "%d bits left in packet %X\n", 8*op_bytes - get_bits_count(&gb), ptype);
2766 if (s->theora < 0x030200)
2770 vp3_decode_init(avctx);
2774 AVCodec vp3_decoder = {
2778 sizeof(Vp3DecodeContext),
2787 #ifndef CONFIG_LIBTHEORA
2788 AVCodec theora_decoder = {
2792 sizeof(Vp3DecodeContext),