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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 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 uint16_t coded_intra_y_dequant[64];
266 uint16_t coded_intra_c_dequant[64];
267 uint16_t coded_inter_dequant[64];
269 /* this is a list of indices into the all_fragments array indicating
270 * which of the fragments are coded */
271 int *coded_fragment_list;
272 int coded_fragment_list_index;
273 int pixel_addresses_inited;
281 VLC superblock_run_length_vlc;
282 VLC fragment_run_length_vlc;
284 VLC motion_vector_vlc;
286 /* these arrays need to be on 16-byte boundaries since SSE2 operations
288 int16_t __align16 intra_y_dequant[64];
289 int16_t __align16 intra_c_dequant[64];
290 int16_t __align16 inter_dequant[64];
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_comments(AVCodecContext *avctx, GetBitContext gb);
332 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext gb);
334 /************************************************************************
335 * VP3 specific functions
336 ************************************************************************/
339 * This function sets up all of the various blocks mappings:
340 * superblocks <-> fragments, macroblocks <-> fragments,
341 * superblocks <-> macroblocks
343 * Returns 0 is successful; returns 1 if *anything* went wrong.
345 static int init_block_mapping(Vp3DecodeContext *s)
348 signed int hilbert_walk_y[16];
349 signed int hilbert_walk_c[16];
350 signed int hilbert_walk_mb[4];
352 int current_fragment = 0;
353 int current_width = 0;
354 int current_height = 0;
357 int superblock_row_inc = 0;
359 int mapping_index = 0;
361 int current_macroblock;
364 signed char travel_width[16] = {
371 signed char travel_height[16] = {
378 signed char travel_width_mb[4] = {
382 signed char travel_height_mb[4] = {
386 debug_vp3(" vp3: initialize block mapping tables\n");
388 /* figure out hilbert pattern per these frame dimensions */
389 hilbert_walk_y[0] = 1;
390 hilbert_walk_y[1] = 1;
391 hilbert_walk_y[2] = s->fragment_width;
392 hilbert_walk_y[3] = -1;
393 hilbert_walk_y[4] = s->fragment_width;
394 hilbert_walk_y[5] = s->fragment_width;
395 hilbert_walk_y[6] = 1;
396 hilbert_walk_y[7] = -s->fragment_width;
397 hilbert_walk_y[8] = 1;
398 hilbert_walk_y[9] = s->fragment_width;
399 hilbert_walk_y[10] = 1;
400 hilbert_walk_y[11] = -s->fragment_width;
401 hilbert_walk_y[12] = -s->fragment_width;
402 hilbert_walk_y[13] = -1;
403 hilbert_walk_y[14] = -s->fragment_width;
404 hilbert_walk_y[15] = 1;
406 hilbert_walk_c[0] = 1;
407 hilbert_walk_c[1] = 1;
408 hilbert_walk_c[2] = s->fragment_width / 2;
409 hilbert_walk_c[3] = -1;
410 hilbert_walk_c[4] = s->fragment_width / 2;
411 hilbert_walk_c[5] = s->fragment_width / 2;
412 hilbert_walk_c[6] = 1;
413 hilbert_walk_c[7] = -s->fragment_width / 2;
414 hilbert_walk_c[8] = 1;
415 hilbert_walk_c[9] = s->fragment_width / 2;
416 hilbert_walk_c[10] = 1;
417 hilbert_walk_c[11] = -s->fragment_width / 2;
418 hilbert_walk_c[12] = -s->fragment_width / 2;
419 hilbert_walk_c[13] = -1;
420 hilbert_walk_c[14] = -s->fragment_width / 2;
421 hilbert_walk_c[15] = 1;
423 hilbert_walk_mb[0] = 1;
424 hilbert_walk_mb[1] = s->macroblock_width;
425 hilbert_walk_mb[2] = 1;
426 hilbert_walk_mb[3] = -s->macroblock_width;
428 /* iterate through each superblock (all planes) and map the fragments */
429 for (i = 0; i < s->superblock_count; i++) {
430 debug_init(" superblock %d (u starts @ %d, v starts @ %d)\n",
431 i, s->u_superblock_start, s->v_superblock_start);
433 /* time to re-assign the limits? */
436 /* start of Y superblocks */
437 right_edge = s->fragment_width;
438 bottom_edge = s->fragment_height;
441 superblock_row_inc = 3 * s->fragment_width -
442 (s->y_superblock_width * 4 - s->fragment_width);
443 hilbert = hilbert_walk_y;
445 /* the first operation for this variable is to advance by 1 */
446 current_fragment = -1;
448 } else if (i == s->u_superblock_start) {
450 /* start of U superblocks */
451 right_edge = s->fragment_width / 2;
452 bottom_edge = s->fragment_height / 2;
455 superblock_row_inc = 3 * (s->fragment_width / 2) -
456 (s->c_superblock_width * 4 - s->fragment_width / 2);
457 hilbert = hilbert_walk_c;
459 /* the first operation for this variable is to advance by 1 */
460 current_fragment = s->u_fragment_start - 1;
462 } else if (i == s->v_superblock_start) {
464 /* start of V superblocks */
465 right_edge = s->fragment_width / 2;
466 bottom_edge = s->fragment_height / 2;
469 superblock_row_inc = 3 * (s->fragment_width / 2) -
470 (s->c_superblock_width * 4 - s->fragment_width / 2);
471 hilbert = hilbert_walk_c;
473 /* the first operation for this variable is to advance by 1 */
474 current_fragment = s->v_fragment_start - 1;
478 if (current_width >= right_edge - 1) {
479 /* reset width and move to next superblock row */
483 /* fragment is now at the start of a new superblock row */
484 current_fragment += superblock_row_inc;
487 /* iterate through all 16 fragments in a superblock */
488 for (j = 0; j < 16; j++) {
489 current_fragment += hilbert[j];
490 current_width += travel_width[j];
491 current_height += travel_height[j];
493 /* check if the fragment is in bounds */
494 if ((current_width < right_edge) &&
495 (current_height < bottom_edge)) {
496 s->superblock_fragments[mapping_index] = current_fragment;
497 debug_init(" mapping fragment %d to superblock %d, position %d (%d/%d x %d/%d)\n",
498 s->superblock_fragments[mapping_index], i, j,
499 current_width, right_edge, current_height, bottom_edge);
501 s->superblock_fragments[mapping_index] = -1;
502 debug_init(" superblock %d, position %d has no fragment (%d/%d x %d/%d)\n",
504 current_width, right_edge, current_height, bottom_edge);
511 /* initialize the superblock <-> macroblock mapping; iterate through
512 * all of the Y plane superblocks to build this mapping */
513 right_edge = s->macroblock_width;
514 bottom_edge = s->macroblock_height;
517 superblock_row_inc = s->macroblock_width -
518 (s->y_superblock_width * 2 - s->macroblock_width);;
519 hilbert = hilbert_walk_mb;
521 current_macroblock = -1;
522 for (i = 0; i < s->u_superblock_start; i++) {
524 if (current_width >= right_edge - 1) {
525 /* reset width and move to next superblock row */
529 /* macroblock is now at the start of a new superblock row */
530 current_macroblock += superblock_row_inc;
533 /* iterate through each potential macroblock in the superblock */
534 for (j = 0; j < 4; j++) {
535 current_macroblock += hilbert_walk_mb[j];
536 current_width += travel_width_mb[j];
537 current_height += travel_height_mb[j];
539 /* check if the macroblock is in bounds */
540 if ((current_width < right_edge) &&
541 (current_height < bottom_edge)) {
542 s->superblock_macroblocks[mapping_index] = current_macroblock;
543 debug_init(" mapping macroblock %d to superblock %d, position %d (%d/%d x %d/%d)\n",
544 s->superblock_macroblocks[mapping_index], i, j,
545 current_width, right_edge, current_height, bottom_edge);
547 s->superblock_macroblocks[mapping_index] = -1;
548 debug_init(" superblock %d, position %d has no macroblock (%d/%d x %d/%d)\n",
550 current_width, right_edge, current_height, bottom_edge);
557 /* initialize the macroblock <-> fragment mapping */
558 current_fragment = 0;
559 current_macroblock = 0;
561 for (i = 0; i < s->fragment_height; i += 2) {
563 for (j = 0; j < s->fragment_width; j += 2) {
565 debug_init(" macroblock %d contains fragments: ", current_macroblock);
566 s->all_fragments[current_fragment].macroblock = current_macroblock;
567 s->macroblock_fragments[mapping_index++] = current_fragment;
568 debug_init("%d ", current_fragment);
570 if (j + 1 < s->fragment_width) {
571 s->all_fragments[current_fragment + 1].macroblock = current_macroblock;
572 s->macroblock_fragments[mapping_index++] = current_fragment + 1;
573 debug_init("%d ", current_fragment + 1);
575 s->macroblock_fragments[mapping_index++] = -1;
577 if (i + 1 < s->fragment_height) {
578 s->all_fragments[current_fragment + s->fragment_width].macroblock =
580 s->macroblock_fragments[mapping_index++] =
581 current_fragment + s->fragment_width;
582 debug_init("%d ", current_fragment + s->fragment_width);
584 s->macroblock_fragments[mapping_index++] = -1;
586 if ((j + 1 < s->fragment_width) && (i + 1 < s->fragment_height)) {
587 s->all_fragments[current_fragment + s->fragment_width + 1].macroblock =
589 s->macroblock_fragments[mapping_index++] =
590 current_fragment + s->fragment_width + 1;
591 debug_init("%d ", current_fragment + s->fragment_width + 1);
593 s->macroblock_fragments[mapping_index++] = -1;
596 c_fragment = s->u_fragment_start +
597 (i * s->fragment_width / 4) + (j / 2);
598 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
599 s->macroblock_fragments[mapping_index++] = c_fragment;
600 debug_init("%d ", c_fragment);
602 c_fragment = s->v_fragment_start +
603 (i * s->fragment_width / 4) + (j / 2);
604 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
605 s->macroblock_fragments[mapping_index++] = c_fragment;
606 debug_init("%d ", c_fragment);
610 if (j + 2 <= s->fragment_width)
611 current_fragment += 2;
614 current_macroblock++;
617 current_fragment += s->fragment_width;
620 return 0; /* successful path out */
624 * This function wipes out all of the fragment data.
626 static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
630 /* zero out all of the fragment information */
631 s->coded_fragment_list_index = 0;
632 for (i = 0; i < s->fragment_count; i++) {
633 s->all_fragments[i].coeff_count = 0;
634 s->all_fragments[i].motion_x = 127;
635 s->all_fragments[i].motion_y = 127;
636 s->all_fragments[i].next_coeff= NULL;
638 s->coeffs[i].coeff=0;
639 s->coeffs[i].next= NULL;
644 * This function sets up the dequantization tables used for a particular
647 static void init_dequantizer(Vp3DecodeContext *s)
650 int ac_scale_factor = s->coded_ac_scale_factor[s->quality_index];
651 int dc_scale_factor = s->coded_dc_scale_factor[s->quality_index];
654 debug_vp3(" vp3: initializing dequantization tables\n");
657 * Scale dequantizers:
663 * where sf = dc_scale_factor for DC quantizer
664 * or ac_scale_factor for AC quantizer
666 * Then, saturate the result to a lower limit of MIN_DEQUANT_VAL.
670 /* scale DC quantizers */
671 s->intra_y_dequant[0] = s->coded_intra_y_dequant[0] * dc_scale_factor / 100;
672 if (s->intra_y_dequant[0] < MIN_DEQUANT_VAL * 2)
673 s->intra_y_dequant[0] = MIN_DEQUANT_VAL * 2;
674 s->intra_y_dequant[0] *= SCALER;
676 s->intra_c_dequant[0] = s->coded_intra_c_dequant[0] * dc_scale_factor / 100;
677 if (s->intra_c_dequant[0] < MIN_DEQUANT_VAL * 2)
678 s->intra_c_dequant[0] = MIN_DEQUANT_VAL * 2;
679 s->intra_c_dequant[0] *= SCALER;
681 s->inter_dequant[0] = s->coded_inter_dequant[0] * dc_scale_factor / 100;
682 if (s->inter_dequant[0] < MIN_DEQUANT_VAL * 4)
683 s->inter_dequant[0] = MIN_DEQUANT_VAL * 4;
684 s->inter_dequant[0] *= SCALER;
686 /* scale AC quantizers, zigzag at the same time in preparation for
687 * the dequantization phase */
688 for (i = 1; i < 64; i++) {
689 int k= s->scantable.scantable[i];
690 j = s->scantable.permutated[i];
692 s->intra_y_dequant[j] = s->coded_intra_y_dequant[k] * ac_scale_factor / 100;
693 if (s->intra_y_dequant[j] < MIN_DEQUANT_VAL)
694 s->intra_y_dequant[j] = MIN_DEQUANT_VAL;
695 s->intra_y_dequant[j] *= SCALER;
697 s->intra_c_dequant[j] = s->coded_intra_c_dequant[k] * ac_scale_factor / 100;
698 if (s->intra_c_dequant[j] < MIN_DEQUANT_VAL)
699 s->intra_c_dequant[j] = MIN_DEQUANT_VAL;
700 s->intra_c_dequant[j] *= SCALER;
702 s->inter_dequant[j] = s->coded_inter_dequant[k] * ac_scale_factor / 100;
703 if (s->inter_dequant[j] < MIN_DEQUANT_VAL * 2)
704 s->inter_dequant[j] = MIN_DEQUANT_VAL * 2;
705 s->inter_dequant[j] *= SCALER;
708 memset(s->qscale_table, (FFMAX(s->intra_y_dequant[1], s->intra_c_dequant[1])+8)/16, 512); //FIXME finetune
710 /* print debug information as requested */
711 debug_dequantizers("intra Y dequantizers:\n");
712 for (i = 0; i < 8; i++) {
713 for (j = i * 8; j < i * 8 + 8; j++) {
714 debug_dequantizers(" %4d,", s->intra_y_dequant[j]);
716 debug_dequantizers("\n");
718 debug_dequantizers("\n");
720 debug_dequantizers("intra C dequantizers:\n");
721 for (i = 0; i < 8; i++) {
722 for (j = i * 8; j < i * 8 + 8; j++) {
723 debug_dequantizers(" %4d,", s->intra_c_dequant[j]);
725 debug_dequantizers("\n");
727 debug_dequantizers("\n");
729 debug_dequantizers("interframe dequantizers:\n");
730 for (i = 0; i < 8; i++) {
731 for (j = i * 8; j < i * 8 + 8; j++) {
732 debug_dequantizers(" %4d,", s->inter_dequant[j]);
734 debug_dequantizers("\n");
736 debug_dequantizers("\n");
740 * This function initializes the loop filter boundary limits if the frame's
741 * quality index is different from the previous frame's.
743 static void init_loop_filter(Vp3DecodeContext *s)
745 int *bounding_values= s->bounding_values_array+127;
749 filter_limit = s->filter_limit_values[s->quality_index];
751 /* set up the bounding values */
752 memset(s->bounding_values_array, 0, 256 * sizeof(int));
753 for (x = 0; x < filter_limit; x++) {
754 bounding_values[-x - filter_limit] = -filter_limit + x;
755 bounding_values[-x] = -x;
756 bounding_values[x] = x;
757 bounding_values[x + filter_limit] = filter_limit - x;
762 * This function unpacks all of the superblock/macroblock/fragment coding
763 * information from the bitstream.
765 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
768 int current_superblock = 0;
770 int decode_fully_flags = 0;
771 int decode_partial_blocks = 0;
772 int first_c_fragment_seen;
775 int current_fragment;
777 debug_vp3(" vp3: unpacking superblock coding\n");
781 debug_vp3(" keyframe-- all superblocks are fully coded\n");
782 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
786 /* unpack the list of partially-coded superblocks */
787 bit = get_bits(gb, 1);
788 /* toggle the bit because as soon as the first run length is
789 * fetched the bit will be toggled again */
791 while (current_superblock < s->superblock_count) {
792 if (current_run-- == 0) {
794 current_run = get_vlc2(gb,
795 s->superblock_run_length_vlc.table, 6, 2);
796 if (current_run == 33)
797 current_run += get_bits(gb, 12);
798 debug_block_coding(" setting superblocks %d..%d to %s\n",
800 current_superblock + current_run - 1,
801 (bit) ? "partially coded" : "not coded");
803 /* if any of the superblocks are not partially coded, flag
804 * a boolean to decode the list of fully-coded superblocks */
806 decode_fully_flags = 1;
809 /* make a note of the fact that there are partially coded
811 decode_partial_blocks = 1;
814 s->superblock_coding[current_superblock++] = bit;
817 /* unpack the list of fully coded superblocks if any of the blocks were
818 * not marked as partially coded in the previous step */
819 if (decode_fully_flags) {
821 current_superblock = 0;
823 bit = get_bits(gb, 1);
824 /* toggle the bit because as soon as the first run length is
825 * fetched the bit will be toggled again */
827 while (current_superblock < s->superblock_count) {
829 /* skip any superblocks already marked as partially coded */
830 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
832 if (current_run-- == 0) {
834 current_run = get_vlc2(gb,
835 s->superblock_run_length_vlc.table, 6, 2);
836 if (current_run == 33)
837 current_run += get_bits(gb, 12);
840 debug_block_coding(" setting superblock %d to %s\n",
842 (bit) ? "fully coded" : "not coded");
843 s->superblock_coding[current_superblock] = 2*bit;
845 current_superblock++;
849 /* if there were partial blocks, initialize bitstream for
850 * unpacking fragment codings */
851 if (decode_partial_blocks) {
854 bit = get_bits(gb, 1);
855 /* toggle the bit because as soon as the first run length is
856 * fetched the bit will be toggled again */
861 /* figure out which fragments are coded; iterate through each
862 * superblock (all planes) */
863 s->coded_fragment_list_index = 0;
864 s->next_coeff= s->coeffs + s->fragment_count;
865 s->first_coded_y_fragment = s->first_coded_c_fragment = 0;
866 s->last_coded_y_fragment = s->last_coded_c_fragment = -1;
867 first_c_fragment_seen = 0;
868 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
869 for (i = 0; i < s->superblock_count; i++) {
871 /* iterate through all 16 fragments in a superblock */
872 for (j = 0; j < 16; j++) {
874 /* if the fragment is in bounds, check its coding status */
875 current_fragment = s->superblock_fragments[i * 16 + j];
876 if (current_fragment >= s->fragment_count) {
877 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
878 current_fragment, s->fragment_count);
881 if (current_fragment != -1) {
882 if (s->superblock_coding[i] == SB_NOT_CODED) {
884 /* copy all the fragments from the prior frame */
885 s->all_fragments[current_fragment].coding_method =
888 } else if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
890 /* fragment may or may not be coded; this is the case
891 * that cares about the fragment coding runs */
892 if (current_run-- == 0) {
894 current_run = get_vlc2(gb,
895 s->fragment_run_length_vlc.table, 5, 2);
899 /* default mode; actual mode will be decoded in
901 s->all_fragments[current_fragment].coding_method =
903 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
904 s->coded_fragment_list[s->coded_fragment_list_index] =
906 if ((current_fragment >= s->u_fragment_start) &&
907 (s->last_coded_y_fragment == -1) &&
908 (!first_c_fragment_seen)) {
909 s->first_coded_c_fragment = s->coded_fragment_list_index;
910 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
911 first_c_fragment_seen = 1;
913 s->coded_fragment_list_index++;
914 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
915 debug_block_coding(" superblock %d is partially coded, fragment %d is coded\n",
916 i, current_fragment);
918 /* not coded; copy this fragment from the prior frame */
919 s->all_fragments[current_fragment].coding_method =
921 debug_block_coding(" superblock %d is partially coded, fragment %d is not coded\n",
922 i, current_fragment);
927 /* fragments are fully coded in this superblock; actual
928 * coding will be determined in next step */
929 s->all_fragments[current_fragment].coding_method =
931 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
932 s->coded_fragment_list[s->coded_fragment_list_index] =
934 if ((current_fragment >= s->u_fragment_start) &&
935 (s->last_coded_y_fragment == -1) &&
936 (!first_c_fragment_seen)) {
937 s->first_coded_c_fragment = s->coded_fragment_list_index;
938 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
939 first_c_fragment_seen = 1;
941 s->coded_fragment_list_index++;
942 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
943 debug_block_coding(" superblock %d is fully coded, fragment %d is coded\n",
944 i, current_fragment);
950 if (!first_c_fragment_seen)
951 /* only Y fragments coded in this frame */
952 s->last_coded_y_fragment = s->coded_fragment_list_index - 1;
954 /* end the list of coded C fragments */
955 s->last_coded_c_fragment = s->coded_fragment_list_index - 1;
957 debug_block_coding(" %d total coded fragments, y: %d -> %d, c: %d -> %d\n",
958 s->coded_fragment_list_index,
959 s->first_coded_y_fragment,
960 s->last_coded_y_fragment,
961 s->first_coded_c_fragment,
962 s->last_coded_c_fragment);
968 * This function unpacks all the coding mode data for individual macroblocks
969 * from the bitstream.
971 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
975 int current_macroblock;
976 int current_fragment;
979 debug_vp3(" vp3: unpacking encoding modes\n");
982 debug_vp3(" keyframe-- all blocks are coded as INTRA\n");
984 for (i = 0; i < s->fragment_count; i++)
985 s->all_fragments[i].coding_method = MODE_INTRA;
989 /* fetch the mode coding scheme for this frame */
990 scheme = get_bits(gb, 3);
991 debug_modes(" using mode alphabet %d\n", scheme);
993 /* is it a custom coding scheme? */
995 debug_modes(" custom mode alphabet ahead:\n");
996 for (i = 0; i < 8; i++)
997 ModeAlphabet[scheme][get_bits(gb, 3)] = i;
1000 for (i = 0; i < 8; i++)
1001 debug_modes(" mode[%d][%d] = %d\n", scheme, i,
1002 ModeAlphabet[scheme][i]);
1004 /* iterate through all of the macroblocks that contain 1 or more
1005 * coded fragments */
1006 for (i = 0; i < s->u_superblock_start; i++) {
1008 for (j = 0; j < 4; j++) {
1009 current_macroblock = s->superblock_macroblocks[i * 4 + j];
1010 if ((current_macroblock == -1) ||
1011 (s->macroblock_coding[current_macroblock] == MODE_COPY))
1013 if (current_macroblock >= s->macroblock_count) {
1014 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad macroblock number (%d >= %d)\n",
1015 current_macroblock, s->macroblock_count);
1019 /* mode 7 means get 3 bits for each coding mode */
1021 coding_mode = get_bits(gb, 3);
1023 coding_mode = ModeAlphabet[scheme]
1024 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
1026 s->macroblock_coding[current_macroblock] = coding_mode;
1027 for (k = 0; k < 6; k++) {
1029 s->macroblock_fragments[current_macroblock * 6 + k];
1030 if (current_fragment == -1)
1032 if (current_fragment >= s->fragment_count) {
1033 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad fragment number (%d >= %d)\n",
1034 current_fragment, s->fragment_count);
1037 if (s->all_fragments[current_fragment].coding_method !=
1039 s->all_fragments[current_fragment].coding_method =
1043 debug_modes(" coding method for macroblock starting @ fragment %d = %d\n",
1044 s->macroblock_fragments[current_macroblock * 6], coding_mode);
1053 * This function unpacks all the motion vectors for the individual
1054 * macroblocks from the bitstream.
1056 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
1062 int last_motion_x = 0;
1063 int last_motion_y = 0;
1064 int prior_last_motion_x = 0;
1065 int prior_last_motion_y = 0;
1066 int current_macroblock;
1067 int current_fragment;
1069 debug_vp3(" vp3: unpacking motion vectors\n");
1072 debug_vp3(" keyframe-- there are no motion vectors\n");
1076 memset(motion_x, 0, 6 * sizeof(int));
1077 memset(motion_y, 0, 6 * sizeof(int));
1079 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
1080 coding_mode = get_bits(gb, 1);
1081 debug_vectors(" using %s scheme for unpacking motion vectors\n",
1082 (coding_mode == 0) ? "VLC" : "fixed-length");
1084 /* iterate through all of the macroblocks that contain 1 or more
1085 * coded fragments */
1086 for (i = 0; i < s->u_superblock_start; i++) {
1088 for (j = 0; j < 4; j++) {
1089 current_macroblock = s->superblock_macroblocks[i * 4 + j];
1090 if ((current_macroblock == -1) ||
1091 (s->macroblock_coding[current_macroblock] == MODE_COPY))
1093 if (current_macroblock >= s->macroblock_count) {
1094 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad macroblock number (%d >= %d)\n",
1095 current_macroblock, s->macroblock_count);
1099 current_fragment = s->macroblock_fragments[current_macroblock * 6];
1100 if (current_fragment >= s->fragment_count) {
1101 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d\n",
1102 current_fragment, s->fragment_count);
1105 switch (s->macroblock_coding[current_macroblock]) {
1107 case MODE_INTER_PLUS_MV:
1108 case MODE_GOLDEN_MV:
1109 /* all 6 fragments use the same motion vector */
1110 if (coding_mode == 0) {
1111 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1112 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1114 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
1115 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
1118 for (k = 1; k < 6; k++) {
1119 motion_x[k] = motion_x[0];
1120 motion_y[k] = motion_y[0];
1123 /* vector maintenance, only on MODE_INTER_PLUS_MV */
1124 if (s->macroblock_coding[current_macroblock] ==
1125 MODE_INTER_PLUS_MV) {
1126 prior_last_motion_x = last_motion_x;
1127 prior_last_motion_y = last_motion_y;
1128 last_motion_x = motion_x[0];
1129 last_motion_y = motion_y[0];
1133 case MODE_INTER_FOURMV:
1134 /* fetch 4 vectors from the bitstream, one for each
1135 * Y fragment, then average for the C fragment vectors */
1136 motion_x[4] = motion_y[4] = 0;
1137 for (k = 0; k < 4; k++) {
1138 if (coding_mode == 0) {
1139 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1140 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1142 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
1143 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
1145 motion_x[4] += motion_x[k];
1146 motion_y[4] += motion_y[k];
1149 if (motion_x[4] >= 0)
1150 motion_x[4] = (motion_x[4] + 2) / 4;
1152 motion_x[4] = (motion_x[4] - 2) / 4;
1153 motion_x[5] = motion_x[4];
1155 if (motion_y[4] >= 0)
1156 motion_y[4] = (motion_y[4] + 2) / 4;
1158 motion_y[4] = (motion_y[4] - 2) / 4;
1159 motion_y[5] = motion_y[4];
1161 /* vector maintenance; vector[3] is treated as the
1162 * last vector in this case */
1163 prior_last_motion_x = last_motion_x;
1164 prior_last_motion_y = last_motion_y;
1165 last_motion_x = motion_x[3];
1166 last_motion_y = motion_y[3];
1169 case MODE_INTER_LAST_MV:
1170 /* all 6 fragments use the last motion vector */
1171 motion_x[0] = last_motion_x;
1172 motion_y[0] = last_motion_y;
1173 for (k = 1; k < 6; k++) {
1174 motion_x[k] = motion_x[0];
1175 motion_y[k] = motion_y[0];
1178 /* no vector maintenance (last vector remains the
1182 case MODE_INTER_PRIOR_LAST:
1183 /* all 6 fragments use the motion vector prior to the
1184 * last motion vector */
1185 motion_x[0] = prior_last_motion_x;
1186 motion_y[0] = prior_last_motion_y;
1187 for (k = 1; k < 6; k++) {
1188 motion_x[k] = motion_x[0];
1189 motion_y[k] = motion_y[0];
1192 /* vector maintenance */
1193 prior_last_motion_x = last_motion_x;
1194 prior_last_motion_y = last_motion_y;
1195 last_motion_x = motion_x[0];
1196 last_motion_y = motion_y[0];
1200 /* covers intra, inter without MV, golden without MV */
1201 memset(motion_x, 0, 6 * sizeof(int));
1202 memset(motion_y, 0, 6 * sizeof(int));
1204 /* no vector maintenance */
1208 /* assign the motion vectors to the correct fragments */
1209 debug_vectors(" vectors for macroblock starting @ fragment %d (coding method %d):\n",
1211 s->macroblock_coding[current_macroblock]);
1212 for (k = 0; k < 6; k++) {
1214 s->macroblock_fragments[current_macroblock * 6 + k];
1215 if (current_fragment == -1)
1217 if (current_fragment >= s->fragment_count) {
1218 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d)\n",
1219 current_fragment, s->fragment_count);
1222 s->all_fragments[current_fragment].motion_x = motion_x[k];
1223 s->all_fragments[current_fragment].motion_y = motion_y[k];
1224 debug_vectors(" vector %d: fragment %d = (%d, %d)\n",
1225 k, current_fragment, motion_x[k], motion_y[k]);
1235 * This function is called by unpack_dct_coeffs() to extract the VLCs from
1236 * the bitstream. The VLCs encode tokens which are used to unpack DCT
1237 * data. This function unpacks all the VLCs for either the Y plane or both
1238 * C planes, and is called for DC coefficients or different AC coefficient
1239 * levels (since different coefficient types require different VLC tables.
1241 * This function returns a residual eob run. E.g, if a particular token gave
1242 * instructions to EOB the next 5 fragments and there were only 2 fragments
1243 * left in the current fragment range, 3 would be returned so that it could
1244 * be passed into the next call to this same function.
1246 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
1247 VLC *table, int coeff_index,
1248 int first_fragment, int last_fragment,
1255 Vp3Fragment *fragment;
1256 uint8_t *perm= s->scantable.permutated;
1259 if ((first_fragment >= s->fragment_count) ||
1260 (last_fragment >= s->fragment_count)) {
1262 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vlcs(): bad fragment number (%d -> %d ?)\n",
1263 first_fragment, last_fragment);
1267 for (i = first_fragment; i <= last_fragment; i++) {
1269 fragment = &s->all_fragments[s->coded_fragment_list[i]];
1270 if (fragment->coeff_count > coeff_index)
1274 /* decode a VLC into a token */
1275 token = get_vlc2(gb, table->table, 5, 3);
1276 debug_vlc(" token = %2d, ", token);
1277 /* use the token to get a zero run, a coefficient, and an eob run */
1279 eob_run = eob_run_base[token];
1280 if (eob_run_get_bits[token])
1281 eob_run += get_bits(gb, eob_run_get_bits[token]);
1282 coeff = zero_run = 0;
1284 bits_to_get = coeff_get_bits[token];
1286 coeff = coeff_tables[token][0];
1288 coeff = coeff_tables[token][get_bits(gb, bits_to_get)];
1290 zero_run = zero_run_base[token];
1291 if (zero_run_get_bits[token])
1292 zero_run += get_bits(gb, zero_run_get_bits[token]);
1297 fragment->coeff_count += zero_run;
1298 if (fragment->coeff_count < 64){
1299 fragment->next_coeff->coeff= coeff;
1300 fragment->next_coeff->index= perm[fragment->coeff_count++]; //FIXME perm here already?
1301 fragment->next_coeff->next= s->next_coeff;
1302 s->next_coeff->next=NULL;
1303 fragment->next_coeff= s->next_coeff++;
1305 debug_vlc(" fragment %d coeff = %d\n",
1306 s->coded_fragment_list[i], fragment->next_coeff[coeff_index]);
1308 fragment->coeff_count |= 128;
1309 debug_vlc(" fragment %d eob with %d coefficients\n",
1310 s->coded_fragment_list[i], fragment->coeff_count&127);
1319 * This function unpacks all of the DCT coefficient data from the
1322 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1329 int residual_eob_run = 0;
1331 /* fetch the DC table indices */
1332 dc_y_table = get_bits(gb, 4);
1333 dc_c_table = get_bits(gb, 4);
1335 /* unpack the Y plane DC coefficients */
1336 debug_vp3(" vp3: unpacking Y plane DC coefficients using table %d\n",
1338 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1339 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1341 /* unpack the C plane DC coefficients */
1342 debug_vp3(" vp3: unpacking C plane DC coefficients using table %d\n",
1344 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1345 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1347 /* fetch the AC table indices */
1348 ac_y_table = get_bits(gb, 4);
1349 ac_c_table = get_bits(gb, 4);
1351 /* unpack the group 1 AC coefficients (coeffs 1-5) */
1352 for (i = 1; i <= 5; i++) {
1354 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1356 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_y_table], i,
1357 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1359 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1361 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_c_table], i,
1362 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1365 /* unpack the group 2 AC coefficients (coeffs 6-14) */
1366 for (i = 6; i <= 14; i++) {
1368 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1370 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_y_table], i,
1371 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1373 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1375 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_c_table], i,
1376 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1379 /* unpack the group 3 AC coefficients (coeffs 15-27) */
1380 for (i = 15; i <= 27; i++) {
1382 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1384 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_y_table], i,
1385 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1387 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1389 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_c_table], i,
1390 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1393 /* unpack the group 4 AC coefficients (coeffs 28-63) */
1394 for (i = 28; i <= 63; i++) {
1396 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1398 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_y_table], i,
1399 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1401 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1403 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_c_table], i,
1404 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1411 * This function reverses the DC prediction for each coded fragment in
1412 * the frame. Much of this function is adapted directly from the original
1415 #define COMPATIBLE_FRAME(x) \
1416 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1417 #define FRAME_CODED(x) (s->all_fragments[x].coding_method != MODE_COPY)
1418 #define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
1419 static inline int iabs (int x) { return ((x < 0) ? -x : x); }
1421 static void reverse_dc_prediction(Vp3DecodeContext *s,
1424 int fragment_height)
1433 int i = first_fragment;
1436 * Fragment prediction groups:
1444 * Note: Groups 5 and 7 do not exist as it would mean that the
1445 * fragment's x coordinate is both 0 and (width - 1) at the same time.
1447 int predictor_group;
1450 /* validity flags for the left, up-left, up, and up-right fragments */
1451 int fl, ful, fu, fur;
1453 /* DC values for the left, up-left, up, and up-right fragments */
1454 int vl, vul, vu, vur;
1456 /* indices for the left, up-left, up, and up-right fragments */
1460 * The 6 fields mean:
1461 * 0: up-left multiplier
1463 * 2: up-right multiplier
1464 * 3: left multiplier
1466 * 5: right bit shift divisor (e.g., 7 means >>=7, a.k.a. div by 128)
1468 int predictor_transform[16][6] = {
1469 { 0, 0, 0, 0, 0, 0 },
1470 { 0, 0, 0, 1, 0, 0 }, // PL
1471 { 0, 0, 1, 0, 0, 0 }, // PUR
1472 { 0, 0, 53, 75, 127, 7 }, // PUR|PL
1473 { 0, 1, 0, 0, 0, 0 }, // PU
1474 { 0, 1, 0, 1, 1, 1 }, // PU|PL
1475 { 0, 1, 0, 0, 0, 0 }, // PU|PUR
1476 { 0, 0, 53, 75, 127, 7 }, // PU|PUR|PL
1477 { 1, 0, 0, 0, 0, 0 }, // PUL
1478 { 0, 0, 0, 1, 0, 0 }, // PUL|PL
1479 { 1, 0, 1, 0, 1, 1 }, // PUL|PUR
1480 { 0, 0, 53, 75, 127, 7 }, // PUL|PUR|PL
1481 { 0, 1, 0, 0, 0, 0 }, // PUL|PU
1482 {-26, 29, 0, 29, 31, 5 }, // PUL|PU|PL
1483 { 3, 10, 3, 0, 15, 4 }, // PUL|PU|PUR
1484 {-26, 29, 0, 29, 31, 5 } // PUL|PU|PUR|PL
1487 /* This table shows which types of blocks can use other blocks for
1488 * prediction. For example, INTRA is the only mode in this table to
1489 * have a frame number of 0. That means INTRA blocks can only predict
1490 * from other INTRA blocks. There are 2 golden frame coding types;
1491 * blocks encoding in these modes can only predict from other blocks
1492 * that were encoded with these 1 of these 2 modes. */
1493 unsigned char compatible_frame[8] = {
1494 1, /* MODE_INTER_NO_MV */
1496 1, /* MODE_INTER_PLUS_MV */
1497 1, /* MODE_INTER_LAST_MV */
1498 1, /* MODE_INTER_PRIOR_MV */
1499 2, /* MODE_USING_GOLDEN */
1500 2, /* MODE_GOLDEN_MV */
1501 1 /* MODE_INTER_FOUR_MV */
1503 int current_frame_type;
1505 /* there is a last DC predictor for each of the 3 frame types */
1510 debug_vp3(" vp3: reversing DC prediction\n");
1512 vul = vu = vur = vl = 0;
1513 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1515 /* for each fragment row... */
1516 for (y = 0; y < fragment_height; y++) {
1518 /* for each fragment in a row... */
1519 for (x = 0; x < fragment_width; x++, i++) {
1521 /* reverse prediction if this block was coded */
1522 if (s->all_fragments[i].coding_method != MODE_COPY) {
1524 current_frame_type =
1525 compatible_frame[s->all_fragments[i].coding_method];
1526 predictor_group = (x == 0) + ((y == 0) << 1) +
1527 ((x + 1 == fragment_width) << 2);
1528 debug_dc_pred(" frag %d: group %d, orig DC = %d, ",
1529 i, predictor_group, DC_COEFF(i));
1531 switch (predictor_group) {
1534 /* main body of fragments; consider all 4 possible
1535 * fragments for prediction */
1537 /* calculate the indices of the predicting fragments */
1538 ul = i - fragment_width - 1;
1539 u = i - fragment_width;
1540 ur = i - fragment_width + 1;
1543 /* fetch the DC values for the predicting fragments */
1549 /* figure out which fragments are valid */
1550 ful = FRAME_CODED(ul) && COMPATIBLE_FRAME(ul);
1551 fu = FRAME_CODED(u) && COMPATIBLE_FRAME(u);
1552 fur = FRAME_CODED(ur) && COMPATIBLE_FRAME(ur);
1553 fl = FRAME_CODED(l) && COMPATIBLE_FRAME(l);
1555 /* decide which predictor transform to use */
1556 transform = (fl*PL) | (fu*PU) | (ful*PUL) | (fur*PUR);
1561 /* left column of fragments, not including top corner;
1562 * only consider up and up-right fragments */
1564 /* calculate the indices of the predicting fragments */
1565 u = i - fragment_width;
1566 ur = i - fragment_width + 1;
1568 /* fetch the DC values for the predicting fragments */
1572 /* figure out which fragments are valid */
1573 fur = FRAME_CODED(ur) && COMPATIBLE_FRAME(ur);
1574 fu = FRAME_CODED(u) && COMPATIBLE_FRAME(u);
1576 /* decide which predictor transform to use */
1577 transform = (fu*PU) | (fur*PUR);
1583 /* top row of fragments, not including top-left frag;
1584 * only consider the left fragment for prediction */
1586 /* calculate the indices of the predicting fragments */
1589 /* fetch the DC values for the predicting fragments */
1592 /* figure out which fragments are valid */
1593 fl = FRAME_CODED(l) && COMPATIBLE_FRAME(l);
1595 /* decide which predictor transform to use */
1596 transform = (fl*PL);
1601 /* top-left fragment */
1603 /* nothing to predict from in this case */
1609 /* right column of fragments, not including top corner;
1610 * consider up-left, up, and left fragments for
1613 /* calculate the indices of the predicting fragments */
1614 ul = i - fragment_width - 1;
1615 u = i - fragment_width;
1618 /* fetch the DC values for the predicting fragments */
1623 /* figure out which fragments are valid */
1624 ful = FRAME_CODED(ul) && COMPATIBLE_FRAME(ul);
1625 fu = FRAME_CODED(u) && COMPATIBLE_FRAME(u);
1626 fl = FRAME_CODED(l) && COMPATIBLE_FRAME(l);
1628 /* decide which predictor transform to use */
1629 transform = (fl*PL) | (fu*PU) | (ful*PUL);
1635 debug_dc_pred("transform = %d, ", transform);
1637 if (transform == 0) {
1639 /* if there were no fragments to predict from, use last
1641 predicted_dc = last_dc[current_frame_type];
1642 debug_dc_pred("from last DC (%d) = %d\n",
1643 current_frame_type, DC_COEFF(i));
1647 /* apply the appropriate predictor transform */
1649 (predictor_transform[transform][0] * vul) +
1650 (predictor_transform[transform][1] * vu) +
1651 (predictor_transform[transform][2] * vur) +
1652 (predictor_transform[transform][3] * vl);
1654 /* if there is a shift value in the transform, add
1655 * the sign bit before the shift */
1656 if (predictor_transform[transform][5] != 0) {
1657 predicted_dc += ((predicted_dc >> 15) &
1658 predictor_transform[transform][4]);
1659 predicted_dc >>= predictor_transform[transform][5];
1662 /* check for outranging on the [ul u l] and
1663 * [ul u ur l] predictors */
1664 if ((transform == 13) || (transform == 15)) {
1665 if (iabs(predicted_dc - vu) > 128)
1667 else if (iabs(predicted_dc - vl) > 128)
1669 else if (iabs(predicted_dc - vul) > 128)
1673 debug_dc_pred("from pred DC = %d\n",
1677 /* at long last, apply the predictor */
1678 if(s->coeffs[i].index){
1679 *s->next_coeff= s->coeffs[i];
1680 s->coeffs[i].index=0;
1681 s->coeffs[i].coeff=0;
1682 s->coeffs[i].next= s->next_coeff++;
1684 s->coeffs[i].coeff += predicted_dc;
1686 last_dc[current_frame_type] = DC_COEFF(i);
1687 if(DC_COEFF(i) && !(s->all_fragments[i].coeff_count&127)){
1688 s->all_fragments[i].coeff_count= 129;
1689 // s->all_fragments[i].next_coeff= s->next_coeff;
1690 s->coeffs[i].next= s->next_coeff;
1691 (s->next_coeff++)->next=NULL;
1699 static void horizontal_filter(unsigned char *first_pixel, int stride,
1700 int *bounding_values);
1701 static void vertical_filter(unsigned char *first_pixel, int stride,
1702 int *bounding_values);
1705 * Perform the final rendering for a particular slice of data.
1706 * The slice number ranges from 0..(macroblock_height - 1).
1708 static void render_slice(Vp3DecodeContext *s, int slice)
1712 int i; /* indicates current fragment */
1713 int16_t *dequantizer;
1714 DCTELEM __align16 block[64];
1715 unsigned char *output_plane;
1716 unsigned char *last_plane;
1717 unsigned char *golden_plane;
1719 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1720 int upper_motion_limit, lower_motion_limit;
1721 int motion_halfpel_index;
1722 uint8_t *motion_source;
1727 int current_macroblock_entry = slice * s->macroblock_width * 6;
1730 if (slice >= s->macroblock_height)
1733 for (plane = 0; plane < 3; plane++) {
1735 /* set up plane-specific parameters */
1737 output_plane = s->current_frame.data[0];
1738 last_plane = s->last_frame.data[0];
1739 golden_plane = s->golden_frame.data[0];
1740 stride = s->current_frame.linesize[0];
1741 if (!s->flipped_image) stride = -stride;
1742 upper_motion_limit = 7 * s->current_frame.linesize[0];
1743 lower_motion_limit = s->height * s->current_frame.linesize[0] + s->width - 8;
1744 y = slice * FRAGMENT_PIXELS * 2;
1745 plane_width = s->width;
1746 plane_height = s->height;
1747 slice_height = y + FRAGMENT_PIXELS * 2;
1748 i = s->macroblock_fragments[current_macroblock_entry + 0];
1749 } else if (plane == 1) {
1750 output_plane = s->current_frame.data[1];
1751 last_plane = s->last_frame.data[1];
1752 golden_plane = s->golden_frame.data[1];
1753 stride = s->current_frame.linesize[1];
1754 if (!s->flipped_image) stride = -stride;
1755 upper_motion_limit = 7 * s->current_frame.linesize[1];
1756 lower_motion_limit = (s->height / 2) * s->current_frame.linesize[1] + (s->width / 2) - 8;
1757 y = slice * FRAGMENT_PIXELS;
1758 plane_width = s->width / 2;
1759 plane_height = s->height / 2;
1760 slice_height = y + FRAGMENT_PIXELS;
1761 i = s->macroblock_fragments[current_macroblock_entry + 4];
1763 output_plane = s->current_frame.data[2];
1764 last_plane = s->last_frame.data[2];
1765 golden_plane = s->golden_frame.data[2];
1766 stride = s->current_frame.linesize[2];
1767 if (!s->flipped_image) stride = -stride;
1768 upper_motion_limit = 7 * s->current_frame.linesize[2];
1769 lower_motion_limit = (s->height / 2) * s->current_frame.linesize[2] + (s->width / 2) - 8;
1770 y = slice * FRAGMENT_PIXELS;
1771 plane_width = s->width / 2;
1772 plane_height = s->height / 2;
1773 slice_height = y + FRAGMENT_PIXELS;
1774 i = s->macroblock_fragments[current_macroblock_entry + 5];
1776 fragment_width = plane_width / FRAGMENT_PIXELS;
1778 if(ABS(stride) > 2048)
1779 return; //various tables are fixed size
1781 /* for each fragment row in the slice (both of them)... */
1782 for (; y < slice_height; y += 8) {
1784 /* for each fragment in a row... */
1785 for (x = 0; x < plane_width; x += 8, i++) {
1787 if ((i < 0) || (i >= s->fragment_count)) {
1788 av_log(s->avctx, AV_LOG_ERROR, " vp3:render_slice(): bad fragment number (%d)\n", i);
1792 /* transform if this block was coded */
1793 if ((s->all_fragments[i].coding_method != MODE_COPY) &&
1794 !((s->avctx->flags & CODEC_FLAG_GRAY) && plane)) {
1796 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1797 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1798 motion_source= golden_plane;
1800 motion_source= last_plane;
1802 motion_source += s->all_fragments[i].first_pixel;
1803 motion_halfpel_index = 0;
1805 /* sort out the motion vector if this fragment is coded
1806 * using a motion vector method */
1807 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1808 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1810 motion_x = s->all_fragments[i].motion_x;
1811 motion_y = s->all_fragments[i].motion_y;
1813 motion_x= (motion_x>>1) | (motion_x&1);
1814 motion_y= (motion_y>>1) | (motion_y&1);
1817 src_x= (motion_x>>1) + x;
1818 src_y= (motion_y>>1) + y;
1819 if ((motion_x == 127) || (motion_y == 127))
1820 av_log(s->avctx, AV_LOG_ERROR, " help! got invalid motion vector! (%X, %X)\n", motion_x, motion_y);
1822 motion_halfpel_index = motion_x & 0x01;
1823 motion_source += (motion_x >> 1);
1825 motion_halfpel_index |= (motion_y & 0x01) << 1;
1826 motion_source += ((motion_y >> 1) * stride);
1828 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1829 uint8_t *temp= s->edge_emu_buffer;
1830 if(stride<0) temp -= 9*stride;
1831 else temp += 9*stride;
1833 ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1834 motion_source= temp;
1839 /* first, take care of copying a block from either the
1840 * previous or the golden frame */
1841 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1842 /* Note, it is possible to implement all MC cases with
1843 put_no_rnd_pixels_l2 which would look more like the
1844 VP3 source but this would be slower as
1845 put_no_rnd_pixels_tab is better optimzed */
1846 if(motion_halfpel_index != 3){
1847 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1848 output_plane + s->all_fragments[i].first_pixel,
1849 motion_source, stride, 8);
1851 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1852 s->dsp.put_no_rnd_pixels_l2[1](
1853 output_plane + s->all_fragments[i].first_pixel,
1855 motion_source + stride + 1 + d,
1858 dequantizer = s->inter_dequant;
1861 dequantizer = s->intra_y_dequant;
1863 dequantizer = s->intra_c_dequant;
1866 /* dequantize the DCT coefficients */
1867 debug_idct("fragment %d, coding mode %d, DC = %d, dequant = %d:\n",
1868 i, s->all_fragments[i].coding_method,
1869 DC_COEFF(i), dequantizer[0]);
1871 if(s->avctx->idct_algo==FF_IDCT_VP3){
1872 Coeff *coeff= s->coeffs + i;
1873 memset(block, 0, sizeof(block));
1875 block[coeff->index]= coeff->coeff * dequantizer[coeff->index];
1879 Coeff *coeff= s->coeffs + i;
1880 memset(block, 0, sizeof(block));
1882 block[coeff->index]= (coeff->coeff * dequantizer[coeff->index] + 2)>>2;
1887 /* invert DCT and place (or add) in final output */
1889 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1890 if(s->avctx->idct_algo!=FF_IDCT_VP3)
1893 output_plane + s->all_fragments[i].first_pixel,
1898 output_plane + s->all_fragments[i].first_pixel,
1903 debug_idct("block after idct_%s():\n",
1904 (s->all_fragments[i].coding_method == MODE_INTRA)?
1906 for (m = 0; m < 8; m++) {
1907 for (n = 0; n < 8; n++) {
1908 debug_idct(" %3d", *(output_plane +
1909 s->all_fragments[i].first_pixel + (m * stride + n)));
1917 /* copy directly from the previous frame */
1918 s->dsp.put_pixels_tab[1][0](
1919 output_plane + s->all_fragments[i].first_pixel,
1920 last_plane + s->all_fragments[i].first_pixel,
1925 /* perform the left edge filter if:
1926 * - the fragment is not on the left column
1927 * - the fragment is coded in this frame
1928 * - the fragment is not coded in this frame but the left
1929 * fragment is coded in this frame (this is done instead
1930 * of a right edge filter when rendering the left fragment
1931 * since this fragment is not available yet) */
1933 ((s->all_fragments[i].coding_method != MODE_COPY) ||
1934 ((s->all_fragments[i].coding_method == MODE_COPY) &&
1935 (s->all_fragments[i - 1].coding_method != MODE_COPY)) )) {
1937 output_plane + s->all_fragments[i].first_pixel + 7*stride,
1938 -stride, bounding_values);
1941 /* perform the top edge filter if:
1942 * - the fragment is not on the top row
1943 * - the fragment is coded in this frame
1944 * - the fragment is not coded in this frame but the above
1945 * fragment is coded in this frame (this is done instead
1946 * of a bottom edge filter when rendering the above
1947 * fragment since this fragment is not available yet) */
1949 ((s->all_fragments[i].coding_method != MODE_COPY) ||
1950 ((s->all_fragments[i].coding_method == MODE_COPY) &&
1951 (s->all_fragments[i - fragment_width].coding_method != MODE_COPY)) )) {
1953 output_plane + s->all_fragments[i].first_pixel - stride,
1954 -stride, bounding_values);
1961 /* this looks like a good place for slice dispatch... */
1963 * if (slice == s->macroblock_height - 1)
1964 * dispatch (both last slice & 2nd-to-last slice);
1965 * else if (slice > 0)
1966 * dispatch (slice - 1);
1972 static void horizontal_filter(unsigned char *first_pixel, int stride,
1973 int *bounding_values)
1978 for (end= first_pixel + 8*stride; first_pixel < end; first_pixel += stride) {
1980 (first_pixel[-2] - first_pixel[ 1])
1981 +3*(first_pixel[ 0] - first_pixel[-1]);
1982 filter_value = bounding_values[(filter_value + 4) >> 3];
1983 first_pixel[-1] = clip_uint8(first_pixel[-1] + filter_value);
1984 first_pixel[ 0] = clip_uint8(first_pixel[ 0] - filter_value);
1988 static void vertical_filter(unsigned char *first_pixel, int stride,
1989 int *bounding_values)
1993 const int nstride= -stride;
1995 for (end= first_pixel + 8; first_pixel < end; first_pixel++) {
1997 (first_pixel[2 * nstride] - first_pixel[ stride])
1998 +3*(first_pixel[0 ] - first_pixel[nstride]);
1999 filter_value = bounding_values[(filter_value + 4) >> 3];
2000 first_pixel[nstride] = clip_uint8(first_pixel[nstride] + filter_value);
2001 first_pixel[0] = clip_uint8(first_pixel[0] - filter_value);
2005 static void apply_loop_filter(Vp3DecodeContext *s)
2011 unsigned char *plane_data;
2012 int *bounding_values= s->bounding_values_array+127;
2015 int bounding_values_array[256];
2018 /* find the right loop limit value */
2019 for (x = 63; x >= 0; x--) {
2020 if (vp31_ac_scale_factor[x] >= s->quality_index)
2023 filter_limit = vp31_filter_limit_values[s->quality_index];
2025 /* set up the bounding values */
2026 memset(bounding_values_array, 0, 256 * sizeof(int));
2027 for (x = 0; x < filter_limit; x++) {
2028 bounding_values[-x - filter_limit] = -filter_limit + x;
2029 bounding_values[-x] = -x;
2030 bounding_values[x] = x;
2031 bounding_values[x + filter_limit] = filter_limit - x;
2035 for (plane = 0; plane < 3; plane++) {
2038 /* Y plane parameters */
2040 width = s->fragment_width;
2041 height = s->fragment_height;
2042 stride = s->current_frame.linesize[0];
2043 plane_data = s->current_frame.data[0];
2044 } else if (plane == 1) {
2045 /* U plane parameters */
2046 fragment = s->u_fragment_start;
2047 width = s->fragment_width / 2;
2048 height = s->fragment_height / 2;
2049 stride = s->current_frame.linesize[1];
2050 plane_data = s->current_frame.data[1];
2052 /* V plane parameters */
2053 fragment = s->v_fragment_start;
2054 width = s->fragment_width / 2;
2055 height = s->fragment_height / 2;
2056 stride = s->current_frame.linesize[2];
2057 plane_data = s->current_frame.data[2];
2060 for (y = 0; y < height; y++) {
2062 for (x = 0; x < width; x++) {
2064 /* do not perform left edge filter for left columns frags */
2066 (s->all_fragments[fragment].coding_method != MODE_COPY)) {
2068 plane_data + s->all_fragments[fragment].first_pixel - 7*stride,
2069 stride, bounding_values);
2072 /* do not perform top edge filter for top row fragments */
2074 (s->all_fragments[fragment].coding_method != MODE_COPY)) {
2076 plane_data + s->all_fragments[fragment].first_pixel + stride,
2077 stride, bounding_values);
2080 /* do not perform right edge filter for right column
2081 * fragments or if right fragment neighbor is also coded
2082 * in this frame (it will be filtered in next iteration) */
2083 if ((x < width - 1) &&
2084 (s->all_fragments[fragment].coding_method != MODE_COPY) &&
2085 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
2087 plane_data + s->all_fragments[fragment + 1].first_pixel - 7*stride,
2088 stride, bounding_values);
2091 /* do not perform bottom edge filter for bottom row
2092 * fragments or if bottom fragment neighbor is also coded
2093 * in this frame (it will be filtered in the next row) */
2094 if ((y < height - 1) &&
2095 (s->all_fragments[fragment].coding_method != MODE_COPY) &&
2096 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
2098 plane_data + s->all_fragments[fragment + width].first_pixel + stride,
2099 stride, bounding_values);
2103 STOP_TIMER("loop filter")
2110 * This function computes the first pixel addresses for each fragment.
2111 * This function needs to be invoked after the first frame is allocated
2112 * so that it has access to the plane strides.
2114 static void vp3_calculate_pixel_addresses(Vp3DecodeContext *s)
2119 /* figure out the first pixel addresses for each of the fragments */
2122 for (y = s->fragment_height; y > 0; y--) {
2123 for (x = 0; x < s->fragment_width; x++) {
2124 s->all_fragments[i++].first_pixel =
2125 s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
2126 s->golden_frame.linesize[0] +
2127 x * FRAGMENT_PIXELS;
2128 debug_init(" fragment %d, first pixel @ %d\n",
2129 i-1, s->all_fragments[i-1].first_pixel);
2134 i = s->u_fragment_start;
2135 for (y = s->fragment_height / 2; y > 0; y--) {
2136 for (x = 0; x < s->fragment_width / 2; x++) {
2137 s->all_fragments[i++].first_pixel =
2138 s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
2139 s->golden_frame.linesize[1] +
2140 x * FRAGMENT_PIXELS;
2141 debug_init(" fragment %d, first pixel @ %d\n",
2142 i-1, s->all_fragments[i-1].first_pixel);
2147 i = s->v_fragment_start;
2148 for (y = s->fragment_height / 2; y > 0; y--) {
2149 for (x = 0; x < s->fragment_width / 2; x++) {
2150 s->all_fragments[i++].first_pixel =
2151 s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
2152 s->golden_frame.linesize[2] +
2153 x * FRAGMENT_PIXELS;
2154 debug_init(" fragment %d, first pixel @ %d\n",
2155 i-1, s->all_fragments[i-1].first_pixel);
2160 /* FIXME: this should be merged with the above! */
2161 static void theora_calculate_pixel_addresses(Vp3DecodeContext *s)
2166 /* figure out the first pixel addresses for each of the fragments */
2169 for (y = 1; y <= s->fragment_height; y++) {
2170 for (x = 0; x < s->fragment_width; x++) {
2171 s->all_fragments[i++].first_pixel =
2172 s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
2173 s->golden_frame.linesize[0] +
2174 x * FRAGMENT_PIXELS;
2175 debug_init(" fragment %d, first pixel @ %d\n",
2176 i-1, s->all_fragments[i-1].first_pixel);
2181 i = s->u_fragment_start;
2182 for (y = 1; y <= s->fragment_height / 2; y++) {
2183 for (x = 0; x < s->fragment_width / 2; x++) {
2184 s->all_fragments[i++].first_pixel =
2185 s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
2186 s->golden_frame.linesize[1] +
2187 x * FRAGMENT_PIXELS;
2188 debug_init(" fragment %d, first pixel @ %d\n",
2189 i-1, s->all_fragments[i-1].first_pixel);
2194 i = s->v_fragment_start;
2195 for (y = 1; y <= s->fragment_height / 2; y++) {
2196 for (x = 0; x < s->fragment_width / 2; x++) {
2197 s->all_fragments[i++].first_pixel =
2198 s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
2199 s->golden_frame.linesize[2] +
2200 x * FRAGMENT_PIXELS;
2201 debug_init(" fragment %d, first pixel @ %d\n",
2202 i-1, s->all_fragments[i-1].first_pixel);
2208 * This is the ffmpeg/libavcodec API init function.
2210 static int vp3_decode_init(AVCodecContext *avctx)
2212 Vp3DecodeContext *s = avctx->priv_data;
2216 int y_superblock_count;
2217 int c_superblock_count;
2219 if (avctx->codec_tag == MKTAG('V','P','3','0'))
2225 s->width = (avctx->width + 15) & 0xFFFFFFF0;
2226 s->height = (avctx->height + 15) & 0xFFFFFFF0;
2227 avctx->pix_fmt = PIX_FMT_YUV420P;
2228 avctx->has_b_frames = 0;
2229 if(avctx->idct_algo==FF_IDCT_AUTO)
2230 avctx->idct_algo=FF_IDCT_VP3;
2231 dsputil_init(&s->dsp, avctx);
2233 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
2235 /* initialize to an impossible value which will force a recalculation
2236 * in the first frame decode */
2237 s->quality_index = -1;
2239 s->y_superblock_width = (s->width + 31) / 32;
2240 s->y_superblock_height = (s->height + 31) / 32;
2241 y_superblock_count = s->y_superblock_width * s->y_superblock_height;
2243 /* work out the dimensions for the C planes */
2244 c_width = s->width / 2;
2245 c_height = s->height / 2;
2246 s->c_superblock_width = (c_width + 31) / 32;
2247 s->c_superblock_height = (c_height + 31) / 32;
2248 c_superblock_count = s->c_superblock_width * s->c_superblock_height;
2250 s->superblock_count = y_superblock_count + (c_superblock_count * 2);
2251 s->u_superblock_start = y_superblock_count;
2252 s->v_superblock_start = s->u_superblock_start + c_superblock_count;
2253 s->superblock_coding = av_malloc(s->superblock_count);
2255 s->macroblock_width = (s->width + 15) / 16;
2256 s->macroblock_height = (s->height + 15) / 16;
2257 s->macroblock_count = s->macroblock_width * s->macroblock_height;
2259 s->fragment_width = s->width / FRAGMENT_PIXELS;
2260 s->fragment_height = s->height / FRAGMENT_PIXELS;
2262 /* fragment count covers all 8x8 blocks for all 3 planes */
2263 s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
2264 s->u_fragment_start = s->fragment_width * s->fragment_height;
2265 s->v_fragment_start = s->fragment_width * s->fragment_height * 5 / 4;
2267 debug_init(" Y plane: %d x %d\n", s->width, s->height);
2268 debug_init(" C plane: %d x %d\n", c_width, c_height);
2269 debug_init(" Y superblocks: %d x %d, %d total\n",
2270 s->y_superblock_width, s->y_superblock_height, y_superblock_count);
2271 debug_init(" C superblocks: %d x %d, %d total\n",
2272 s->c_superblock_width, s->c_superblock_height, c_superblock_count);
2273 debug_init(" total superblocks = %d, U starts @ %d, V starts @ %d\n",
2274 s->superblock_count, s->u_superblock_start, s->v_superblock_start);
2275 debug_init(" macroblocks: %d x %d, %d total\n",
2276 s->macroblock_width, s->macroblock_height, s->macroblock_count);
2277 debug_init(" %d fragments, %d x %d, u starts @ %d, v starts @ %d\n",
2281 s->u_fragment_start,
2282 s->v_fragment_start);
2284 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
2285 s->coeffs = av_malloc(s->fragment_count * sizeof(Coeff) * 65);
2286 s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int));
2287 s->pixel_addresses_inited = 0;
2289 if (!s->theora_tables)
2291 for (i = 0; i < 64; i++)
2292 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
2293 for (i = 0; i < 64; i++)
2294 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
2295 for (i = 0; i < 64; i++)
2296 s->coded_intra_y_dequant[i] = vp31_intra_y_dequant[i];
2297 for (i = 0; i < 64; i++)
2298 s->coded_intra_c_dequant[i] = vp31_intra_c_dequant[i];
2299 for (i = 0; i < 64; i++)
2300 s->coded_inter_dequant[i] = vp31_inter_dequant[i];
2301 for (i = 0; i < 64; i++)
2302 s->filter_limit_values[i] = vp31_filter_limit_values[i];
2304 /* init VLC tables */
2305 for (i = 0; i < 16; i++) {
2308 init_vlc(&s->dc_vlc[i], 5, 32,
2309 &dc_bias[i][0][1], 4, 2,
2310 &dc_bias[i][0][0], 4, 2, 0);
2312 /* group 1 AC histograms */
2313 init_vlc(&s->ac_vlc_1[i], 5, 32,
2314 &ac_bias_0[i][0][1], 4, 2,
2315 &ac_bias_0[i][0][0], 4, 2, 0);
2317 /* group 2 AC histograms */
2318 init_vlc(&s->ac_vlc_2[i], 5, 32,
2319 &ac_bias_1[i][0][1], 4, 2,
2320 &ac_bias_1[i][0][0], 4, 2, 0);
2322 /* group 3 AC histograms */
2323 init_vlc(&s->ac_vlc_3[i], 5, 32,
2324 &ac_bias_2[i][0][1], 4, 2,
2325 &ac_bias_2[i][0][0], 4, 2, 0);
2327 /* group 4 AC histograms */
2328 init_vlc(&s->ac_vlc_4[i], 5, 32,
2329 &ac_bias_3[i][0][1], 4, 2,
2330 &ac_bias_3[i][0][0], 4, 2, 0);
2333 for (i = 0; i < 16; i++) {
2336 init_vlc(&s->dc_vlc[i], 5, 32,
2337 &s->huffman_table[i][0][1], 4, 2,
2338 &s->huffman_table[i][0][0], 4, 2, 0);
2340 /* group 1 AC histograms */
2341 init_vlc(&s->ac_vlc_1[i], 5, 32,
2342 &s->huffman_table[i+16][0][1], 4, 2,
2343 &s->huffman_table[i+16][0][0], 4, 2, 0);
2345 /* group 2 AC histograms */
2346 init_vlc(&s->ac_vlc_2[i], 5, 32,
2347 &s->huffman_table[i+16*2][0][1], 4, 2,
2348 &s->huffman_table[i+16*2][0][0], 4, 2, 0);
2350 /* group 3 AC histograms */
2351 init_vlc(&s->ac_vlc_3[i], 5, 32,
2352 &s->huffman_table[i+16*3][0][1], 4, 2,
2353 &s->huffman_table[i+16*3][0][0], 4, 2, 0);
2355 /* group 4 AC histograms */
2356 init_vlc(&s->ac_vlc_4[i], 5, 32,
2357 &s->huffman_table[i+16*4][0][1], 4, 2,
2358 &s->huffman_table[i+16*4][0][0], 4, 2, 0);
2362 init_vlc(&s->superblock_run_length_vlc, 6, 34,
2363 &superblock_run_length_vlc_table[0][1], 4, 2,
2364 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
2366 init_vlc(&s->fragment_run_length_vlc, 5, 30,
2367 &fragment_run_length_vlc_table[0][1], 4, 2,
2368 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
2370 init_vlc(&s->mode_code_vlc, 3, 8,
2371 &mode_code_vlc_table[0][1], 2, 1,
2372 &mode_code_vlc_table[0][0], 2, 1, 0);
2374 init_vlc(&s->motion_vector_vlc, 6, 63,
2375 &motion_vector_vlc_table[0][1], 2, 1,
2376 &motion_vector_vlc_table[0][0], 2, 1, 0);
2378 /* work out the block mapping tables */
2379 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
2380 s->superblock_macroblocks = av_malloc(s->superblock_count * 4 * sizeof(int));
2381 s->macroblock_fragments = av_malloc(s->macroblock_count * 6 * sizeof(int));
2382 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
2383 init_block_mapping(s);
2385 for (i = 0; i < 3; i++) {
2386 s->current_frame.data[i] = NULL;
2387 s->last_frame.data[i] = NULL;
2388 s->golden_frame.data[i] = NULL;
2395 * This is the ffmpeg/libavcodec API frame decode function.
2397 static int vp3_decode_frame(AVCodecContext *avctx,
2398 void *data, int *data_size,
2399 uint8_t *buf, int buf_size)
2401 Vp3DecodeContext *s = avctx->priv_data;
2403 static int counter = 0;
2406 init_get_bits(&gb, buf, buf_size * 8);
2408 if (s->theora && get_bits1(&gb))
2411 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
2414 int ptype = get_bits(&gb, 7);
2416 skip_bits(&gb, 6*8); /* "theora" */
2421 theora_decode_comments(avctx, gb);
2424 theora_decode_tables(avctx, gb);
2425 init_dequantizer(s);
2428 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype);
2434 s->keyframe = !get_bits1(&gb);
2437 s->last_quality_index = s->quality_index;
2438 s->quality_index = get_bits(&gb, 6);
2439 if (s->theora >= 0x030200)
2442 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
2443 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
2444 s->keyframe?"key":"", counter, s->quality_index);
2447 if (s->quality_index != s->last_quality_index) {
2448 init_dequantizer(s);
2449 init_loop_filter(s);
2455 skip_bits(&gb, 4); /* width code */
2456 skip_bits(&gb, 4); /* height code */
2459 s->version = get_bits(&gb, 5);
2461 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
2464 if (s->version || s->theora)
2467 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
2468 skip_bits(&gb, 2); /* reserved? */
2471 if (s->last_frame.data[0] == s->golden_frame.data[0]) {
2472 if (s->golden_frame.data[0])
2473 avctx->release_buffer(avctx, &s->golden_frame);
2474 s->last_frame= s->golden_frame; /* ensure that we catch any access to this released frame */
2476 if (s->golden_frame.data[0])
2477 avctx->release_buffer(avctx, &s->golden_frame);
2478 if (s->last_frame.data[0])
2479 avctx->release_buffer(avctx, &s->last_frame);
2482 s->golden_frame.reference = 3;
2483 if(avctx->get_buffer(avctx, &s->golden_frame) < 0) {
2484 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
2488 /* golden frame is also the current frame */
2489 memcpy(&s->current_frame, &s->golden_frame, sizeof(AVFrame));
2491 /* time to figure out pixel addresses? */
2492 if (!s->pixel_addresses_inited)
2494 if (!s->flipped_image)
2495 vp3_calculate_pixel_addresses(s);
2497 theora_calculate_pixel_addresses(s);
2500 /* allocate a new current frame */
2501 s->current_frame.reference = 3;
2502 if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
2503 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
2508 s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
2509 s->current_frame.qstride= 0;
2513 STOP_TIMER("init_frame")}
2518 memcpy(s->current_frame.data[0], s->golden_frame.data[0],
2519 s->current_frame.linesize[0] * s->height);
2520 memcpy(s->current_frame.data[1], s->golden_frame.data[1],
2521 s->current_frame.linesize[1] * s->height / 2);
2522 memcpy(s->current_frame.data[2], s->golden_frame.data[2],
2523 s->current_frame.linesize[2] * s->height / 2);
2529 if (unpack_superblocks(s, &gb)){
2530 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
2533 STOP_TIMER("unpack_superblocks")}
2535 if (unpack_modes(s, &gb)){
2536 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
2539 STOP_TIMER("unpack_modes")}
2541 if (unpack_vectors(s, &gb)){
2542 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2545 STOP_TIMER("unpack_vectors")}
2547 if (unpack_dct_coeffs(s, &gb)){
2548 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
2551 STOP_TIMER("unpack_dct_coeffs")}
2554 reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
2555 if ((avctx->flags & CODEC_FLAG_GRAY) == 0) {
2556 reverse_dc_prediction(s, s->u_fragment_start,
2557 s->fragment_width / 2, s->fragment_height / 2);
2558 reverse_dc_prediction(s, s->v_fragment_start,
2559 s->fragment_width / 2, s->fragment_height / 2);
2561 STOP_TIMER("reverse_dc_prediction")}
2564 for (i = 0; i < s->macroblock_height; i++)
2566 STOP_TIMER("render_fragments")}
2569 apply_loop_filter(s);
2570 STOP_TIMER("apply_loop_filter")}
2575 *data_size=sizeof(AVFrame);
2576 *(AVFrame*)data= s->current_frame;
2578 /* release the last frame, if it is allocated and if it is not the
2580 if ((s->last_frame.data[0]) &&
2581 (s->last_frame.data[0] != s->golden_frame.data[0]))
2582 avctx->release_buffer(avctx, &s->last_frame);
2584 /* shuffle frames (last = current) */
2585 memcpy(&s->last_frame, &s->current_frame, sizeof(AVFrame));
2586 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
2592 * This is the ffmpeg/libavcodec API module cleanup function.
2594 static int vp3_decode_end(AVCodecContext *avctx)
2596 Vp3DecodeContext *s = avctx->priv_data;
2598 av_free(s->all_fragments);
2600 av_free(s->coded_fragment_list);
2601 av_free(s->superblock_fragments);
2602 av_free(s->superblock_macroblocks);
2603 av_free(s->macroblock_fragments);
2604 av_free(s->macroblock_coding);
2606 /* release all frames */
2607 if (s->golden_frame.data[0] && s->golden_frame.data[0] != s->last_frame.data[0])
2608 avctx->release_buffer(avctx, &s->golden_frame);
2609 if (s->last_frame.data[0])
2610 avctx->release_buffer(avctx, &s->last_frame);
2611 /* no need to release the current_frame since it will always be pointing
2612 * to the same frame as either the golden or last frame */
2617 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2619 Vp3DecodeContext *s = avctx->priv_data;
2621 if (get_bits(gb, 1)) {
2623 if (s->entries >= 32) { /* overflow */
2624 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2627 token = get_bits(gb, 5);
2628 //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);
2629 s->huffman_table[s->hti][token][0] = s->hbits;
2630 s->huffman_table[s->hti][token][1] = s->huff_code_size;
2634 if (s->huff_code_size >= 32) {/* overflow */
2635 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2638 s->huff_code_size++;
2640 read_huffman_tree(avctx, gb);
2642 read_huffman_tree(avctx, gb);
2644 s->huff_code_size--;
2649 static int theora_decode_header(AVCodecContext *avctx, GetBitContext gb)
2651 Vp3DecodeContext *s = avctx->priv_data;
2652 int major, minor, micro;
2654 major = get_bits(&gb, 8); /* version major */
2655 minor = get_bits(&gb, 8); /* version minor */
2656 micro = get_bits(&gb, 8); /* version micro */
2657 av_log(avctx, AV_LOG_INFO, "Theora bitstream version %d.%d.%d\n",
2658 major, minor, micro);
2660 /* FIXME: endianess? */
2661 s->theora = (major << 16) | (minor << 8) | micro;
2663 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2664 /* but previous versions have the image flipped relative to vp3 */
2665 if (s->theora < 0x030200)
2667 s->flipped_image = 1;
2668 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2671 s->width = get_bits(&gb, 16) << 4;
2672 s->height = get_bits(&gb, 16) << 4;
2674 if(avcodec_check_dimensions(avctx, s->width, s->height)){
2675 av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2676 s->width= s->height= 0;
2680 if (s->theora >= 0x030400)
2682 skip_bits(&gb, 32); /* total number of superblocks in a frame */
2683 // fixme, the next field is 36bits long
2684 skip_bits(&gb, 32); /* total number of blocks in a frame */
2685 skip_bits(&gb, 4); /* total number of blocks in a frame */
2686 skip_bits(&gb, 32); /* total number of macroblocks in a frame */
2688 skip_bits(&gb, 24); /* frame width */
2689 skip_bits(&gb, 24); /* frame height */
2693 skip_bits(&gb, 24); /* frame width */
2694 skip_bits(&gb, 24); /* frame height */
2697 skip_bits(&gb, 8); /* offset x */
2698 skip_bits(&gb, 8); /* offset y */
2700 skip_bits(&gb, 32); /* fps numerator */
2701 skip_bits(&gb, 32); /* fps denumerator */
2702 skip_bits(&gb, 24); /* aspect numerator */
2703 skip_bits(&gb, 24); /* aspect denumerator */
2705 if (s->theora < 0x030200)
2706 skip_bits(&gb, 5); /* keyframe frequency force */
2707 skip_bits(&gb, 8); /* colorspace */
2708 if (s->theora >= 0x030400)
2709 skip_bits(&gb, 2); /* pixel format: 420,res,422,444 */
2710 skip_bits(&gb, 24); /* bitrate */
2712 skip_bits(&gb, 6); /* quality hint */
2714 if (s->theora >= 0x030200)
2716 skip_bits(&gb, 5); /* keyframe frequency force */
2718 if (s->theora < 0x030400)
2719 skip_bits(&gb, 5); /* spare bits */
2722 // align_get_bits(&gb);
2724 avctx->width = s->width;
2725 avctx->height = s->height;
2730 static inline int theora_get_32bit(GetBitContext gb)
2732 int ret = get_bits(&gb, 8);
2733 ret += get_bits(&gb, 8) << 8;
2734 ret += get_bits(&gb, 8) << 16;
2735 ret += get_bits(&gb, 8) << 24;
2740 static int theora_decode_comments(AVCodecContext *avctx, GetBitContext gb)
2742 Vp3DecodeContext *s = avctx->priv_data;
2745 if (s->theora <= 0x030200)
2750 len = get_bits_long(&gb, 32);
2751 len = le2me_32(len);
2756 comments = get_bits_long(&gb, 32);
2757 comments = le2me_32(comments);
2758 for (i = 0; i < comments; i++)
2760 len = get_bits_long(&gb, 32);
2761 len = be2me_32(len);
2769 len = get_bits_long(&gb, 32);
2770 len = le2me_32(len);
2780 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext gb)
2782 Vp3DecodeContext *s = avctx->priv_data;
2785 if (s->theora >= 0x030200) {
2786 n = get_bits(&gb, 3);
2787 /* loop filter limit values table */
2788 for (i = 0; i < 64; i++)
2789 s->filter_limit_values[i] = get_bits(&gb, n);
2792 if (s->theora >= 0x030200)
2793 n = get_bits(&gb, 4) + 1;
2796 /* quality threshold table */
2797 for (i = 0; i < 64; i++)
2798 s->coded_ac_scale_factor[i] = get_bits(&gb, n);
2800 if (s->theora >= 0x030200)
2801 n = get_bits(&gb, 4) + 1;
2804 /* dc scale factor table */
2805 for (i = 0; i < 64; i++)
2806 s->coded_dc_scale_factor[i] = get_bits(&gb, n);
2808 if (s->theora >= 0x030200)
2809 matrices = get_bits(&gb, 9) + 1;
2812 if (matrices != 3) {
2813 av_log(avctx,AV_LOG_ERROR, "unsupported matrices: %d\n", matrices);
2817 for (i = 0; i < 64; i++)
2818 s->coded_intra_y_dequant[i] = get_bits(&gb, 8);
2821 for (i = 0; i < 64; i++)
2822 s->coded_intra_c_dequant[i] = get_bits(&gb, 8);
2825 for (i = 0; i < 64; i++)
2826 s->coded_inter_dequant[i] = get_bits(&gb, 8);
2828 /* skip unknown matrices */
2831 for (i = 0; i < 64; i++)
2834 for (i = 0; i <= 1; i++) {
2835 for (n = 0; n <= 2; n++) {
2838 newqr = get_bits(&gb, 1);
2847 skip_bits(&gb, av_log2(matrices-1)+1);
2849 qi += get_bits(&gb, av_log2(63-qi)+1) + 1;
2850 skip_bits(&gb, av_log2(matrices-1)+1);
2853 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2860 /* Huffman tables */
2861 for (s->hti = 0; s->hti < 80; s->hti++) {
2863 s->huff_code_size = 1;
2864 if (!get_bits(&gb, 1)) {
2866 read_huffman_tree(avctx, &gb);
2868 read_huffman_tree(avctx, &gb);
2872 s->theora_tables = 1;
2877 static int theora_decode_init(AVCodecContext *avctx)
2879 Vp3DecodeContext *s = avctx->priv_data;
2882 uint8_t *p= avctx->extradata;
2887 if (!avctx->extradata_size)
2889 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2894 op_bytes = *(p++)<<8;
2897 init_get_bits(&gb, p, op_bytes);
2900 ptype = get_bits(&gb, 8);
2901 debug_vp3("Theora headerpacket type: %x\n", ptype);
2903 if (!(ptype & 0x80))
2905 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2909 // FIXME: check for this aswell
2910 skip_bits(&gb, 6*8); /* "theora" */
2915 theora_decode_header(avctx, gb);
2918 // FIXME: is this needed? it breaks sometimes
2919 // theora_decode_comments(avctx, gb);
2922 theora_decode_tables(avctx, gb);
2925 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2930 vp3_decode_init(avctx);
2934 AVCodec vp3_decoder = {
2938 sizeof(Vp3DecodeContext),
2947 #ifndef CONFIG_LIBTHEORA
2948 AVCodec theora_decoder = {
2952 sizeof(Vp3DecodeContext),