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 printf
80 static inline void debug_vp3(const char *format, ...) { }
84 #define debug_init printf
86 static inline void debug_init(const char *format, ...) { }
89 #if DEBUG_DEQUANTIZERS
90 #define debug_dequantizers printf
92 static inline void debug_dequantizers(const char *format, ...) { }
95 #if DEBUG_BLOCK_CODING
96 #define debug_block_coding printf
98 static inline void debug_block_coding(const char *format, ...) { }
102 #define debug_modes printf
104 static inline void debug_modes(const char *format, ...) { }
108 #define debug_vectors printf
110 static inline void debug_vectors(const char *format, ...) { }
114 #define debug_token printf
116 static inline void debug_token(const char *format, ...) { }
120 #define debug_vlc printf
122 static inline void debug_vlc(const char *format, ...) { }
126 #define debug_dc_pred printf
128 static inline void debug_dc_pred(const char *format, ...) { }
132 #define debug_idct printf
134 static inline void debug_idct(const char *format, ...) { }
137 typedef struct Vp3Fragment {
139 /* address of first pixel taking into account which plane the fragment
140 * lives on as well as the plane stride */
142 /* this is the macroblock that the fragment belongs to */
144 uint8_t coding_method;
151 #define SB_NOT_CODED 0
152 #define SB_PARTIALLY_CODED 1
153 #define SB_FULLY_CODED 2
155 #define MODE_INTER_NO_MV 0
157 #define MODE_INTER_PLUS_MV 2
158 #define MODE_INTER_LAST_MV 3
159 #define MODE_INTER_PRIOR_LAST 4
160 #define MODE_USING_GOLDEN 5
161 #define MODE_GOLDEN_MV 6
162 #define MODE_INTER_FOURMV 7
163 #define CODING_MODE_COUNT 8
165 /* special internal mode */
168 /* There are 6 preset schemes, plus a free-form scheme */
169 static int ModeAlphabet[7][CODING_MODE_COUNT] =
171 /* this is the custom scheme */
172 { 0, 0, 0, 0, 0, 0, 0, 0 },
174 /* scheme 1: Last motion vector dominates */
175 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
176 MODE_INTER_PLUS_MV, MODE_INTER_NO_MV,
177 MODE_INTRA, MODE_USING_GOLDEN,
178 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
181 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
182 MODE_INTER_NO_MV, MODE_INTER_PLUS_MV,
183 MODE_INTRA, MODE_USING_GOLDEN,
184 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
187 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
188 MODE_INTER_PRIOR_LAST, MODE_INTER_NO_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_NO_MV, MODE_INTER_PRIOR_LAST,
195 MODE_INTRA, MODE_USING_GOLDEN,
196 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
198 /* scheme 5: No motion vector dominates */
199 { MODE_INTER_NO_MV, MODE_INTER_LAST_MV,
200 MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
201 MODE_INTRA, MODE_USING_GOLDEN,
202 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
205 { MODE_INTER_NO_MV, MODE_USING_GOLDEN,
206 MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
207 MODE_INTER_PLUS_MV, MODE_INTRA,
208 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
212 #define MIN_DEQUANT_VAL 2
214 typedef struct Vp3DecodeContext {
215 AVCodecContext *avctx;
216 int theora, theora_tables;
219 AVFrame golden_frame;
221 AVFrame current_frame;
227 int last_quality_index;
229 int superblock_count;
230 int superblock_width;
231 int superblock_height;
232 int y_superblock_width;
233 int y_superblock_height;
234 int c_superblock_width;
235 int c_superblock_height;
236 int u_superblock_start;
237 int v_superblock_start;
238 unsigned char *superblock_coding;
240 int macroblock_count;
241 int macroblock_width;
242 int macroblock_height;
248 Vp3Fragment *all_fragments;
250 int u_fragment_start;
251 int v_fragment_start;
256 uint16_t coded_dc_scale_factor[64];
257 uint32_t coded_ac_scale_factor[64];
258 uint16_t coded_intra_y_dequant[64];
259 uint16_t coded_intra_c_dequant[64];
260 uint16_t coded_inter_dequant[64];
262 /* this is a list of indices into the all_fragments array indicating
263 * which of the fragments are coded */
264 int *coded_fragment_list;
265 int coded_fragment_list_index;
266 int pixel_addresses_inited;
274 VLC superblock_run_length_vlc;
275 VLC fragment_run_length_vlc;
277 VLC motion_vector_vlc;
279 /* these arrays need to be on 16-byte boundaries since SSE2 operations
281 int16_t __align16 intra_y_dequant[64];
282 int16_t __align16 intra_c_dequant[64];
283 int16_t __align16 inter_dequant[64];
285 /* This table contains superblock_count * 16 entries. Each set of 16
286 * numbers corresponds to the fragment indices 0..15 of the superblock.
287 * An entry will be -1 to indicate that no entry corresponds to that
289 int *superblock_fragments;
291 /* This table contains superblock_count * 4 entries. Each set of 4
292 * numbers corresponds to the macroblock indices 0..3 of the superblock.
293 * An entry will be -1 to indicate that no entry corresponds to that
295 int *superblock_macroblocks;
297 /* This table contains macroblock_count * 6 entries. Each set of 6
298 * numbers corresponds to the fragment indices 0..5 which comprise
299 * the macroblock (4 Y fragments and 2 C fragments). */
300 int *macroblock_fragments;
301 /* This is an array that indicates how a particular macroblock
303 unsigned char *macroblock_coding;
305 int first_coded_y_fragment;
306 int first_coded_c_fragment;
307 int last_coded_y_fragment;
308 int last_coded_c_fragment;
310 uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
311 uint8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
314 static int theora_decode_comments(AVCodecContext *avctx, GetBitContext gb);
315 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext gb);
317 /************************************************************************
318 * VP3 specific functions
319 ************************************************************************/
322 * This function sets up all of the various blocks mappings:
323 * superblocks <-> fragments, macroblocks <-> fragments,
324 * superblocks <-> macroblocks
326 * Returns 0 is successful; returns 1 if *anything* went wrong.
328 static int init_block_mapping(Vp3DecodeContext *s)
331 signed int hilbert_walk_y[16];
332 signed int hilbert_walk_c[16];
333 signed int hilbert_walk_mb[4];
335 int current_fragment = 0;
336 int current_width = 0;
337 int current_height = 0;
340 int superblock_row_inc = 0;
342 int mapping_index = 0;
344 int current_macroblock;
347 signed char travel_width[16] = {
354 signed char travel_height[16] = {
361 signed char travel_width_mb[4] = {
365 signed char travel_height_mb[4] = {
369 debug_vp3(" vp3: initialize block mapping tables\n");
371 /* figure out hilbert pattern per these frame dimensions */
372 hilbert_walk_y[0] = 1;
373 hilbert_walk_y[1] = 1;
374 hilbert_walk_y[2] = s->fragment_width;
375 hilbert_walk_y[3] = -1;
376 hilbert_walk_y[4] = s->fragment_width;
377 hilbert_walk_y[5] = s->fragment_width;
378 hilbert_walk_y[6] = 1;
379 hilbert_walk_y[7] = -s->fragment_width;
380 hilbert_walk_y[8] = 1;
381 hilbert_walk_y[9] = s->fragment_width;
382 hilbert_walk_y[10] = 1;
383 hilbert_walk_y[11] = -s->fragment_width;
384 hilbert_walk_y[12] = -s->fragment_width;
385 hilbert_walk_y[13] = -1;
386 hilbert_walk_y[14] = -s->fragment_width;
387 hilbert_walk_y[15] = 1;
389 hilbert_walk_c[0] = 1;
390 hilbert_walk_c[1] = 1;
391 hilbert_walk_c[2] = s->fragment_width / 2;
392 hilbert_walk_c[3] = -1;
393 hilbert_walk_c[4] = s->fragment_width / 2;
394 hilbert_walk_c[5] = s->fragment_width / 2;
395 hilbert_walk_c[6] = 1;
396 hilbert_walk_c[7] = -s->fragment_width / 2;
397 hilbert_walk_c[8] = 1;
398 hilbert_walk_c[9] = s->fragment_width / 2;
399 hilbert_walk_c[10] = 1;
400 hilbert_walk_c[11] = -s->fragment_width / 2;
401 hilbert_walk_c[12] = -s->fragment_width / 2;
402 hilbert_walk_c[13] = -1;
403 hilbert_walk_c[14] = -s->fragment_width / 2;
404 hilbert_walk_c[15] = 1;
406 hilbert_walk_mb[0] = 1;
407 hilbert_walk_mb[1] = s->macroblock_width;
408 hilbert_walk_mb[2] = 1;
409 hilbert_walk_mb[3] = -s->macroblock_width;
411 /* iterate through each superblock (all planes) and map the fragments */
412 for (i = 0; i < s->superblock_count; i++) {
413 debug_init(" superblock %d (u starts @ %d, v starts @ %d)\n",
414 i, s->u_superblock_start, s->v_superblock_start);
416 /* time to re-assign the limits? */
419 /* start of Y superblocks */
420 right_edge = s->fragment_width;
421 bottom_edge = s->fragment_height;
424 superblock_row_inc = 3 * s->fragment_width -
425 (s->y_superblock_width * 4 - s->fragment_width);
426 hilbert = hilbert_walk_y;
428 /* the first operation for this variable is to advance by 1 */
429 current_fragment = -1;
431 } else if (i == s->u_superblock_start) {
433 /* start of U superblocks */
434 right_edge = s->fragment_width / 2;
435 bottom_edge = s->fragment_height / 2;
438 superblock_row_inc = 3 * (s->fragment_width / 2) -
439 (s->c_superblock_width * 4 - s->fragment_width / 2);
440 hilbert = hilbert_walk_c;
442 /* the first operation for this variable is to advance by 1 */
443 current_fragment = s->u_fragment_start - 1;
445 } else if (i == s->v_superblock_start) {
447 /* start of V superblocks */
448 right_edge = s->fragment_width / 2;
449 bottom_edge = s->fragment_height / 2;
452 superblock_row_inc = 3 * (s->fragment_width / 2) -
453 (s->c_superblock_width * 4 - s->fragment_width / 2);
454 hilbert = hilbert_walk_c;
456 /* the first operation for this variable is to advance by 1 */
457 current_fragment = s->v_fragment_start - 1;
461 if (current_width >= right_edge - 1) {
462 /* reset width and move to next superblock row */
466 /* fragment is now at the start of a new superblock row */
467 current_fragment += superblock_row_inc;
470 /* iterate through all 16 fragments in a superblock */
471 for (j = 0; j < 16; j++) {
472 current_fragment += hilbert[j];
473 current_width += travel_width[j];
474 current_height += travel_height[j];
476 /* check if the fragment is in bounds */
477 if ((current_width < right_edge) &&
478 (current_height < bottom_edge)) {
479 s->superblock_fragments[mapping_index] = current_fragment;
480 debug_init(" mapping fragment %d to superblock %d, position %d (%d/%d x %d/%d)\n",
481 s->superblock_fragments[mapping_index], i, j,
482 current_width, right_edge, current_height, bottom_edge);
484 s->superblock_fragments[mapping_index] = -1;
485 debug_init(" superblock %d, position %d has no fragment (%d/%d x %d/%d)\n",
487 current_width, right_edge, current_height, bottom_edge);
494 /* initialize the superblock <-> macroblock mapping; iterate through
495 * all of the Y plane superblocks to build this mapping */
496 right_edge = s->macroblock_width;
497 bottom_edge = s->macroblock_height;
500 superblock_row_inc = s->macroblock_width -
501 (s->y_superblock_width * 2 - s->macroblock_width);;
502 hilbert = hilbert_walk_mb;
504 current_macroblock = -1;
505 for (i = 0; i < s->u_superblock_start; i++) {
507 if (current_width >= right_edge - 1) {
508 /* reset width and move to next superblock row */
512 /* macroblock is now at the start of a new superblock row */
513 current_macroblock += superblock_row_inc;
516 /* iterate through each potential macroblock in the superblock */
517 for (j = 0; j < 4; j++) {
518 current_macroblock += hilbert_walk_mb[j];
519 current_width += travel_width_mb[j];
520 current_height += travel_height_mb[j];
522 /* check if the macroblock is in bounds */
523 if ((current_width < right_edge) &&
524 (current_height < bottom_edge)) {
525 s->superblock_macroblocks[mapping_index] = current_macroblock;
526 debug_init(" mapping macroblock %d to superblock %d, position %d (%d/%d x %d/%d)\n",
527 s->superblock_macroblocks[mapping_index], i, j,
528 current_width, right_edge, current_height, bottom_edge);
530 s->superblock_macroblocks[mapping_index] = -1;
531 debug_init(" superblock %d, position %d has no macroblock (%d/%d x %d/%d)\n",
533 current_width, right_edge, current_height, bottom_edge);
540 /* initialize the macroblock <-> fragment mapping */
541 current_fragment = 0;
542 current_macroblock = 0;
544 for (i = 0; i < s->fragment_height; i += 2) {
546 for (j = 0; j < s->fragment_width; j += 2) {
548 debug_init(" macroblock %d contains fragments: ", current_macroblock);
549 s->all_fragments[current_fragment].macroblock = current_macroblock;
550 s->macroblock_fragments[mapping_index++] = current_fragment;
551 debug_init("%d ", current_fragment);
553 if (j + 1 < s->fragment_width) {
554 s->all_fragments[current_fragment + 1].macroblock = current_macroblock;
555 s->macroblock_fragments[mapping_index++] = current_fragment + 1;
556 debug_init("%d ", current_fragment + 1);
558 s->macroblock_fragments[mapping_index++] = -1;
560 if (i + 1 < s->fragment_height) {
561 s->all_fragments[current_fragment + s->fragment_width].macroblock =
563 s->macroblock_fragments[mapping_index++] =
564 current_fragment + s->fragment_width;
565 debug_init("%d ", current_fragment + s->fragment_width);
567 s->macroblock_fragments[mapping_index++] = -1;
569 if ((j + 1 < s->fragment_width) && (i + 1 < s->fragment_height)) {
570 s->all_fragments[current_fragment + s->fragment_width + 1].macroblock =
572 s->macroblock_fragments[mapping_index++] =
573 current_fragment + s->fragment_width + 1;
574 debug_init("%d ", current_fragment + s->fragment_width + 1);
576 s->macroblock_fragments[mapping_index++] = -1;
579 c_fragment = s->u_fragment_start +
580 (i * s->fragment_width / 4) + (j / 2);
581 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
582 s->macroblock_fragments[mapping_index++] = c_fragment;
583 debug_init("%d ", c_fragment);
585 c_fragment = s->v_fragment_start +
586 (i * s->fragment_width / 4) + (j / 2);
587 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
588 s->macroblock_fragments[mapping_index++] = c_fragment;
589 debug_init("%d ", c_fragment);
593 if (j + 2 <= s->fragment_width)
594 current_fragment += 2;
597 current_macroblock++;
600 current_fragment += s->fragment_width;
603 return 0; /* successful path out */
607 * This function unpacks a single token (which should be in the range 0..31)
608 * and returns a zero run (number of zero coefficients in current DCT matrix
609 * before next non-zero coefficient), the next DCT coefficient, and the
610 * number of consecutive, non-EOB'd DCT blocks to EOB.
612 static void unpack_token(GetBitContext *gb, int token, int *zero_run,
613 DCTELEM *coeff, int *eob_run)
621 debug_token(" vp3 token %d: ", token);
625 debug_token("DCT_EOB_TOKEN, EOB next block\n");
630 debug_token("DCT_EOB_PAIR_TOKEN, EOB next 2 blocks\n");
635 debug_token("DCT_EOB_TRIPLE_TOKEN, EOB next 3 blocks\n");
640 debug_token("DCT_REPEAT_RUN_TOKEN, ");
641 *eob_run = get_bits(gb, 2) + 4;
642 debug_token("EOB the next %d blocks\n", *eob_run);
646 debug_token("DCT_REPEAT_RUN2_TOKEN, ");
647 *eob_run = get_bits(gb, 3) + 8;
648 debug_token("EOB the next %d blocks\n", *eob_run);
652 debug_token("DCT_REPEAT_RUN3_TOKEN, ");
653 *eob_run = get_bits(gb, 4) + 16;
654 debug_token("EOB the next %d blocks\n", *eob_run);
658 debug_token("DCT_REPEAT_RUN4_TOKEN, ");
659 *eob_run = get_bits(gb, 12);
660 debug_token("EOB the next %d blocks\n", *eob_run);
664 debug_token("DCT_SHORT_ZRL_TOKEN, ");
665 /* note that this token actually indicates that (3 extra bits) + 1 0s
666 * should be output; this case specifies a run of (3 EBs) 0s and a
667 * coefficient of 0. */
668 *zero_run = get_bits(gb, 3);
670 debug_token("skip the next %d positions in output matrix\n", *zero_run + 1);
674 debug_token("DCT_ZRL_TOKEN, ");
675 /* note that this token actually indicates that (6 extra bits) + 1 0s
676 * should be output; this case specifies a run of (6 EBs) 0s and a
677 * coefficient of 0. */
678 *zero_run = get_bits(gb, 6);
680 debug_token("skip the next %d positions in output matrix\n", *zero_run + 1);
684 debug_token("ONE_TOKEN, output 1\n");
689 debug_token("MINUS_ONE_TOKEN, output -1\n");
694 debug_token("TWO_TOKEN, output 2\n");
699 debug_token("MINUS_TWO_TOKEN, output -2\n");
707 debug_token("LOW_VAL_TOKENS, ");
709 *coeff = -(3 + (token - 13));
711 *coeff = 3 + (token - 13);
712 debug_token("output %d\n", *coeff);
716 debug_token("DCT_VAL_CATEGORY3, ");
717 sign = get_bits(gb, 1);
718 *coeff = 7 + get_bits(gb, 1);
721 debug_token("output %d\n", *coeff);
725 debug_token("DCT_VAL_CATEGORY4, ");
726 sign = get_bits(gb, 1);
727 *coeff = 9 + get_bits(gb, 2);
730 debug_token("output %d\n", *coeff);
734 debug_token("DCT_VAL_CATEGORY5, ");
735 sign = get_bits(gb, 1);
736 *coeff = 13 + get_bits(gb, 3);
739 debug_token("output %d\n", *coeff);
743 debug_token("DCT_VAL_CATEGORY6, ");
744 sign = get_bits(gb, 1);
745 *coeff = 21 + get_bits(gb, 4);
748 debug_token("output %d\n", *coeff);
752 debug_token("DCT_VAL_CATEGORY7, ");
753 sign = get_bits(gb, 1);
754 *coeff = 37 + get_bits(gb, 5);
757 debug_token("output %d\n", *coeff);
761 debug_token("DCT_VAL_CATEGORY8, ");
762 sign = get_bits(gb, 1);
763 *coeff = 69 + get_bits(gb, 9);
766 debug_token("output %d\n", *coeff);
774 debug_token("DCT_RUN_CATEGORY1, ");
775 *zero_run = token - 22;
780 debug_token("output %d 0s, then %d\n", *zero_run, *coeff);
784 debug_token("DCT_RUN_CATEGORY1B, ");
789 *zero_run = 6 + get_bits(gb, 2);
790 debug_token("output %d 0s, then %d\n", *zero_run, *coeff);
794 debug_token("DCT_RUN_CATEGORY1C, ");
799 *zero_run = 10 + get_bits(gb, 3);
800 debug_token("output %d 0s, then %d\n", *zero_run, *coeff);
804 debug_token("DCT_RUN_CATEGORY2, ");
805 sign = get_bits(gb, 1);
806 *coeff = 2 + get_bits(gb, 1);
810 debug_token("output %d 0s, then %d\n", *zero_run, *coeff);
814 debug_token("DCT_RUN_CATEGORY2, ");
815 sign = get_bits(gb, 1);
816 *coeff = 2 + get_bits(gb, 1);
819 *zero_run = 2 + get_bits(gb, 1);
820 debug_token("output %d 0s, then %d\n", *zero_run, *coeff);
824 av_log(NULL, AV_LOG_ERROR, " vp3: help! Got a bad token: %d > 31\n", token);
831 * This function wipes out all of the fragment data.
833 static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
836 static const DCTELEM zero_block[64];
838 /* zero out all of the fragment information */
839 s->coded_fragment_list_index = 0;
840 for (i = 0; i < s->fragment_count; i++) {
841 s->all_fragments[i].coeffs = zero_block;
842 s->all_fragments[i].coeff_count = 0;
843 s->all_fragments[i].last_coeff = -1;
844 s->all_fragments[i].motion_x = 0xbeef;
845 s->all_fragments[i].motion_y = 0xbeef;
850 * This function sets of the dequantization tables used for a particular
853 static void init_dequantizer(Vp3DecodeContext *s)
856 int ac_scale_factor = s->coded_ac_scale_factor[s->quality_index];
857 int dc_scale_factor = s->coded_dc_scale_factor[s->quality_index];
860 debug_vp3(" vp3: initializing dequantization tables\n");
863 * Scale dequantizers:
869 * where sf = dc_scale_factor for DC quantizer
870 * or ac_scale_factor for AC quantizer
872 * Then, saturate the result to a lower limit of MIN_DEQUANT_VAL.
876 /* scale DC quantizers */
877 s->intra_y_dequant[0] = s->coded_intra_y_dequant[0] * dc_scale_factor / 100;
878 if (s->intra_y_dequant[0] < MIN_DEQUANT_VAL * 2)
879 s->intra_y_dequant[0] = MIN_DEQUANT_VAL * 2;
880 s->intra_y_dequant[0] *= SCALER;
882 s->intra_c_dequant[0] = s->coded_intra_c_dequant[0] * dc_scale_factor / 100;
883 if (s->intra_c_dequant[0] < MIN_DEQUANT_VAL * 2)
884 s->intra_c_dequant[0] = MIN_DEQUANT_VAL * 2;
885 s->intra_c_dequant[0] *= SCALER;
887 s->inter_dequant[0] = s->coded_inter_dequant[0] * dc_scale_factor / 100;
888 if (s->inter_dequant[0] < MIN_DEQUANT_VAL * 4)
889 s->inter_dequant[0] = MIN_DEQUANT_VAL * 4;
890 s->inter_dequant[0] *= SCALER;
892 /* scale AC quantizers, zigzag at the same time in preparation for
893 * the dequantization phase */
894 for (i = 1; i < 64; i++) {
895 int k= s->scantable.scantable[i];
896 j = s->scantable.permutated[i];
898 s->intra_y_dequant[j] = s->coded_intra_y_dequant[k] * ac_scale_factor / 100;
899 if (s->intra_y_dequant[j] < MIN_DEQUANT_VAL)
900 s->intra_y_dequant[j] = MIN_DEQUANT_VAL;
901 s->intra_y_dequant[j] *= SCALER;
903 s->intra_c_dequant[j] = s->coded_intra_c_dequant[k] * ac_scale_factor / 100;
904 if (s->intra_c_dequant[j] < MIN_DEQUANT_VAL)
905 s->intra_c_dequant[j] = MIN_DEQUANT_VAL;
906 s->intra_c_dequant[j] *= SCALER;
908 s->inter_dequant[j] = s->coded_inter_dequant[k] * ac_scale_factor / 100;
909 if (s->inter_dequant[j] < MIN_DEQUANT_VAL * 2)
910 s->inter_dequant[j] = MIN_DEQUANT_VAL * 2;
911 s->inter_dequant[j] *= SCALER;
914 memset(s->qscale_table, (FFMAX(s->intra_y_dequant[1], s->intra_c_dequant[1])+8)/16, 512); //FIXME finetune
916 /* print debug information as requested */
917 debug_dequantizers("intra Y dequantizers:\n");
918 for (i = 0; i < 8; i++) {
919 for (j = i * 8; j < i * 8 + 8; j++) {
920 debug_dequantizers(" %4d,", s->intra_y_dequant[j]);
922 debug_dequantizers("\n");
924 debug_dequantizers("\n");
926 debug_dequantizers("intra C dequantizers:\n");
927 for (i = 0; i < 8; i++) {
928 for (j = i * 8; j < i * 8 + 8; j++) {
929 debug_dequantizers(" %4d,", s->intra_c_dequant[j]);
931 debug_dequantizers("\n");
933 debug_dequantizers("\n");
935 debug_dequantizers("interframe dequantizers:\n");
936 for (i = 0; i < 8; i++) {
937 for (j = i * 8; j < i * 8 + 8; j++) {
938 debug_dequantizers(" %4d,", s->inter_dequant[j]);
940 debug_dequantizers("\n");
942 debug_dequantizers("\n");
946 * This function is used to fetch runs of 1s or 0s from the bitstream for
947 * use in determining which superblocks are fully and partially coded.
956 * 111111xxxxxxxxxxxx 34-4129
958 static int get_superblock_run_length(GetBitContext *gb)
961 if (get_bits(gb, 1) == 0)
964 else if (get_bits(gb, 1) == 0)
965 return (2 + get_bits(gb, 1));
967 else if (get_bits(gb, 1) == 0)
968 return (4 + get_bits(gb, 1));
970 else if (get_bits(gb, 1) == 0)
971 return (6 + get_bits(gb, 2));
973 else if (get_bits(gb, 1) == 0)
974 return (10 + get_bits(gb, 3));
976 else if (get_bits(gb, 1) == 0)
977 return (18 + get_bits(gb, 4));
980 return (34 + get_bits(gb, 12));
985 * This function is used to fetch runs of 1s or 0s from the bitstream for
986 * use in determining which particular fragments are coded.
996 static int get_fragment_run_length(GetBitContext *gb)
999 if (get_bits(gb, 1) == 0)
1000 return (1 + get_bits(gb, 1));
1002 else if (get_bits(gb, 1) == 0)
1003 return (3 + get_bits(gb, 1));
1005 else if (get_bits(gb, 1) == 0)
1006 return (5 + get_bits(gb, 1));
1008 else if (get_bits(gb, 1) == 0)
1009 return (7 + get_bits(gb, 2));
1011 else if (get_bits(gb, 1) == 0)
1012 return (11 + get_bits(gb, 2));
1015 return (15 + get_bits(gb, 4));
1020 * This function decodes a VLC from the bitstream and returns a number
1021 * that ranges from 0..7. The number indicates which of the 8 coding
1035 static int get_mode_code(GetBitContext *gb)
1038 if (get_bits(gb, 1) == 0)
1041 else if (get_bits(gb, 1) == 0)
1044 else if (get_bits(gb, 1) == 0)
1047 else if (get_bits(gb, 1) == 0)
1050 else if (get_bits(gb, 1) == 0)
1053 else if (get_bits(gb, 1) == 0)
1056 else if (get_bits(gb, 1) == 0)
1065 * This function extracts a motion vector from the bitstream using a VLC
1066 * scheme. 3 bits are fetched from the bitstream and 1 of 8 actions is
1067 * taken depending on the value on those 3 bits:
1072 * 3: if (next bit is 1) return -2, else return 2
1073 * 4: if (next bit is 1) return -3, else return 3
1074 * 5: return 4 + (next 2 bits), next bit is sign
1075 * 6: return 8 + (next 3 bits), next bit is sign
1076 * 7: return 16 + (next 4 bits), next bit is sign
1078 static int get_motion_vector_vlc(GetBitContext *gb)
1082 bits = get_bits(gb, 3);
1099 if (get_bits(gb, 1) == 0)
1106 if (get_bits(gb, 1) == 0)
1113 bits = 4 + get_bits(gb, 2);
1114 if (get_bits(gb, 1) == 1)
1119 bits = 8 + get_bits(gb, 3);
1120 if (get_bits(gb, 1) == 1)
1125 bits = 16 + get_bits(gb, 4);
1126 if (get_bits(gb, 1) == 1)
1136 * This function fetches a 5-bit number from the stream followed by
1137 * a sign and calls it a motion vector.
1139 static int get_motion_vector_fixed(GetBitContext *gb)
1144 bits = get_bits(gb, 5);
1146 if (get_bits(gb, 1) == 1)
1153 * This function unpacks all of the superblock/macroblock/fragment coding
1154 * information from the bitstream.
1156 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
1159 int current_superblock = 0;
1160 int current_run = 0;
1161 int decode_fully_flags = 0;
1162 int decode_partial_blocks = 0;
1163 int first_c_fragment_seen;
1166 int current_fragment;
1168 debug_vp3(" vp3: unpacking superblock coding\n");
1172 debug_vp3(" keyframe-- all superblocks are fully coded\n");
1173 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
1177 /* unpack the list of partially-coded superblocks */
1178 bit = get_bits(gb, 1);
1179 /* toggle the bit because as soon as the first run length is
1180 * fetched the bit will be toggled again */
1182 while (current_superblock < s->superblock_count) {
1183 if (current_run-- == 0) {
1186 current_run = get_vlc2(gb,
1187 s->superblock_run_length_vlc.table, 6, 2);
1188 if (current_run == 33)
1189 current_run += get_bits(gb, 12);
1191 current_run = get_superblock_run_length(gb);
1193 debug_block_coding(" setting superblocks %d..%d to %s\n",
1195 current_superblock + current_run - 1,
1196 (bit) ? "partially coded" : "not coded");
1198 /* if any of the superblocks are not partially coded, flag
1199 * a boolean to decode the list of fully-coded superblocks */
1201 decode_fully_flags = 1;
1204 /* make a note of the fact that there are partially coded
1206 decode_partial_blocks = 1;
1209 s->superblock_coding[current_superblock++] = bit;
1212 /* unpack the list of fully coded superblocks if any of the blocks were
1213 * not marked as partially coded in the previous step */
1214 if (decode_fully_flags) {
1216 current_superblock = 0;
1218 bit = get_bits(gb, 1);
1219 /* toggle the bit because as soon as the first run length is
1220 * fetched the bit will be toggled again */
1222 while (current_superblock < s->superblock_count) {
1224 /* skip any superblocks already marked as partially coded */
1225 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
1227 if (current_run-- == 0) {
1230 current_run = get_vlc2(gb,
1231 s->superblock_run_length_vlc.table, 6, 2);
1232 if (current_run == 33)
1233 current_run += get_bits(gb, 12);
1235 current_run = get_superblock_run_length(gb);
1239 debug_block_coding(" setting superblock %d to %s\n",
1241 (bit) ? "fully coded" : "not coded");
1242 s->superblock_coding[current_superblock] = 2*bit;
1244 current_superblock++;
1248 /* if there were partial blocks, initialize bitstream for
1249 * unpacking fragment codings */
1250 if (decode_partial_blocks) {
1253 bit = get_bits(gb, 1);
1254 /* toggle the bit because as soon as the first run length is
1255 * fetched the bit will be toggled again */
1260 /* figure out which fragments are coded; iterate through each
1261 * superblock (all planes) */
1262 s->coded_fragment_list_index = 0;
1263 s->first_coded_y_fragment = s->first_coded_c_fragment = 0;
1264 s->last_coded_y_fragment = s->last_coded_c_fragment = -1;
1265 first_c_fragment_seen = 0;
1266 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
1267 for (i = 0; i < s->superblock_count; i++) {
1269 /* iterate through all 16 fragments in a superblock */
1270 for (j = 0; j < 16; j++) {
1272 /* if the fragment is in bounds, check its coding status */
1273 current_fragment = s->superblock_fragments[i * 16 + j];
1274 if (current_fragment >= s->fragment_count) {
1275 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
1276 current_fragment, s->fragment_count);
1279 if (current_fragment != -1) {
1280 if (s->superblock_coding[i] == SB_NOT_CODED) {
1282 /* copy all the fragments from the prior frame */
1283 s->all_fragments[current_fragment].coding_method =
1286 } else if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
1288 /* fragment may or may not be coded; this is the case
1289 * that cares about the fragment coding runs */
1290 if (current_run-- == 0) {
1293 current_run = get_vlc2(gb,
1294 s->fragment_run_length_vlc.table, 5, 2);
1296 current_run = get_fragment_run_length(gb);
1301 /* default mode; actual mode will be decoded in
1303 s->all_fragments[current_fragment].coding_method =
1305 s->all_fragments[current_fragment].coeffs= s->coeffs + 64*s->coded_fragment_list_index;
1306 s->coded_fragment_list[s->coded_fragment_list_index] =
1308 if ((current_fragment >= s->u_fragment_start) &&
1309 (s->last_coded_y_fragment == -1) &&
1310 (!first_c_fragment_seen)) {
1311 s->first_coded_c_fragment = s->coded_fragment_list_index;
1312 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
1313 first_c_fragment_seen = 1;
1315 s->coded_fragment_list_index++;
1316 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
1317 debug_block_coding(" superblock %d is partially coded, fragment %d is coded\n",
1318 i, current_fragment);
1320 /* not coded; copy this fragment from the prior frame */
1321 s->all_fragments[current_fragment].coding_method =
1323 debug_block_coding(" superblock %d is partially coded, fragment %d is not coded\n",
1324 i, current_fragment);
1329 /* fragments are fully coded in this superblock; actual
1330 * coding will be determined in next step */
1331 s->all_fragments[current_fragment].coding_method =
1333 s->all_fragments[current_fragment].coeffs= s->coeffs + 64*s->coded_fragment_list_index;
1334 s->coded_fragment_list[s->coded_fragment_list_index] =
1336 if ((current_fragment >= s->u_fragment_start) &&
1337 (s->last_coded_y_fragment == -1) &&
1338 (!first_c_fragment_seen)) {
1339 s->first_coded_c_fragment = s->coded_fragment_list_index;
1340 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
1341 first_c_fragment_seen = 1;
1343 s->coded_fragment_list_index++;
1344 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
1345 debug_block_coding(" superblock %d is fully coded, fragment %d is coded\n",
1346 i, current_fragment);
1352 if (!first_c_fragment_seen)
1353 /* only Y fragments coded in this frame */
1354 s->last_coded_y_fragment = s->coded_fragment_list_index - 1;
1356 /* end the list of coded C fragments */
1357 s->last_coded_c_fragment = s->coded_fragment_list_index - 1;
1359 debug_block_coding(" %d total coded fragments, y: %d -> %d, c: %d -> %d\n",
1360 s->coded_fragment_list_index,
1361 s->first_coded_y_fragment,
1362 s->last_coded_y_fragment,
1363 s->first_coded_c_fragment,
1364 s->last_coded_c_fragment);
1370 * This function unpacks all the coding mode data for individual macroblocks
1371 * from the bitstream.
1373 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
1377 int current_macroblock;
1378 int current_fragment;
1381 debug_vp3(" vp3: unpacking encoding modes\n");
1384 debug_vp3(" keyframe-- all blocks are coded as INTRA\n");
1386 for (i = 0; i < s->fragment_count; i++)
1387 s->all_fragments[i].coding_method = MODE_INTRA;
1391 /* fetch the mode coding scheme for this frame */
1392 scheme = get_bits(gb, 3);
1393 debug_modes(" using mode alphabet %d\n", scheme);
1395 /* is it a custom coding scheme? */
1397 debug_modes(" custom mode alphabet ahead:\n");
1398 for (i = 0; i < 8; i++)
1399 ModeAlphabet[scheme][get_bits(gb, 3)] = i;
1402 for (i = 0; i < 8; i++)
1403 debug_modes(" mode[%d][%d] = %d\n", scheme, i,
1404 ModeAlphabet[scheme][i]);
1406 /* iterate through all of the macroblocks that contain 1 or more
1407 * coded fragments */
1408 for (i = 0; i < s->u_superblock_start; i++) {
1410 for (j = 0; j < 4; j++) {
1411 current_macroblock = s->superblock_macroblocks[i * 4 + j];
1412 if ((current_macroblock == -1) ||
1413 (s->macroblock_coding[current_macroblock] == MODE_COPY))
1415 if (current_macroblock >= s->macroblock_count) {
1416 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad macroblock number (%d >= %d)\n",
1417 current_macroblock, s->macroblock_count);
1421 /* mode 7 means get 3 bits for each coding mode */
1423 coding_mode = get_bits(gb, 3);
1427 coding_mode = ModeAlphabet[scheme]
1428 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
1430 coding_mode = ModeAlphabet[scheme][get_mode_code(gb)];
1434 s->macroblock_coding[current_macroblock] = coding_mode;
1435 for (k = 0; k < 6; k++) {
1437 s->macroblock_fragments[current_macroblock * 6 + k];
1438 if (current_fragment == -1)
1440 if (current_fragment >= s->fragment_count) {
1441 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad fragment number (%d >= %d)\n",
1442 current_fragment, s->fragment_count);
1445 if (s->all_fragments[current_fragment].coding_method !=
1447 s->all_fragments[current_fragment].coding_method =
1451 debug_modes(" coding method for macroblock starting @ fragment %d = %d\n",
1452 s->macroblock_fragments[current_macroblock * 6], coding_mode);
1461 * This function unpacks all the motion vectors for the individual
1462 * macroblocks from the bitstream.
1464 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
1470 int last_motion_x = 0;
1471 int last_motion_y = 0;
1472 int prior_last_motion_x = 0;
1473 int prior_last_motion_y = 0;
1474 int current_macroblock;
1475 int current_fragment;
1477 debug_vp3(" vp3: unpacking motion vectors\n");
1480 debug_vp3(" keyframe-- there are no motion vectors\n");
1484 memset(motion_x, 0, 6 * sizeof(int));
1485 memset(motion_y, 0, 6 * sizeof(int));
1487 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
1488 coding_mode = get_bits(gb, 1);
1489 debug_vectors(" using %s scheme for unpacking motion vectors\n",
1490 (coding_mode == 0) ? "VLC" : "fixed-length");
1492 /* iterate through all of the macroblocks that contain 1 or more
1493 * coded fragments */
1494 for (i = 0; i < s->u_superblock_start; i++) {
1496 for (j = 0; j < 4; j++) {
1497 current_macroblock = s->superblock_macroblocks[i * 4 + j];
1498 if ((current_macroblock == -1) ||
1499 (s->macroblock_coding[current_macroblock] == MODE_COPY))
1501 if (current_macroblock >= s->macroblock_count) {
1502 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad macroblock number (%d >= %d)\n",
1503 current_macroblock, s->macroblock_count);
1507 current_fragment = s->macroblock_fragments[current_macroblock * 6];
1508 if (current_fragment >= s->fragment_count) {
1509 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d\n",
1510 current_fragment, s->fragment_count);
1513 switch (s->macroblock_coding[current_macroblock]) {
1515 case MODE_INTER_PLUS_MV:
1516 case MODE_GOLDEN_MV:
1517 /* all 6 fragments use the same motion vector */
1518 if (coding_mode == 0) {
1520 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1521 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1523 motion_x[0] = get_motion_vector_vlc(gb);
1524 motion_y[0] = get_motion_vector_vlc(gb);
1527 motion_x[0] = get_motion_vector_fixed(gb);
1528 motion_y[0] = get_motion_vector_fixed(gb);
1530 for (k = 1; k < 6; k++) {
1531 motion_x[k] = motion_x[0];
1532 motion_y[k] = motion_y[0];
1535 /* vector maintenance, only on MODE_INTER_PLUS_MV */
1536 if (s->macroblock_coding[current_macroblock] ==
1537 MODE_INTER_PLUS_MV) {
1538 prior_last_motion_x = last_motion_x;
1539 prior_last_motion_y = last_motion_y;
1540 last_motion_x = motion_x[0];
1541 last_motion_y = motion_y[0];
1545 case MODE_INTER_FOURMV:
1546 /* fetch 4 vectors from the bitstream, one for each
1547 * Y fragment, then average for the C fragment vectors */
1548 motion_x[4] = motion_y[4] = 0;
1549 for (k = 0; k < 4; k++) {
1550 if (coding_mode == 0) {
1552 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1553 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1555 motion_x[k] = get_motion_vector_vlc(gb);
1556 motion_y[k] = get_motion_vector_vlc(gb);
1559 motion_x[k] = get_motion_vector_fixed(gb);
1560 motion_y[k] = get_motion_vector_fixed(gb);
1562 motion_x[4] += motion_x[k];
1563 motion_y[4] += motion_y[k];
1566 if (motion_x[4] >= 0)
1567 motion_x[4] = (motion_x[4] + 2) / 4;
1569 motion_x[4] = (motion_x[4] - 2) / 4;
1570 motion_x[5] = motion_x[4];
1572 if (motion_y[4] >= 0)
1573 motion_y[4] = (motion_y[4] + 2) / 4;
1575 motion_y[4] = (motion_y[4] - 2) / 4;
1576 motion_y[5] = motion_y[4];
1578 /* vector maintenance; vector[3] is treated as the
1579 * last vector in this case */
1580 prior_last_motion_x = last_motion_x;
1581 prior_last_motion_y = last_motion_y;
1582 last_motion_x = motion_x[3];
1583 last_motion_y = motion_y[3];
1586 case MODE_INTER_LAST_MV:
1587 /* all 6 fragments use the last motion vector */
1588 motion_x[0] = last_motion_x;
1589 motion_y[0] = last_motion_y;
1590 for (k = 1; k < 6; k++) {
1591 motion_x[k] = motion_x[0];
1592 motion_y[k] = motion_y[0];
1595 /* no vector maintenance (last vector remains the
1599 case MODE_INTER_PRIOR_LAST:
1600 /* all 6 fragments use the motion vector prior to the
1601 * last motion vector */
1602 motion_x[0] = prior_last_motion_x;
1603 motion_y[0] = prior_last_motion_y;
1604 for (k = 1; k < 6; k++) {
1605 motion_x[k] = motion_x[0];
1606 motion_y[k] = motion_y[0];
1609 /* vector maintenance */
1610 prior_last_motion_x = last_motion_x;
1611 prior_last_motion_y = last_motion_y;
1612 last_motion_x = motion_x[0];
1613 last_motion_y = motion_y[0];
1617 /* covers intra, inter without MV, golden without MV */
1618 memset(motion_x, 0, 6 * sizeof(int));
1619 memset(motion_y, 0, 6 * sizeof(int));
1621 /* no vector maintenance */
1625 /* assign the motion vectors to the correct fragments */
1626 debug_vectors(" vectors for macroblock starting @ fragment %d (coding method %d):\n",
1628 s->macroblock_coding[current_macroblock]);
1629 for (k = 0; k < 6; k++) {
1631 s->macroblock_fragments[current_macroblock * 6 + k];
1632 if (current_fragment == -1)
1634 if (current_fragment >= s->fragment_count) {
1635 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d)\n",
1636 current_fragment, s->fragment_count);
1639 s->all_fragments[current_fragment].motion_x = motion_x[k];
1640 s->all_fragments[current_fragment].motion_y = motion_y[k];
1641 debug_vectors(" vector %d: fragment %d = (%d, %d)\n",
1642 k, current_fragment, motion_x[k], motion_y[k]);
1652 * This function is called by unpack_dct_coeffs() to extract the VLCs from
1653 * the bitstream. The VLCs encode tokens which are used to unpack DCT
1654 * data. This function unpacks all the VLCs for either the Y plane or both
1655 * C planes, and is called for DC coefficients or different AC coefficient
1656 * levels (since different coefficient types require different VLC tables.
1658 * This function returns a residual eob run. E.g, if a particular token gave
1659 * instructions to EOB the next 5 fragments and there were only 2 fragments
1660 * left in the current fragment range, 3 would be returned so that it could
1661 * be passed into the next call to this same function.
1663 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
1664 VLC *table, int coeff_index,
1665 int first_fragment, int last_fragment,
1672 Vp3Fragment *fragment;
1673 uint8_t *perm= s->scantable.permutated;
1675 if ((first_fragment >= s->fragment_count) ||
1676 (last_fragment >= s->fragment_count)) {
1678 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vlcs(): bad fragment number (%d -> %d ?)\n",
1679 first_fragment, last_fragment);
1683 for (i = first_fragment; i <= last_fragment; i++) {
1685 fragment = &s->all_fragments[s->coded_fragment_list[i]];
1686 if (fragment->coeff_count > coeff_index)
1690 /* decode a VLC into a token */
1691 token = get_vlc2(gb, table->table, 5, 3);
1692 debug_vlc(" token = %2d, ", token);
1693 /* use the token to get a zero run, a coefficient, and an eob run */
1694 unpack_token(gb, token, &zero_run, &coeff, &eob_run);
1698 fragment->coeff_count += zero_run;
1699 if (fragment->coeff_count < 64)
1700 fragment->coeffs[perm[fragment->coeff_count++]] = coeff;
1701 debug_vlc(" fragment %d coeff = %d\n",
1702 s->coded_fragment_list[i], fragment->coeffs[coeff_index]);
1704 fragment->last_coeff = fragment->coeff_count;
1705 fragment->coeff_count = 64;
1706 debug_vlc(" fragment %d eob with %d coefficients\n",
1707 s->coded_fragment_list[i], fragment->last_coeff);
1716 * This function unpacks all of the DCT coefficient data from the
1719 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1726 int residual_eob_run = 0;
1728 /* fetch the DC table indices */
1729 dc_y_table = get_bits(gb, 4);
1730 dc_c_table = get_bits(gb, 4);
1732 /* unpack the Y plane DC coefficients */
1733 debug_vp3(" vp3: unpacking Y plane DC coefficients using table %d\n",
1735 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1736 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1738 /* unpack the C plane DC coefficients */
1739 debug_vp3(" vp3: unpacking C plane DC coefficients using table %d\n",
1741 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1742 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1744 /* fetch the AC table indices */
1745 ac_y_table = get_bits(gb, 4);
1746 ac_c_table = get_bits(gb, 4);
1748 /* unpack the group 1 AC coefficients (coeffs 1-5) */
1749 for (i = 1; i <= 5; i++) {
1751 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1753 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_y_table], i,
1754 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1756 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1758 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_c_table], i,
1759 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1762 /* unpack the group 2 AC coefficients (coeffs 6-14) */
1763 for (i = 6; i <= 14; i++) {
1765 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1767 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_y_table], i,
1768 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1770 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1772 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_c_table], i,
1773 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1776 /* unpack the group 3 AC coefficients (coeffs 15-27) */
1777 for (i = 15; i <= 27; i++) {
1779 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1781 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_y_table], i,
1782 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1784 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1786 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_c_table], i,
1787 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1790 /* unpack the group 4 AC coefficients (coeffs 28-63) */
1791 for (i = 28; i <= 63; i++) {
1793 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1795 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_y_table], i,
1796 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1798 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1800 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_c_table], i,
1801 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1808 * This function reverses the DC prediction for each coded fragment in
1809 * the frame. Much of this function is adapted directly from the original
1812 #define COMPATIBLE_FRAME(x) \
1813 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1814 #define FRAME_CODED(x) (s->all_fragments[x].coding_method != MODE_COPY)
1815 static inline int iabs (int x) { return ((x < 0) ? -x : x); }
1817 static void reverse_dc_prediction(Vp3DecodeContext *s,
1820 int fragment_height)
1829 int i = first_fragment;
1832 * Fragment prediction groups:
1840 * Note: Groups 5 and 7 do not exist as it would mean that the
1841 * fragment's x coordinate is both 0 and (width - 1) at the same time.
1843 int predictor_group;
1846 /* validity flags for the left, up-left, up, and up-right fragments */
1847 int fl, ful, fu, fur;
1849 /* DC values for the left, up-left, up, and up-right fragments */
1850 int vl, vul, vu, vur;
1852 /* indices for the left, up-left, up, and up-right fragments */
1856 * The 6 fields mean:
1857 * 0: up-left multiplier
1859 * 2: up-right multiplier
1860 * 3: left multiplier
1862 * 5: right bit shift divisor (e.g., 7 means >>=7, a.k.a. div by 128)
1864 int predictor_transform[16][6] = {
1865 { 0, 0, 0, 0, 0, 0 },
1866 { 0, 0, 0, 1, 0, 0 }, // PL
1867 { 0, 0, 1, 0, 0, 0 }, // PUR
1868 { 0, 0, 53, 75, 127, 7 }, // PUR|PL
1869 { 0, 1, 0, 0, 0, 0 }, // PU
1870 { 0, 1, 0, 1, 1, 1 }, // PU|PL
1871 { 0, 1, 0, 0, 0, 0 }, // PU|PUR
1872 { 0, 0, 53, 75, 127, 7 }, // PU|PUR|PL
1873 { 1, 0, 0, 0, 0, 0 }, // PUL
1874 { 0, 0, 0, 1, 0, 0 }, // PUL|PL
1875 { 1, 0, 1, 0, 1, 1 }, // PUL|PUR
1876 { 0, 0, 53, 75, 127, 7 }, // PUL|PUR|PL
1877 { 0, 1, 0, 0, 0, 0 }, // PUL|PU
1878 {-26, 29, 0, 29, 31, 5 }, // PUL|PU|PL
1879 { 3, 10, 3, 0, 15, 4 }, // PUL|PU|PUR
1880 {-26, 29, 0, 29, 31, 5 } // PUL|PU|PUR|PL
1883 /* This table shows which types of blocks can use other blocks for
1884 * prediction. For example, INTRA is the only mode in this table to
1885 * have a frame number of 0. That means INTRA blocks can only predict
1886 * from other INTRA blocks. There are 2 golden frame coding types;
1887 * blocks encoding in these modes can only predict from other blocks
1888 * that were encoded with these 1 of these 2 modes. */
1889 unsigned char compatible_frame[8] = {
1890 1, /* MODE_INTER_NO_MV */
1892 1, /* MODE_INTER_PLUS_MV */
1893 1, /* MODE_INTER_LAST_MV */
1894 1, /* MODE_INTER_PRIOR_MV */
1895 2, /* MODE_USING_GOLDEN */
1896 2, /* MODE_GOLDEN_MV */
1897 1 /* MODE_INTER_FOUR_MV */
1899 int current_frame_type;
1901 /* there is a last DC predictor for each of the 3 frame types */
1906 debug_vp3(" vp3: reversing DC prediction\n");
1908 vul = vu = vur = vl = 0;
1909 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1911 /* for each fragment row... */
1912 for (y = 0; y < fragment_height; y++) {
1914 /* for each fragment in a row... */
1915 for (x = 0; x < fragment_width; x++, i++) {
1917 /* reverse prediction if this block was coded */
1918 if (s->all_fragments[i].coding_method != MODE_COPY) {
1920 current_frame_type =
1921 compatible_frame[s->all_fragments[i].coding_method];
1922 predictor_group = (x == 0) + ((y == 0) << 1) +
1923 ((x + 1 == fragment_width) << 2);
1924 debug_dc_pred(" frag %d: group %d, orig DC = %d, ",
1925 i, predictor_group, s->all_fragments[i].coeffs[0]);
1927 switch (predictor_group) {
1930 /* main body of fragments; consider all 4 possible
1931 * fragments for prediction */
1933 /* calculate the indices of the predicting fragments */
1934 ul = i - fragment_width - 1;
1935 u = i - fragment_width;
1936 ur = i - fragment_width + 1;
1939 /* fetch the DC values for the predicting fragments */
1940 vul = s->all_fragments[ul].coeffs[0];
1941 vu = s->all_fragments[u].coeffs[0];
1942 vur = s->all_fragments[ur].coeffs[0];
1943 vl = s->all_fragments[l].coeffs[0];
1945 /* figure out which fragments are valid */
1946 ful = FRAME_CODED(ul) && COMPATIBLE_FRAME(ul);
1947 fu = FRAME_CODED(u) && COMPATIBLE_FRAME(u);
1948 fur = FRAME_CODED(ur) && COMPATIBLE_FRAME(ur);
1949 fl = FRAME_CODED(l) && COMPATIBLE_FRAME(l);
1951 /* decide which predictor transform to use */
1952 transform = (fl*PL) | (fu*PU) | (ful*PUL) | (fur*PUR);
1957 /* left column of fragments, not including top corner;
1958 * only consider up and up-right fragments */
1960 /* calculate the indices of the predicting fragments */
1961 u = i - fragment_width;
1962 ur = i - fragment_width + 1;
1964 /* fetch the DC values for the predicting fragments */
1965 vu = s->all_fragments[u].coeffs[0];
1966 vur = s->all_fragments[ur].coeffs[0];
1968 /* figure out which fragments are valid */
1969 fur = FRAME_CODED(ur) && COMPATIBLE_FRAME(ur);
1970 fu = FRAME_CODED(u) && COMPATIBLE_FRAME(u);
1972 /* decide which predictor transform to use */
1973 transform = (fu*PU) | (fur*PUR);
1979 /* top row of fragments, not including top-left frag;
1980 * only consider the left fragment for prediction */
1982 /* calculate the indices of the predicting fragments */
1985 /* fetch the DC values for the predicting fragments */
1986 vl = s->all_fragments[l].coeffs[0];
1988 /* figure out which fragments are valid */
1989 fl = FRAME_CODED(l) && COMPATIBLE_FRAME(l);
1991 /* decide which predictor transform to use */
1992 transform = (fl*PL);
1997 /* top-left fragment */
1999 /* nothing to predict from in this case */
2005 /* right column of fragments, not including top corner;
2006 * consider up-left, up, and left fragments for
2009 /* calculate the indices of the predicting fragments */
2010 ul = i - fragment_width - 1;
2011 u = i - fragment_width;
2014 /* fetch the DC values for the predicting fragments */
2015 vul = s->all_fragments[ul].coeffs[0];
2016 vu = s->all_fragments[u].coeffs[0];
2017 vl = s->all_fragments[l].coeffs[0];
2019 /* figure out which fragments are valid */
2020 ful = FRAME_CODED(ul) && COMPATIBLE_FRAME(ul);
2021 fu = FRAME_CODED(u) && COMPATIBLE_FRAME(u);
2022 fl = FRAME_CODED(l) && COMPATIBLE_FRAME(l);
2024 /* decide which predictor transform to use */
2025 transform = (fl*PL) | (fu*PU) | (ful*PUL);
2031 debug_dc_pred("transform = %d, ", transform);
2033 if (transform == 0) {
2035 /* if there were no fragments to predict from, use last
2037 s->all_fragments[i].coeffs[0] += last_dc[current_frame_type];
2038 debug_dc_pred("from last DC (%d) = %d\n",
2039 current_frame_type, s->all_fragments[i].coeffs[0]);
2043 /* apply the appropriate predictor transform */
2045 (predictor_transform[transform][0] * vul) +
2046 (predictor_transform[transform][1] * vu) +
2047 (predictor_transform[transform][2] * vur) +
2048 (predictor_transform[transform][3] * vl);
2050 /* if there is a shift value in the transform, add
2051 * the sign bit before the shift */
2052 if (predictor_transform[transform][5] != 0) {
2053 predicted_dc += ((predicted_dc >> 15) &
2054 predictor_transform[transform][4]);
2055 predicted_dc >>= predictor_transform[transform][5];
2058 /* check for outranging on the [ul u l] and
2059 * [ul u ur l] predictors */
2060 if ((transform == 13) || (transform == 15)) {
2061 if (iabs(predicted_dc - vu) > 128)
2063 else if (iabs(predicted_dc - vl) > 128)
2065 else if (iabs(predicted_dc - vul) > 128)
2069 /* at long last, apply the predictor */
2070 s->all_fragments[i].coeffs[0] += predicted_dc;
2071 debug_dc_pred("from pred DC = %d\n",
2072 s->all_fragments[i].coeffs[0]);
2076 last_dc[current_frame_type] = s->all_fragments[i].coeffs[0];
2077 if(s->all_fragments[i].coeffs[0] && s->all_fragments[i].last_coeff<0)
2078 s->all_fragments[i].last_coeff= 0;
2085 * This function performs the final rendering of each fragment's data
2086 * onto the output frame.
2088 static void render_fragments(Vp3DecodeContext *s,
2092 int plane /* 0 = Y, 1 = U, 2 = V */)
2096 int i = first_fragment;
2097 int16_t *dequantizer;
2098 DCTELEM __align16 output_samples[64];
2099 unsigned char *output_plane;
2100 unsigned char *last_plane;
2101 unsigned char *golden_plane;
2103 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
2104 int upper_motion_limit, lower_motion_limit;
2105 int motion_halfpel_index;
2106 uint8_t *motion_source;
2108 debug_vp3(" vp3: rendering final fragments for %s\n",
2109 (plane == 0) ? "Y plane" : (plane == 1) ? "U plane" : "V plane");
2111 /* set up plane-specific parameters */
2113 output_plane = s->current_frame.data[0];
2114 last_plane = s->last_frame.data[0];
2115 golden_plane = s->golden_frame.data[0];
2116 stride = s->current_frame.linesize[0];
2117 if (!s->flipped_image) stride = -stride;
2118 upper_motion_limit = 7 * s->current_frame.linesize[0];
2119 lower_motion_limit = height * s->current_frame.linesize[0] + width - 8;
2120 } else if (plane == 1) {
2121 output_plane = s->current_frame.data[1];
2122 last_plane = s->last_frame.data[1];
2123 golden_plane = s->golden_frame.data[1];
2124 stride = s->current_frame.linesize[1];
2125 if (!s->flipped_image) stride = -stride;
2126 upper_motion_limit = 7 * s->current_frame.linesize[1];
2127 lower_motion_limit = height * s->current_frame.linesize[1] + width - 8;
2129 output_plane = s->current_frame.data[2];
2130 last_plane = s->last_frame.data[2];
2131 golden_plane = s->golden_frame.data[2];
2132 stride = s->current_frame.linesize[2];
2133 if (!s->flipped_image) stride = -stride;
2134 upper_motion_limit = 7 * s->current_frame.linesize[2];
2135 lower_motion_limit = height * s->current_frame.linesize[2] + width - 8;
2138 if(ABS(stride) > 2048)
2139 return; //various tables are fixed size
2141 /* for each fragment row... */
2142 for (y = 0; y < height; y += 8) {
2144 /* for each fragment in a row... */
2145 for (x = 0; x < width; x += 8, i++) {
2147 if ((i < 0) || (i >= s->fragment_count)) {
2148 av_log(s->avctx, AV_LOG_ERROR, " vp3:render_fragments(): bad fragment number (%d)\n", i);
2152 /* transform if this block was coded */
2153 if ((s->all_fragments[i].coding_method != MODE_COPY) &&
2154 !((s->avctx->flags & CODEC_FLAG_GRAY) && plane)) {
2156 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
2157 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
2158 motion_source= golden_plane;
2160 motion_source= last_plane;
2162 motion_source += s->all_fragments[i].first_pixel;
2163 motion_halfpel_index = 0;
2165 /* sort out the motion vector if this fragment is coded
2166 * using a motion vector method */
2167 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
2168 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
2170 motion_x = s->all_fragments[i].motion_x;
2171 motion_y = s->all_fragments[i].motion_y;
2173 motion_x= (motion_x>>1) | (motion_x&1);
2174 motion_y= (motion_y>>1) | (motion_y&1);
2177 src_x= (motion_x>>1) + x;
2178 src_y= (motion_y>>1) + y;
2179 if ((motion_x == 0xbeef) || (motion_y == 0xbeef))
2180 av_log(s->avctx, AV_LOG_ERROR, " help! got beefy vector! (%X, %X)\n", motion_x, motion_y);
2182 motion_halfpel_index = motion_x & 0x01;
2183 motion_source += (motion_x >> 1);
2185 // motion_y = -motion_y;
2186 motion_halfpel_index |= (motion_y & 0x01) << 1;
2187 motion_source += ((motion_y >> 1) * stride);
2189 if(src_x<0 || src_y<0 || src_x + 9 >= width || src_y + 9 >= height){
2190 uint8_t *temp= s->edge_emu_buffer;
2191 if(stride<0) temp -= 9*stride;
2192 else temp += 9*stride;
2194 ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, width, height);
2195 motion_source= temp;
2200 /* first, take care of copying a block from either the
2201 * previous or the golden frame */
2202 if (s->all_fragments[i].coding_method != MODE_INTRA) {
2203 //Note, it is possible to implement all MC cases with put_no_rnd_pixels_l2 which would look more like the VP3 source but this would be slower as put_no_rnd_pixels_tab is better optimzed
2204 if(motion_halfpel_index != 3){
2205 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
2206 output_plane + s->all_fragments[i].first_pixel,
2207 motion_source, stride, 8);
2209 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
2210 s->dsp.put_no_rnd_pixels_l2[1](
2211 output_plane + s->all_fragments[i].first_pixel,
2213 motion_source + stride + 1 + d,
2216 dequantizer = s->inter_dequant;
2219 dequantizer = s->intra_y_dequant;
2221 dequantizer = s->intra_c_dequant;
2224 /* dequantize the DCT coefficients */
2225 debug_idct("fragment %d, coding mode %d, DC = %d, dequant = %d:\n",
2226 i, s->all_fragments[i].coding_method,
2227 s->all_fragments[i].coeffs[0], dequantizer[0]);
2229 if(s->avctx->idct_algo==FF_IDCT_VP3){
2230 for (j = 0; j < 64; j++) {
2231 s->all_fragments[i].coeffs[j] *= dequantizer[j];
2234 for (j = 0; j < 64; j++) {
2235 s->all_fragments[i].coeffs[j]= (dequantizer[j] * s->all_fragments[i].coeffs[j] + 2) >> 2;
2239 /* invert DCT and place (or add) in final output */
2241 if (s->all_fragments[i].coding_method == MODE_INTRA) {
2242 if(s->avctx->idct_algo!=FF_IDCT_VP3)
2243 s->all_fragments[i].coeffs[0] += 128<<3;
2245 output_plane + s->all_fragments[i].first_pixel,
2247 s->all_fragments[i].coeffs);
2250 output_plane + s->all_fragments[i].first_pixel,
2252 s->all_fragments[i].coeffs);
2254 memset(s->all_fragments[i].coeffs, 0, 64*sizeof(DCTELEM));
2256 debug_idct("block after idct_%s():\n",
2257 (s->all_fragments[i].coding_method == MODE_INTRA)?
2259 for (m = 0; m < 8; m++) {
2260 for (n = 0; n < 8; n++) {
2261 debug_idct(" %3d", *(output_plane +
2262 s->all_fragments[i].first_pixel + (m * stride + n)));
2270 /* copy directly from the previous frame */
2271 s->dsp.put_pixels_tab[1][0](
2272 output_plane + s->all_fragments[i].first_pixel,
2273 last_plane + s->all_fragments[i].first_pixel,
2283 static void horizontal_filter(unsigned char *first_pixel, int stride,
2284 int *bounding_values)
2289 for (end= first_pixel + 8*stride; first_pixel < end; first_pixel += stride) {
2291 (first_pixel[-2] - first_pixel[ 1])
2292 +3*(first_pixel[ 0] - first_pixel[-1]);
2293 filter_value = bounding_values[(filter_value + 4) >> 3];
2294 first_pixel[-1] = clip_uint8(first_pixel[-1] + filter_value);
2295 first_pixel[ 0] = clip_uint8(first_pixel[ 0] - filter_value);
2299 static void vertical_filter(unsigned char *first_pixel, int stride,
2300 int *bounding_values)
2304 const int nstride= -stride;
2306 for (end= first_pixel + 8; first_pixel < end; first_pixel++) {
2308 (first_pixel[2 * nstride] - first_pixel[ stride])
2309 +3*(first_pixel[0 ] - first_pixel[nstride]);
2310 filter_value = bounding_values[(filter_value + 4) >> 3];
2311 first_pixel[nstride] = clip_uint8(first_pixel[nstride] + filter_value);
2312 first_pixel[0] = clip_uint8(first_pixel[0] - filter_value);
2316 static void apply_loop_filter(Vp3DecodeContext *s)
2322 unsigned char *plane_data;
2324 int bounding_values_array[256];
2325 int *bounding_values= bounding_values_array+127;
2328 /* find the right loop limit value */
2329 for (x = 63; x >= 0; x--) {
2330 if (vp31_ac_scale_factor[x] >= s->quality_index)
2333 filter_limit = vp31_filter_limit_values[s->quality_index];
2335 /* set up the bounding values */
2336 memset(bounding_values_array, 0, 256 * sizeof(int));
2337 for (x = 0; x < filter_limit; x++) {
2338 bounding_values[-x - filter_limit] = -filter_limit + x;
2339 bounding_values[-x] = -x;
2340 bounding_values[x] = x;
2341 bounding_values[x + filter_limit] = filter_limit - x;
2344 for (plane = 0; plane < 3; plane++) {
2347 /* Y plane parameters */
2349 width = s->fragment_width;
2350 height = s->fragment_height;
2351 stride = s->current_frame.linesize[0];
2352 plane_data = s->current_frame.data[0];
2353 } else if (plane == 1) {
2354 /* U plane parameters */
2355 fragment = s->u_fragment_start;
2356 width = s->fragment_width / 2;
2357 height = s->fragment_height / 2;
2358 stride = s->current_frame.linesize[1];
2359 plane_data = s->current_frame.data[1];
2361 /* V plane parameters */
2362 fragment = s->v_fragment_start;
2363 width = s->fragment_width / 2;
2364 height = s->fragment_height / 2;
2365 stride = s->current_frame.linesize[2];
2366 plane_data = s->current_frame.data[2];
2369 for (y = 0; y < height; y++) {
2371 for (x = 0; x < width; x++) {
2373 /* do not perform left edge filter for left columns frags */
2375 (s->all_fragments[fragment].coding_method != MODE_COPY)) {
2377 plane_data + s->all_fragments[fragment].first_pixel - 7*stride,
2378 stride, bounding_values);
2381 /* do not perform top edge filter for top row fragments */
2383 (s->all_fragments[fragment].coding_method != MODE_COPY)) {
2385 plane_data + s->all_fragments[fragment].first_pixel + stride,
2386 stride, bounding_values);
2389 /* do not perform right edge filter for right column
2390 * fragments or if right fragment neighbor is also coded
2391 * in this frame (it will be filtered in next iteration) */
2392 if ((x < width - 1) &&
2393 (s->all_fragments[fragment].coding_method != MODE_COPY) &&
2394 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
2396 plane_data + s->all_fragments[fragment + 1].first_pixel - 7*stride,
2397 stride, bounding_values);
2400 /* do not perform bottom edge filter for bottom row
2401 * fragments or if bottom fragment neighbor is also coded
2402 * in this frame (it will be filtered in the next row) */
2403 if ((y < height - 1) &&
2404 (s->all_fragments[fragment].coding_method != MODE_COPY) &&
2405 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
2407 plane_data + s->all_fragments[fragment + width].first_pixel + stride,
2408 stride, bounding_values);
2412 STOP_TIMER("loop filter")
2419 * This function computes the first pixel addresses for each fragment.
2420 * This function needs to be invoked after the first frame is allocated
2421 * so that it has access to the plane strides.
2423 static void vp3_calculate_pixel_addresses(Vp3DecodeContext *s)
2428 /* figure out the first pixel addresses for each of the fragments */
2431 for (y = s->fragment_height; y > 0; y--) {
2432 for (x = 0; x < s->fragment_width; x++) {
2433 s->all_fragments[i++].first_pixel =
2434 s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
2435 s->golden_frame.linesize[0] +
2436 x * FRAGMENT_PIXELS;
2437 debug_init(" fragment %d, first pixel @ %d\n",
2438 i-1, s->all_fragments[i-1].first_pixel);
2443 i = s->u_fragment_start;
2444 for (y = s->fragment_height / 2; y > 0; y--) {
2445 for (x = 0; x < s->fragment_width / 2; x++) {
2446 s->all_fragments[i++].first_pixel =
2447 s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
2448 s->golden_frame.linesize[1] +
2449 x * FRAGMENT_PIXELS;
2450 debug_init(" fragment %d, first pixel @ %d\n",
2451 i-1, s->all_fragments[i-1].first_pixel);
2456 i = s->v_fragment_start;
2457 for (y = s->fragment_height / 2; y > 0; y--) {
2458 for (x = 0; x < s->fragment_width / 2; x++) {
2459 s->all_fragments[i++].first_pixel =
2460 s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
2461 s->golden_frame.linesize[2] +
2462 x * FRAGMENT_PIXELS;
2463 debug_init(" fragment %d, first pixel @ %d\n",
2464 i-1, s->all_fragments[i-1].first_pixel);
2469 /* FIXME: this should be merged with the above! */
2470 static void theora_calculate_pixel_addresses(Vp3DecodeContext *s)
2475 /* figure out the first pixel addresses for each of the fragments */
2478 for (y = 1; y <= s->fragment_height; y++) {
2479 for (x = 0; x < s->fragment_width; x++) {
2480 s->all_fragments[i++].first_pixel =
2481 s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
2482 s->golden_frame.linesize[0] +
2483 x * FRAGMENT_PIXELS;
2484 debug_init(" fragment %d, first pixel @ %d\n",
2485 i-1, s->all_fragments[i-1].first_pixel);
2490 i = s->u_fragment_start;
2491 for (y = 1; y <= s->fragment_height / 2; y++) {
2492 for (x = 0; x < s->fragment_width / 2; x++) {
2493 s->all_fragments[i++].first_pixel =
2494 s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
2495 s->golden_frame.linesize[1] +
2496 x * FRAGMENT_PIXELS;
2497 debug_init(" fragment %d, first pixel @ %d\n",
2498 i-1, s->all_fragments[i-1].first_pixel);
2503 i = s->v_fragment_start;
2504 for (y = 1; y <= s->fragment_height / 2; y++) {
2505 for (x = 0; x < s->fragment_width / 2; x++) {
2506 s->all_fragments[i++].first_pixel =
2507 s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
2508 s->golden_frame.linesize[2] +
2509 x * FRAGMENT_PIXELS;
2510 debug_init(" fragment %d, first pixel @ %d\n",
2511 i-1, s->all_fragments[i-1].first_pixel);
2517 * This is the ffmpeg/libavcodec API init function.
2519 static int vp3_decode_init(AVCodecContext *avctx)
2521 Vp3DecodeContext *s = avctx->priv_data;
2525 int y_superblock_count;
2526 int c_superblock_count;
2528 if (avctx->codec_tag == MKTAG('V','P','3','0'))
2534 s->width = (avctx->width + 15) & 0xFFFFFFF0;
2535 s->height = (avctx->height + 15) & 0xFFFFFFF0;
2536 avctx->pix_fmt = PIX_FMT_YUV420P;
2537 avctx->has_b_frames = 0;
2538 if(avctx->idct_algo==FF_IDCT_AUTO)
2539 avctx->idct_algo=FF_IDCT_VP3;
2540 dsputil_init(&s->dsp, avctx);
2542 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
2544 /* initialize to an impossible value which will force a recalculation
2545 * in the first frame decode */
2546 s->quality_index = -1;
2548 s->y_superblock_width = (s->width + 31) / 32;
2549 s->y_superblock_height = (s->height + 31) / 32;
2550 y_superblock_count = s->y_superblock_width * s->y_superblock_height;
2552 /* work out the dimensions for the C planes */
2553 c_width = s->width / 2;
2554 c_height = s->height / 2;
2555 s->c_superblock_width = (c_width + 31) / 32;
2556 s->c_superblock_height = (c_height + 31) / 32;
2557 c_superblock_count = s->c_superblock_width * s->c_superblock_height;
2559 s->superblock_count = y_superblock_count + (c_superblock_count * 2);
2560 s->u_superblock_start = y_superblock_count;
2561 s->v_superblock_start = s->u_superblock_start + c_superblock_count;
2562 s->superblock_coding = av_malloc(s->superblock_count);
2564 s->macroblock_width = (s->width + 15) / 16;
2565 s->macroblock_height = (s->height + 15) / 16;
2566 s->macroblock_count = s->macroblock_width * s->macroblock_height;
2568 s->fragment_width = s->width / FRAGMENT_PIXELS;
2569 s->fragment_height = s->height / FRAGMENT_PIXELS;
2571 /* fragment count covers all 8x8 blocks for all 3 planes */
2572 s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
2573 s->u_fragment_start = s->fragment_width * s->fragment_height;
2574 s->v_fragment_start = s->fragment_width * s->fragment_height * 5 / 4;
2576 debug_init(" Y plane: %d x %d\n", s->width, s->height);
2577 debug_init(" C plane: %d x %d\n", c_width, c_height);
2578 debug_init(" Y superblocks: %d x %d, %d total\n",
2579 s->y_superblock_width, s->y_superblock_height, y_superblock_count);
2580 debug_init(" C superblocks: %d x %d, %d total\n",
2581 s->c_superblock_width, s->c_superblock_height, c_superblock_count);
2582 debug_init(" total superblocks = %d, U starts @ %d, V starts @ %d\n",
2583 s->superblock_count, s->u_superblock_start, s->v_superblock_start);
2584 debug_init(" macroblocks: %d x %d, %d total\n",
2585 s->macroblock_width, s->macroblock_height, s->macroblock_count);
2586 debug_init(" %d fragments, %d x %d, u starts @ %d, v starts @ %d\n",
2590 s->u_fragment_start,
2591 s->v_fragment_start);
2593 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
2594 s->coeffs = av_malloc(s->fragment_count * sizeof(DCTELEM) * 64);
2595 s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int));
2596 s->pixel_addresses_inited = 0;
2598 if (!s->theora_tables)
2600 for (i = 0; i < 64; i++)
2601 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
2602 for (i = 0; i < 64; i++)
2603 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
2604 for (i = 0; i < 64; i++)
2605 s->coded_intra_y_dequant[i] = vp31_intra_y_dequant[i];
2606 for (i = 0; i < 64; i++)
2607 s->coded_intra_c_dequant[i] = vp31_intra_c_dequant[i];
2608 for (i = 0; i < 64; i++)
2609 s->coded_inter_dequant[i] = vp31_inter_dequant[i];
2612 /* init VLC tables */
2613 for (i = 0; i < 16; i++) {
2616 init_vlc(&s->dc_vlc[i], 5, 32,
2617 &dc_bias[i][0][1], 4, 2,
2618 &dc_bias[i][0][0], 4, 2, 0);
2620 /* group 1 AC histograms */
2621 init_vlc(&s->ac_vlc_1[i], 5, 32,
2622 &ac_bias_0[i][0][1], 4, 2,
2623 &ac_bias_0[i][0][0], 4, 2, 0);
2625 /* group 2 AC histograms */
2626 init_vlc(&s->ac_vlc_2[i], 5, 32,
2627 &ac_bias_1[i][0][1], 4, 2,
2628 &ac_bias_1[i][0][0], 4, 2, 0);
2630 /* group 3 AC histograms */
2631 init_vlc(&s->ac_vlc_3[i], 5, 32,
2632 &ac_bias_2[i][0][1], 4, 2,
2633 &ac_bias_2[i][0][0], 4, 2, 0);
2635 /* group 4 AC histograms */
2636 init_vlc(&s->ac_vlc_4[i], 5, 32,
2637 &ac_bias_3[i][0][1], 4, 2,
2638 &ac_bias_3[i][0][0], 4, 2, 0);
2641 init_vlc(&s->superblock_run_length_vlc, 6, 34,
2642 &superblock_run_length_vlc_table[0][1], 4, 2,
2643 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
2645 init_vlc(&s->fragment_run_length_vlc, 5, 31,
2646 &fragment_run_length_vlc_table[0][1], 4, 2,
2647 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
2649 init_vlc(&s->mode_code_vlc, 3, 8,
2650 &mode_code_vlc_table[0][1], 2, 1,
2651 &mode_code_vlc_table[0][0], 2, 1, 0);
2653 init_vlc(&s->motion_vector_vlc, 6, 63,
2654 &motion_vector_vlc_table[0][1], 2, 1,
2655 &motion_vector_vlc_table[0][0], 2, 1, 0);
2657 /* work out the block mapping tables */
2658 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
2659 s->superblock_macroblocks = av_malloc(s->superblock_count * 4 * sizeof(int));
2660 s->macroblock_fragments = av_malloc(s->macroblock_count * 6 * sizeof(int));
2661 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
2662 init_block_mapping(s);
2664 for (i = 0; i < 3; i++) {
2665 s->current_frame.data[i] = NULL;
2666 s->last_frame.data[i] = NULL;
2667 s->golden_frame.data[i] = NULL;
2674 * This is the ffmpeg/libavcodec API frame decode function.
2676 static int vp3_decode_frame(AVCodecContext *avctx,
2677 void *data, int *data_size,
2678 uint8_t *buf, int buf_size)
2680 Vp3DecodeContext *s = avctx->priv_data;
2682 static int counter = 0;
2684 init_get_bits(&gb, buf, buf_size * 8);
2686 if (s->theora && get_bits1(&gb))
2688 int ptype = get_bits(&gb, 7);
2690 skip_bits(&gb, 6*8); /* "theora" */
2695 theora_decode_comments(avctx, gb);
2698 theora_decode_tables(avctx, gb);
2699 init_dequantizer(s);
2702 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype);
2707 s->keyframe = !get_bits1(&gb);
2710 s->last_quality_index = s->quality_index;
2711 s->quality_index = get_bits(&gb, 6);
2712 if (s->theora >= 0x030200)
2715 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
2716 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
2717 s->keyframe?"key":"", counter, s->quality_index);
2720 if (s->quality_index != s->last_quality_index)
2721 init_dequantizer(s);
2726 skip_bits(&gb, 4); /* width code */
2727 skip_bits(&gb, 4); /* height code */
2730 s->version = get_bits(&gb, 5);
2732 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
2735 if (s->version || s->theora)
2738 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
2739 skip_bits(&gb, 2); /* reserved? */
2742 if (s->last_frame.data[0] == s->golden_frame.data[0]) {
2743 if (s->golden_frame.data[0])
2744 avctx->release_buffer(avctx, &s->golden_frame);
2745 s->last_frame= s->golden_frame; /* ensure that we catch any access to this released frame */
2747 if (s->golden_frame.data[0])
2748 avctx->release_buffer(avctx, &s->golden_frame);
2749 if (s->last_frame.data[0])
2750 avctx->release_buffer(avctx, &s->last_frame);
2753 s->golden_frame.reference = 3;
2754 if(avctx->get_buffer(avctx, &s->golden_frame) < 0) {
2755 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
2759 /* golden frame is also the current frame */
2760 memcpy(&s->current_frame, &s->golden_frame, sizeof(AVFrame));
2762 /* time to figure out pixel addresses? */
2763 if (!s->pixel_addresses_inited)
2765 if (!s->flipped_image)
2766 vp3_calculate_pixel_addresses(s);
2768 theora_calculate_pixel_addresses(s);
2771 /* allocate a new current frame */
2772 s->current_frame.reference = 3;
2773 if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
2774 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
2779 s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
2780 s->current_frame.qstride= 0;
2784 STOP_TIMER("init_frame")}
2789 memcpy(s->current_frame.data[0], s->golden_frame.data[0],
2790 s->current_frame.linesize[0] * s->height);
2791 memcpy(s->current_frame.data[1], s->golden_frame.data[1],
2792 s->current_frame.linesize[1] * s->height / 2);
2793 memcpy(s->current_frame.data[2], s->golden_frame.data[2],
2794 s->current_frame.linesize[2] * s->height / 2);
2800 if (unpack_superblocks(s, &gb)){
2801 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
2804 STOP_TIMER("unpack_superblocks")}
2806 if (unpack_modes(s, &gb)){
2807 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
2810 STOP_TIMER("unpack_modes")}
2812 if (unpack_vectors(s, &gb)){
2813 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2816 STOP_TIMER("unpack_vectors")}
2818 if (unpack_dct_coeffs(s, &gb)){
2819 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
2822 STOP_TIMER("unpack_dct_coeffs")}
2825 reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
2826 STOP_TIMER("reverse_dc_prediction")}
2828 render_fragments(s, 0, s->width, s->height, 0);
2829 STOP_TIMER("render_fragments")}
2831 if ((avctx->flags & CODEC_FLAG_GRAY) == 0) {
2832 reverse_dc_prediction(s, s->u_fragment_start,
2833 s->fragment_width / 2, s->fragment_height / 2);
2834 reverse_dc_prediction(s, s->v_fragment_start,
2835 s->fragment_width / 2, s->fragment_height / 2);
2836 render_fragments(s, s->u_fragment_start, s->width / 2, s->height / 2, 1);
2837 render_fragments(s, s->v_fragment_start, s->width / 2, s->height / 2, 2);
2839 memset(s->current_frame.data[1], 0x80, s->width * s->height / 4);
2840 memset(s->current_frame.data[2], 0x80, s->width * s->height / 4);
2844 apply_loop_filter(s);
2845 STOP_TIMER("apply_loop_filter")}
2850 *data_size=sizeof(AVFrame);
2851 *(AVFrame*)data= s->current_frame;
2853 /* release the last frame, if it is allocated and if it is not the
2855 if ((s->last_frame.data[0]) &&
2856 (s->last_frame.data[0] != s->golden_frame.data[0]))
2857 avctx->release_buffer(avctx, &s->last_frame);
2859 /* shuffle frames (last = current) */
2860 memcpy(&s->last_frame, &s->current_frame, sizeof(AVFrame));
2861 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
2867 * This is the ffmpeg/libavcodec API module cleanup function.
2869 static int vp3_decode_end(AVCodecContext *avctx)
2871 Vp3DecodeContext *s = avctx->priv_data;
2873 av_free(s->all_fragments);
2875 av_free(s->coded_fragment_list);
2876 av_free(s->superblock_fragments);
2877 av_free(s->superblock_macroblocks);
2878 av_free(s->macroblock_fragments);
2879 av_free(s->macroblock_coding);
2881 /* release all frames */
2882 if (s->golden_frame.data[0] && s->golden_frame.data[0] != s->last_frame.data[0])
2883 avctx->release_buffer(avctx, &s->golden_frame);
2884 if (s->last_frame.data[0])
2885 avctx->release_buffer(avctx, &s->last_frame);
2886 /* no need to release the current_frame since it will always be pointing
2887 * to the same frame as either the golden or last frame */
2892 static int theora_decode_header(AVCodecContext *avctx, GetBitContext gb)
2894 Vp3DecodeContext *s = avctx->priv_data;
2895 int major, minor, micro;
2897 major = get_bits(&gb, 8); /* version major */
2898 minor = get_bits(&gb, 8); /* version minor */
2899 micro = get_bits(&gb, 8); /* version micro */
2900 av_log(avctx, AV_LOG_INFO, "Theora bitstream version %d.%d.%d\n",
2901 major, minor, micro);
2903 /* FIXME: endianess? */
2904 s->theora = (major << 16) | (minor << 8) | micro;
2906 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2907 /* but previous versions have the image flipped relative to vp3 */
2908 if (s->theora < 0x030200)
2910 s->flipped_image = 1;
2911 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2914 s->width = get_bits(&gb, 16) << 4;
2915 s->height = get_bits(&gb, 16) << 4;
2917 if(avcodec_check_dimensions(avctx, s->width, s->height)){
2918 s->width= s->height= 0;
2922 skip_bits(&gb, 24); /* frame width */
2923 skip_bits(&gb, 24); /* frame height */
2925 skip_bits(&gb, 8); /* offset x */
2926 skip_bits(&gb, 8); /* offset y */
2928 skip_bits(&gb, 32); /* fps numerator */
2929 skip_bits(&gb, 32); /* fps denumerator */
2930 skip_bits(&gb, 24); /* aspect numerator */
2931 skip_bits(&gb, 24); /* aspect denumerator */
2933 if (s->theora < 0x030200)
2934 skip_bits(&gb, 5); /* keyframe frequency force */
2935 skip_bits(&gb, 8); /* colorspace */
2936 skip_bits(&gb, 24); /* bitrate */
2938 skip_bits(&gb, 6); /* last(?) quality index */
2940 if (s->theora >= 0x030200)
2942 skip_bits(&gb, 5); /* keyframe frequency force */
2943 skip_bits(&gb, 5); /* spare bits */
2946 // align_get_bits(&gb);
2948 avctx->width = s->width;
2949 avctx->height = s->height;
2954 static int theora_decode_comments(AVCodecContext *avctx, GetBitContext gb)
2956 int nb_comments, i, tmp;
2958 tmp = get_bits_long(&gb, 32);
2959 tmp = be2me_32(tmp);
2963 nb_comments = get_bits_long(&gb, 32);
2964 nb_comments = be2me_32(nb_comments);
2965 for (i = 0; i < nb_comments; i++)
2967 tmp = get_bits_long(&gb, 32);
2968 tmp = be2me_32(tmp);
2976 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext gb)
2978 Vp3DecodeContext *s = avctx->priv_data;
2981 if (s->theora >= 0x030200) {
2982 n = get_bits(&gb, 3);
2983 /* loop filter table */
2984 for (i = 0; i < 64; i++)
2988 if (s->theora >= 0x030200)
2989 n = get_bits(&gb, 4) + 1;
2992 /* quality threshold table */
2993 for (i = 0; i < 64; i++)
2994 s->coded_ac_scale_factor[i] = get_bits(&gb, n);
2996 if (s->theora >= 0x030200)
2997 n = get_bits(&gb, 4) + 1;
3000 /* dc scale factor table */
3001 for (i = 0; i < 64; i++)
3002 s->coded_dc_scale_factor[i] = get_bits(&gb, n);
3004 if (s->theora >= 0x030200)
3005 n = get_bits(&gb, 9) + 1;
3009 av_log(NULL,AV_LOG_ERROR, "unsupported nbms : %d\n", n);
3013 for (i = 0; i < 64; i++)
3014 s->coded_intra_y_dequant[i] = get_bits(&gb, 8);
3017 for (i = 0; i < 64; i++)
3018 s->coded_intra_c_dequant[i] = get_bits(&gb, 8);
3021 for (i = 0; i < 64; i++)
3022 s->coded_inter_dequant[i] = get_bits(&gb, 8);
3024 /* FIXME: read huffmann tree.. */
3026 s->theora_tables = 1;
3031 static int theora_decode_init(AVCodecContext *avctx)
3033 Vp3DecodeContext *s = avctx->priv_data;
3036 uint8_t *p= avctx->extradata;
3041 if (!avctx->extradata_size)
3045 op_bytes = *(p++)<<8;
3048 init_get_bits(&gb, p, op_bytes);
3051 ptype = get_bits(&gb, 8);
3052 debug_vp3("Theora headerpacket type: %x\n", ptype);
3054 if (!(ptype & 0x80))
3057 skip_bits(&gb, 6*8); /* "theora" */
3062 theora_decode_header(avctx, gb);
3065 theora_decode_comments(avctx, gb);
3068 theora_decode_tables(avctx, gb);
3073 vp3_decode_init(avctx);
3077 AVCodec vp3_decoder = {
3081 sizeof(Vp3DecodeContext),
3090 #ifndef CONFIG_LIBTHEORA
3091 AVCodec theora_decoder = {
3095 sizeof(Vp3DecodeContext),