2 * Copyright (C) 2003-2004 the ffmpeg project
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 * On2 VP3 Video Decoder
25 * VP3 Video Decoder by Mike Melanson (mike at multimedia.cx)
26 * For more information about the VP3 coding process, visit:
27 * http://wiki.multimedia.cx/index.php?title=On2_VP3
29 * Theora decoder by Alex Beregszaszi
39 #include "bitstream.h"
44 #define FRAGMENT_PIXELS 8
49 * Define one or more of the following compile-time variables to 1 to obtain
50 * elaborate information about certain aspects of the decoding process.
52 * KEYFRAMES_ONLY: set this to 1 to only see keyframes (VP3 slideshow mode)
53 * DEBUG_VP3: high-level decoding flow
54 * DEBUG_INIT: initialization parameters
55 * DEBUG_DEQUANTIZERS: display how the dequanization tables are built
56 * DEBUG_BLOCK_CODING: unpacking the superblock/macroblock/fragment coding
57 * DEBUG_MODES: unpacking the coding modes for individual fragments
58 * DEBUG_VECTORS: display the motion vectors
59 * DEBUG_TOKEN: display exhaustive information about each DCT token
60 * DEBUG_VLC: display the VLCs as they are extracted from the stream
61 * DEBUG_DC_PRED: display the process of reversing DC prediction
62 * DEBUG_IDCT: show every detail of the IDCT process
65 #define KEYFRAMES_ONLY 0
69 #define DEBUG_DEQUANTIZERS 0
70 #define DEBUG_BLOCK_CODING 0
72 #define DEBUG_VECTORS 0
75 #define DEBUG_DC_PRED 0
79 #define debug_vp3(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
81 static inline void debug_vp3(const char *format
, ...) { }
85 #define debug_init(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
87 static inline void debug_init(const char *format
, ...) { }
90 #if DEBUG_DEQUANTIZERS
91 #define debug_dequantizers(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
93 static inline void debug_dequantizers(const char *format
, ...) { }
96 #if DEBUG_BLOCK_CODING
97 #define debug_block_coding(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
99 static inline void debug_block_coding(const char *format
, ...) { }
103 #define debug_modes(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
105 static inline void debug_modes(const char *format
, ...) { }
109 #define debug_vectors(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
111 static inline void debug_vectors(const char *format
, ...) { }
115 #define debug_token(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
117 static inline void debug_token(const char *format
, ...) { }
121 #define debug_vlc(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
123 static inline void debug_vlc(const char *format
, ...) { }
127 #define debug_dc_pred(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
129 static inline void debug_dc_pred(const char *format
, ...) { }
133 #define debug_idct(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
135 static inline void debug_idct(const char *format
, ...) { }
138 typedef struct Coeff
{
144 //FIXME split things out into their own arrays
145 typedef struct Vp3Fragment
{
147 /* address of first pixel taking into account which plane the fragment
148 * lives on as well as the plane stride */
150 /* this is the macroblock that the fragment belongs to */
152 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 const int ModeAlphabet
[6][CODING_MODE_COUNT
] =
177 /* scheme 1: Last motion vector dominates */
178 { MODE_INTER_LAST_MV
, MODE_INTER_PRIOR_LAST
,
179 MODE_INTER_PLUS_MV
, MODE_INTER_NO_MV
,
180 MODE_INTRA
, MODE_USING_GOLDEN
,
181 MODE_GOLDEN_MV
, MODE_INTER_FOURMV
},
184 { MODE_INTER_LAST_MV
, MODE_INTER_PRIOR_LAST
,
185 MODE_INTER_NO_MV
, MODE_INTER_PLUS_MV
,
186 MODE_INTRA
, MODE_USING_GOLDEN
,
187 MODE_GOLDEN_MV
, MODE_INTER_FOURMV
},
190 { MODE_INTER_LAST_MV
, MODE_INTER_PLUS_MV
,
191 MODE_INTER_PRIOR_LAST
, MODE_INTER_NO_MV
,
192 MODE_INTRA
, MODE_USING_GOLDEN
,
193 MODE_GOLDEN_MV
, MODE_INTER_FOURMV
},
196 { MODE_INTER_LAST_MV
, MODE_INTER_PLUS_MV
,
197 MODE_INTER_NO_MV
, MODE_INTER_PRIOR_LAST
,
198 MODE_INTRA
, MODE_USING_GOLDEN
,
199 MODE_GOLDEN_MV
, MODE_INTER_FOURMV
},
201 /* scheme 5: No motion vector dominates */
202 { MODE_INTER_NO_MV
, MODE_INTER_LAST_MV
,
203 MODE_INTER_PRIOR_LAST
, MODE_INTER_PLUS_MV
,
204 MODE_INTRA
, MODE_USING_GOLDEN
,
205 MODE_GOLDEN_MV
, MODE_INTER_FOURMV
},
208 { MODE_INTER_NO_MV
, MODE_USING_GOLDEN
,
209 MODE_INTER_LAST_MV
, MODE_INTER_PRIOR_LAST
,
210 MODE_INTER_PLUS_MV
, MODE_INTRA
,
211 MODE_GOLDEN_MV
, MODE_INTER_FOURMV
},
215 #define MIN_DEQUANT_VAL 2
217 typedef struct Vp3DecodeContext
{
218 AVCodecContext
*avctx
;
219 int theora
, theora_tables
;
222 AVFrame golden_frame
;
224 AVFrame current_frame
;
232 int last_quality_index
;
234 int superblock_count
;
235 int superblock_width
;
236 int superblock_height
;
237 int y_superblock_width
;
238 int y_superblock_height
;
239 int c_superblock_width
;
240 int c_superblock_height
;
241 int u_superblock_start
;
242 int v_superblock_start
;
243 unsigned char *superblock_coding
;
245 int macroblock_count
;
246 int macroblock_width
;
247 int macroblock_height
;
253 Vp3Fragment
*all_fragments
;
254 uint8_t *coeff_counts
;
257 int fragment_start
[3];
262 uint16_t coded_dc_scale_factor
[64];
263 uint32_t coded_ac_scale_factor
[64];
264 uint8_t base_matrix
[384][64];
265 uint8_t qr_count
[2][3];
266 uint8_t qr_size
[2][3][64];
267 uint16_t qr_base
[2][3][64];
269 /* this is a list of indexes 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_initialized
;
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 DECLARE_ALIGNED_16(int16_t, qmat
[2][4][64]); //<qmat[is_inter][plane]
290 /* This table contains superblock_count * 16 entries. Each set of 16
291 * numbers corresponds to the fragment indexes 0..15 of the superblock.
292 * An entry will be -1 to indicate that no entry corresponds to that
294 int *superblock_fragments
;
296 /* This table contains superblock_count * 4 entries. Each set of 4
297 * numbers corresponds to the macroblock indexes 0..3 of the superblock.
298 * An entry will be -1 to indicate that no entry corresponds to that
300 int *superblock_macroblocks
;
302 /* This table contains macroblock_count * 6 entries. Each set of 6
303 * numbers corresponds to the fragment indexes 0..5 which comprise
304 * the macroblock (4 Y fragments and 2 C fragments). */
305 int *macroblock_fragments
;
306 /* This is an array that indicates how a particular macroblock
308 unsigned char *macroblock_coding
;
310 int first_coded_y_fragment
;
311 int first_coded_c_fragment
;
312 int last_coded_y_fragment
;
313 int last_coded_c_fragment
;
315 uint8_t edge_emu_buffer
[9*2048]; //FIXME dynamic alloc
316 int8_t qscale_table
[2048]; //FIXME dynamic alloc (width+15)/16
323 uint16_t huffman_table
[80][32][2];
325 uint32_t filter_limit_values
[64];
326 int bounding_values_array
[256];
329 /************************************************************************
330 * VP3 specific functions
331 ************************************************************************/
334 * This function sets up all of the various blocks mappings:
335 * superblocks <-> fragments, macroblocks <-> fragments,
336 * superblocks <-> macroblocks
338 * Returns 0 is successful; returns 1 if *anything* went wrong.
340 static int init_block_mapping(Vp3DecodeContext
*s
)
343 signed int hilbert_walk_mb
[4];
345 int current_fragment
= 0;
346 int current_width
= 0;
347 int current_height
= 0;
350 int superblock_row_inc
= 0;
352 int mapping_index
= 0;
354 int current_macroblock
;
357 signed char travel_width
[16] = {
364 signed char travel_height
[16] = {
371 signed char travel_width_mb
[4] = {
375 signed char travel_height_mb
[4] = {
379 debug_vp3(" vp3: initialize block mapping tables\n");
381 hilbert_walk_mb
[0] = 1;
382 hilbert_walk_mb
[1] = s
->macroblock_width
;
383 hilbert_walk_mb
[2] = 1;
384 hilbert_walk_mb
[3] = -s
->macroblock_width
;
386 /* iterate through each superblock (all planes) and map the fragments */
387 for (i
= 0; i
< s
->superblock_count
; i
++) {
388 debug_init(" superblock %d (u starts @ %d, v starts @ %d)\n",
389 i
, s
->u_superblock_start
, s
->v_superblock_start
);
391 /* time to re-assign the limits? */
394 /* start of Y superblocks */
395 right_edge
= s
->fragment_width
;
396 bottom_edge
= s
->fragment_height
;
399 superblock_row_inc
= 3 * s
->fragment_width
-
400 (s
->y_superblock_width
* 4 - s
->fragment_width
);
402 /* the first operation for this variable is to advance by 1 */
403 current_fragment
= -1;
405 } else if (i
== s
->u_superblock_start
) {
407 /* start of U superblocks */
408 right_edge
= s
->fragment_width
/ 2;
409 bottom_edge
= s
->fragment_height
/ 2;
412 superblock_row_inc
= 3 * (s
->fragment_width
/ 2) -
413 (s
->c_superblock_width
* 4 - s
->fragment_width
/ 2);
415 /* the first operation for this variable is to advance by 1 */
416 current_fragment
= s
->fragment_start
[1] - 1;
418 } else if (i
== s
->v_superblock_start
) {
420 /* start of V superblocks */
421 right_edge
= s
->fragment_width
/ 2;
422 bottom_edge
= s
->fragment_height
/ 2;
425 superblock_row_inc
= 3 * (s
->fragment_width
/ 2) -
426 (s
->c_superblock_width
* 4 - s
->fragment_width
/ 2);
428 /* the first operation for this variable is to advance by 1 */
429 current_fragment
= s
->fragment_start
[2] - 1;
433 if (current_width
>= right_edge
- 1) {
434 /* reset width and move to next superblock row */
438 /* fragment is now at the start of a new superblock row */
439 current_fragment
+= superblock_row_inc
;
442 /* iterate through all 16 fragments in a superblock */
443 for (j
= 0; j
< 16; j
++) {
444 current_fragment
+= travel_width
[j
] + right_edge
* travel_height
[j
];
445 current_width
+= travel_width
[j
];
446 current_height
+= travel_height
[j
];
448 /* check if the fragment is in bounds */
449 if ((current_width
< right_edge
) &&
450 (current_height
< bottom_edge
)) {
451 s
->superblock_fragments
[mapping_index
] = current_fragment
;
452 debug_init(" mapping fragment %d to superblock %d, position %d (%d/%d x %d/%d)\n",
453 s
->superblock_fragments
[mapping_index
], i
, j
,
454 current_width
, right_edge
, current_height
, bottom_edge
);
456 s
->superblock_fragments
[mapping_index
] = -1;
457 debug_init(" superblock %d, position %d has no fragment (%d/%d x %d/%d)\n",
459 current_width
, right_edge
, current_height
, bottom_edge
);
466 /* initialize the superblock <-> macroblock mapping; iterate through
467 * all of the Y plane superblocks to build this mapping */
468 right_edge
= s
->macroblock_width
;
469 bottom_edge
= s
->macroblock_height
;
472 superblock_row_inc
= s
->macroblock_width
-
473 (s
->y_superblock_width
* 2 - s
->macroblock_width
);
474 hilbert
= hilbert_walk_mb
;
476 current_macroblock
= -1;
477 for (i
= 0; i
< s
->u_superblock_start
; i
++) {
479 if (current_width
>= right_edge
- 1) {
480 /* reset width and move to next superblock row */
484 /* macroblock is now at the start of a new superblock row */
485 current_macroblock
+= superblock_row_inc
;
488 /* iterate through each potential macroblock in the superblock */
489 for (j
= 0; j
< 4; j
++) {
490 current_macroblock
+= hilbert_walk_mb
[j
];
491 current_width
+= travel_width_mb
[j
];
492 current_height
+= travel_height_mb
[j
];
494 /* check if the macroblock is in bounds */
495 if ((current_width
< right_edge
) &&
496 (current_height
< bottom_edge
)) {
497 s
->superblock_macroblocks
[mapping_index
] = current_macroblock
;
498 debug_init(" mapping macroblock %d to superblock %d, position %d (%d/%d x %d/%d)\n",
499 s
->superblock_macroblocks
[mapping_index
], i
, j
,
500 current_width
, right_edge
, current_height
, bottom_edge
);
502 s
->superblock_macroblocks
[mapping_index
] = -1;
503 debug_init(" superblock %d, position %d has no macroblock (%d/%d x %d/%d)\n",
505 current_width
, right_edge
, current_height
, bottom_edge
);
512 /* initialize the macroblock <-> fragment mapping */
513 current_fragment
= 0;
514 current_macroblock
= 0;
516 for (i
= 0; i
< s
->fragment_height
; i
+= 2) {
518 for (j
= 0; j
< s
->fragment_width
; j
+= 2) {
520 debug_init(" macroblock %d contains fragments: ", current_macroblock
);
521 s
->all_fragments
[current_fragment
].macroblock
= current_macroblock
;
522 s
->macroblock_fragments
[mapping_index
++] = current_fragment
;
523 debug_init("%d ", current_fragment
);
525 if (j
+ 1 < s
->fragment_width
) {
526 s
->all_fragments
[current_fragment
+ 1].macroblock
= current_macroblock
;
527 s
->macroblock_fragments
[mapping_index
++] = current_fragment
+ 1;
528 debug_init("%d ", current_fragment
+ 1);
530 s
->macroblock_fragments
[mapping_index
++] = -1;
532 if (i
+ 1 < s
->fragment_height
) {
533 s
->all_fragments
[current_fragment
+ s
->fragment_width
].macroblock
=
535 s
->macroblock_fragments
[mapping_index
++] =
536 current_fragment
+ s
->fragment_width
;
537 debug_init("%d ", current_fragment
+ s
->fragment_width
);
539 s
->macroblock_fragments
[mapping_index
++] = -1;
541 if ((j
+ 1 < s
->fragment_width
) && (i
+ 1 < s
->fragment_height
)) {
542 s
->all_fragments
[current_fragment
+ s
->fragment_width
+ 1].macroblock
=
544 s
->macroblock_fragments
[mapping_index
++] =
545 current_fragment
+ s
->fragment_width
+ 1;
546 debug_init("%d ", current_fragment
+ s
->fragment_width
+ 1);
548 s
->macroblock_fragments
[mapping_index
++] = -1;
551 c_fragment
= s
->fragment_start
[1] +
552 (i
* s
->fragment_width
/ 4) + (j
/ 2);
553 s
->all_fragments
[c_fragment
].macroblock
= s
->macroblock_count
;
554 s
->macroblock_fragments
[mapping_index
++] = c_fragment
;
555 debug_init("%d ", c_fragment
);
557 c_fragment
= s
->fragment_start
[2] +
558 (i
* s
->fragment_width
/ 4) + (j
/ 2);
559 s
->all_fragments
[c_fragment
].macroblock
= s
->macroblock_count
;
560 s
->macroblock_fragments
[mapping_index
++] = c_fragment
;
561 debug_init("%d ", c_fragment
);
565 if (j
+ 2 <= s
->fragment_width
)
566 current_fragment
+= 2;
569 current_macroblock
++;
572 current_fragment
+= s
->fragment_width
;
575 return 0; /* successful path out */
579 * This function wipes out all of the fragment data.
581 static void init_frame(Vp3DecodeContext
*s
, GetBitContext
*gb
)
585 /* zero out all of the fragment information */
586 s
->coded_fragment_list_index
= 0;
587 for (i
= 0; i
< s
->fragment_count
; i
++) {
588 s
->coeff_counts
[i
] = 0;
589 s
->all_fragments
[i
].motion_x
= 127;
590 s
->all_fragments
[i
].motion_y
= 127;
591 s
->all_fragments
[i
].next_coeff
= NULL
;
593 s
->coeffs
[i
].coeff
=0;
594 s
->coeffs
[i
].next
= NULL
;
599 * This function sets up the dequantization tables used for a particular
602 static void init_dequantizer(Vp3DecodeContext
*s
)
604 int ac_scale_factor
= s
->coded_ac_scale_factor
[s
->quality_index
];
605 int dc_scale_factor
= s
->coded_dc_scale_factor
[s
->quality_index
];
606 int i
, plane
, inter
, qri
, bmi
, bmj
, qistart
;
608 debug_vp3(" vp3: initializing dequantization tables\n");
610 for(inter
=0; inter
<2; inter
++){
611 for(plane
=0; plane
<3; plane
++){
613 for(qri
=0; qri
<s
->qr_count
[inter
][plane
]; qri
++){
614 sum
+= s
->qr_size
[inter
][plane
][qri
];
615 if(s
->quality_index
<= sum
)
618 qistart
= sum
- s
->qr_size
[inter
][plane
][qri
];
619 bmi
= s
->qr_base
[inter
][plane
][qri
];
620 bmj
= s
->qr_base
[inter
][plane
][qri
+1];
622 int coeff
= ( 2*(sum
-s
->quality_index
)*s
->base_matrix
[bmi
][i
]
623 - 2*(qistart
-s
->quality_index
)*s
->base_matrix
[bmj
][i
]
624 + s
->qr_size
[inter
][plane
][qri
])
625 / (2*s
->qr_size
[inter
][plane
][qri
]);
627 int qmin
= 8<<(inter
+ !i
);
628 int qscale
= i
? ac_scale_factor
: dc_scale_factor
;
630 s
->qmat
[inter
][plane
][i
]= av_clip((qscale
* coeff
)/100 * 4, qmin
, 4096);
635 memset(s
->qscale_table
, (FFMAX(s
->qmat
[0][0][1], s
->qmat
[0][1][1])+8)/16, 512); //FIXME finetune
639 * This function initializes the loop filter boundary limits if the frame's
640 * quality index is different from the previous frame's.
642 static void init_loop_filter(Vp3DecodeContext
*s
)
644 int *bounding_values
= s
->bounding_values_array
+127;
648 filter_limit
= s
->filter_limit_values
[s
->quality_index
];
650 /* set up the bounding values */
651 memset(s
->bounding_values_array
, 0, 256 * sizeof(int));
652 for (x
= 0; x
< filter_limit
; x
++) {
653 bounding_values
[-x
- filter_limit
] = -filter_limit
+ x
;
654 bounding_values
[-x
] = -x
;
655 bounding_values
[x
] = x
;
656 bounding_values
[x
+ filter_limit
] = filter_limit
- x
;
661 * This function unpacks all of the superblock/macroblock/fragment coding
662 * information from the bitstream.
664 static int unpack_superblocks(Vp3DecodeContext
*s
, GetBitContext
*gb
)
667 int current_superblock
= 0;
669 int decode_fully_flags
= 0;
670 int decode_partial_blocks
= 0;
671 int first_c_fragment_seen
;
674 int current_fragment
;
676 debug_vp3(" vp3: unpacking superblock coding\n");
680 debug_vp3(" keyframe-- all superblocks are fully coded\n");
681 memset(s
->superblock_coding
, SB_FULLY_CODED
, s
->superblock_count
);
685 /* unpack the list of partially-coded superblocks */
687 /* toggle the bit because as soon as the first run length is
688 * fetched the bit will be toggled again */
690 while (current_superblock
< s
->superblock_count
) {
691 if (current_run
-- == 0) {
693 current_run
= get_vlc2(gb
,
694 s
->superblock_run_length_vlc
.table
, 6, 2);
695 if (current_run
== 33)
696 current_run
+= get_bits(gb
, 12);
697 debug_block_coding(" setting superblocks %d..%d to %s\n",
699 current_superblock
+ current_run
- 1,
700 (bit
) ? "partially coded" : "not coded");
702 /* if any of the superblocks are not partially coded, flag
703 * a boolean to decode the list of fully-coded superblocks */
705 decode_fully_flags
= 1;
708 /* make a note of the fact that there are partially coded
710 decode_partial_blocks
= 1;
713 s
->superblock_coding
[current_superblock
++] = bit
;
716 /* unpack the list of fully coded superblocks if any of the blocks were
717 * not marked as partially coded in the previous step */
718 if (decode_fully_flags
) {
720 current_superblock
= 0;
723 /* toggle the bit because as soon as the first run length is
724 * fetched the bit will be toggled again */
726 while (current_superblock
< s
->superblock_count
) {
728 /* skip any superblocks already marked as partially coded */
729 if (s
->superblock_coding
[current_superblock
] == SB_NOT_CODED
) {
731 if (current_run
-- == 0) {
733 current_run
= get_vlc2(gb
,
734 s
->superblock_run_length_vlc
.table
, 6, 2);
735 if (current_run
== 33)
736 current_run
+= get_bits(gb
, 12);
739 debug_block_coding(" setting superblock %d to %s\n",
741 (bit
) ? "fully coded" : "not coded");
742 s
->superblock_coding
[current_superblock
] = 2*bit
;
744 current_superblock
++;
748 /* if there were partial blocks, initialize bitstream for
749 * unpacking fragment codings */
750 if (decode_partial_blocks
) {
754 /* toggle the bit because as soon as the first run length is
755 * fetched the bit will be toggled again */
760 /* figure out which fragments are coded; iterate through each
761 * superblock (all planes) */
762 s
->coded_fragment_list_index
= 0;
763 s
->next_coeff
= s
->coeffs
+ s
->fragment_count
;
764 s
->first_coded_y_fragment
= s
->first_coded_c_fragment
= 0;
765 s
->last_coded_y_fragment
= s
->last_coded_c_fragment
= -1;
766 first_c_fragment_seen
= 0;
767 memset(s
->macroblock_coding
, MODE_COPY
, s
->macroblock_count
);
768 for (i
= 0; i
< s
->superblock_count
; i
++) {
770 /* iterate through all 16 fragments in a superblock */
771 for (j
= 0; j
< 16; j
++) {
773 /* if the fragment is in bounds, check its coding status */
774 current_fragment
= s
->superblock_fragments
[i
* 16 + j
];
775 if (current_fragment
>= s
->fragment_count
) {
776 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
777 current_fragment
, s
->fragment_count
);
780 if (current_fragment
!= -1) {
781 if (s
->superblock_coding
[i
] == SB_NOT_CODED
) {
783 /* copy all the fragments from the prior frame */
784 s
->all_fragments
[current_fragment
].coding_method
=
787 } else if (s
->superblock_coding
[i
] == SB_PARTIALLY_CODED
) {
789 /* fragment may or may not be coded; this is the case
790 * that cares about the fragment coding runs */
791 if (current_run
-- == 0) {
793 current_run
= get_vlc2(gb
,
794 s
->fragment_run_length_vlc
.table
, 5, 2);
798 /* default mode; actual mode will be decoded in
800 s
->all_fragments
[current_fragment
].coding_method
=
802 s
->all_fragments
[current_fragment
].next_coeff
= s
->coeffs
+ current_fragment
;
803 s
->coded_fragment_list
[s
->coded_fragment_list_index
] =
805 if ((current_fragment
>= s
->fragment_start
[1]) &&
806 (s
->last_coded_y_fragment
== -1) &&
807 (!first_c_fragment_seen
)) {
808 s
->first_coded_c_fragment
= s
->coded_fragment_list_index
;
809 s
->last_coded_y_fragment
= s
->first_coded_c_fragment
- 1;
810 first_c_fragment_seen
= 1;
812 s
->coded_fragment_list_index
++;
813 s
->macroblock_coding
[s
->all_fragments
[current_fragment
].macroblock
] = MODE_INTER_NO_MV
;
814 debug_block_coding(" superblock %d is partially coded, fragment %d is coded\n",
815 i
, current_fragment
);
817 /* not coded; copy this fragment from the prior frame */
818 s
->all_fragments
[current_fragment
].coding_method
=
820 debug_block_coding(" superblock %d is partially coded, fragment %d is not coded\n",
821 i
, current_fragment
);
826 /* fragments are fully coded in this superblock; actual
827 * coding will be determined in next step */
828 s
->all_fragments
[current_fragment
].coding_method
=
830 s
->all_fragments
[current_fragment
].next_coeff
= s
->coeffs
+ current_fragment
;
831 s
->coded_fragment_list
[s
->coded_fragment_list_index
] =
833 if ((current_fragment
>= s
->fragment_start
[1]) &&
834 (s
->last_coded_y_fragment
== -1) &&
835 (!first_c_fragment_seen
)) {
836 s
->first_coded_c_fragment
= s
->coded_fragment_list_index
;
837 s
->last_coded_y_fragment
= s
->first_coded_c_fragment
- 1;
838 first_c_fragment_seen
= 1;
840 s
->coded_fragment_list_index
++;
841 s
->macroblock_coding
[s
->all_fragments
[current_fragment
].macroblock
] = MODE_INTER_NO_MV
;
842 debug_block_coding(" superblock %d is fully coded, fragment %d is coded\n",
843 i
, current_fragment
);
849 if (!first_c_fragment_seen
)
850 /* only Y fragments coded in this frame */
851 s
->last_coded_y_fragment
= s
->coded_fragment_list_index
- 1;
853 /* end the list of coded C fragments */
854 s
->last_coded_c_fragment
= s
->coded_fragment_list_index
- 1;
856 debug_block_coding(" %d total coded fragments, y: %d -> %d, c: %d -> %d\n",
857 s
->coded_fragment_list_index
,
858 s
->first_coded_y_fragment
,
859 s
->last_coded_y_fragment
,
860 s
->first_coded_c_fragment
,
861 s
->last_coded_c_fragment
);
867 * This function unpacks all the coding mode data for individual macroblocks
868 * from the bitstream.
870 static int unpack_modes(Vp3DecodeContext
*s
, GetBitContext
*gb
)
874 int current_macroblock
;
875 int current_fragment
;
877 int custom_mode_alphabet
[CODING_MODE_COUNT
];
879 debug_vp3(" vp3: unpacking encoding modes\n");
882 debug_vp3(" keyframe-- all blocks are coded as INTRA\n");
884 for (i
= 0; i
< s
->fragment_count
; i
++)
885 s
->all_fragments
[i
].coding_method
= MODE_INTRA
;
889 /* fetch the mode coding scheme for this frame */
890 scheme
= get_bits(gb
, 3);
891 debug_modes(" using mode alphabet %d\n", scheme
);
893 /* is it a custom coding scheme? */
895 debug_modes(" custom mode alphabet ahead:\n");
896 for (i
= 0; i
< 8; i
++)
897 custom_mode_alphabet
[get_bits(gb
, 3)] = i
;
900 for (i
= 0; i
< 8; i
++) {
902 debug_modes(" mode[%d][%d] = %d\n", scheme
, i
,
903 ModeAlphabet
[scheme
-1][i
]);
905 debug_modes(" mode[0][%d] = %d\n", i
,
906 custom_mode_alphabet
[i
]);
909 /* iterate through all of the macroblocks that contain 1 or more
911 for (i
= 0; i
< s
->u_superblock_start
; i
++) {
913 for (j
= 0; j
< 4; j
++) {
914 current_macroblock
= s
->superblock_macroblocks
[i
* 4 + j
];
915 if ((current_macroblock
== -1) ||
916 (s
->macroblock_coding
[current_macroblock
] == MODE_COPY
))
918 if (current_macroblock
>= s
->macroblock_count
) {
919 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_modes(): bad macroblock number (%d >= %d)\n",
920 current_macroblock
, s
->macroblock_count
);
924 /* mode 7 means get 3 bits for each coding mode */
926 coding_mode
= get_bits(gb
, 3);
928 coding_mode
= custom_mode_alphabet
929 [get_vlc2(gb
, s
->mode_code_vlc
.table
, 3, 3)];
931 coding_mode
= ModeAlphabet
[scheme
-1]
932 [get_vlc2(gb
, s
->mode_code_vlc
.table
, 3, 3)];
934 s
->macroblock_coding
[current_macroblock
] = coding_mode
;
935 for (k
= 0; k
< 6; k
++) {
937 s
->macroblock_fragments
[current_macroblock
* 6 + k
];
938 if (current_fragment
== -1)
940 if (current_fragment
>= s
->fragment_count
) {
941 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_modes(): bad fragment number (%d >= %d)\n",
942 current_fragment
, s
->fragment_count
);
945 if (s
->all_fragments
[current_fragment
].coding_method
!=
947 s
->all_fragments
[current_fragment
].coding_method
=
951 debug_modes(" coding method for macroblock starting @ fragment %d = %d\n",
952 s
->macroblock_fragments
[current_macroblock
* 6], coding_mode
);
961 * This function unpacks all the motion vectors for the individual
962 * macroblocks from the bitstream.
964 static int unpack_vectors(Vp3DecodeContext
*s
, GetBitContext
*gb
)
970 int last_motion_x
= 0;
971 int last_motion_y
= 0;
972 int prior_last_motion_x
= 0;
973 int prior_last_motion_y
= 0;
974 int current_macroblock
;
975 int current_fragment
;
977 debug_vp3(" vp3: unpacking motion vectors\n");
980 debug_vp3(" keyframe-- there are no motion vectors\n");
984 memset(motion_x
, 0, 6 * sizeof(int));
985 memset(motion_y
, 0, 6 * sizeof(int));
987 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
988 coding_mode
= get_bits1(gb
);
989 debug_vectors(" using %s scheme for unpacking motion vectors\n",
990 (coding_mode
== 0) ? "VLC" : "fixed-length");
992 /* iterate through all of the macroblocks that contain 1 or more
994 for (i
= 0; i
< s
->u_superblock_start
; i
++) {
996 for (j
= 0; j
< 4; j
++) {
997 current_macroblock
= s
->superblock_macroblocks
[i
* 4 + j
];
998 if ((current_macroblock
== -1) ||
999 (s
->macroblock_coding
[current_macroblock
] == MODE_COPY
))
1001 if (current_macroblock
>= s
->macroblock_count
) {
1002 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_vectors(): bad macroblock number (%d >= %d)\n",
1003 current_macroblock
, s
->macroblock_count
);
1007 current_fragment
= s
->macroblock_fragments
[current_macroblock
* 6];
1008 if (current_fragment
>= s
->fragment_count
) {
1009 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_vectors(): bad fragment number (%d >= %d\n",
1010 current_fragment
, s
->fragment_count
);
1013 switch (s
->macroblock_coding
[current_macroblock
]) {
1015 case MODE_INTER_PLUS_MV
:
1016 case MODE_GOLDEN_MV
:
1017 /* all 6 fragments use the same motion vector */
1018 if (coding_mode
== 0) {
1019 motion_x
[0] = motion_vector_table
[get_vlc2(gb
, s
->motion_vector_vlc
.table
, 6, 2)];
1020 motion_y
[0] = motion_vector_table
[get_vlc2(gb
, s
->motion_vector_vlc
.table
, 6, 2)];
1022 motion_x
[0] = fixed_motion_vector_table
[get_bits(gb
, 6)];
1023 motion_y
[0] = fixed_motion_vector_table
[get_bits(gb
, 6)];
1026 for (k
= 1; k
< 6; k
++) {
1027 motion_x
[k
] = motion_x
[0];
1028 motion_y
[k
] = motion_y
[0];
1031 /* vector maintenance, only on MODE_INTER_PLUS_MV */
1032 if (s
->macroblock_coding
[current_macroblock
] ==
1033 MODE_INTER_PLUS_MV
) {
1034 prior_last_motion_x
= last_motion_x
;
1035 prior_last_motion_y
= last_motion_y
;
1036 last_motion_x
= motion_x
[0];
1037 last_motion_y
= motion_y
[0];
1041 case MODE_INTER_FOURMV
:
1042 /* vector maintenance */
1043 prior_last_motion_x
= last_motion_x
;
1044 prior_last_motion_y
= last_motion_y
;
1046 /* fetch 4 vectors from the bitstream, one for each
1047 * Y fragment, then average for the C fragment vectors */
1048 motion_x
[4] = motion_y
[4] = 0;
1049 for (k
= 0; k
< 4; k
++) {
1050 for (l
= 0; l
< s
->coded_fragment_list_index
; l
++)
1051 if (s
->coded_fragment_list
[l
] == s
->macroblock_fragments
[6*current_macroblock
+ k
])
1053 if (l
< s
->coded_fragment_list_index
) {
1054 if (coding_mode
== 0) {
1055 motion_x
[k
] = motion_vector_table
[get_vlc2(gb
, s
->motion_vector_vlc
.table
, 6, 2)];
1056 motion_y
[k
] = motion_vector_table
[get_vlc2(gb
, s
->motion_vector_vlc
.table
, 6, 2)];
1058 motion_x
[k
] = fixed_motion_vector_table
[get_bits(gb
, 6)];
1059 motion_y
[k
] = fixed_motion_vector_table
[get_bits(gb
, 6)];
1061 last_motion_x
= motion_x
[k
];
1062 last_motion_y
= motion_y
[k
];
1067 motion_x
[4] += motion_x
[k
];
1068 motion_y
[4] += motion_y
[k
];
1072 motion_x
[4]= RSHIFT(motion_x
[4], 2);
1074 motion_y
[4]= RSHIFT(motion_y
[4], 2);
1077 case MODE_INTER_LAST_MV
:
1078 /* all 6 fragments use the last motion vector */
1079 motion_x
[0] = last_motion_x
;
1080 motion_y
[0] = last_motion_y
;
1081 for (k
= 1; k
< 6; k
++) {
1082 motion_x
[k
] = motion_x
[0];
1083 motion_y
[k
] = motion_y
[0];
1086 /* no vector maintenance (last vector remains the
1090 case MODE_INTER_PRIOR_LAST
:
1091 /* all 6 fragments use the motion vector prior to the
1092 * last motion vector */
1093 motion_x
[0] = prior_last_motion_x
;
1094 motion_y
[0] = prior_last_motion_y
;
1095 for (k
= 1; k
< 6; k
++) {
1096 motion_x
[k
] = motion_x
[0];
1097 motion_y
[k
] = motion_y
[0];
1100 /* vector maintenance */
1101 prior_last_motion_x
= last_motion_x
;
1102 prior_last_motion_y
= last_motion_y
;
1103 last_motion_x
= motion_x
[0];
1104 last_motion_y
= motion_y
[0];
1108 /* covers intra, inter without MV, golden without MV */
1109 memset(motion_x
, 0, 6 * sizeof(int));
1110 memset(motion_y
, 0, 6 * sizeof(int));
1112 /* no vector maintenance */
1116 /* assign the motion vectors to the correct fragments */
1117 debug_vectors(" vectors for macroblock starting @ fragment %d (coding method %d):\n",
1119 s
->macroblock_coding
[current_macroblock
]);
1120 for (k
= 0; k
< 6; k
++) {
1122 s
->macroblock_fragments
[current_macroblock
* 6 + k
];
1123 if (current_fragment
== -1)
1125 if (current_fragment
>= s
->fragment_count
) {
1126 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_vectors(): bad fragment number (%d >= %d)\n",
1127 current_fragment
, s
->fragment_count
);
1130 s
->all_fragments
[current_fragment
].motion_x
= motion_x
[k
];
1131 s
->all_fragments
[current_fragment
].motion_y
= motion_y
[k
];
1132 debug_vectors(" vector %d: fragment %d = (%d, %d)\n",
1133 k
, current_fragment
, motion_x
[k
], motion_y
[k
]);
1143 * This function is called by unpack_dct_coeffs() to extract the VLCs from
1144 * the bitstream. The VLCs encode tokens which are used to unpack DCT
1145 * data. This function unpacks all the VLCs for either the Y plane or both
1146 * C planes, and is called for DC coefficients or different AC coefficient
1147 * levels (since different coefficient types require different VLC tables.
1149 * This function returns a residual eob run. E.g, if a particular token gave
1150 * instructions to EOB the next 5 fragments and there were only 2 fragments
1151 * left in the current fragment range, 3 would be returned so that it could
1152 * be passed into the next call to this same function.
1154 static int unpack_vlcs(Vp3DecodeContext
*s
, GetBitContext
*gb
,
1155 VLC
*table
, int coeff_index
,
1156 int first_fragment
, int last_fragment
,
1163 Vp3Fragment
*fragment
;
1164 uint8_t *perm
= s
->scantable
.permutated
;
1167 if ((first_fragment
>= s
->fragment_count
) ||
1168 (last_fragment
>= s
->fragment_count
)) {
1170 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_vlcs(): bad fragment number (%d -> %d ?)\n",
1171 first_fragment
, last_fragment
);
1175 for (i
= first_fragment
; i
<= last_fragment
; i
++) {
1176 int fragment_num
= s
->coded_fragment_list
[i
];
1178 if (s
->coeff_counts
[fragment_num
] > coeff_index
)
1180 fragment
= &s
->all_fragments
[fragment_num
];
1183 /* decode a VLC into a token */
1184 token
= get_vlc2(gb
, table
->table
, 5, 3);
1185 debug_vlc(" token = %2d, ", token
);
1186 /* use the token to get a zero run, a coefficient, and an eob run */
1188 eob_run
= eob_run_base
[token
];
1189 if (eob_run_get_bits
[token
])
1190 eob_run
+= get_bits(gb
, eob_run_get_bits
[token
]);
1191 coeff
= zero_run
= 0;
1193 bits_to_get
= coeff_get_bits
[token
];
1195 coeff
= coeff_tables
[token
][0];
1197 coeff
= coeff_tables
[token
][get_bits(gb
, bits_to_get
)];
1199 zero_run
= zero_run_base
[token
];
1200 if (zero_run_get_bits
[token
])
1201 zero_run
+= get_bits(gb
, zero_run_get_bits
[token
]);
1206 s
->coeff_counts
[fragment_num
] += zero_run
;
1207 if (s
->coeff_counts
[fragment_num
] < 64){
1208 fragment
->next_coeff
->coeff
= coeff
;
1209 fragment
->next_coeff
->index
= perm
[s
->coeff_counts
[fragment_num
]++]; //FIXME perm here already?
1210 fragment
->next_coeff
->next
= s
->next_coeff
;
1211 s
->next_coeff
->next
=NULL
;
1212 fragment
->next_coeff
= s
->next_coeff
++;
1214 debug_vlc(" fragment %d coeff = %d\n",
1215 s
->coded_fragment_list
[i
], fragment
->next_coeff
[coeff_index
]);
1217 s
->coeff_counts
[fragment_num
] |= 128;
1218 debug_vlc(" fragment %d eob with %d coefficients\n",
1219 s
->coded_fragment_list
[i
], s
->coeff_counts
[fragment_num
]&127);
1228 * This function unpacks all of the DCT coefficient data from the
1231 static int unpack_dct_coeffs(Vp3DecodeContext
*s
, GetBitContext
*gb
)
1238 int residual_eob_run
= 0;
1240 /* fetch the DC table indexes */
1241 dc_y_table
= get_bits(gb
, 4);
1242 dc_c_table
= get_bits(gb
, 4);
1244 /* unpack the Y plane DC coefficients */
1245 debug_vp3(" vp3: unpacking Y plane DC coefficients using table %d\n",
1247 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->dc_vlc
[dc_y_table
], 0,
1248 s
->first_coded_y_fragment
, s
->last_coded_y_fragment
, residual_eob_run
);
1250 /* unpack the C plane DC coefficients */
1251 debug_vp3(" vp3: unpacking C plane DC coefficients using table %d\n",
1253 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->dc_vlc
[dc_c_table
], 0,
1254 s
->first_coded_c_fragment
, s
->last_coded_c_fragment
, residual_eob_run
);
1256 /* fetch the AC table indexes */
1257 ac_y_table
= get_bits(gb
, 4);
1258 ac_c_table
= get_bits(gb
, 4);
1260 /* unpack the group 1 AC coefficients (coeffs 1-5) */
1261 for (i
= 1; i
<= 5; i
++) {
1263 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1265 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_1
[ac_y_table
], i
,
1266 s
->first_coded_y_fragment
, s
->last_coded_y_fragment
, residual_eob_run
);
1268 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1270 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_1
[ac_c_table
], i
,
1271 s
->first_coded_c_fragment
, s
->last_coded_c_fragment
, residual_eob_run
);
1274 /* unpack the group 2 AC coefficients (coeffs 6-14) */
1275 for (i
= 6; i
<= 14; i
++) {
1277 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1279 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_2
[ac_y_table
], i
,
1280 s
->first_coded_y_fragment
, s
->last_coded_y_fragment
, residual_eob_run
);
1282 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1284 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_2
[ac_c_table
], i
,
1285 s
->first_coded_c_fragment
, s
->last_coded_c_fragment
, residual_eob_run
);
1288 /* unpack the group 3 AC coefficients (coeffs 15-27) */
1289 for (i
= 15; i
<= 27; i
++) {
1291 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1293 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_3
[ac_y_table
], i
,
1294 s
->first_coded_y_fragment
, s
->last_coded_y_fragment
, residual_eob_run
);
1296 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1298 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_3
[ac_c_table
], i
,
1299 s
->first_coded_c_fragment
, s
->last_coded_c_fragment
, residual_eob_run
);
1302 /* unpack the group 4 AC coefficients (coeffs 28-63) */
1303 for (i
= 28; i
<= 63; i
++) {
1305 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1307 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_4
[ac_y_table
], i
,
1308 s
->first_coded_y_fragment
, s
->last_coded_y_fragment
, residual_eob_run
);
1310 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1312 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_4
[ac_c_table
], i
,
1313 s
->first_coded_c_fragment
, s
->last_coded_c_fragment
, residual_eob_run
);
1320 * This function reverses the DC prediction for each coded fragment in
1321 * the frame. Much of this function is adapted directly from the original
1324 #define COMPATIBLE_FRAME(x) \
1325 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1326 #define FRAME_CODED(x) (s->all_fragments[x].coding_method != MODE_COPY)
1327 #define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
1329 static void reverse_dc_prediction(Vp3DecodeContext
*s
,
1332 int fragment_height
)
1341 int i
= first_fragment
;
1345 /* DC values for the left, up-left, up, and up-right fragments */
1346 int vl
, vul
, vu
, vur
;
1348 /* indexes for the left, up-left, up, and up-right fragments */
1352 * The 6 fields mean:
1353 * 0: up-left multiplier
1355 * 2: up-right multiplier
1356 * 3: left multiplier
1358 int predictor_transform
[16][4] = {
1360 { 0, 0, 0,128}, // PL
1361 { 0, 0,128, 0}, // PUR
1362 { 0, 0, 53, 75}, // PUR|PL
1363 { 0,128, 0, 0}, // PU
1364 { 0, 64, 0, 64}, // PU|PL
1365 { 0,128, 0, 0}, // PU|PUR
1366 { 0, 0, 53, 75}, // PU|PUR|PL
1367 {128, 0, 0, 0}, // PUL
1368 { 0, 0, 0,128}, // PUL|PL
1369 { 64, 0, 64, 0}, // PUL|PUR
1370 { 0, 0, 53, 75}, // PUL|PUR|PL
1371 { 0,128, 0, 0}, // PUL|PU
1372 {-104,116, 0,116}, // PUL|PU|PL
1373 { 24, 80, 24, 0}, // PUL|PU|PUR
1374 {-104,116, 0,116} // PUL|PU|PUR|PL
1377 /* This table shows which types of blocks can use other blocks for
1378 * prediction. For example, INTRA is the only mode in this table to
1379 * have a frame number of 0. That means INTRA blocks can only predict
1380 * from other INTRA blocks. There are 2 golden frame coding types;
1381 * blocks encoding in these modes can only predict from other blocks
1382 * that were encoded with these 1 of these 2 modes. */
1383 unsigned char compatible_frame
[8] = {
1384 1, /* MODE_INTER_NO_MV */
1386 1, /* MODE_INTER_PLUS_MV */
1387 1, /* MODE_INTER_LAST_MV */
1388 1, /* MODE_INTER_PRIOR_MV */
1389 2, /* MODE_USING_GOLDEN */
1390 2, /* MODE_GOLDEN_MV */
1391 1 /* MODE_INTER_FOUR_MV */
1393 int current_frame_type
;
1395 /* there is a last DC predictor for each of the 3 frame types */
1400 debug_vp3(" vp3: reversing DC prediction\n");
1402 vul
= vu
= vur
= vl
= 0;
1403 last_dc
[0] = last_dc
[1] = last_dc
[2] = 0;
1405 /* for each fragment row... */
1406 for (y
= 0; y
< fragment_height
; y
++) {
1408 /* for each fragment in a row... */
1409 for (x
= 0; x
< fragment_width
; x
++, i
++) {
1411 /* reverse prediction if this block was coded */
1412 if (s
->all_fragments
[i
].coding_method
!= MODE_COPY
) {
1414 current_frame_type
=
1415 compatible_frame
[s
->all_fragments
[i
].coding_method
];
1416 debug_dc_pred(" frag %d: orig DC = %d, ",
1423 if(FRAME_CODED(l
) && COMPATIBLE_FRAME(l
))
1427 u
= i
-fragment_width
;
1429 if(FRAME_CODED(u
) && COMPATIBLE_FRAME(u
))
1432 ul
= i
-fragment_width
-1;
1434 if(FRAME_CODED(ul
) && COMPATIBLE_FRAME(ul
))
1437 if(x
+ 1 < fragment_width
){
1438 ur
= i
-fragment_width
+1;
1440 if(FRAME_CODED(ur
) && COMPATIBLE_FRAME(ur
))
1445 debug_dc_pred("transform = %d, ", transform
);
1447 if (transform
== 0) {
1449 /* if there were no fragments to predict from, use last
1451 predicted_dc
= last_dc
[current_frame_type
];
1452 debug_dc_pred("from last DC (%d) = %d\n",
1453 current_frame_type
, DC_COEFF(i
));
1457 /* apply the appropriate predictor transform */
1459 (predictor_transform
[transform
][0] * vul
) +
1460 (predictor_transform
[transform
][1] * vu
) +
1461 (predictor_transform
[transform
][2] * vur
) +
1462 (predictor_transform
[transform
][3] * vl
);
1464 predicted_dc
/= 128;
1466 /* check for outranging on the [ul u l] and
1467 * [ul u ur l] predictors */
1468 if ((transform
== 13) || (transform
== 15)) {
1469 if (FFABS(predicted_dc
- vu
) > 128)
1471 else if (FFABS(predicted_dc
- vl
) > 128)
1473 else if (FFABS(predicted_dc
- vul
) > 128)
1477 debug_dc_pred("from pred DC = %d\n",
1481 /* at long last, apply the predictor */
1482 if(s
->coeffs
[i
].index
){
1483 *s
->next_coeff
= s
->coeffs
[i
];
1484 s
->coeffs
[i
].index
=0;
1485 s
->coeffs
[i
].coeff
=0;
1486 s
->coeffs
[i
].next
= s
->next_coeff
++;
1488 s
->coeffs
[i
].coeff
+= predicted_dc
;
1490 last_dc
[current_frame_type
] = DC_COEFF(i
);
1491 if(DC_COEFF(i
) && !(s
->coeff_counts
[i
]&127)){
1492 s
->coeff_counts
[i
]= 129;
1493 // s->all_fragments[i].next_coeff= s->next_coeff;
1494 s
->coeffs
[i
].next
= s
->next_coeff
;
1495 (s
->next_coeff
++)->next
=NULL
;
1503 static void horizontal_filter(unsigned char *first_pixel
, int stride
,
1504 int *bounding_values
);
1505 static void vertical_filter(unsigned char *first_pixel
, int stride
,
1506 int *bounding_values
);
1509 * Perform the final rendering for a particular slice of data.
1510 * The slice number ranges from 0..(macroblock_height - 1).
1512 static void render_slice(Vp3DecodeContext
*s
, int slice
)
1516 int16_t *dequantizer
;
1517 DECLARE_ALIGNED_16(DCTELEM
, block
[64]);
1518 int motion_x
= 0xdeadbeef, motion_y
= 0xdeadbeef;
1519 int motion_halfpel_index
;
1520 uint8_t *motion_source
;
1522 int current_macroblock_entry
= slice
* s
->macroblock_width
* 6;
1524 if (slice
>= s
->macroblock_height
)
1527 for (plane
= 0; plane
< 3; plane
++) {
1528 uint8_t *output_plane
= s
->current_frame
.data
[plane
];
1529 uint8_t * last_plane
= s
-> last_frame
.data
[plane
];
1530 uint8_t *golden_plane
= s
-> golden_frame
.data
[plane
];
1531 int stride
= s
->current_frame
.linesize
[plane
];
1532 int plane_width
= s
->width
>> !!plane
;
1533 int plane_height
= s
->height
>> !!plane
;
1534 int y
= slice
* FRAGMENT_PIXELS
<< !plane
;
1535 int slice_height
= y
+ (FRAGMENT_PIXELS
<< !plane
);
1536 int i
= s
->macroblock_fragments
[current_macroblock_entry
+ plane
+ 3*!!plane
];
1538 if (!s
->flipped_image
) stride
= -stride
;
1541 if(FFABS(stride
) > 2048)
1542 return; //various tables are fixed size
1544 /* for each fragment row in the slice (both of them)... */
1545 for (; y
< slice_height
; y
+= 8) {
1547 /* for each fragment in a row... */
1548 for (x
= 0; x
< plane_width
; x
+= 8, i
++) {
1550 if ((i
< 0) || (i
>= s
->fragment_count
)) {
1551 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:render_slice(): bad fragment number (%d)\n", i
);
1555 /* transform if this block was coded */
1556 if ((s
->all_fragments
[i
].coding_method
!= MODE_COPY
) &&
1557 !((s
->avctx
->flags
& CODEC_FLAG_GRAY
) && plane
)) {
1559 if ((s
->all_fragments
[i
].coding_method
== MODE_USING_GOLDEN
) ||
1560 (s
->all_fragments
[i
].coding_method
== MODE_GOLDEN_MV
))
1561 motion_source
= golden_plane
;
1563 motion_source
= last_plane
;
1565 motion_source
+= s
->all_fragments
[i
].first_pixel
;
1566 motion_halfpel_index
= 0;
1568 /* sort out the motion vector if this fragment is coded
1569 * using a motion vector method */
1570 if ((s
->all_fragments
[i
].coding_method
> MODE_INTRA
) &&
1571 (s
->all_fragments
[i
].coding_method
!= MODE_USING_GOLDEN
)) {
1573 motion_x
= s
->all_fragments
[i
].motion_x
;
1574 motion_y
= s
->all_fragments
[i
].motion_y
;
1576 motion_x
= (motion_x
>>1) | (motion_x
&1);
1577 motion_y
= (motion_y
>>1) | (motion_y
&1);
1580 src_x
= (motion_x
>>1) + x
;
1581 src_y
= (motion_y
>>1) + y
;
1582 if ((motion_x
== 127) || (motion_y
== 127))
1583 av_log(s
->avctx
, AV_LOG_ERROR
, " help! got invalid motion vector! (%X, %X)\n", motion_x
, motion_y
);
1585 motion_halfpel_index
= motion_x
& 0x01;
1586 motion_source
+= (motion_x
>> 1);
1588 motion_halfpel_index
|= (motion_y
& 0x01) << 1;
1589 motion_source
+= ((motion_y
>> 1) * stride
);
1591 if(src_x
<0 || src_y
<0 || src_x
+ 9 >= plane_width
|| src_y
+ 9 >= plane_height
){
1592 uint8_t *temp
= s
->edge_emu_buffer
;
1593 if(stride
<0) temp
-= 9*stride
;
1594 else temp
+= 9*stride
;
1596 ff_emulated_edge_mc(temp
, motion_source
, stride
, 9, 9, src_x
, src_y
, plane_width
, plane_height
);
1597 motion_source
= temp
;
1602 /* first, take care of copying a block from either the
1603 * previous or the golden frame */
1604 if (s
->all_fragments
[i
].coding_method
!= MODE_INTRA
) {
1605 /* Note, it is possible to implement all MC cases with
1606 put_no_rnd_pixels_l2 which would look more like the
1607 VP3 source but this would be slower as
1608 put_no_rnd_pixels_tab is better optimzed */
1609 if(motion_halfpel_index
!= 3){
1610 s
->dsp
.put_no_rnd_pixels_tab
[1][motion_halfpel_index
](
1611 output_plane
+ s
->all_fragments
[i
].first_pixel
,
1612 motion_source
, stride
, 8);
1614 int d
= (motion_x
^ motion_y
)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1615 s
->dsp
.put_no_rnd_pixels_l2
[1](
1616 output_plane
+ s
->all_fragments
[i
].first_pixel
,
1618 motion_source
+ stride
+ 1 + d
,
1621 dequantizer
= s
->qmat
[1][plane
];
1623 dequantizer
= s
->qmat
[0][plane
];
1626 /* dequantize the DCT coefficients */
1627 debug_idct("fragment %d, coding mode %d, DC = %d, dequant = %d:\n",
1628 i
, s
->all_fragments
[i
].coding_method
,
1629 DC_COEFF(i
), dequantizer
[0]);
1631 if(s
->avctx
->idct_algo
==FF_IDCT_VP3
){
1632 Coeff
*coeff
= s
->coeffs
+ i
;
1633 memset(block
, 0, sizeof(block
));
1635 block
[coeff
->index
]= coeff
->coeff
* dequantizer
[coeff
->index
];
1639 Coeff
*coeff
= s
->coeffs
+ i
;
1640 memset(block
, 0, sizeof(block
));
1642 block
[coeff
->index
]= (coeff
->coeff
* dequantizer
[coeff
->index
] + 2)>>2;
1647 /* invert DCT and place (or add) in final output */
1649 if (s
->all_fragments
[i
].coding_method
== MODE_INTRA
) {
1650 if(s
->avctx
->idct_algo
!=FF_IDCT_VP3
)
1653 output_plane
+ s
->all_fragments
[i
].first_pixel
,
1658 output_plane
+ s
->all_fragments
[i
].first_pixel
,
1663 debug_idct("block after idct_%s():\n",
1664 (s
->all_fragments
[i
].coding_method
== MODE_INTRA
)?
1666 for (m
= 0; m
< 8; m
++) {
1667 for (n
= 0; n
< 8; n
++) {
1668 debug_idct(" %3d", *(output_plane
+
1669 s
->all_fragments
[i
].first_pixel
+ (m
* stride
+ n
)));
1677 /* copy directly from the previous frame */
1678 s
->dsp
.put_pixels_tab
[1][0](
1679 output_plane
+ s
->all_fragments
[i
].first_pixel
,
1680 last_plane
+ s
->all_fragments
[i
].first_pixel
,
1685 /* perform the left edge filter if:
1686 * - the fragment is not on the left column
1687 * - the fragment is coded in this frame
1688 * - the fragment is not coded in this frame but the left
1689 * fragment is coded in this frame (this is done instead
1690 * of a right edge filter when rendering the left fragment
1691 * since this fragment is not available yet) */
1693 ((s
->all_fragments
[i
].coding_method
!= MODE_COPY
) ||
1694 ((s
->all_fragments
[i
].coding_method
== MODE_COPY
) &&
1695 (s
->all_fragments
[i
- 1].coding_method
!= MODE_COPY
)) )) {
1697 output_plane
+ s
->all_fragments
[i
].first_pixel
+ 7*stride
,
1698 -stride
, s
->bounding_values_array
+ 127);
1701 /* perform the top edge filter if:
1702 * - the fragment is not on the top row
1703 * - the fragment is coded in this frame
1704 * - the fragment is not coded in this frame but the above
1705 * fragment is coded in this frame (this is done instead
1706 * of a bottom edge filter when rendering the above
1707 * fragment since this fragment is not available yet) */
1709 ((s
->all_fragments
[i
].coding_method
!= MODE_COPY
) ||
1710 ((s
->all_fragments
[i
].coding_method
== MODE_COPY
) &&
1711 (s
->all_fragments
[i
- fragment_width
].coding_method
!= MODE_COPY
)) )) {
1713 output_plane
+ s
->all_fragments
[i
].first_pixel
- stride
,
1714 -stride
, s
->bounding_values_array
+ 127);
1721 /* this looks like a good place for slice dispatch... */
1723 * if (slice == s->macroblock_height - 1)
1724 * dispatch (both last slice & 2nd-to-last slice);
1725 * else if (slice > 0)
1726 * dispatch (slice - 1);
1732 static void horizontal_filter(unsigned char *first_pixel
, int stride
,
1733 int *bounding_values
)
1738 for (end
= first_pixel
+ 8*stride
; first_pixel
!= end
; first_pixel
+= stride
) {
1740 (first_pixel
[-2] - first_pixel
[ 1])
1741 +3*(first_pixel
[ 0] - first_pixel
[-1]);
1742 filter_value
= bounding_values
[(filter_value
+ 4) >> 3];
1743 first_pixel
[-1] = av_clip_uint8(first_pixel
[-1] + filter_value
);
1744 first_pixel
[ 0] = av_clip_uint8(first_pixel
[ 0] - filter_value
);
1748 static void vertical_filter(unsigned char *first_pixel
, int stride
,
1749 int *bounding_values
)
1753 const int nstride
= -stride
;
1755 for (end
= first_pixel
+ 8; first_pixel
< end
; first_pixel
++) {
1757 (first_pixel
[2 * nstride
] - first_pixel
[ stride
])
1758 +3*(first_pixel
[0 ] - first_pixel
[nstride
]);
1759 filter_value
= bounding_values
[(filter_value
+ 4) >> 3];
1760 first_pixel
[nstride
] = av_clip_uint8(first_pixel
[nstride
] + filter_value
);
1761 first_pixel
[0] = av_clip_uint8(first_pixel
[0] - filter_value
);
1765 static void apply_loop_filter(Vp3DecodeContext
*s
)
1769 int *bounding_values
= s
->bounding_values_array
+127;
1772 int bounding_values_array
[256];
1775 /* find the right loop limit value */
1776 for (x
= 63; x
>= 0; x
--) {
1777 if (vp31_ac_scale_factor
[x
] >= s
->quality_index
)
1780 filter_limit
= vp31_filter_limit_values
[s
->quality_index
];
1782 /* set up the bounding values */
1783 memset(bounding_values_array
, 0, 256 * sizeof(int));
1784 for (x
= 0; x
< filter_limit
; x
++) {
1785 bounding_values
[-x
- filter_limit
] = -filter_limit
+ x
;
1786 bounding_values
[-x
] = -x
;
1787 bounding_values
[x
] = x
;
1788 bounding_values
[x
+ filter_limit
] = filter_limit
- x
;
1792 for (plane
= 0; plane
< 3; plane
++) {
1793 int width
= s
->fragment_width
>> !!plane
;
1794 int height
= s
->fragment_height
>> !!plane
;
1795 int fragment
= s
->fragment_start
[plane
];
1796 int stride
= s
->current_frame
.linesize
[plane
];
1797 uint8_t *plane_data
= s
->current_frame
.data
[plane
];
1798 if (!s
->flipped_image
) stride
= -stride
;
1800 for (y
= 0; y
< height
; y
++) {
1802 for (x
= 0; x
< width
; x
++) {
1803 /* do not perform left edge filter for left columns frags */
1805 (s
->all_fragments
[fragment
].coding_method
!= MODE_COPY
)) {
1807 plane_data
+ s
->all_fragments
[fragment
].first_pixel
,
1808 stride
, bounding_values
);
1811 /* do not perform top edge filter for top row fragments */
1813 (s
->all_fragments
[fragment
].coding_method
!= MODE_COPY
)) {
1815 plane_data
+ s
->all_fragments
[fragment
].first_pixel
,
1816 stride
, bounding_values
);
1819 /* do not perform right edge filter for right column
1820 * fragments or if right fragment neighbor is also coded
1821 * in this frame (it will be filtered in next iteration) */
1822 if ((x
< width
- 1) &&
1823 (s
->all_fragments
[fragment
].coding_method
!= MODE_COPY
) &&
1824 (s
->all_fragments
[fragment
+ 1].coding_method
== MODE_COPY
)) {
1826 plane_data
+ s
->all_fragments
[fragment
+ 1].first_pixel
,
1827 stride
, bounding_values
);
1830 /* do not perform bottom edge filter for bottom row
1831 * fragments or if bottom fragment neighbor is also coded
1832 * in this frame (it will be filtered in the next row) */
1833 if ((y
< height
- 1) &&
1834 (s
->all_fragments
[fragment
].coding_method
!= MODE_COPY
) &&
1835 (s
->all_fragments
[fragment
+ width
].coding_method
== MODE_COPY
)) {
1837 plane_data
+ s
->all_fragments
[fragment
+ width
].first_pixel
,
1838 stride
, bounding_values
);
1848 * This function computes the first pixel addresses for each fragment.
1849 * This function needs to be invoked after the first frame is allocated
1850 * so that it has access to the plane strides.
1852 static void vp3_calculate_pixel_addresses(Vp3DecodeContext
*s
)
1857 /* figure out the first pixel addresses for each of the fragments */
1860 for (y
= s
->fragment_height
; y
> 0; y
--) {
1861 for (x
= 0; x
< s
->fragment_width
; x
++) {
1862 s
->all_fragments
[i
++].first_pixel
=
1863 s
->golden_frame
.linesize
[0] * y
* FRAGMENT_PIXELS
-
1864 s
->golden_frame
.linesize
[0] +
1865 x
* FRAGMENT_PIXELS
;
1866 debug_init(" fragment %d, first pixel @ %d\n",
1867 i
-1, s
->all_fragments
[i
-1].first_pixel
);
1872 i
= s
->fragment_start
[1];
1873 for (y
= s
->fragment_height
/ 2; y
> 0; y
--) {
1874 for (x
= 0; x
< s
->fragment_width
/ 2; x
++) {
1875 s
->all_fragments
[i
++].first_pixel
=
1876 s
->golden_frame
.linesize
[1] * y
* FRAGMENT_PIXELS
-
1877 s
->golden_frame
.linesize
[1] +
1878 x
* FRAGMENT_PIXELS
;
1879 debug_init(" fragment %d, first pixel @ %d\n",
1880 i
-1, s
->all_fragments
[i
-1].first_pixel
);
1885 i
= s
->fragment_start
[2];
1886 for (y
= s
->fragment_height
/ 2; y
> 0; y
--) {
1887 for (x
= 0; x
< s
->fragment_width
/ 2; x
++) {
1888 s
->all_fragments
[i
++].first_pixel
=
1889 s
->golden_frame
.linesize
[2] * y
* FRAGMENT_PIXELS
-
1890 s
->golden_frame
.linesize
[2] +
1891 x
* FRAGMENT_PIXELS
;
1892 debug_init(" fragment %d, first pixel @ %d\n",
1893 i
-1, s
->all_fragments
[i
-1].first_pixel
);
1898 /* FIXME: this should be merged with the above! */
1899 static void theora_calculate_pixel_addresses(Vp3DecodeContext
*s
)
1904 /* figure out the first pixel addresses for each of the fragments */
1907 for (y
= 1; y
<= s
->fragment_height
; y
++) {
1908 for (x
= 0; x
< s
->fragment_width
; x
++) {
1909 s
->all_fragments
[i
++].first_pixel
=
1910 s
->golden_frame
.linesize
[0] * y
* FRAGMENT_PIXELS
-
1911 s
->golden_frame
.linesize
[0] +
1912 x
* FRAGMENT_PIXELS
;
1913 debug_init(" fragment %d, first pixel @ %d\n",
1914 i
-1, s
->all_fragments
[i
-1].first_pixel
);
1919 i
= s
->fragment_start
[1];
1920 for (y
= 1; y
<= s
->fragment_height
/ 2; y
++) {
1921 for (x
= 0; x
< s
->fragment_width
/ 2; x
++) {
1922 s
->all_fragments
[i
++].first_pixel
=
1923 s
->golden_frame
.linesize
[1] * y
* FRAGMENT_PIXELS
-
1924 s
->golden_frame
.linesize
[1] +
1925 x
* FRAGMENT_PIXELS
;
1926 debug_init(" fragment %d, first pixel @ %d\n",
1927 i
-1, s
->all_fragments
[i
-1].first_pixel
);
1932 i
= s
->fragment_start
[2];
1933 for (y
= 1; y
<= s
->fragment_height
/ 2; y
++) {
1934 for (x
= 0; x
< s
->fragment_width
/ 2; x
++) {
1935 s
->all_fragments
[i
++].first_pixel
=
1936 s
->golden_frame
.linesize
[2] * y
* FRAGMENT_PIXELS
-
1937 s
->golden_frame
.linesize
[2] +
1938 x
* FRAGMENT_PIXELS
;
1939 debug_init(" fragment %d, first pixel @ %d\n",
1940 i
-1, s
->all_fragments
[i
-1].first_pixel
);
1946 * This is the ffmpeg/libavcodec API init function.
1948 static av_cold
int vp3_decode_init(AVCodecContext
*avctx
)
1950 Vp3DecodeContext
*s
= avctx
->priv_data
;
1951 int i
, inter
, plane
;
1954 int y_superblock_count
;
1955 int c_superblock_count
;
1957 if (avctx
->codec_tag
== MKTAG('V','P','3','0'))
1963 s
->width
= (avctx
->width
+ 15) & 0xFFFFFFF0;
1964 s
->height
= (avctx
->height
+ 15) & 0xFFFFFFF0;
1965 avctx
->pix_fmt
= PIX_FMT_YUV420P
;
1966 if(avctx
->idct_algo
==FF_IDCT_AUTO
)
1967 avctx
->idct_algo
=FF_IDCT_VP3
;
1968 dsputil_init(&s
->dsp
, avctx
);
1970 ff_init_scantable(s
->dsp
.idct_permutation
, &s
->scantable
, ff_zigzag_direct
);
1972 /* initialize to an impossible value which will force a recalculation
1973 * in the first frame decode */
1974 s
->quality_index
= -1;
1976 s
->y_superblock_width
= (s
->width
+ 31) / 32;
1977 s
->y_superblock_height
= (s
->height
+ 31) / 32;
1978 y_superblock_count
= s
->y_superblock_width
* s
->y_superblock_height
;
1980 /* work out the dimensions for the C planes */
1981 c_width
= s
->width
/ 2;
1982 c_height
= s
->height
/ 2;
1983 s
->c_superblock_width
= (c_width
+ 31) / 32;
1984 s
->c_superblock_height
= (c_height
+ 31) / 32;
1985 c_superblock_count
= s
->c_superblock_width
* s
->c_superblock_height
;
1987 s
->superblock_count
= y_superblock_count
+ (c_superblock_count
* 2);
1988 s
->u_superblock_start
= y_superblock_count
;
1989 s
->v_superblock_start
= s
->u_superblock_start
+ c_superblock_count
;
1990 s
->superblock_coding
= av_malloc(s
->superblock_count
);
1992 s
->macroblock_width
= (s
->width
+ 15) / 16;
1993 s
->macroblock_height
= (s
->height
+ 15) / 16;
1994 s
->macroblock_count
= s
->macroblock_width
* s
->macroblock_height
;
1996 s
->fragment_width
= s
->width
/ FRAGMENT_PIXELS
;
1997 s
->fragment_height
= s
->height
/ FRAGMENT_PIXELS
;
1999 /* fragment count covers all 8x8 blocks for all 3 planes */
2000 s
->fragment_count
= s
->fragment_width
* s
->fragment_height
* 3 / 2;
2001 s
->fragment_start
[1] = s
->fragment_width
* s
->fragment_height
;
2002 s
->fragment_start
[2] = s
->fragment_width
* s
->fragment_height
* 5 / 4;
2004 debug_init(" Y plane: %d x %d\n", s
->width
, s
->height
);
2005 debug_init(" C plane: %d x %d\n", c_width
, c_height
);
2006 debug_init(" Y superblocks: %d x %d, %d total\n",
2007 s
->y_superblock_width
, s
->y_superblock_height
, y_superblock_count
);
2008 debug_init(" C superblocks: %d x %d, %d total\n",
2009 s
->c_superblock_width
, s
->c_superblock_height
, c_superblock_count
);
2010 debug_init(" total superblocks = %d, U starts @ %d, V starts @ %d\n",
2011 s
->superblock_count
, s
->u_superblock_start
, s
->v_superblock_start
);
2012 debug_init(" macroblocks: %d x %d, %d total\n",
2013 s
->macroblock_width
, s
->macroblock_height
, s
->macroblock_count
);
2014 debug_init(" %d fragments, %d x %d, u starts @ %d, v starts @ %d\n",
2018 s
->fragment_start
[1],
2019 s
->fragment_start
[2]);
2021 s
->all_fragments
= av_malloc(s
->fragment_count
* sizeof(Vp3Fragment
));
2022 s
->coeff_counts
= av_malloc(s
->fragment_count
* sizeof(*s
->coeff_counts
));
2023 s
->coeffs
= av_malloc(s
->fragment_count
* sizeof(Coeff
) * 65);
2024 s
->coded_fragment_list
= av_malloc(s
->fragment_count
* sizeof(int));
2025 s
->pixel_addresses_initialized
= 0;
2027 if (!s
->theora_tables
)
2029 for (i
= 0; i
< 64; i
++) {
2030 s
->coded_dc_scale_factor
[i
] = vp31_dc_scale_factor
[i
];
2031 s
->coded_ac_scale_factor
[i
] = vp31_ac_scale_factor
[i
];
2032 s
->base_matrix
[0][i
] = vp31_intra_y_dequant
[i
];
2033 s
->base_matrix
[1][i
] = vp31_intra_c_dequant
[i
];
2034 s
->base_matrix
[2][i
] = vp31_inter_dequant
[i
];
2035 s
->filter_limit_values
[i
] = vp31_filter_limit_values
[i
];
2038 for(inter
=0; inter
<2; inter
++){
2039 for(plane
=0; plane
<3; plane
++){
2040 s
->qr_count
[inter
][plane
]= 1;
2041 s
->qr_size
[inter
][plane
][0]= 63;
2042 s
->qr_base
[inter
][plane
][0]=
2043 s
->qr_base
[inter
][plane
][1]= 2*inter
+ (!!plane
)*!inter
;
2047 /* init VLC tables */
2048 for (i
= 0; i
< 16; i
++) {
2051 init_vlc(&s
->dc_vlc
[i
], 5, 32,
2052 &dc_bias
[i
][0][1], 4, 2,
2053 &dc_bias
[i
][0][0], 4, 2, 0);
2055 /* group 1 AC histograms */
2056 init_vlc(&s
->ac_vlc_1
[i
], 5, 32,
2057 &ac_bias_0
[i
][0][1], 4, 2,
2058 &ac_bias_0
[i
][0][0], 4, 2, 0);
2060 /* group 2 AC histograms */
2061 init_vlc(&s
->ac_vlc_2
[i
], 5, 32,
2062 &ac_bias_1
[i
][0][1], 4, 2,
2063 &ac_bias_1
[i
][0][0], 4, 2, 0);
2065 /* group 3 AC histograms */
2066 init_vlc(&s
->ac_vlc_3
[i
], 5, 32,
2067 &ac_bias_2
[i
][0][1], 4, 2,
2068 &ac_bias_2
[i
][0][0], 4, 2, 0);
2070 /* group 4 AC histograms */
2071 init_vlc(&s
->ac_vlc_4
[i
], 5, 32,
2072 &ac_bias_3
[i
][0][1], 4, 2,
2073 &ac_bias_3
[i
][0][0], 4, 2, 0);
2076 for (i
= 0; i
< 16; i
++) {
2079 init_vlc(&s
->dc_vlc
[i
], 5, 32,
2080 &s
->huffman_table
[i
][0][1], 4, 2,
2081 &s
->huffman_table
[i
][0][0], 4, 2, 0);
2083 /* group 1 AC histograms */
2084 init_vlc(&s
->ac_vlc_1
[i
], 5, 32,
2085 &s
->huffman_table
[i
+16][0][1], 4, 2,
2086 &s
->huffman_table
[i
+16][0][0], 4, 2, 0);
2088 /* group 2 AC histograms */
2089 init_vlc(&s
->ac_vlc_2
[i
], 5, 32,
2090 &s
->huffman_table
[i
+16*2][0][1], 4, 2,
2091 &s
->huffman_table
[i
+16*2][0][0], 4, 2, 0);
2093 /* group 3 AC histograms */
2094 init_vlc(&s
->ac_vlc_3
[i
], 5, 32,
2095 &s
->huffman_table
[i
+16*3][0][1], 4, 2,
2096 &s
->huffman_table
[i
+16*3][0][0], 4, 2, 0);
2098 /* group 4 AC histograms */
2099 init_vlc(&s
->ac_vlc_4
[i
], 5, 32,
2100 &s
->huffman_table
[i
+16*4][0][1], 4, 2,
2101 &s
->huffman_table
[i
+16*4][0][0], 4, 2, 0);
2105 init_vlc(&s
->superblock_run_length_vlc
, 6, 34,
2106 &superblock_run_length_vlc_table
[0][1], 4, 2,
2107 &superblock_run_length_vlc_table
[0][0], 4, 2, 0);
2109 init_vlc(&s
->fragment_run_length_vlc
, 5, 30,
2110 &fragment_run_length_vlc_table
[0][1], 4, 2,
2111 &fragment_run_length_vlc_table
[0][0], 4, 2, 0);
2113 init_vlc(&s
->mode_code_vlc
, 3, 8,
2114 &mode_code_vlc_table
[0][1], 2, 1,
2115 &mode_code_vlc_table
[0][0], 2, 1, 0);
2117 init_vlc(&s
->motion_vector_vlc
, 6, 63,
2118 &motion_vector_vlc_table
[0][1], 2, 1,
2119 &motion_vector_vlc_table
[0][0], 2, 1, 0);
2121 /* work out the block mapping tables */
2122 s
->superblock_fragments
= av_malloc(s
->superblock_count
* 16 * sizeof(int));
2123 s
->superblock_macroblocks
= av_malloc(s
->superblock_count
* 4 * sizeof(int));
2124 s
->macroblock_fragments
= av_malloc(s
->macroblock_count
* 6 * sizeof(int));
2125 s
->macroblock_coding
= av_malloc(s
->macroblock_count
+ 1);
2126 init_block_mapping(s
);
2128 for (i
= 0; i
< 3; i
++) {
2129 s
->current_frame
.data
[i
] = NULL
;
2130 s
->last_frame
.data
[i
] = NULL
;
2131 s
->golden_frame
.data
[i
] = NULL
;
2138 * This is the ffmpeg/libavcodec API frame decode function.
2140 static int vp3_decode_frame(AVCodecContext
*avctx
,
2141 void *data
, int *data_size
,
2142 const uint8_t *buf
, int buf_size
)
2144 Vp3DecodeContext
*s
= avctx
->priv_data
;
2146 static int counter
= 0;
2149 init_get_bits(&gb
, buf
, buf_size
* 8);
2151 if (s
->theora
&& get_bits1(&gb
))
2153 av_log(avctx
, AV_LOG_ERROR
, "Header packet passed to frame decoder, skipping\n");
2157 s
->keyframe
= !get_bits1(&gb
);
2160 s
->last_quality_index
= s
->quality_index
;
2164 s
->qis
[s
->nqis
++]= get_bits(&gb
, 6);
2165 } while(s
->theora
>= 0x030200 && s
->nqis
<3 && get_bits1(&gb
));
2167 s
->quality_index
= s
->qis
[0];
2169 if (s
->avctx
->debug
& FF_DEBUG_PICT_INFO
)
2170 av_log(s
->avctx
, AV_LOG_INFO
, " VP3 %sframe #%d: Q index = %d\n",
2171 s
->keyframe
?"key":"", counter
, s
->quality_index
);
2174 if (s
->quality_index
!= s
->last_quality_index
) {
2175 init_dequantizer(s
);
2176 init_loop_filter(s
);
2182 skip_bits(&gb
, 4); /* width code */
2183 skip_bits(&gb
, 4); /* height code */
2186 s
->version
= get_bits(&gb
, 5);
2188 av_log(s
->avctx
, AV_LOG_DEBUG
, "VP version: %d\n", s
->version
);
2191 if (s
->version
|| s
->theora
)
2194 av_log(s
->avctx
, AV_LOG_ERROR
, "Warning, unsupported keyframe coding type?!\n");
2195 skip_bits(&gb
, 2); /* reserved? */
2198 if (s
->last_frame
.data
[0] == s
->golden_frame
.data
[0]) {
2199 if (s
->golden_frame
.data
[0])
2200 avctx
->release_buffer(avctx
, &s
->golden_frame
);
2201 s
->last_frame
= s
->golden_frame
; /* ensure that we catch any access to this released frame */
2203 if (s
->golden_frame
.data
[0])
2204 avctx
->release_buffer(avctx
, &s
->golden_frame
);
2205 if (s
->last_frame
.data
[0])
2206 avctx
->release_buffer(avctx
, &s
->last_frame
);
2209 s
->golden_frame
.reference
= 3;
2210 if(avctx
->get_buffer(avctx
, &s
->golden_frame
) < 0) {
2211 av_log(s
->avctx
, AV_LOG_ERROR
, "vp3: get_buffer() failed\n");
2215 /* golden frame is also the current frame */
2216 s
->current_frame
= s
->golden_frame
;
2218 /* time to figure out pixel addresses? */
2219 if (!s
->pixel_addresses_initialized
)
2221 if (!s
->flipped_image
)
2222 vp3_calculate_pixel_addresses(s
);
2224 theora_calculate_pixel_addresses(s
);
2225 s
->pixel_addresses_initialized
= 1;
2228 /* allocate a new current frame */
2229 s
->current_frame
.reference
= 3;
2230 if (!s
->pixel_addresses_initialized
) {
2231 av_log(s
->avctx
, AV_LOG_ERROR
, "vp3: first frame not a keyframe\n");
2234 if(avctx
->get_buffer(avctx
, &s
->current_frame
) < 0) {
2235 av_log(s
->avctx
, AV_LOG_ERROR
, "vp3: get_buffer() failed\n");
2240 s
->current_frame
.qscale_table
= s
->qscale_table
; //FIXME allocate individual tables per AVFrame
2241 s
->current_frame
.qstride
= 0;
2248 memcpy(s
->current_frame
.data
[0], s
->golden_frame
.data
[0],
2249 s
->current_frame
.linesize
[0] * s
->height
);
2250 memcpy(s
->current_frame
.data
[1], s
->golden_frame
.data
[1],
2251 s
->current_frame
.linesize
[1] * s
->height
/ 2);
2252 memcpy(s
->current_frame
.data
[2], s
->golden_frame
.data
[2],
2253 s
->current_frame
.linesize
[2] * s
->height
/ 2);
2258 if (unpack_superblocks(s
, &gb
)){
2259 av_log(s
->avctx
, AV_LOG_ERROR
, "error in unpack_superblocks\n");
2262 if (unpack_modes(s
, &gb
)){
2263 av_log(s
->avctx
, AV_LOG_ERROR
, "error in unpack_modes\n");
2266 if (unpack_vectors(s
, &gb
)){
2267 av_log(s
->avctx
, AV_LOG_ERROR
, "error in unpack_vectors\n");
2270 if (unpack_dct_coeffs(s
, &gb
)){
2271 av_log(s
->avctx
, AV_LOG_ERROR
, "error in unpack_dct_coeffs\n");
2275 reverse_dc_prediction(s
, 0, s
->fragment_width
, s
->fragment_height
);
2276 if ((avctx
->flags
& CODEC_FLAG_GRAY
) == 0) {
2277 reverse_dc_prediction(s
, s
->fragment_start
[1],
2278 s
->fragment_width
/ 2, s
->fragment_height
/ 2);
2279 reverse_dc_prediction(s
, s
->fragment_start
[2],
2280 s
->fragment_width
/ 2, s
->fragment_height
/ 2);
2283 for (i
= 0; i
< s
->macroblock_height
; i
++)
2286 apply_loop_filter(s
);
2291 *data_size
=sizeof(AVFrame
);
2292 *(AVFrame
*)data
= s
->current_frame
;
2294 /* release the last frame, if it is allocated and if it is not the
2296 if ((s
->last_frame
.data
[0]) &&
2297 (s
->last_frame
.data
[0] != s
->golden_frame
.data
[0]))
2298 avctx
->release_buffer(avctx
, &s
->last_frame
);
2300 /* shuffle frames (last = current) */
2301 s
->last_frame
= s
->current_frame
;
2302 s
->current_frame
.data
[0]= NULL
; /* ensure that we catch any access to this released frame */
2308 * This is the ffmpeg/libavcodec API module cleanup function.
2310 static av_cold
int vp3_decode_end(AVCodecContext
*avctx
)
2312 Vp3DecodeContext
*s
= avctx
->priv_data
;
2315 av_free(s
->superblock_coding
);
2316 av_free(s
->all_fragments
);
2317 av_free(s
->coeff_counts
);
2319 av_free(s
->coded_fragment_list
);
2320 av_free(s
->superblock_fragments
);
2321 av_free(s
->superblock_macroblocks
);
2322 av_free(s
->macroblock_fragments
);
2323 av_free(s
->macroblock_coding
);
2325 for (i
= 0; i
< 16; i
++) {
2326 free_vlc(&s
->dc_vlc
[i
]);
2327 free_vlc(&s
->ac_vlc_1
[i
]);
2328 free_vlc(&s
->ac_vlc_2
[i
]);
2329 free_vlc(&s
->ac_vlc_3
[i
]);
2330 free_vlc(&s
->ac_vlc_4
[i
]);
2333 free_vlc(&s
->superblock_run_length_vlc
);
2334 free_vlc(&s
->fragment_run_length_vlc
);
2335 free_vlc(&s
->mode_code_vlc
);
2336 free_vlc(&s
->motion_vector_vlc
);
2338 /* release all frames */
2339 if (s
->golden_frame
.data
[0] && s
->golden_frame
.data
[0] != s
->last_frame
.data
[0])
2340 avctx
->release_buffer(avctx
, &s
->golden_frame
);
2341 if (s
->last_frame
.data
[0])
2342 avctx
->release_buffer(avctx
, &s
->last_frame
);
2343 /* no need to release the current_frame since it will always be pointing
2344 * to the same frame as either the golden or last frame */
2349 static int read_huffman_tree(AVCodecContext
*avctx
, GetBitContext
*gb
)
2351 Vp3DecodeContext
*s
= avctx
->priv_data
;
2353 if (get_bits1(gb
)) {
2355 if (s
->entries
>= 32) { /* overflow */
2356 av_log(avctx
, AV_LOG_ERROR
, "huffman tree overflow\n");
2359 token
= get_bits(gb
, 5);
2360 //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);
2361 s
->huffman_table
[s
->hti
][token
][0] = s
->hbits
;
2362 s
->huffman_table
[s
->hti
][token
][1] = s
->huff_code_size
;
2366 if (s
->huff_code_size
>= 32) {/* overflow */
2367 av_log(avctx
, AV_LOG_ERROR
, "huffman tree overflow\n");
2370 s
->huff_code_size
++;
2372 read_huffman_tree(avctx
, gb
);
2374 read_huffman_tree(avctx
, gb
);
2376 s
->huff_code_size
--;
2381 #ifdef CONFIG_THEORA_DECODER
2382 static int theora_decode_header(AVCodecContext
*avctx
, GetBitContext
*gb
)
2384 Vp3DecodeContext
*s
= avctx
->priv_data
;
2385 int visible_width
, visible_height
;
2387 s
->theora
= get_bits_long(gb
, 24);
2388 av_log(avctx
, AV_LOG_DEBUG
, "Theora bitstream version %X\n", s
->theora
);
2390 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2391 /* but previous versions have the image flipped relative to vp3 */
2392 if (s
->theora
< 0x030200)
2394 s
->flipped_image
= 1;
2395 av_log(avctx
, AV_LOG_DEBUG
, "Old (<alpha3) Theora bitstream, flipped image\n");
2398 s
->width
= get_bits(gb
, 16) << 4;
2399 s
->height
= get_bits(gb
, 16) << 4;
2401 if(avcodec_check_dimensions(avctx
, s
->width
, s
->height
)){
2402 av_log(avctx
, AV_LOG_ERROR
, "Invalid dimensions (%dx%d)\n", s
->width
, s
->height
);
2403 s
->width
= s
->height
= 0;
2407 if (s
->theora
>= 0x030400)
2409 skip_bits(gb
, 32); /* total number of superblocks in a frame */
2410 // fixme, the next field is 36bits long
2411 skip_bits(gb
, 32); /* total number of blocks in a frame */
2412 skip_bits(gb
, 4); /* total number of blocks in a frame */
2413 skip_bits(gb
, 32); /* total number of macroblocks in a frame */
2416 visible_width
= get_bits_long(gb
, 24);
2417 visible_height
= get_bits_long(gb
, 24);
2419 if (s
->theora
>= 0x030200) {
2420 skip_bits(gb
, 8); /* offset x */
2421 skip_bits(gb
, 8); /* offset y */
2424 skip_bits(gb
, 32); /* fps numerator */
2425 skip_bits(gb
, 32); /* fps denumerator */
2426 skip_bits(gb
, 24); /* aspect numerator */
2427 skip_bits(gb
, 24); /* aspect denumerator */
2429 if (s
->theora
< 0x030200)
2430 skip_bits(gb
, 5); /* keyframe frequency force */
2431 skip_bits(gb
, 8); /* colorspace */
2432 if (s
->theora
>= 0x030400)
2433 skip_bits(gb
, 2); /* pixel format: 420,res,422,444 */
2434 skip_bits(gb
, 24); /* bitrate */
2436 skip_bits(gb
, 6); /* quality hint */
2438 if (s
->theora
>= 0x030200)
2440 skip_bits(gb
, 5); /* keyframe frequency force */
2442 if (s
->theora
< 0x030400)
2443 skip_bits(gb
, 5); /* spare bits */
2446 // align_get_bits(gb);
2448 if ( visible_width
<= s
->width
&& visible_width
> s
->width
-16
2449 && visible_height
<= s
->height
&& visible_height
> s
->height
-16)
2450 avcodec_set_dimensions(avctx
, visible_width
, visible_height
);
2452 avcodec_set_dimensions(avctx
, s
->width
, s
->height
);
2457 static int theora_decode_tables(AVCodecContext
*avctx
, GetBitContext
*gb
)
2459 Vp3DecodeContext
*s
= avctx
->priv_data
;
2460 int i
, n
, matrices
, inter
, plane
;
2462 if (s
->theora
>= 0x030200) {
2463 n
= get_bits(gb
, 3);
2464 /* loop filter limit values table */
2465 for (i
= 0; i
< 64; i
++)
2466 s
->filter_limit_values
[i
] = get_bits(gb
, n
);
2469 if (s
->theora
>= 0x030200)
2470 n
= get_bits(gb
, 4) + 1;
2473 /* quality threshold table */
2474 for (i
= 0; i
< 64; i
++)
2475 s
->coded_ac_scale_factor
[i
] = get_bits(gb
, n
);
2477 if (s
->theora
>= 0x030200)
2478 n
= get_bits(gb
, 4) + 1;
2481 /* dc scale factor table */
2482 for (i
= 0; i
< 64; i
++)
2483 s
->coded_dc_scale_factor
[i
] = get_bits(gb
, n
);
2485 if (s
->theora
>= 0x030200)
2486 matrices
= get_bits(gb
, 9) + 1;
2491 av_log(avctx
, AV_LOG_ERROR
, "invalid number of base matrixes\n");
2495 for(n
=0; n
<matrices
; n
++){
2496 for (i
= 0; i
< 64; i
++)
2497 s
->base_matrix
[n
][i
]= get_bits(gb
, 8);
2500 for (inter
= 0; inter
<= 1; inter
++) {
2501 for (plane
= 0; plane
<= 2; plane
++) {
2503 if (inter
|| plane
> 0)
2504 newqr
= get_bits1(gb
);
2507 if(inter
&& get_bits1(gb
)){
2511 qtj
= (3*inter
+ plane
- 1) / 3;
2512 plj
= (plane
+ 2) % 3;
2514 s
->qr_count
[inter
][plane
]= s
->qr_count
[qtj
][plj
];
2515 memcpy(s
->qr_size
[inter
][plane
], s
->qr_size
[qtj
][plj
], sizeof(s
->qr_size
[0][0]));
2516 memcpy(s
->qr_base
[inter
][plane
], s
->qr_base
[qtj
][plj
], sizeof(s
->qr_base
[0][0]));
2522 i
= get_bits(gb
, av_log2(matrices
-1)+1);
2524 av_log(avctx
, AV_LOG_ERROR
, "invalid base matrix index\n");
2527 s
->qr_base
[inter
][plane
][qri
]= i
;
2530 i
= get_bits(gb
, av_log2(63-qi
)+1) + 1;
2531 s
->qr_size
[inter
][plane
][qri
++]= i
;
2536 av_log(avctx
, AV_LOG_ERROR
, "invalid qi %d > 63\n", qi
);
2539 s
->qr_count
[inter
][plane
]= qri
;
2544 /* Huffman tables */
2545 for (s
->hti
= 0; s
->hti
< 80; s
->hti
++) {
2547 s
->huff_code_size
= 1;
2548 if (!get_bits1(gb
)) {
2550 read_huffman_tree(avctx
, gb
);
2552 read_huffman_tree(avctx
, gb
);
2556 s
->theora_tables
= 1;
2561 static int theora_decode_init(AVCodecContext
*avctx
)
2563 Vp3DecodeContext
*s
= avctx
->priv_data
;
2566 uint8_t *header_start
[3];
2572 if (!avctx
->extradata_size
)
2574 av_log(avctx
, AV_LOG_ERROR
, "Missing extradata!\n");
2578 if (ff_split_xiph_headers(avctx
->extradata
, avctx
->extradata_size
,
2579 42, header_start
, header_len
) < 0) {
2580 av_log(avctx
, AV_LOG_ERROR
, "Corrupt extradata\n");
2585 init_get_bits(&gb
, header_start
[i
], header_len
[i
]);
2587 ptype
= get_bits(&gb
, 8);
2588 debug_vp3("Theora headerpacket type: %x\n", ptype
);
2590 if (!(ptype
& 0x80))
2592 av_log(avctx
, AV_LOG_ERROR
, "Invalid extradata!\n");
2596 // FIXME: Check for this as well.
2597 skip_bits(&gb
, 6*8); /* "theora" */
2602 theora_decode_header(avctx
, &gb
);
2605 // FIXME: is this needed? it breaks sometimes
2606 // theora_decode_comments(avctx, gb);
2609 theora_decode_tables(avctx
, &gb
);
2612 av_log(avctx
, AV_LOG_ERROR
, "Unknown Theora config packet: %d\n", ptype
&~0x80);
2615 if(8*header_len
[i
] != get_bits_count(&gb
))
2616 av_log(avctx
, AV_LOG_ERROR
, "%d bits left in packet %X\n", 8*header_len
[i
] - get_bits_count(&gb
), ptype
);
2617 if (s
->theora
< 0x030200)
2621 vp3_decode_init(avctx
);
2625 AVCodec theora_decoder
= {
2629 sizeof(Vp3DecodeContext
),
2636 .long_name
= NULL_IF_CONFIG_SMALL("Theora"),
2640 AVCodec vp3_decoder
= {
2644 sizeof(Vp3DecodeContext
),
2651 .long_name
= NULL_IF_CONFIG_SMALL("On2 VP3"),