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
22 * @file libavcodec/vp3.c
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
43 #define FRAGMENT_PIXELS 8
45 static av_cold
int vp3_decode_end(AVCodecContext
*avctx
);
47 typedef struct Coeff
{
53 //FIXME split things out into their own arrays
54 typedef struct Vp3Fragment
{
56 /* address of first pixel taking into account which plane the fragment
57 * lives on as well as the plane stride */
59 /* this is the macroblock that the fragment belongs to */
61 uint8_t coding_method
;
67 #define SB_NOT_CODED 0
68 #define SB_PARTIALLY_CODED 1
69 #define SB_FULLY_CODED 2
71 #define MODE_INTER_NO_MV 0
73 #define MODE_INTER_PLUS_MV 2
74 #define MODE_INTER_LAST_MV 3
75 #define MODE_INTER_PRIOR_LAST 4
76 #define MODE_USING_GOLDEN 5
77 #define MODE_GOLDEN_MV 6
78 #define MODE_INTER_FOURMV 7
79 #define CODING_MODE_COUNT 8
81 /* special internal mode */
84 /* There are 6 preset schemes, plus a free-form scheme */
85 static const int ModeAlphabet
[6][CODING_MODE_COUNT
] =
87 /* scheme 1: Last motion vector dominates */
88 { MODE_INTER_LAST_MV
, MODE_INTER_PRIOR_LAST
,
89 MODE_INTER_PLUS_MV
, MODE_INTER_NO_MV
,
90 MODE_INTRA
, MODE_USING_GOLDEN
,
91 MODE_GOLDEN_MV
, MODE_INTER_FOURMV
},
94 { MODE_INTER_LAST_MV
, MODE_INTER_PRIOR_LAST
,
95 MODE_INTER_NO_MV
, MODE_INTER_PLUS_MV
,
96 MODE_INTRA
, MODE_USING_GOLDEN
,
97 MODE_GOLDEN_MV
, MODE_INTER_FOURMV
},
100 { MODE_INTER_LAST_MV
, MODE_INTER_PLUS_MV
,
101 MODE_INTER_PRIOR_LAST
, MODE_INTER_NO_MV
,
102 MODE_INTRA
, MODE_USING_GOLDEN
,
103 MODE_GOLDEN_MV
, MODE_INTER_FOURMV
},
106 { MODE_INTER_LAST_MV
, MODE_INTER_PLUS_MV
,
107 MODE_INTER_NO_MV
, MODE_INTER_PRIOR_LAST
,
108 MODE_INTRA
, MODE_USING_GOLDEN
,
109 MODE_GOLDEN_MV
, MODE_INTER_FOURMV
},
111 /* scheme 5: No motion vector dominates */
112 { MODE_INTER_NO_MV
, MODE_INTER_LAST_MV
,
113 MODE_INTER_PRIOR_LAST
, MODE_INTER_PLUS_MV
,
114 MODE_INTRA
, MODE_USING_GOLDEN
,
115 MODE_GOLDEN_MV
, MODE_INTER_FOURMV
},
118 { MODE_INTER_NO_MV
, MODE_USING_GOLDEN
,
119 MODE_INTER_LAST_MV
, MODE_INTER_PRIOR_LAST
,
120 MODE_INTER_PLUS_MV
, MODE_INTRA
,
121 MODE_GOLDEN_MV
, MODE_INTER_FOURMV
},
125 #define MIN_DEQUANT_VAL 2
127 typedef struct Vp3DecodeContext
{
128 AVCodecContext
*avctx
;
129 int theora
, theora_tables
;
132 AVFrame golden_frame
;
134 AVFrame current_frame
;
143 int superblock_count
;
144 int y_superblock_width
;
145 int y_superblock_height
;
146 int c_superblock_width
;
147 int c_superblock_height
;
148 int u_superblock_start
;
149 int v_superblock_start
;
150 unsigned char *superblock_coding
;
152 int macroblock_count
;
153 int macroblock_width
;
154 int macroblock_height
;
160 Vp3Fragment
*all_fragments
;
161 uint8_t *coeff_counts
;
164 int fragment_start
[3];
169 uint16_t coded_dc_scale_factor
[64];
170 uint32_t coded_ac_scale_factor
[64];
171 uint8_t base_matrix
[384][64];
172 uint8_t qr_count
[2][3];
173 uint8_t qr_size
[2][3][64];
174 uint16_t qr_base
[2][3][64];
176 /* this is a list of indexes into the all_fragments array indicating
177 * which of the fragments are coded */
178 int *coded_fragment_list
;
179 int coded_fragment_list_index
;
180 int pixel_addresses_initialized
;
188 VLC superblock_run_length_vlc
;
189 VLC fragment_run_length_vlc
;
191 VLC motion_vector_vlc
;
193 /* these arrays need to be on 16-byte boundaries since SSE2 operations
195 DECLARE_ALIGNED_16(int16_t, qmat
[3][2][3][64]); //<qmat[qpi][is_inter][plane]
197 /* This table contains superblock_count * 16 entries. Each set of 16
198 * numbers corresponds to the fragment indexes 0..15 of the superblock.
199 * An entry will be -1 to indicate that no entry corresponds to that
201 int *superblock_fragments
;
203 /* This table contains superblock_count * 4 entries. Each set of 4
204 * numbers corresponds to the macroblock indexes 0..3 of the superblock.
205 * An entry will be -1 to indicate that no entry corresponds to that
207 int *superblock_macroblocks
;
209 /* This table contains macroblock_count * 6 entries. Each set of 6
210 * numbers corresponds to the fragment indexes 0..5 which comprise
211 * the macroblock (4 Y fragments and 2 C fragments). */
212 int *macroblock_fragments
;
213 /* This is an array that indicates how a particular macroblock
215 unsigned char *macroblock_coding
;
217 int first_coded_y_fragment
;
218 int first_coded_c_fragment
;
219 int last_coded_y_fragment
;
220 int last_coded_c_fragment
;
222 uint8_t edge_emu_buffer
[9*2048]; //FIXME dynamic alloc
223 int8_t qscale_table
[2048]; //FIXME dynamic alloc (width+15)/16
230 uint16_t huffman_table
[80][32][2];
232 uint8_t filter_limit_values
[64];
233 DECLARE_ALIGNED_8(int, bounding_values_array
[256+2]);
236 /************************************************************************
237 * VP3 specific functions
238 ************************************************************************/
241 * This function sets up all of the various blocks mappings:
242 * superblocks <-> fragments, macroblocks <-> fragments,
243 * superblocks <-> macroblocks
245 * Returns 0 is successful; returns 1 if *anything* went wrong.
247 static int init_block_mapping(Vp3DecodeContext
*s
)
250 signed int hilbert_walk_mb
[4];
252 int current_fragment
= 0;
253 int current_width
= 0;
254 int current_height
= 0;
257 int superblock_row_inc
= 0;
258 int mapping_index
= 0;
260 int current_macroblock
;
263 static const signed char travel_width
[16] = {
270 static const signed char travel_height
[16] = {
277 static const signed char travel_width_mb
[4] = {
281 static const signed char travel_height_mb
[4] = {
285 hilbert_walk_mb
[0] = 1;
286 hilbert_walk_mb
[1] = s
->macroblock_width
;
287 hilbert_walk_mb
[2] = 1;
288 hilbert_walk_mb
[3] = -s
->macroblock_width
;
290 /* iterate through each superblock (all planes) and map the fragments */
291 for (i
= 0; i
< s
->superblock_count
; i
++) {
292 /* time to re-assign the limits? */
295 /* start of Y superblocks */
296 right_edge
= s
->fragment_width
;
297 bottom_edge
= s
->fragment_height
;
300 superblock_row_inc
= 3 * s
->fragment_width
-
301 (s
->y_superblock_width
* 4 - s
->fragment_width
);
303 /* the first operation for this variable is to advance by 1 */
304 current_fragment
= -1;
306 } else if (i
== s
->u_superblock_start
) {
308 /* start of U superblocks */
309 right_edge
= s
->fragment_width
/ 2;
310 bottom_edge
= s
->fragment_height
/ 2;
313 superblock_row_inc
= 3 * (s
->fragment_width
/ 2) -
314 (s
->c_superblock_width
* 4 - s
->fragment_width
/ 2);
316 /* the first operation for this variable is to advance by 1 */
317 current_fragment
= s
->fragment_start
[1] - 1;
319 } else if (i
== s
->v_superblock_start
) {
321 /* start of V superblocks */
322 right_edge
= s
->fragment_width
/ 2;
323 bottom_edge
= s
->fragment_height
/ 2;
326 superblock_row_inc
= 3 * (s
->fragment_width
/ 2) -
327 (s
->c_superblock_width
* 4 - s
->fragment_width
/ 2);
329 /* the first operation for this variable is to advance by 1 */
330 current_fragment
= s
->fragment_start
[2] - 1;
334 if (current_width
>= right_edge
- 1) {
335 /* reset width and move to next superblock row */
339 /* fragment is now at the start of a new superblock row */
340 current_fragment
+= superblock_row_inc
;
343 /* iterate through all 16 fragments in a superblock */
344 for (j
= 0; j
< 16; j
++) {
345 current_fragment
+= travel_width
[j
] + right_edge
* travel_height
[j
];
346 current_width
+= travel_width
[j
];
347 current_height
+= travel_height
[j
];
349 /* check if the fragment is in bounds */
350 if ((current_width
< right_edge
) &&
351 (current_height
< bottom_edge
)) {
352 s
->superblock_fragments
[mapping_index
] = current_fragment
;
354 s
->superblock_fragments
[mapping_index
] = -1;
361 /* initialize the superblock <-> macroblock mapping; iterate through
362 * all of the Y plane superblocks to build this mapping */
363 right_edge
= s
->macroblock_width
;
364 bottom_edge
= s
->macroblock_height
;
367 superblock_row_inc
= s
->macroblock_width
-
368 (s
->y_superblock_width
* 2 - s
->macroblock_width
);
370 current_macroblock
= -1;
371 for (i
= 0; i
< s
->u_superblock_start
; i
++) {
373 if (current_width
>= right_edge
- 1) {
374 /* reset width and move to next superblock row */
378 /* macroblock is now at the start of a new superblock row */
379 current_macroblock
+= superblock_row_inc
;
382 /* iterate through each potential macroblock in the superblock */
383 for (j
= 0; j
< 4; j
++) {
384 current_macroblock
+= hilbert_walk_mb
[j
];
385 current_width
+= travel_width_mb
[j
];
386 current_height
+= travel_height_mb
[j
];
388 /* check if the macroblock is in bounds */
389 if ((current_width
< right_edge
) &&
390 (current_height
< bottom_edge
)) {
391 s
->superblock_macroblocks
[mapping_index
] = current_macroblock
;
393 s
->superblock_macroblocks
[mapping_index
] = -1;
400 /* initialize the macroblock <-> fragment mapping */
401 current_fragment
= 0;
402 current_macroblock
= 0;
404 for (i
= 0; i
< s
->fragment_height
; i
+= 2) {
406 for (j
= 0; j
< s
->fragment_width
; j
+= 2) {
408 s
->all_fragments
[current_fragment
].macroblock
= current_macroblock
;
409 s
->macroblock_fragments
[mapping_index
++] = current_fragment
;
411 if (j
+ 1 < s
->fragment_width
) {
412 s
->all_fragments
[current_fragment
+ 1].macroblock
= current_macroblock
;
413 s
->macroblock_fragments
[mapping_index
++] = current_fragment
+ 1;
415 s
->macroblock_fragments
[mapping_index
++] = -1;
417 if (i
+ 1 < s
->fragment_height
) {
418 s
->all_fragments
[current_fragment
+ s
->fragment_width
].macroblock
=
420 s
->macroblock_fragments
[mapping_index
++] =
421 current_fragment
+ s
->fragment_width
;
423 s
->macroblock_fragments
[mapping_index
++] = -1;
425 if ((j
+ 1 < s
->fragment_width
) && (i
+ 1 < s
->fragment_height
)) {
426 s
->all_fragments
[current_fragment
+ s
->fragment_width
+ 1].macroblock
=
428 s
->macroblock_fragments
[mapping_index
++] =
429 current_fragment
+ s
->fragment_width
+ 1;
431 s
->macroblock_fragments
[mapping_index
++] = -1;
434 c_fragment
= s
->fragment_start
[1] +
435 (i
* s
->fragment_width
/ 4) + (j
/ 2);
436 s
->all_fragments
[c_fragment
].macroblock
= s
->macroblock_count
;
437 s
->macroblock_fragments
[mapping_index
++] = c_fragment
;
439 c_fragment
= s
->fragment_start
[2] +
440 (i
* s
->fragment_width
/ 4) + (j
/ 2);
441 s
->all_fragments
[c_fragment
].macroblock
= s
->macroblock_count
;
442 s
->macroblock_fragments
[mapping_index
++] = c_fragment
;
444 if (j
+ 2 <= s
->fragment_width
)
445 current_fragment
+= 2;
448 current_macroblock
++;
451 current_fragment
+= s
->fragment_width
;
454 return 0; /* successful path out */
458 * This function wipes out all of the fragment data.
460 static void init_frame(Vp3DecodeContext
*s
, GetBitContext
*gb
)
464 /* zero out all of the fragment information */
465 s
->coded_fragment_list_index
= 0;
466 for (i
= 0; i
< s
->fragment_count
; i
++) {
467 s
->coeff_counts
[i
] = 0;
468 s
->all_fragments
[i
].motion_x
= 127;
469 s
->all_fragments
[i
].motion_y
= 127;
470 s
->all_fragments
[i
].next_coeff
= NULL
;
471 s
->all_fragments
[i
].qpi
= 0;
473 s
->coeffs
[i
].coeff
=0;
474 s
->coeffs
[i
].next
= NULL
;
479 * This function sets up the dequantization tables used for a particular
482 static void init_dequantizer(Vp3DecodeContext
*s
, int qpi
)
484 int ac_scale_factor
= s
->coded_ac_scale_factor
[s
->qps
[qpi
]];
485 int dc_scale_factor
= s
->coded_dc_scale_factor
[s
->qps
[qpi
]];
486 int i
, plane
, inter
, qri
, bmi
, bmj
, qistart
;
488 for(inter
=0; inter
<2; inter
++){
489 for(plane
=0; plane
<3; plane
++){
491 for(qri
=0; qri
<s
->qr_count
[inter
][plane
]; qri
++){
492 sum
+= s
->qr_size
[inter
][plane
][qri
];
493 if(s
->qps
[qpi
] <= sum
)
496 qistart
= sum
- s
->qr_size
[inter
][plane
][qri
];
497 bmi
= s
->qr_base
[inter
][plane
][qri
];
498 bmj
= s
->qr_base
[inter
][plane
][qri
+1];
500 int coeff
= ( 2*(sum
-s
->qps
[qpi
])*s
->base_matrix
[bmi
][i
]
501 - 2*(qistart
-s
->qps
[qpi
])*s
->base_matrix
[bmj
][i
]
502 + s
->qr_size
[inter
][plane
][qri
])
503 / (2*s
->qr_size
[inter
][plane
][qri
]);
505 int qmin
= 8<<(inter
+ !i
);
506 int qscale
= i
? ac_scale_factor
: dc_scale_factor
;
508 s
->qmat
[qpi
][inter
][plane
][s
->dsp
.idct_permutation
[i
]]= av_clip((qscale
* coeff
)/100 * 4, qmin
, 4096);
510 // all DC coefficients use the same quant so as not to interfere with DC prediction
511 s
->qmat
[qpi
][inter
][plane
][0] = s
->qmat
[0][inter
][plane
][0];
515 memset(s
->qscale_table
, (FFMAX(s
->qmat
[0][0][0][1], s
->qmat
[0][0][1][1])+8)/16, 512); //FIXME finetune
519 * This function initializes the loop filter boundary limits if the frame's
520 * quality index is different from the previous frame's.
522 * The filter_limit_values may not be larger than 127.
524 static void init_loop_filter(Vp3DecodeContext
*s
)
526 int *bounding_values
= s
->bounding_values_array
+127;
531 filter_limit
= s
->filter_limit_values
[s
->qps
[0]];
533 /* set up the bounding values */
534 memset(s
->bounding_values_array
, 0, 256 * sizeof(int));
535 for (x
= 0; x
< filter_limit
; x
++) {
536 bounding_values
[-x
] = -x
;
537 bounding_values
[x
] = x
;
539 for (x
= value
= filter_limit
; x
< 128 && value
; x
++, value
--) {
540 bounding_values
[ x
] = value
;
541 bounding_values
[-x
] = -value
;
544 bounding_values
[128] = value
;
545 bounding_values
[129] = bounding_values
[130] = filter_limit
* 0x02020202;
549 * This function unpacks all of the superblock/macroblock/fragment coding
550 * information from the bitstream.
552 static int unpack_superblocks(Vp3DecodeContext
*s
, GetBitContext
*gb
)
555 int current_superblock
= 0;
557 int decode_fully_flags
= 0;
558 int decode_partial_blocks
= 0;
559 int first_c_fragment_seen
;
562 int current_fragment
;
565 memset(s
->superblock_coding
, SB_FULLY_CODED
, s
->superblock_count
);
569 /* unpack the list of partially-coded superblocks */
571 /* toggle the bit because as soon as the first run length is
572 * fetched the bit will be toggled again */
574 while (current_superblock
< s
->superblock_count
) {
575 if (current_run
-- == 0) {
577 current_run
= get_vlc2(gb
,
578 s
->superblock_run_length_vlc
.table
, 6, 2);
579 if (current_run
== 33)
580 current_run
+= get_bits(gb
, 12);
582 /* if any of the superblocks are not partially coded, flag
583 * a boolean to decode the list of fully-coded superblocks */
585 decode_fully_flags
= 1;
588 /* make a note of the fact that there are partially coded
590 decode_partial_blocks
= 1;
593 s
->superblock_coding
[current_superblock
++] = bit
;
596 /* unpack the list of fully coded superblocks if any of the blocks were
597 * not marked as partially coded in the previous step */
598 if (decode_fully_flags
) {
600 current_superblock
= 0;
603 /* toggle the bit because as soon as the first run length is
604 * fetched the bit will be toggled again */
606 while (current_superblock
< s
->superblock_count
) {
608 /* skip any superblocks already marked as partially coded */
609 if (s
->superblock_coding
[current_superblock
] == SB_NOT_CODED
) {
611 if (current_run
-- == 0) {
613 current_run
= get_vlc2(gb
,
614 s
->superblock_run_length_vlc
.table
, 6, 2);
615 if (current_run
== 33)
616 current_run
+= get_bits(gb
, 12);
618 s
->superblock_coding
[current_superblock
] = 2*bit
;
620 current_superblock
++;
624 /* if there were partial blocks, initialize bitstream for
625 * unpacking fragment codings */
626 if (decode_partial_blocks
) {
630 /* toggle the bit because as soon as the first run length is
631 * fetched the bit will be toggled again */
636 /* figure out which fragments are coded; iterate through each
637 * superblock (all planes) */
638 s
->coded_fragment_list_index
= 0;
639 s
->next_coeff
= s
->coeffs
+ s
->fragment_count
;
640 s
->first_coded_y_fragment
= s
->first_coded_c_fragment
= 0;
641 s
->last_coded_y_fragment
= s
->last_coded_c_fragment
= -1;
642 first_c_fragment_seen
= 0;
643 memset(s
->macroblock_coding
, MODE_COPY
, s
->macroblock_count
);
644 for (i
= 0; i
< s
->superblock_count
; i
++) {
646 /* iterate through all 16 fragments in a superblock */
647 for (j
= 0; j
< 16; j
++) {
649 /* if the fragment is in bounds, check its coding status */
650 current_fragment
= s
->superblock_fragments
[i
* 16 + j
];
651 if (current_fragment
>= s
->fragment_count
) {
652 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
653 current_fragment
, s
->fragment_count
);
656 if (current_fragment
!= -1) {
657 if (s
->superblock_coding
[i
] == SB_NOT_CODED
) {
659 /* copy all the fragments from the prior frame */
660 s
->all_fragments
[current_fragment
].coding_method
=
663 } else if (s
->superblock_coding
[i
] == SB_PARTIALLY_CODED
) {
665 /* fragment may or may not be coded; this is the case
666 * that cares about the fragment coding runs */
667 if (current_run
-- == 0) {
669 current_run
= get_vlc2(gb
,
670 s
->fragment_run_length_vlc
.table
, 5, 2);
674 /* default mode; actual mode will be decoded in
676 s
->all_fragments
[current_fragment
].coding_method
=
678 s
->all_fragments
[current_fragment
].next_coeff
= s
->coeffs
+ current_fragment
;
679 s
->coded_fragment_list
[s
->coded_fragment_list_index
] =
681 if ((current_fragment
>= s
->fragment_start
[1]) &&
682 (s
->last_coded_y_fragment
== -1) &&
683 (!first_c_fragment_seen
)) {
684 s
->first_coded_c_fragment
= s
->coded_fragment_list_index
;
685 s
->last_coded_y_fragment
= s
->first_coded_c_fragment
- 1;
686 first_c_fragment_seen
= 1;
688 s
->coded_fragment_list_index
++;
689 s
->macroblock_coding
[s
->all_fragments
[current_fragment
].macroblock
] = MODE_INTER_NO_MV
;
691 /* not coded; copy this fragment from the prior frame */
692 s
->all_fragments
[current_fragment
].coding_method
=
698 /* fragments are fully coded in this superblock; actual
699 * coding will be determined in next step */
700 s
->all_fragments
[current_fragment
].coding_method
=
702 s
->all_fragments
[current_fragment
].next_coeff
= s
->coeffs
+ current_fragment
;
703 s
->coded_fragment_list
[s
->coded_fragment_list_index
] =
705 if ((current_fragment
>= s
->fragment_start
[1]) &&
706 (s
->last_coded_y_fragment
== -1) &&
707 (!first_c_fragment_seen
)) {
708 s
->first_coded_c_fragment
= s
->coded_fragment_list_index
;
709 s
->last_coded_y_fragment
= s
->first_coded_c_fragment
- 1;
710 first_c_fragment_seen
= 1;
712 s
->coded_fragment_list_index
++;
713 s
->macroblock_coding
[s
->all_fragments
[current_fragment
].macroblock
] = MODE_INTER_NO_MV
;
719 if (!first_c_fragment_seen
)
720 /* only Y fragments coded in this frame */
721 s
->last_coded_y_fragment
= s
->coded_fragment_list_index
- 1;
723 /* end the list of coded C fragments */
724 s
->last_coded_c_fragment
= s
->coded_fragment_list_index
- 1;
730 * This function unpacks all the coding mode data for individual macroblocks
731 * from the bitstream.
733 static int unpack_modes(Vp3DecodeContext
*s
, GetBitContext
*gb
)
737 int current_macroblock
;
738 int current_fragment
;
740 int custom_mode_alphabet
[CODING_MODE_COUNT
];
743 for (i
= 0; i
< s
->fragment_count
; i
++)
744 s
->all_fragments
[i
].coding_method
= MODE_INTRA
;
748 /* fetch the mode coding scheme for this frame */
749 scheme
= get_bits(gb
, 3);
751 /* is it a custom coding scheme? */
753 for (i
= 0; i
< 8; i
++)
754 custom_mode_alphabet
[i
] = MODE_INTER_NO_MV
;
755 for (i
= 0; i
< 8; i
++)
756 custom_mode_alphabet
[get_bits(gb
, 3)] = i
;
759 /* iterate through all of the macroblocks that contain 1 or more
761 for (i
= 0; i
< s
->u_superblock_start
; i
++) {
763 for (j
= 0; j
< 4; j
++) {
764 current_macroblock
= s
->superblock_macroblocks
[i
* 4 + j
];
765 if ((current_macroblock
== -1) ||
766 (s
->macroblock_coding
[current_macroblock
] == MODE_COPY
))
768 if (current_macroblock
>= s
->macroblock_count
) {
769 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_modes(): bad macroblock number (%d >= %d)\n",
770 current_macroblock
, s
->macroblock_count
);
774 /* mode 7 means get 3 bits for each coding mode */
776 coding_mode
= get_bits(gb
, 3);
778 coding_mode
= custom_mode_alphabet
779 [get_vlc2(gb
, s
->mode_code_vlc
.table
, 3, 3)];
781 coding_mode
= ModeAlphabet
[scheme
-1]
782 [get_vlc2(gb
, s
->mode_code_vlc
.table
, 3, 3)];
784 s
->macroblock_coding
[current_macroblock
] = coding_mode
;
785 for (k
= 0; k
< 6; k
++) {
787 s
->macroblock_fragments
[current_macroblock
* 6 + k
];
788 if (current_fragment
== -1)
790 if (current_fragment
>= s
->fragment_count
) {
791 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_modes(): bad fragment number (%d >= %d)\n",
792 current_fragment
, s
->fragment_count
);
795 if (s
->all_fragments
[current_fragment
].coding_method
!=
797 s
->all_fragments
[current_fragment
].coding_method
=
808 * This function unpacks all the motion vectors for the individual
809 * macroblocks from the bitstream.
811 static int unpack_vectors(Vp3DecodeContext
*s
, GetBitContext
*gb
)
817 int last_motion_x
= 0;
818 int last_motion_y
= 0;
819 int prior_last_motion_x
= 0;
820 int prior_last_motion_y
= 0;
821 int current_macroblock
;
822 int current_fragment
;
827 memset(motion_x
, 0, 6 * sizeof(int));
828 memset(motion_y
, 0, 6 * sizeof(int));
830 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
831 coding_mode
= get_bits1(gb
);
833 /* iterate through all of the macroblocks that contain 1 or more
835 for (i
= 0; i
< s
->u_superblock_start
; i
++) {
837 for (j
= 0; j
< 4; j
++) {
838 current_macroblock
= s
->superblock_macroblocks
[i
* 4 + j
];
839 if ((current_macroblock
== -1) ||
840 (s
->macroblock_coding
[current_macroblock
] == MODE_COPY
))
842 if (current_macroblock
>= s
->macroblock_count
) {
843 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_vectors(): bad macroblock number (%d >= %d)\n",
844 current_macroblock
, s
->macroblock_count
);
848 current_fragment
= s
->macroblock_fragments
[current_macroblock
* 6];
849 if (current_fragment
>= s
->fragment_count
) {
850 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_vectors(): bad fragment number (%d >= %d\n",
851 current_fragment
, s
->fragment_count
);
854 switch (s
->macroblock_coding
[current_macroblock
]) {
856 case MODE_INTER_PLUS_MV
:
858 /* all 6 fragments use the same motion vector */
859 if (coding_mode
== 0) {
860 motion_x
[0] = motion_vector_table
[get_vlc2(gb
, s
->motion_vector_vlc
.table
, 6, 2)];
861 motion_y
[0] = motion_vector_table
[get_vlc2(gb
, s
->motion_vector_vlc
.table
, 6, 2)];
863 motion_x
[0] = fixed_motion_vector_table
[get_bits(gb
, 6)];
864 motion_y
[0] = fixed_motion_vector_table
[get_bits(gb
, 6)];
867 for (k
= 1; k
< 6; k
++) {
868 motion_x
[k
] = motion_x
[0];
869 motion_y
[k
] = motion_y
[0];
872 /* vector maintenance, only on MODE_INTER_PLUS_MV */
873 if (s
->macroblock_coding
[current_macroblock
] ==
874 MODE_INTER_PLUS_MV
) {
875 prior_last_motion_x
= last_motion_x
;
876 prior_last_motion_y
= last_motion_y
;
877 last_motion_x
= motion_x
[0];
878 last_motion_y
= motion_y
[0];
882 case MODE_INTER_FOURMV
:
883 /* vector maintenance */
884 prior_last_motion_x
= last_motion_x
;
885 prior_last_motion_y
= last_motion_y
;
887 /* fetch 4 vectors from the bitstream, one for each
888 * Y fragment, then average for the C fragment vectors */
889 motion_x
[4] = motion_y
[4] = 0;
890 for (k
= 0; k
< 4; k
++) {
891 for (l
= 0; l
< s
->coded_fragment_list_index
; l
++)
892 if (s
->coded_fragment_list
[l
] == s
->macroblock_fragments
[6*current_macroblock
+ k
])
894 if (l
< s
->coded_fragment_list_index
) {
895 if (coding_mode
== 0) {
896 motion_x
[k
] = motion_vector_table
[get_vlc2(gb
, s
->motion_vector_vlc
.table
, 6, 2)];
897 motion_y
[k
] = motion_vector_table
[get_vlc2(gb
, s
->motion_vector_vlc
.table
, 6, 2)];
899 motion_x
[k
] = fixed_motion_vector_table
[get_bits(gb
, 6)];
900 motion_y
[k
] = fixed_motion_vector_table
[get_bits(gb
, 6)];
902 last_motion_x
= motion_x
[k
];
903 last_motion_y
= motion_y
[k
];
908 motion_x
[4] += motion_x
[k
];
909 motion_y
[4] += motion_y
[k
];
913 motion_x
[4]= RSHIFT(motion_x
[4], 2);
915 motion_y
[4]= RSHIFT(motion_y
[4], 2);
918 case MODE_INTER_LAST_MV
:
919 /* all 6 fragments use the last motion vector */
920 motion_x
[0] = last_motion_x
;
921 motion_y
[0] = last_motion_y
;
922 for (k
= 1; k
< 6; k
++) {
923 motion_x
[k
] = motion_x
[0];
924 motion_y
[k
] = motion_y
[0];
927 /* no vector maintenance (last vector remains the
931 case MODE_INTER_PRIOR_LAST
:
932 /* all 6 fragments use the motion vector prior to the
933 * last motion vector */
934 motion_x
[0] = prior_last_motion_x
;
935 motion_y
[0] = prior_last_motion_y
;
936 for (k
= 1; k
< 6; k
++) {
937 motion_x
[k
] = motion_x
[0];
938 motion_y
[k
] = motion_y
[0];
941 /* vector maintenance */
942 prior_last_motion_x
= last_motion_x
;
943 prior_last_motion_y
= last_motion_y
;
944 last_motion_x
= motion_x
[0];
945 last_motion_y
= motion_y
[0];
949 /* covers intra, inter without MV, golden without MV */
950 memset(motion_x
, 0, 6 * sizeof(int));
951 memset(motion_y
, 0, 6 * sizeof(int));
953 /* no vector maintenance */
957 /* assign the motion vectors to the correct fragments */
958 for (k
= 0; k
< 6; k
++) {
960 s
->macroblock_fragments
[current_macroblock
* 6 + k
];
961 if (current_fragment
== -1)
963 if (current_fragment
>= s
->fragment_count
) {
964 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_vectors(): bad fragment number (%d >= %d)\n",
965 current_fragment
, s
->fragment_count
);
968 s
->all_fragments
[current_fragment
].motion_x
= motion_x
[k
];
969 s
->all_fragments
[current_fragment
].motion_y
= motion_y
[k
];
977 static int unpack_block_qpis(Vp3DecodeContext
*s
, GetBitContext
*gb
)
979 int qpi
, i
, j
, bit
, run_length
, blocks_decoded
, num_blocks_at_qpi
;
980 int num_blocks
= s
->coded_fragment_list_index
;
982 for (qpi
= 0; qpi
< s
->nqps
-1 && num_blocks
> 0; qpi
++) {
983 i
= blocks_decoded
= num_blocks_at_qpi
= 0;
988 run_length
= get_vlc2(gb
, s
->superblock_run_length_vlc
.table
, 6, 2) + 1;
989 if (run_length
== 34)
990 run_length
+= get_bits(gb
, 12);
991 blocks_decoded
+= run_length
;
994 num_blocks_at_qpi
+= run_length
;
996 for (j
= 0; j
< run_length
; i
++) {
997 if (i
>= s
->coded_fragment_list_index
)
1000 if (s
->all_fragments
[s
->coded_fragment_list
[i
]].qpi
== qpi
) {
1001 s
->all_fragments
[s
->coded_fragment_list
[i
]].qpi
+= bit
;
1006 if (run_length
== 4129)
1007 bit
= get_bits1(gb
);
1010 } while (blocks_decoded
< num_blocks
);
1012 num_blocks
-= num_blocks_at_qpi
;
1019 * This function is called by unpack_dct_coeffs() to extract the VLCs from
1020 * the bitstream. The VLCs encode tokens which are used to unpack DCT
1021 * data. This function unpacks all the VLCs for either the Y plane or both
1022 * C planes, and is called for DC coefficients or different AC coefficient
1023 * levels (since different coefficient types require different VLC tables.
1025 * This function returns a residual eob run. E.g, if a particular token gave
1026 * instructions to EOB the next 5 fragments and there were only 2 fragments
1027 * left in the current fragment range, 3 would be returned so that it could
1028 * be passed into the next call to this same function.
1030 static int unpack_vlcs(Vp3DecodeContext
*s
, GetBitContext
*gb
,
1031 VLC
*table
, int coeff_index
,
1032 int first_fragment
, int last_fragment
,
1039 Vp3Fragment
*fragment
;
1042 /* local references to structure members to avoid repeated deferences */
1043 uint8_t *perm
= s
->scantable
.permutated
;
1044 int *coded_fragment_list
= s
->coded_fragment_list
;
1045 Vp3Fragment
*all_fragments
= s
->all_fragments
;
1046 uint8_t *coeff_counts
= s
->coeff_counts
;
1047 VLC_TYPE (*vlc_table
)[2] = table
->table
;
1049 if ((first_fragment
>= s
->fragment_count
) ||
1050 (last_fragment
>= s
->fragment_count
)) {
1052 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_vlcs(): bad fragment number (%d -> %d ?)\n",
1053 first_fragment
, last_fragment
);
1057 for (i
= first_fragment
; i
<= last_fragment
; i
++) {
1058 int fragment_num
= coded_fragment_list
[i
];
1060 if (coeff_counts
[fragment_num
] > coeff_index
)
1062 fragment
= &all_fragments
[fragment_num
];
1065 /* decode a VLC into a token */
1066 token
= get_vlc2(gb
, vlc_table
, 5, 3);
1067 /* use the token to get a zero run, a coefficient, and an eob run */
1069 eob_run
= eob_run_base
[token
];
1070 if (eob_run_get_bits
[token
])
1071 eob_run
+= get_bits(gb
, eob_run_get_bits
[token
]);
1072 coeff
= zero_run
= 0;
1074 bits_to_get
= coeff_get_bits
[token
];
1076 bits_to_get
= get_bits(gb
, bits_to_get
);
1077 coeff
= coeff_tables
[token
][bits_to_get
];
1079 zero_run
= zero_run_base
[token
];
1080 if (zero_run_get_bits
[token
])
1081 zero_run
+= get_bits(gb
, zero_run_get_bits
[token
]);
1086 coeff_counts
[fragment_num
] += zero_run
;
1087 if (coeff_counts
[fragment_num
] < 64){
1088 fragment
->next_coeff
->coeff
= coeff
;
1089 fragment
->next_coeff
->index
= perm
[coeff_counts
[fragment_num
]++]; //FIXME perm here already?
1090 fragment
->next_coeff
->next
= s
->next_coeff
;
1091 s
->next_coeff
->next
=NULL
;
1092 fragment
->next_coeff
= s
->next_coeff
++;
1095 coeff_counts
[fragment_num
] |= 128;
1103 static void reverse_dc_prediction(Vp3DecodeContext
*s
,
1106 int fragment_height
);
1108 * This function unpacks all of the DCT coefficient data from the
1111 static int unpack_dct_coeffs(Vp3DecodeContext
*s
, GetBitContext
*gb
)
1118 int residual_eob_run
= 0;
1120 /* fetch the DC table indexes */
1121 dc_y_table
= get_bits(gb
, 4);
1122 dc_c_table
= get_bits(gb
, 4);
1124 /* unpack the Y plane DC coefficients */
1125 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->dc_vlc
[dc_y_table
], 0,
1126 s
->first_coded_y_fragment
, s
->last_coded_y_fragment
, residual_eob_run
);
1128 /* reverse prediction of the Y-plane DC coefficients */
1129 reverse_dc_prediction(s
, 0, s
->fragment_width
, s
->fragment_height
);
1131 /* unpack the C plane DC coefficients */
1132 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->dc_vlc
[dc_c_table
], 0,
1133 s
->first_coded_c_fragment
, s
->last_coded_c_fragment
, residual_eob_run
);
1135 /* reverse prediction of the C-plane DC coefficients */
1136 if (!(s
->avctx
->flags
& CODEC_FLAG_GRAY
))
1138 reverse_dc_prediction(s
, s
->fragment_start
[1],
1139 s
->fragment_width
/ 2, s
->fragment_height
/ 2);
1140 reverse_dc_prediction(s
, s
->fragment_start
[2],
1141 s
->fragment_width
/ 2, s
->fragment_height
/ 2);
1144 /* fetch the AC table indexes */
1145 ac_y_table
= get_bits(gb
, 4);
1146 ac_c_table
= get_bits(gb
, 4);
1148 /* unpack the group 1 AC coefficients (coeffs 1-5) */
1149 for (i
= 1; i
<= 5; i
++) {
1150 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_1
[ac_y_table
], i
,
1151 s
->first_coded_y_fragment
, s
->last_coded_y_fragment
, residual_eob_run
);
1153 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_1
[ac_c_table
], i
,
1154 s
->first_coded_c_fragment
, s
->last_coded_c_fragment
, residual_eob_run
);
1157 /* unpack the group 2 AC coefficients (coeffs 6-14) */
1158 for (i
= 6; i
<= 14; i
++) {
1159 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_2
[ac_y_table
], i
,
1160 s
->first_coded_y_fragment
, s
->last_coded_y_fragment
, residual_eob_run
);
1162 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_2
[ac_c_table
], i
,
1163 s
->first_coded_c_fragment
, s
->last_coded_c_fragment
, residual_eob_run
);
1166 /* unpack the group 3 AC coefficients (coeffs 15-27) */
1167 for (i
= 15; i
<= 27; i
++) {
1168 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_3
[ac_y_table
], i
,
1169 s
->first_coded_y_fragment
, s
->last_coded_y_fragment
, residual_eob_run
);
1171 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_3
[ac_c_table
], i
,
1172 s
->first_coded_c_fragment
, s
->last_coded_c_fragment
, residual_eob_run
);
1175 /* unpack the group 4 AC coefficients (coeffs 28-63) */
1176 for (i
= 28; i
<= 63; i
++) {
1177 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_4
[ac_y_table
], i
,
1178 s
->first_coded_y_fragment
, s
->last_coded_y_fragment
, residual_eob_run
);
1180 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_4
[ac_c_table
], i
,
1181 s
->first_coded_c_fragment
, s
->last_coded_c_fragment
, residual_eob_run
);
1188 * This function reverses the DC prediction for each coded fragment in
1189 * the frame. Much of this function is adapted directly from the original
1192 #define COMPATIBLE_FRAME(x) \
1193 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1194 #define FRAME_CODED(x) (s->all_fragments[x].coding_method != MODE_COPY)
1195 #define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
1197 static void reverse_dc_prediction(Vp3DecodeContext
*s
,
1200 int fragment_height
)
1209 int i
= first_fragment
;
1213 /* DC values for the left, up-left, up, and up-right fragments */
1214 int vl
, vul
, vu
, vur
;
1216 /* indexes for the left, up-left, up, and up-right fragments */
1220 * The 6 fields mean:
1221 * 0: up-left multiplier
1223 * 2: up-right multiplier
1224 * 3: left multiplier
1226 static const int predictor_transform
[16][4] = {
1228 { 0, 0, 0,128}, // PL
1229 { 0, 0,128, 0}, // PUR
1230 { 0, 0, 53, 75}, // PUR|PL
1231 { 0,128, 0, 0}, // PU
1232 { 0, 64, 0, 64}, // PU|PL
1233 { 0,128, 0, 0}, // PU|PUR
1234 { 0, 0, 53, 75}, // PU|PUR|PL
1235 {128, 0, 0, 0}, // PUL
1236 { 0, 0, 0,128}, // PUL|PL
1237 { 64, 0, 64, 0}, // PUL|PUR
1238 { 0, 0, 53, 75}, // PUL|PUR|PL
1239 { 0,128, 0, 0}, // PUL|PU
1240 {-104,116, 0,116}, // PUL|PU|PL
1241 { 24, 80, 24, 0}, // PUL|PU|PUR
1242 {-104,116, 0,116} // PUL|PU|PUR|PL
1245 /* This table shows which types of blocks can use other blocks for
1246 * prediction. For example, INTRA is the only mode in this table to
1247 * have a frame number of 0. That means INTRA blocks can only predict
1248 * from other INTRA blocks. There are 2 golden frame coding types;
1249 * blocks encoding in these modes can only predict from other blocks
1250 * that were encoded with these 1 of these 2 modes. */
1251 static const unsigned char compatible_frame
[8] = {
1252 1, /* MODE_INTER_NO_MV */
1254 1, /* MODE_INTER_PLUS_MV */
1255 1, /* MODE_INTER_LAST_MV */
1256 1, /* MODE_INTER_PRIOR_MV */
1257 2, /* MODE_USING_GOLDEN */
1258 2, /* MODE_GOLDEN_MV */
1259 1 /* MODE_INTER_FOUR_MV */
1261 int current_frame_type
;
1263 /* there is a last DC predictor for each of the 3 frame types */
1268 vul
= vu
= vur
= vl
= 0;
1269 last_dc
[0] = last_dc
[1] = last_dc
[2] = 0;
1271 /* for each fragment row... */
1272 for (y
= 0; y
< fragment_height
; y
++) {
1274 /* for each fragment in a row... */
1275 for (x
= 0; x
< fragment_width
; x
++, i
++) {
1277 /* reverse prediction if this block was coded */
1278 if (s
->all_fragments
[i
].coding_method
!= MODE_COPY
) {
1280 current_frame_type
=
1281 compatible_frame
[s
->all_fragments
[i
].coding_method
];
1287 if(FRAME_CODED(l
) && COMPATIBLE_FRAME(l
))
1291 u
= i
-fragment_width
;
1293 if(FRAME_CODED(u
) && COMPATIBLE_FRAME(u
))
1296 ul
= i
-fragment_width
-1;
1298 if(FRAME_CODED(ul
) && COMPATIBLE_FRAME(ul
))
1301 if(x
+ 1 < fragment_width
){
1302 ur
= i
-fragment_width
+1;
1304 if(FRAME_CODED(ur
) && COMPATIBLE_FRAME(ur
))
1309 if (transform
== 0) {
1311 /* if there were no fragments to predict from, use last
1313 predicted_dc
= last_dc
[current_frame_type
];
1316 /* apply the appropriate predictor transform */
1318 (predictor_transform
[transform
][0] * vul
) +
1319 (predictor_transform
[transform
][1] * vu
) +
1320 (predictor_transform
[transform
][2] * vur
) +
1321 (predictor_transform
[transform
][3] * vl
);
1323 predicted_dc
/= 128;
1325 /* check for outranging on the [ul u l] and
1326 * [ul u ur l] predictors */
1327 if ((transform
== 13) || (transform
== 15)) {
1328 if (FFABS(predicted_dc
- vu
) > 128)
1330 else if (FFABS(predicted_dc
- vl
) > 128)
1332 else if (FFABS(predicted_dc
- vul
) > 128)
1337 /* at long last, apply the predictor */
1338 if(s
->coeffs
[i
].index
){
1339 *s
->next_coeff
= s
->coeffs
[i
];
1340 s
->coeffs
[i
].index
=0;
1341 s
->coeffs
[i
].coeff
=0;
1342 s
->coeffs
[i
].next
= s
->next_coeff
++;
1344 s
->coeffs
[i
].coeff
+= predicted_dc
;
1346 last_dc
[current_frame_type
] = DC_COEFF(i
);
1347 if(DC_COEFF(i
) && !(s
->coeff_counts
[i
]&127)){
1348 s
->coeff_counts
[i
]= 129;
1349 // s->all_fragments[i].next_coeff= s->next_coeff;
1350 s
->coeffs
[i
].next
= s
->next_coeff
;
1351 (s
->next_coeff
++)->next
=NULL
;
1359 * Perform the final rendering for a particular slice of data.
1360 * The slice number ranges from 0..(macroblock_height - 1).
1362 static void render_slice(Vp3DecodeContext
*s
, int slice
)
1365 int16_t *dequantizer
;
1366 DECLARE_ALIGNED_16(DCTELEM
, block
[64]);
1367 int motion_x
= 0xdeadbeef, motion_y
= 0xdeadbeef;
1368 int motion_halfpel_index
;
1369 uint8_t *motion_source
;
1371 int current_macroblock_entry
= slice
* s
->macroblock_width
* 6;
1373 if (slice
>= s
->macroblock_height
)
1376 for (plane
= 0; plane
< 3; plane
++) {
1377 uint8_t *output_plane
= s
->current_frame
.data
[plane
];
1378 uint8_t * last_plane
= s
-> last_frame
.data
[plane
];
1379 uint8_t *golden_plane
= s
-> golden_frame
.data
[plane
];
1380 int stride
= s
->current_frame
.linesize
[plane
];
1381 int plane_width
= s
->width
>> !!plane
;
1382 int plane_height
= s
->height
>> !!plane
;
1383 int y
= slice
* FRAGMENT_PIXELS
<< !plane
;
1384 int slice_height
= y
+ (FRAGMENT_PIXELS
<< !plane
);
1385 int i
= s
->macroblock_fragments
[current_macroblock_entry
+ plane
+ 3*!!plane
];
1387 if (!s
->flipped_image
) stride
= -stride
;
1390 if(FFABS(stride
) > 2048)
1391 return; //various tables are fixed size
1393 /* for each fragment row in the slice (both of them)... */
1394 for (; y
< slice_height
; y
+= 8) {
1396 /* for each fragment in a row... */
1397 for (x
= 0; x
< plane_width
; x
+= 8, i
++) {
1399 if ((i
< 0) || (i
>= s
->fragment_count
)) {
1400 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:render_slice(): bad fragment number (%d)\n", i
);
1404 /* transform if this block was coded */
1405 if ((s
->all_fragments
[i
].coding_method
!= MODE_COPY
) &&
1406 !((s
->avctx
->flags
& CODEC_FLAG_GRAY
) && plane
)) {
1408 if ((s
->all_fragments
[i
].coding_method
== MODE_USING_GOLDEN
) ||
1409 (s
->all_fragments
[i
].coding_method
== MODE_GOLDEN_MV
))
1410 motion_source
= golden_plane
;
1412 motion_source
= last_plane
;
1414 motion_source
+= s
->all_fragments
[i
].first_pixel
;
1415 motion_halfpel_index
= 0;
1417 /* sort out the motion vector if this fragment is coded
1418 * using a motion vector method */
1419 if ((s
->all_fragments
[i
].coding_method
> MODE_INTRA
) &&
1420 (s
->all_fragments
[i
].coding_method
!= MODE_USING_GOLDEN
)) {
1422 motion_x
= s
->all_fragments
[i
].motion_x
;
1423 motion_y
= s
->all_fragments
[i
].motion_y
;
1425 motion_x
= (motion_x
>>1) | (motion_x
&1);
1426 motion_y
= (motion_y
>>1) | (motion_y
&1);
1429 src_x
= (motion_x
>>1) + x
;
1430 src_y
= (motion_y
>>1) + y
;
1431 if ((motion_x
== 127) || (motion_y
== 127))
1432 av_log(s
->avctx
, AV_LOG_ERROR
, " help! got invalid motion vector! (%X, %X)\n", motion_x
, motion_y
);
1434 motion_halfpel_index
= motion_x
& 0x01;
1435 motion_source
+= (motion_x
>> 1);
1437 motion_halfpel_index
|= (motion_y
& 0x01) << 1;
1438 motion_source
+= ((motion_y
>> 1) * stride
);
1440 if(src_x
<0 || src_y
<0 || src_x
+ 9 >= plane_width
|| src_y
+ 9 >= plane_height
){
1441 uint8_t *temp
= s
->edge_emu_buffer
;
1442 if(stride
<0) temp
-= 9*stride
;
1443 else temp
+= 9*stride
;
1445 ff_emulated_edge_mc(temp
, motion_source
, stride
, 9, 9, src_x
, src_y
, plane_width
, plane_height
);
1446 motion_source
= temp
;
1451 /* first, take care of copying a block from either the
1452 * previous or the golden frame */
1453 if (s
->all_fragments
[i
].coding_method
!= MODE_INTRA
) {
1454 /* Note, it is possible to implement all MC cases with
1455 put_no_rnd_pixels_l2 which would look more like the
1456 VP3 source but this would be slower as
1457 put_no_rnd_pixels_tab is better optimzed */
1458 if(motion_halfpel_index
!= 3){
1459 s
->dsp
.put_no_rnd_pixels_tab
[1][motion_halfpel_index
](
1460 output_plane
+ s
->all_fragments
[i
].first_pixel
,
1461 motion_source
, stride
, 8);
1463 int d
= (motion_x
^ motion_y
)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1464 s
->dsp
.put_no_rnd_pixels_l2
[1](
1465 output_plane
+ s
->all_fragments
[i
].first_pixel
,
1467 motion_source
+ stride
+ 1 + d
,
1470 dequantizer
= s
->qmat
[s
->all_fragments
[i
].qpi
][1][plane
];
1472 dequantizer
= s
->qmat
[s
->all_fragments
[i
].qpi
][0][plane
];
1475 /* dequantize the DCT coefficients */
1476 if(s
->avctx
->idct_algo
==FF_IDCT_VP3
){
1477 Coeff
*coeff
= s
->coeffs
+ i
;
1478 s
->dsp
.clear_block(block
);
1480 block
[coeff
->index
]= coeff
->coeff
* dequantizer
[coeff
->index
];
1484 Coeff
*coeff
= s
->coeffs
+ i
;
1485 s
->dsp
.clear_block(block
);
1487 block
[coeff
->index
]= (coeff
->coeff
* dequantizer
[coeff
->index
] + 2)>>2;
1492 /* invert DCT and place (or add) in final output */
1494 if (s
->all_fragments
[i
].coding_method
== MODE_INTRA
) {
1495 if(s
->avctx
->idct_algo
!=FF_IDCT_VP3
)
1498 output_plane
+ s
->all_fragments
[i
].first_pixel
,
1503 output_plane
+ s
->all_fragments
[i
].first_pixel
,
1509 /* copy directly from the previous frame */
1510 s
->dsp
.put_pixels_tab
[1][0](
1511 output_plane
+ s
->all_fragments
[i
].first_pixel
,
1512 last_plane
+ s
->all_fragments
[i
].first_pixel
,
1517 /* perform the left edge filter if:
1518 * - the fragment is not on the left column
1519 * - the fragment is coded in this frame
1520 * - the fragment is not coded in this frame but the left
1521 * fragment is coded in this frame (this is done instead
1522 * of a right edge filter when rendering the left fragment
1523 * since this fragment is not available yet) */
1525 ((s
->all_fragments
[i
].coding_method
!= MODE_COPY
) ||
1526 ((s
->all_fragments
[i
].coding_method
== MODE_COPY
) &&
1527 (s
->all_fragments
[i
- 1].coding_method
!= MODE_COPY
)) )) {
1529 output_plane
+ s
->all_fragments
[i
].first_pixel
+ 7*stride
,
1530 -stride
, s
->bounding_values_array
+ 127);
1533 /* perform the top edge filter if:
1534 * - the fragment is not on the top row
1535 * - the fragment is coded in this frame
1536 * - the fragment is not coded in this frame but the above
1537 * fragment is coded in this frame (this is done instead
1538 * of a bottom edge filter when rendering the above
1539 * fragment since this fragment is not available yet) */
1541 ((s
->all_fragments
[i
].coding_method
!= MODE_COPY
) ||
1542 ((s
->all_fragments
[i
].coding_method
== MODE_COPY
) &&
1543 (s
->all_fragments
[i
- fragment_width
].coding_method
!= MODE_COPY
)) )) {
1545 output_plane
+ s
->all_fragments
[i
].first_pixel
- stride
,
1546 -stride
, s
->bounding_values_array
+ 127);
1553 /* this looks like a good place for slice dispatch... */
1555 * if (slice == s->macroblock_height - 1)
1556 * dispatch (both last slice & 2nd-to-last slice);
1557 * else if (slice > 0)
1558 * dispatch (slice - 1);
1564 static void apply_loop_filter(Vp3DecodeContext
*s
)
1568 int *bounding_values
= s
->bounding_values_array
+127;
1571 int bounding_values_array
[256];
1574 /* find the right loop limit value */
1575 for (x
= 63; x
>= 0; x
--) {
1576 if (vp31_ac_scale_factor
[x
] >= s
->quality_index
)
1579 filter_limit
= vp31_filter_limit_values
[s
->quality_index
];
1581 /* set up the bounding values */
1582 memset(bounding_values_array
, 0, 256 * sizeof(int));
1583 for (x
= 0; x
< filter_limit
; x
++) {
1584 bounding_values
[-x
- filter_limit
] = -filter_limit
+ x
;
1585 bounding_values
[-x
] = -x
;
1586 bounding_values
[x
] = x
;
1587 bounding_values
[x
+ filter_limit
] = filter_limit
- x
;
1591 for (plane
= 0; plane
< 3; plane
++) {
1592 int width
= s
->fragment_width
>> !!plane
;
1593 int height
= s
->fragment_height
>> !!plane
;
1594 int fragment
= s
->fragment_start
[plane
];
1595 int stride
= s
->current_frame
.linesize
[plane
];
1596 uint8_t *plane_data
= s
->current_frame
.data
[plane
];
1597 if (!s
->flipped_image
) stride
= -stride
;
1599 for (y
= 0; y
< height
; y
++) {
1601 for (x
= 0; x
< width
; x
++) {
1602 /* do not perform left edge filter for left columns frags */
1604 (s
->all_fragments
[fragment
].coding_method
!= MODE_COPY
)) {
1605 s
->dsp
.vp3_h_loop_filter(
1606 plane_data
+ s
->all_fragments
[fragment
].first_pixel
,
1607 stride
, bounding_values
);
1610 /* do not perform top edge filter for top row fragments */
1612 (s
->all_fragments
[fragment
].coding_method
!= MODE_COPY
)) {
1613 s
->dsp
.vp3_v_loop_filter(
1614 plane_data
+ s
->all_fragments
[fragment
].first_pixel
,
1615 stride
, bounding_values
);
1618 /* do not perform right edge filter for right column
1619 * fragments or if right fragment neighbor is also coded
1620 * in this frame (it will be filtered in next iteration) */
1621 if ((x
< width
- 1) &&
1622 (s
->all_fragments
[fragment
].coding_method
!= MODE_COPY
) &&
1623 (s
->all_fragments
[fragment
+ 1].coding_method
== MODE_COPY
)) {
1624 s
->dsp
.vp3_h_loop_filter(
1625 plane_data
+ s
->all_fragments
[fragment
+ 1].first_pixel
,
1626 stride
, bounding_values
);
1629 /* do not perform bottom edge filter for bottom row
1630 * fragments or if bottom fragment neighbor is also coded
1631 * in this frame (it will be filtered in the next row) */
1632 if ((y
< height
- 1) &&
1633 (s
->all_fragments
[fragment
].coding_method
!= MODE_COPY
) &&
1634 (s
->all_fragments
[fragment
+ width
].coding_method
== MODE_COPY
)) {
1635 s
->dsp
.vp3_v_loop_filter(
1636 plane_data
+ s
->all_fragments
[fragment
+ width
].first_pixel
,
1637 stride
, bounding_values
);
1647 * This function computes the first pixel addresses for each fragment.
1648 * This function needs to be invoked after the first frame is allocated
1649 * so that it has access to the plane strides.
1651 static void vp3_calculate_pixel_addresses(Vp3DecodeContext
*s
)
1653 #define Y_INITIAL(chroma_shift) s->flipped_image ? 1 : s->fragment_height >> chroma_shift
1654 #define Y_FINISHED(chroma_shift) s->flipped_image ? y <= s->fragment_height >> chroma_shift : y > 0
1657 const int y_inc
= s
->flipped_image
? 1 : -1;
1659 /* figure out the first pixel addresses for each of the fragments */
1662 for (y
= Y_INITIAL(0); Y_FINISHED(0); y
+= y_inc
) {
1663 for (x
= 0; x
< s
->fragment_width
; x
++) {
1664 s
->all_fragments
[i
++].first_pixel
=
1665 s
->golden_frame
.linesize
[0] * y
* FRAGMENT_PIXELS
-
1666 s
->golden_frame
.linesize
[0] +
1667 x
* FRAGMENT_PIXELS
;
1672 i
= s
->fragment_start
[1];
1673 for (y
= Y_INITIAL(1); Y_FINISHED(1); y
+= y_inc
) {
1674 for (x
= 0; x
< s
->fragment_width
/ 2; x
++) {
1675 s
->all_fragments
[i
++].first_pixel
=
1676 s
->golden_frame
.linesize
[1] * y
* FRAGMENT_PIXELS
-
1677 s
->golden_frame
.linesize
[1] +
1678 x
* FRAGMENT_PIXELS
;
1683 i
= s
->fragment_start
[2];
1684 for (y
= Y_INITIAL(1); Y_FINISHED(1); y
+= y_inc
) {
1685 for (x
= 0; x
< s
->fragment_width
/ 2; x
++) {
1686 s
->all_fragments
[i
++].first_pixel
=
1687 s
->golden_frame
.linesize
[2] * y
* FRAGMENT_PIXELS
-
1688 s
->golden_frame
.linesize
[2] +
1689 x
* FRAGMENT_PIXELS
;
1695 * This is the ffmpeg/libavcodec API init function.
1697 static av_cold
int vp3_decode_init(AVCodecContext
*avctx
)
1699 Vp3DecodeContext
*s
= avctx
->priv_data
;
1700 int i
, inter
, plane
;
1703 int y_superblock_count
;
1704 int c_superblock_count
;
1706 if (avctx
->codec_tag
== MKTAG('V','P','3','0'))
1712 s
->width
= FFALIGN(avctx
->width
, 16);
1713 s
->height
= FFALIGN(avctx
->height
, 16);
1714 avctx
->pix_fmt
= PIX_FMT_YUV420P
;
1715 avctx
->chroma_sample_location
= AVCHROMA_LOC_CENTER
;
1716 if(avctx
->idct_algo
==FF_IDCT_AUTO
)
1717 avctx
->idct_algo
=FF_IDCT_VP3
;
1718 dsputil_init(&s
->dsp
, avctx
);
1720 ff_init_scantable(s
->dsp
.idct_permutation
, &s
->scantable
, ff_zigzag_direct
);
1722 /* initialize to an impossible value which will force a recalculation
1723 * in the first frame decode */
1724 for (i
= 0; i
< 3; i
++)
1727 s
->y_superblock_width
= (s
->width
+ 31) / 32;
1728 s
->y_superblock_height
= (s
->height
+ 31) / 32;
1729 y_superblock_count
= s
->y_superblock_width
* s
->y_superblock_height
;
1731 /* work out the dimensions for the C planes */
1732 c_width
= s
->width
/ 2;
1733 c_height
= s
->height
/ 2;
1734 s
->c_superblock_width
= (c_width
+ 31) / 32;
1735 s
->c_superblock_height
= (c_height
+ 31) / 32;
1736 c_superblock_count
= s
->c_superblock_width
* s
->c_superblock_height
;
1738 s
->superblock_count
= y_superblock_count
+ (c_superblock_count
* 2);
1739 s
->u_superblock_start
= y_superblock_count
;
1740 s
->v_superblock_start
= s
->u_superblock_start
+ c_superblock_count
;
1741 s
->superblock_coding
= av_malloc(s
->superblock_count
);
1743 s
->macroblock_width
= (s
->width
+ 15) / 16;
1744 s
->macroblock_height
= (s
->height
+ 15) / 16;
1745 s
->macroblock_count
= s
->macroblock_width
* s
->macroblock_height
;
1747 s
->fragment_width
= s
->width
/ FRAGMENT_PIXELS
;
1748 s
->fragment_height
= s
->height
/ FRAGMENT_PIXELS
;
1750 /* fragment count covers all 8x8 blocks for all 3 planes */
1751 s
->fragment_count
= s
->fragment_width
* s
->fragment_height
* 3 / 2;
1752 s
->fragment_start
[1] = s
->fragment_width
* s
->fragment_height
;
1753 s
->fragment_start
[2] = s
->fragment_width
* s
->fragment_height
* 5 / 4;
1755 s
->all_fragments
= av_malloc(s
->fragment_count
* sizeof(Vp3Fragment
));
1756 s
->coeff_counts
= av_malloc(s
->fragment_count
* sizeof(*s
->coeff_counts
));
1757 s
->coeffs
= av_malloc(s
->fragment_count
* sizeof(Coeff
) * 65);
1758 s
->coded_fragment_list
= av_malloc(s
->fragment_count
* sizeof(int));
1759 s
->pixel_addresses_initialized
= 0;
1760 if (!s
->superblock_coding
|| !s
->all_fragments
|| !s
->coeff_counts
||
1761 !s
->coeffs
|| !s
->coded_fragment_list
) {
1762 vp3_decode_end(avctx
);
1766 if (!s
->theora_tables
)
1768 for (i
= 0; i
< 64; i
++) {
1769 s
->coded_dc_scale_factor
[i
] = vp31_dc_scale_factor
[i
];
1770 s
->coded_ac_scale_factor
[i
] = vp31_ac_scale_factor
[i
];
1771 s
->base_matrix
[0][i
] = vp31_intra_y_dequant
[i
];
1772 s
->base_matrix
[1][i
] = vp31_intra_c_dequant
[i
];
1773 s
->base_matrix
[2][i
] = vp31_inter_dequant
[i
];
1774 s
->filter_limit_values
[i
] = vp31_filter_limit_values
[i
];
1777 for(inter
=0; inter
<2; inter
++){
1778 for(plane
=0; plane
<3; plane
++){
1779 s
->qr_count
[inter
][plane
]= 1;
1780 s
->qr_size
[inter
][plane
][0]= 63;
1781 s
->qr_base
[inter
][plane
][0]=
1782 s
->qr_base
[inter
][plane
][1]= 2*inter
+ (!!plane
)*!inter
;
1786 /* init VLC tables */
1787 for (i
= 0; i
< 16; i
++) {
1790 init_vlc(&s
->dc_vlc
[i
], 5, 32,
1791 &dc_bias
[i
][0][1], 4, 2,
1792 &dc_bias
[i
][0][0], 4, 2, 0);
1794 /* group 1 AC histograms */
1795 init_vlc(&s
->ac_vlc_1
[i
], 5, 32,
1796 &ac_bias_0
[i
][0][1], 4, 2,
1797 &ac_bias_0
[i
][0][0], 4, 2, 0);
1799 /* group 2 AC histograms */
1800 init_vlc(&s
->ac_vlc_2
[i
], 5, 32,
1801 &ac_bias_1
[i
][0][1], 4, 2,
1802 &ac_bias_1
[i
][0][0], 4, 2, 0);
1804 /* group 3 AC histograms */
1805 init_vlc(&s
->ac_vlc_3
[i
], 5, 32,
1806 &ac_bias_2
[i
][0][1], 4, 2,
1807 &ac_bias_2
[i
][0][0], 4, 2, 0);
1809 /* group 4 AC histograms */
1810 init_vlc(&s
->ac_vlc_4
[i
], 5, 32,
1811 &ac_bias_3
[i
][0][1], 4, 2,
1812 &ac_bias_3
[i
][0][0], 4, 2, 0);
1815 for (i
= 0; i
< 16; i
++) {
1818 if (init_vlc(&s
->dc_vlc
[i
], 5, 32,
1819 &s
->huffman_table
[i
][0][1], 4, 2,
1820 &s
->huffman_table
[i
][0][0], 4, 2, 0) < 0)
1823 /* group 1 AC histograms */
1824 if (init_vlc(&s
->ac_vlc_1
[i
], 5, 32,
1825 &s
->huffman_table
[i
+16][0][1], 4, 2,
1826 &s
->huffman_table
[i
+16][0][0], 4, 2, 0) < 0)
1829 /* group 2 AC histograms */
1830 if (init_vlc(&s
->ac_vlc_2
[i
], 5, 32,
1831 &s
->huffman_table
[i
+16*2][0][1], 4, 2,
1832 &s
->huffman_table
[i
+16*2][0][0], 4, 2, 0) < 0)
1835 /* group 3 AC histograms */
1836 if (init_vlc(&s
->ac_vlc_3
[i
], 5, 32,
1837 &s
->huffman_table
[i
+16*3][0][1], 4, 2,
1838 &s
->huffman_table
[i
+16*3][0][0], 4, 2, 0) < 0)
1841 /* group 4 AC histograms */
1842 if (init_vlc(&s
->ac_vlc_4
[i
], 5, 32,
1843 &s
->huffman_table
[i
+16*4][0][1], 4, 2,
1844 &s
->huffman_table
[i
+16*4][0][0], 4, 2, 0) < 0)
1849 init_vlc(&s
->superblock_run_length_vlc
, 6, 34,
1850 &superblock_run_length_vlc_table
[0][1], 4, 2,
1851 &superblock_run_length_vlc_table
[0][0], 4, 2, 0);
1853 init_vlc(&s
->fragment_run_length_vlc
, 5, 30,
1854 &fragment_run_length_vlc_table
[0][1], 4, 2,
1855 &fragment_run_length_vlc_table
[0][0], 4, 2, 0);
1857 init_vlc(&s
->mode_code_vlc
, 3, 8,
1858 &mode_code_vlc_table
[0][1], 2, 1,
1859 &mode_code_vlc_table
[0][0], 2, 1, 0);
1861 init_vlc(&s
->motion_vector_vlc
, 6, 63,
1862 &motion_vector_vlc_table
[0][1], 2, 1,
1863 &motion_vector_vlc_table
[0][0], 2, 1, 0);
1865 /* work out the block mapping tables */
1866 s
->superblock_fragments
= av_malloc(s
->superblock_count
* 16 * sizeof(int));
1867 s
->superblock_macroblocks
= av_malloc(s
->superblock_count
* 4 * sizeof(int));
1868 s
->macroblock_fragments
= av_malloc(s
->macroblock_count
* 6 * sizeof(int));
1869 s
->macroblock_coding
= av_malloc(s
->macroblock_count
+ 1);
1870 if (!s
->superblock_fragments
|| !s
->superblock_macroblocks
||
1871 !s
->macroblock_fragments
|| !s
->macroblock_coding
) {
1872 vp3_decode_end(avctx
);
1875 init_block_mapping(s
);
1877 for (i
= 0; i
< 3; i
++) {
1878 s
->current_frame
.data
[i
] = NULL
;
1879 s
->last_frame
.data
[i
] = NULL
;
1880 s
->golden_frame
.data
[i
] = NULL
;
1886 av_log(avctx
, AV_LOG_FATAL
, "Invalid huffman table\n");
1891 * This is the ffmpeg/libavcodec API frame decode function.
1893 static int vp3_decode_frame(AVCodecContext
*avctx
,
1894 void *data
, int *data_size
,
1897 const uint8_t *buf
= avpkt
->data
;
1898 int buf_size
= avpkt
->size
;
1899 Vp3DecodeContext
*s
= avctx
->priv_data
;
1901 static int counter
= 0;
1904 init_get_bits(&gb
, buf
, buf_size
* 8);
1906 if (s
->theora
&& get_bits1(&gb
))
1908 av_log(avctx
, AV_LOG_ERROR
, "Header packet passed to frame decoder, skipping\n");
1912 s
->keyframe
= !get_bits1(&gb
);
1915 for (i
= 0; i
< 3; i
++)
1916 s
->last_qps
[i
] = s
->qps
[i
];
1920 s
->qps
[s
->nqps
++]= get_bits(&gb
, 6);
1921 } while(s
->theora
>= 0x030200 && s
->nqps
<3 && get_bits1(&gb
));
1922 for (i
= s
->nqps
; i
< 3; i
++)
1925 if (s
->avctx
->debug
& FF_DEBUG_PICT_INFO
)
1926 av_log(s
->avctx
, AV_LOG_INFO
, " VP3 %sframe #%d: Q index = %d\n",
1927 s
->keyframe
?"key":"", counter
, s
->qps
[0]);
1930 if (s
->qps
[0] != s
->last_qps
[0])
1931 init_loop_filter(s
);
1933 for (i
= 0; i
< s
->nqps
; i
++)
1934 // reinit all dequantizers if the first one changed, because
1935 // the DC of the first quantizer must be used for all matrices
1936 if (s
->qps
[i
] != s
->last_qps
[i
] || s
->qps
[0] != s
->last_qps
[0])
1937 init_dequantizer(s
, i
);
1939 if (avctx
->skip_frame
>= AVDISCARD_NONKEY
&& !s
->keyframe
)
1945 skip_bits(&gb
, 4); /* width code */
1946 skip_bits(&gb
, 4); /* height code */
1949 s
->version
= get_bits(&gb
, 5);
1951 av_log(s
->avctx
, AV_LOG_DEBUG
, "VP version: %d\n", s
->version
);
1954 if (s
->version
|| s
->theora
)
1957 av_log(s
->avctx
, AV_LOG_ERROR
, "Warning, unsupported keyframe coding type?!\n");
1958 skip_bits(&gb
, 2); /* reserved? */
1961 if (s
->last_frame
.data
[0] == s
->golden_frame
.data
[0]) {
1962 if (s
->golden_frame
.data
[0])
1963 avctx
->release_buffer(avctx
, &s
->golden_frame
);
1964 s
->last_frame
= s
->golden_frame
; /* ensure that we catch any access to this released frame */
1966 if (s
->golden_frame
.data
[0])
1967 avctx
->release_buffer(avctx
, &s
->golden_frame
);
1968 if (s
->last_frame
.data
[0])
1969 avctx
->release_buffer(avctx
, &s
->last_frame
);
1972 s
->golden_frame
.reference
= 3;
1973 if(avctx
->get_buffer(avctx
, &s
->golden_frame
) < 0) {
1974 av_log(s
->avctx
, AV_LOG_ERROR
, "vp3: get_buffer() failed\n");
1978 /* golden frame is also the current frame */
1979 s
->current_frame
= s
->golden_frame
;
1981 /* time to figure out pixel addresses? */
1982 if (!s
->pixel_addresses_initialized
)
1984 vp3_calculate_pixel_addresses(s
);
1985 s
->pixel_addresses_initialized
= 1;
1988 /* allocate a new current frame */
1989 s
->current_frame
.reference
= 3;
1990 if (!s
->pixel_addresses_initialized
) {
1991 av_log(s
->avctx
, AV_LOG_ERROR
, "vp3: first frame not a keyframe\n");
1994 if(avctx
->get_buffer(avctx
, &s
->current_frame
) < 0) {
1995 av_log(s
->avctx
, AV_LOG_ERROR
, "vp3: get_buffer() failed\n");
2000 s
->current_frame
.qscale_table
= s
->qscale_table
; //FIXME allocate individual tables per AVFrame
2001 s
->current_frame
.qstride
= 0;
2005 if (unpack_superblocks(s
, &gb
)){
2006 av_log(s
->avctx
, AV_LOG_ERROR
, "error in unpack_superblocks\n");
2009 if (unpack_modes(s
, &gb
)){
2010 av_log(s
->avctx
, AV_LOG_ERROR
, "error in unpack_modes\n");
2013 if (unpack_vectors(s
, &gb
)){
2014 av_log(s
->avctx
, AV_LOG_ERROR
, "error in unpack_vectors\n");
2017 if (unpack_block_qpis(s
, &gb
)){
2018 av_log(s
->avctx
, AV_LOG_ERROR
, "error in unpack_block_qpis\n");
2021 if (unpack_dct_coeffs(s
, &gb
)){
2022 av_log(s
->avctx
, AV_LOG_ERROR
, "error in unpack_dct_coeffs\n");
2026 for (i
= 0; i
< s
->macroblock_height
; i
++)
2029 apply_loop_filter(s
);
2031 *data_size
=sizeof(AVFrame
);
2032 *(AVFrame
*)data
= s
->current_frame
;
2034 /* release the last frame, if it is allocated and if it is not the
2036 if ((s
->last_frame
.data
[0]) &&
2037 (s
->last_frame
.data
[0] != s
->golden_frame
.data
[0]))
2038 avctx
->release_buffer(avctx
, &s
->last_frame
);
2040 /* shuffle frames (last = current) */
2041 s
->last_frame
= s
->current_frame
;
2042 s
->current_frame
.data
[0]= NULL
; /* ensure that we catch any access to this released frame */
2048 * This is the ffmpeg/libavcodec API module cleanup function.
2050 static av_cold
int vp3_decode_end(AVCodecContext
*avctx
)
2052 Vp3DecodeContext
*s
= avctx
->priv_data
;
2055 av_free(s
->superblock_coding
);
2056 av_free(s
->all_fragments
);
2057 av_free(s
->coeff_counts
);
2059 av_free(s
->coded_fragment_list
);
2060 av_free(s
->superblock_fragments
);
2061 av_free(s
->superblock_macroblocks
);
2062 av_free(s
->macroblock_fragments
);
2063 av_free(s
->macroblock_coding
);
2065 for (i
= 0; i
< 16; i
++) {
2066 free_vlc(&s
->dc_vlc
[i
]);
2067 free_vlc(&s
->ac_vlc_1
[i
]);
2068 free_vlc(&s
->ac_vlc_2
[i
]);
2069 free_vlc(&s
->ac_vlc_3
[i
]);
2070 free_vlc(&s
->ac_vlc_4
[i
]);
2073 free_vlc(&s
->superblock_run_length_vlc
);
2074 free_vlc(&s
->fragment_run_length_vlc
);
2075 free_vlc(&s
->mode_code_vlc
);
2076 free_vlc(&s
->motion_vector_vlc
);
2078 /* release all frames */
2079 if (s
->golden_frame
.data
[0] && s
->golden_frame
.data
[0] != s
->last_frame
.data
[0])
2080 avctx
->release_buffer(avctx
, &s
->golden_frame
);
2081 if (s
->last_frame
.data
[0])
2082 avctx
->release_buffer(avctx
, &s
->last_frame
);
2083 /* no need to release the current_frame since it will always be pointing
2084 * to the same frame as either the golden or last frame */
2089 static int read_huffman_tree(AVCodecContext
*avctx
, GetBitContext
*gb
)
2091 Vp3DecodeContext
*s
= avctx
->priv_data
;
2093 if (get_bits1(gb
)) {
2095 if (s
->entries
>= 32) { /* overflow */
2096 av_log(avctx
, AV_LOG_ERROR
, "huffman tree overflow\n");
2099 token
= get_bits(gb
, 5);
2100 //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);
2101 s
->huffman_table
[s
->hti
][token
][0] = s
->hbits
;
2102 s
->huffman_table
[s
->hti
][token
][1] = s
->huff_code_size
;
2106 if (s
->huff_code_size
>= 32) {/* overflow */
2107 av_log(avctx
, AV_LOG_ERROR
, "huffman tree overflow\n");
2110 s
->huff_code_size
++;
2112 if (read_huffman_tree(avctx
, gb
))
2115 if (read_huffman_tree(avctx
, gb
))
2118 s
->huff_code_size
--;
2123 #if CONFIG_THEORA_DECODER
2124 static int theora_decode_header(AVCodecContext
*avctx
, GetBitContext
*gb
)
2126 Vp3DecodeContext
*s
= avctx
->priv_data
;
2127 int visible_width
, visible_height
;
2129 s
->theora
= get_bits_long(gb
, 24);
2130 av_log(avctx
, AV_LOG_DEBUG
, "Theora bitstream version %X\n", s
->theora
);
2132 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2133 /* but previous versions have the image flipped relative to vp3 */
2134 if (s
->theora
< 0x030200)
2136 s
->flipped_image
= 1;
2137 av_log(avctx
, AV_LOG_DEBUG
, "Old (<alpha3) Theora bitstream, flipped image\n");
2140 visible_width
= s
->width
= get_bits(gb
, 16) << 4;
2141 visible_height
= s
->height
= get_bits(gb
, 16) << 4;
2143 if(avcodec_check_dimensions(avctx
, s
->width
, s
->height
)){
2144 av_log(avctx
, AV_LOG_ERROR
, "Invalid dimensions (%dx%d)\n", s
->width
, s
->height
);
2145 s
->width
= s
->height
= 0;
2149 if (s
->theora
>= 0x030400)
2151 skip_bits(gb
, 32); /* total number of superblocks in a frame */
2152 // fixme, the next field is 36bits long
2153 skip_bits(gb
, 32); /* total number of blocks in a frame */
2154 skip_bits(gb
, 4); /* total number of blocks in a frame */
2155 skip_bits(gb
, 32); /* total number of macroblocks in a frame */
2158 if (s
->theora
>= 0x030200) {
2159 visible_width
= get_bits_long(gb
, 24);
2160 visible_height
= get_bits_long(gb
, 24);
2162 skip_bits(gb
, 8); /* offset x */
2163 skip_bits(gb
, 8); /* offset y */
2166 skip_bits(gb
, 32); /* fps numerator */
2167 skip_bits(gb
, 32); /* fps denumerator */
2168 skip_bits(gb
, 24); /* aspect numerator */
2169 skip_bits(gb
, 24); /* aspect denumerator */
2171 if (s
->theora
< 0x030200)
2172 skip_bits(gb
, 5); /* keyframe frequency force */
2173 skip_bits(gb
, 8); /* colorspace */
2174 if (s
->theora
>= 0x030400)
2175 skip_bits(gb
, 2); /* pixel format: 420,res,422,444 */
2176 skip_bits(gb
, 24); /* bitrate */
2178 skip_bits(gb
, 6); /* quality hint */
2180 if (s
->theora
>= 0x030200)
2182 skip_bits(gb
, 5); /* keyframe frequency force */
2184 if (s
->theora
< 0x030400)
2185 skip_bits(gb
, 5); /* spare bits */
2188 // align_get_bits(gb);
2190 if ( visible_width
<= s
->width
&& visible_width
> s
->width
-16
2191 && visible_height
<= s
->height
&& visible_height
> s
->height
-16)
2192 avcodec_set_dimensions(avctx
, visible_width
, visible_height
);
2194 avcodec_set_dimensions(avctx
, s
->width
, s
->height
);
2199 static int theora_decode_tables(AVCodecContext
*avctx
, GetBitContext
*gb
)
2201 Vp3DecodeContext
*s
= avctx
->priv_data
;
2202 int i
, n
, matrices
, inter
, plane
;
2204 if (s
->theora
>= 0x030200) {
2205 n
= get_bits(gb
, 3);
2206 /* loop filter limit values table */
2207 for (i
= 0; i
< 64; i
++) {
2208 s
->filter_limit_values
[i
] = get_bits(gb
, n
);
2209 if (s
->filter_limit_values
[i
] > 127) {
2210 av_log(avctx
, AV_LOG_ERROR
, "filter limit value too large (%i > 127), clamping\n", s
->filter_limit_values
[i
]);
2211 s
->filter_limit_values
[i
] = 127;
2216 if (s
->theora
>= 0x030200)
2217 n
= get_bits(gb
, 4) + 1;
2220 /* quality threshold table */
2221 for (i
= 0; i
< 64; i
++)
2222 s
->coded_ac_scale_factor
[i
] = get_bits(gb
, n
);
2224 if (s
->theora
>= 0x030200)
2225 n
= get_bits(gb
, 4) + 1;
2228 /* dc scale factor table */
2229 for (i
= 0; i
< 64; i
++)
2230 s
->coded_dc_scale_factor
[i
] = get_bits(gb
, n
);
2232 if (s
->theora
>= 0x030200)
2233 matrices
= get_bits(gb
, 9) + 1;
2238 av_log(avctx
, AV_LOG_ERROR
, "invalid number of base matrixes\n");
2242 for(n
=0; n
<matrices
; n
++){
2243 for (i
= 0; i
< 64; i
++)
2244 s
->base_matrix
[n
][i
]= get_bits(gb
, 8);
2247 for (inter
= 0; inter
<= 1; inter
++) {
2248 for (plane
= 0; plane
<= 2; plane
++) {
2250 if (inter
|| plane
> 0)
2251 newqr
= get_bits1(gb
);
2254 if(inter
&& get_bits1(gb
)){
2258 qtj
= (3*inter
+ plane
- 1) / 3;
2259 plj
= (plane
+ 2) % 3;
2261 s
->qr_count
[inter
][plane
]= s
->qr_count
[qtj
][plj
];
2262 memcpy(s
->qr_size
[inter
][plane
], s
->qr_size
[qtj
][plj
], sizeof(s
->qr_size
[0][0]));
2263 memcpy(s
->qr_base
[inter
][plane
], s
->qr_base
[qtj
][plj
], sizeof(s
->qr_base
[0][0]));
2269 i
= get_bits(gb
, av_log2(matrices
-1)+1);
2271 av_log(avctx
, AV_LOG_ERROR
, "invalid base matrix index\n");
2274 s
->qr_base
[inter
][plane
][qri
]= i
;
2277 i
= get_bits(gb
, av_log2(63-qi
)+1) + 1;
2278 s
->qr_size
[inter
][plane
][qri
++]= i
;
2283 av_log(avctx
, AV_LOG_ERROR
, "invalid qi %d > 63\n", qi
);
2286 s
->qr_count
[inter
][plane
]= qri
;
2291 /* Huffman tables */
2292 for (s
->hti
= 0; s
->hti
< 80; s
->hti
++) {
2294 s
->huff_code_size
= 1;
2295 if (!get_bits1(gb
)) {
2297 if(read_huffman_tree(avctx
, gb
))
2300 if(read_huffman_tree(avctx
, gb
))
2305 s
->theora_tables
= 1;
2310 static av_cold
int theora_decode_init(AVCodecContext
*avctx
)
2312 Vp3DecodeContext
*s
= avctx
->priv_data
;
2315 uint8_t *header_start
[3];
2321 if (!avctx
->extradata_size
)
2323 av_log(avctx
, AV_LOG_ERROR
, "Missing extradata!\n");
2327 if (ff_split_xiph_headers(avctx
->extradata
, avctx
->extradata_size
,
2328 42, header_start
, header_len
) < 0) {
2329 av_log(avctx
, AV_LOG_ERROR
, "Corrupt extradata\n");
2334 init_get_bits(&gb
, header_start
[i
], header_len
[i
] * 8);
2336 ptype
= get_bits(&gb
, 8);
2338 if (!(ptype
& 0x80))
2340 av_log(avctx
, AV_LOG_ERROR
, "Invalid extradata!\n");
2344 // FIXME: Check for this as well.
2345 skip_bits_long(&gb
, 6*8); /* "theora" */
2350 theora_decode_header(avctx
, &gb
);
2353 // FIXME: is this needed? it breaks sometimes
2354 // theora_decode_comments(avctx, gb);
2357 if (theora_decode_tables(avctx
, &gb
))
2361 av_log(avctx
, AV_LOG_ERROR
, "Unknown Theora config packet: %d\n", ptype
&~0x80);
2364 if(ptype
!= 0x81 && 8*header_len
[i
] != get_bits_count(&gb
))
2365 av_log(avctx
, AV_LOG_WARNING
, "%d bits left in packet %X\n", 8*header_len
[i
] - get_bits_count(&gb
), ptype
);
2366 if (s
->theora
< 0x030200)
2370 return vp3_decode_init(avctx
);
2373 AVCodec theora_decoder
= {
2377 sizeof(Vp3DecodeContext
),
2384 .long_name
= NULL_IF_CONFIG_SMALL("Theora"),
2388 AVCodec vp3_decoder
= {
2392 sizeof(Vp3DecodeContext
),
2399 .long_name
= NULL_IF_CONFIG_SMALL("On2 VP3"),