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
44 #define FRAGMENT_PIXELS 8
46 typedef struct Coeff
{
52 //FIXME split things out into their own arrays
53 typedef struct Vp3Fragment
{
55 /* address of first pixel taking into account which plane the fragment
56 * lives on as well as the plane stride */
58 /* this is the macroblock that the fragment belongs to */
60 uint8_t coding_method
;
66 #define SB_NOT_CODED 0
67 #define SB_PARTIALLY_CODED 1
68 #define SB_FULLY_CODED 2
70 #define MODE_INTER_NO_MV 0
72 #define MODE_INTER_PLUS_MV 2
73 #define MODE_INTER_LAST_MV 3
74 #define MODE_INTER_PRIOR_LAST 4
75 #define MODE_USING_GOLDEN 5
76 #define MODE_GOLDEN_MV 6
77 #define MODE_INTER_FOURMV 7
78 #define CODING_MODE_COUNT 8
80 /* special internal mode */
83 /* There are 6 preset schemes, plus a free-form scheme */
84 static const int ModeAlphabet
[6][CODING_MODE_COUNT
] =
86 /* scheme 1: Last motion vector dominates */
87 { MODE_INTER_LAST_MV
, MODE_INTER_PRIOR_LAST
,
88 MODE_INTER_PLUS_MV
, MODE_INTER_NO_MV
,
89 MODE_INTRA
, MODE_USING_GOLDEN
,
90 MODE_GOLDEN_MV
, MODE_INTER_FOURMV
},
93 { MODE_INTER_LAST_MV
, MODE_INTER_PRIOR_LAST
,
94 MODE_INTER_NO_MV
, MODE_INTER_PLUS_MV
,
95 MODE_INTRA
, MODE_USING_GOLDEN
,
96 MODE_GOLDEN_MV
, MODE_INTER_FOURMV
},
99 { MODE_INTER_LAST_MV
, MODE_INTER_PLUS_MV
,
100 MODE_INTER_PRIOR_LAST
, MODE_INTER_NO_MV
,
101 MODE_INTRA
, MODE_USING_GOLDEN
,
102 MODE_GOLDEN_MV
, MODE_INTER_FOURMV
},
105 { MODE_INTER_LAST_MV
, MODE_INTER_PLUS_MV
,
106 MODE_INTER_NO_MV
, MODE_INTER_PRIOR_LAST
,
107 MODE_INTRA
, MODE_USING_GOLDEN
,
108 MODE_GOLDEN_MV
, MODE_INTER_FOURMV
},
110 /* scheme 5: No motion vector dominates */
111 { MODE_INTER_NO_MV
, MODE_INTER_LAST_MV
,
112 MODE_INTER_PRIOR_LAST
, MODE_INTER_PLUS_MV
,
113 MODE_INTRA
, MODE_USING_GOLDEN
,
114 MODE_GOLDEN_MV
, MODE_INTER_FOURMV
},
117 { MODE_INTER_NO_MV
, MODE_USING_GOLDEN
,
118 MODE_INTER_LAST_MV
, MODE_INTER_PRIOR_LAST
,
119 MODE_INTER_PLUS_MV
, MODE_INTRA
,
120 MODE_GOLDEN_MV
, MODE_INTER_FOURMV
},
124 #define MIN_DEQUANT_VAL 2
126 typedef struct Vp3DecodeContext
{
127 AVCodecContext
*avctx
;
128 int theora
, theora_tables
;
131 AVFrame golden_frame
;
133 AVFrame current_frame
;
142 int superblock_count
;
143 int y_superblock_width
;
144 int y_superblock_height
;
145 int c_superblock_width
;
146 int c_superblock_height
;
147 int u_superblock_start
;
148 int v_superblock_start
;
149 unsigned char *superblock_coding
;
151 int macroblock_count
;
152 int macroblock_width
;
153 int macroblock_height
;
159 Vp3Fragment
*all_fragments
;
160 uint8_t *coeff_counts
;
163 int fragment_start
[3];
168 uint16_t coded_dc_scale_factor
[64];
169 uint32_t coded_ac_scale_factor
[64];
170 uint8_t base_matrix
[384][64];
171 uint8_t qr_count
[2][3];
172 uint8_t qr_size
[2][3][64];
173 uint16_t qr_base
[2][3][64];
175 /* this is a list of indexes into the all_fragments array indicating
176 * which of the fragments are coded */
177 int *coded_fragment_list
;
178 int coded_fragment_list_index
;
179 int pixel_addresses_initialized
;
187 VLC superblock_run_length_vlc
;
188 VLC fragment_run_length_vlc
;
190 VLC motion_vector_vlc
;
192 /* these arrays need to be on 16-byte boundaries since SSE2 operations
194 DECLARE_ALIGNED_16(int16_t, qmat
[3][2][3][64]); //<qmat[qpi][is_inter][plane]
196 /* This table contains superblock_count * 16 entries. Each set of 16
197 * numbers corresponds to the fragment indexes 0..15 of the superblock.
198 * An entry will be -1 to indicate that no entry corresponds to that
200 int *superblock_fragments
;
202 /* This table contains superblock_count * 4 entries. Each set of 4
203 * numbers corresponds to the macroblock indexes 0..3 of the superblock.
204 * An entry will be -1 to indicate that no entry corresponds to that
206 int *superblock_macroblocks
;
208 /* This table contains macroblock_count * 6 entries. Each set of 6
209 * numbers corresponds to the fragment indexes 0..5 which comprise
210 * the macroblock (4 Y fragments and 2 C fragments). */
211 int *macroblock_fragments
;
212 /* This is an array that indicates how a particular macroblock
214 unsigned char *macroblock_coding
;
216 int first_coded_y_fragment
;
217 int first_coded_c_fragment
;
218 int last_coded_y_fragment
;
219 int last_coded_c_fragment
;
221 uint8_t edge_emu_buffer
[9*2048]; //FIXME dynamic alloc
222 int8_t qscale_table
[2048]; //FIXME dynamic alloc (width+15)/16
229 uint16_t huffman_table
[80][32][2];
231 uint8_t filter_limit_values
[64];
232 DECLARE_ALIGNED_8(int, bounding_values_array
[256+2]);
235 /************************************************************************
236 * VP3 specific functions
237 ************************************************************************/
240 * This function sets up all of the various blocks mappings:
241 * superblocks <-> fragments, macroblocks <-> fragments,
242 * superblocks <-> macroblocks
244 * Returns 0 is successful; returns 1 if *anything* went wrong.
246 static int init_block_mapping(Vp3DecodeContext
*s
)
249 signed int hilbert_walk_mb
[4];
251 int current_fragment
= 0;
252 int current_width
= 0;
253 int current_height
= 0;
256 int superblock_row_inc
= 0;
257 int mapping_index
= 0;
259 int current_macroblock
;
262 signed char travel_width
[16] = {
269 signed char travel_height
[16] = {
276 signed char travel_width_mb
[4] = {
280 signed char travel_height_mb
[4] = {
284 hilbert_walk_mb
[0] = 1;
285 hilbert_walk_mb
[1] = s
->macroblock_width
;
286 hilbert_walk_mb
[2] = 1;
287 hilbert_walk_mb
[3] = -s
->macroblock_width
;
289 /* iterate through each superblock (all planes) and map the fragments */
290 for (i
= 0; i
< s
->superblock_count
; i
++) {
291 /* time to re-assign the limits? */
294 /* start of Y superblocks */
295 right_edge
= s
->fragment_width
;
296 bottom_edge
= s
->fragment_height
;
299 superblock_row_inc
= 3 * s
->fragment_width
-
300 (s
->y_superblock_width
* 4 - s
->fragment_width
);
302 /* the first operation for this variable is to advance by 1 */
303 current_fragment
= -1;
305 } else if (i
== s
->u_superblock_start
) {
307 /* start of U superblocks */
308 right_edge
= s
->fragment_width
/ 2;
309 bottom_edge
= s
->fragment_height
/ 2;
312 superblock_row_inc
= 3 * (s
->fragment_width
/ 2) -
313 (s
->c_superblock_width
* 4 - s
->fragment_width
/ 2);
315 /* the first operation for this variable is to advance by 1 */
316 current_fragment
= s
->fragment_start
[1] - 1;
318 } else if (i
== s
->v_superblock_start
) {
320 /* start of V superblocks */
321 right_edge
= s
->fragment_width
/ 2;
322 bottom_edge
= s
->fragment_height
/ 2;
325 superblock_row_inc
= 3 * (s
->fragment_width
/ 2) -
326 (s
->c_superblock_width
* 4 - s
->fragment_width
/ 2);
328 /* the first operation for this variable is to advance by 1 */
329 current_fragment
= s
->fragment_start
[2] - 1;
333 if (current_width
>= right_edge
- 1) {
334 /* reset width and move to next superblock row */
338 /* fragment is now at the start of a new superblock row */
339 current_fragment
+= superblock_row_inc
;
342 /* iterate through all 16 fragments in a superblock */
343 for (j
= 0; j
< 16; j
++) {
344 current_fragment
+= travel_width
[j
] + right_edge
* travel_height
[j
];
345 current_width
+= travel_width
[j
];
346 current_height
+= travel_height
[j
];
348 /* check if the fragment is in bounds */
349 if ((current_width
< right_edge
) &&
350 (current_height
< bottom_edge
)) {
351 s
->superblock_fragments
[mapping_index
] = current_fragment
;
353 s
->superblock_fragments
[mapping_index
] = -1;
360 /* initialize the superblock <-> macroblock mapping; iterate through
361 * all of the Y plane superblocks to build this mapping */
362 right_edge
= s
->macroblock_width
;
363 bottom_edge
= s
->macroblock_height
;
366 superblock_row_inc
= s
->macroblock_width
-
367 (s
->y_superblock_width
* 2 - s
->macroblock_width
);
369 current_macroblock
= -1;
370 for (i
= 0; i
< s
->u_superblock_start
; i
++) {
372 if (current_width
>= right_edge
- 1) {
373 /* reset width and move to next superblock row */
377 /* macroblock is now at the start of a new superblock row */
378 current_macroblock
+= superblock_row_inc
;
381 /* iterate through each potential macroblock in the superblock */
382 for (j
= 0; j
< 4; j
++) {
383 current_macroblock
+= hilbert_walk_mb
[j
];
384 current_width
+= travel_width_mb
[j
];
385 current_height
+= travel_height_mb
[j
];
387 /* check if the macroblock is in bounds */
388 if ((current_width
< right_edge
) &&
389 (current_height
< bottom_edge
)) {
390 s
->superblock_macroblocks
[mapping_index
] = current_macroblock
;
392 s
->superblock_macroblocks
[mapping_index
] = -1;
399 /* initialize the macroblock <-> fragment mapping */
400 current_fragment
= 0;
401 current_macroblock
= 0;
403 for (i
= 0; i
< s
->fragment_height
; i
+= 2) {
405 for (j
= 0; j
< s
->fragment_width
; j
+= 2) {
407 s
->all_fragments
[current_fragment
].macroblock
= current_macroblock
;
408 s
->macroblock_fragments
[mapping_index
++] = current_fragment
;
410 if (j
+ 1 < s
->fragment_width
) {
411 s
->all_fragments
[current_fragment
+ 1].macroblock
= current_macroblock
;
412 s
->macroblock_fragments
[mapping_index
++] = current_fragment
+ 1;
414 s
->macroblock_fragments
[mapping_index
++] = -1;
416 if (i
+ 1 < s
->fragment_height
) {
417 s
->all_fragments
[current_fragment
+ s
->fragment_width
].macroblock
=
419 s
->macroblock_fragments
[mapping_index
++] =
420 current_fragment
+ s
->fragment_width
;
422 s
->macroblock_fragments
[mapping_index
++] = -1;
424 if ((j
+ 1 < s
->fragment_width
) && (i
+ 1 < s
->fragment_height
)) {
425 s
->all_fragments
[current_fragment
+ s
->fragment_width
+ 1].macroblock
=
427 s
->macroblock_fragments
[mapping_index
++] =
428 current_fragment
+ s
->fragment_width
+ 1;
430 s
->macroblock_fragments
[mapping_index
++] = -1;
433 c_fragment
= s
->fragment_start
[1] +
434 (i
* s
->fragment_width
/ 4) + (j
/ 2);
435 s
->all_fragments
[c_fragment
].macroblock
= s
->macroblock_count
;
436 s
->macroblock_fragments
[mapping_index
++] = c_fragment
;
438 c_fragment
= s
->fragment_start
[2] +
439 (i
* s
->fragment_width
/ 4) + (j
/ 2);
440 s
->all_fragments
[c_fragment
].macroblock
= s
->macroblock_count
;
441 s
->macroblock_fragments
[mapping_index
++] = c_fragment
;
443 if (j
+ 2 <= s
->fragment_width
)
444 current_fragment
+= 2;
447 current_macroblock
++;
450 current_fragment
+= s
->fragment_width
;
453 return 0; /* successful path out */
457 * This function wipes out all of the fragment data.
459 static void init_frame(Vp3DecodeContext
*s
, GetBitContext
*gb
)
463 /* zero out all of the fragment information */
464 s
->coded_fragment_list_index
= 0;
465 for (i
= 0; i
< s
->fragment_count
; i
++) {
466 s
->coeff_counts
[i
] = 0;
467 s
->all_fragments
[i
].motion_x
= 127;
468 s
->all_fragments
[i
].motion_y
= 127;
469 s
->all_fragments
[i
].next_coeff
= NULL
;
470 s
->all_fragments
[i
].qpi
= 0;
472 s
->coeffs
[i
].coeff
=0;
473 s
->coeffs
[i
].next
= NULL
;
478 * This function sets up the dequantization tables used for a particular
481 static void init_dequantizer(Vp3DecodeContext
*s
, int qpi
)
483 int ac_scale_factor
= s
->coded_ac_scale_factor
[s
->qps
[qpi
]];
484 int dc_scale_factor
= s
->coded_dc_scale_factor
[s
->qps
[qpi
]];
485 int i
, plane
, inter
, qri
, bmi
, bmj
, qistart
;
487 for(inter
=0; inter
<2; inter
++){
488 for(plane
=0; plane
<3; plane
++){
490 for(qri
=0; qri
<s
->qr_count
[inter
][plane
]; qri
++){
491 sum
+= s
->qr_size
[inter
][plane
][qri
];
492 if(s
->qps
[qpi
] <= sum
)
495 qistart
= sum
- s
->qr_size
[inter
][plane
][qri
];
496 bmi
= s
->qr_base
[inter
][plane
][qri
];
497 bmj
= s
->qr_base
[inter
][plane
][qri
+1];
499 int coeff
= ( 2*(sum
-s
->qps
[qpi
])*s
->base_matrix
[bmi
][i
]
500 - 2*(qistart
-s
->qps
[qpi
])*s
->base_matrix
[bmj
][i
]
501 + s
->qr_size
[inter
][plane
][qri
])
502 / (2*s
->qr_size
[inter
][plane
][qri
]);
504 int qmin
= 8<<(inter
+ !i
);
505 int qscale
= i
? ac_scale_factor
: dc_scale_factor
;
507 s
->qmat
[qpi
][inter
][plane
][s
->dsp
.idct_permutation
[i
]]= av_clip((qscale
* coeff
)/100 * 4, qmin
, 4096);
509 // all DC coefficients use the same quant so as not to interfere with DC prediction
510 s
->qmat
[qpi
][inter
][plane
][0] = s
->qmat
[0][inter
][plane
][0];
514 memset(s
->qscale_table
, (FFMAX(s
->qmat
[0][0][0][1], s
->qmat
[0][0][1][1])+8)/16, 512); //FIXME finetune
518 * This function initializes the loop filter boundary limits if the frame's
519 * quality index is different from the previous frame's.
521 static void init_loop_filter(Vp3DecodeContext
*s
)
523 int *bounding_values
= s
->bounding_values_array
+127;
527 filter_limit
= s
->filter_limit_values
[s
->qps
[0]];
529 /* set up the bounding values */
530 memset(s
->bounding_values_array
, 0, 256 * sizeof(int));
531 for (x
= 0; x
< filter_limit
; x
++) {
532 bounding_values
[-x
- filter_limit
] = -filter_limit
+ x
;
533 bounding_values
[-x
] = -x
;
534 bounding_values
[x
] = x
;
535 bounding_values
[x
+ filter_limit
] = filter_limit
- x
;
537 bounding_values
[129] = bounding_values
[130] = filter_limit
* 0x02020202;
541 * This function unpacks all of the superblock/macroblock/fragment coding
542 * information from the bitstream.
544 static int unpack_superblocks(Vp3DecodeContext
*s
, GetBitContext
*gb
)
547 int current_superblock
= 0;
549 int decode_fully_flags
= 0;
550 int decode_partial_blocks
= 0;
551 int first_c_fragment_seen
;
554 int current_fragment
;
557 memset(s
->superblock_coding
, SB_FULLY_CODED
, s
->superblock_count
);
561 /* unpack the list of partially-coded superblocks */
563 /* toggle the bit because as soon as the first run length is
564 * fetched the bit will be toggled again */
566 while (current_superblock
< s
->superblock_count
) {
567 if (current_run
-- == 0) {
569 current_run
= get_vlc2(gb
,
570 s
->superblock_run_length_vlc
.table
, 6, 2);
571 if (current_run
== 33)
572 current_run
+= get_bits(gb
, 12);
574 /* if any of the superblocks are not partially coded, flag
575 * a boolean to decode the list of fully-coded superblocks */
577 decode_fully_flags
= 1;
580 /* make a note of the fact that there are partially coded
582 decode_partial_blocks
= 1;
585 s
->superblock_coding
[current_superblock
++] = bit
;
588 /* unpack the list of fully coded superblocks if any of the blocks were
589 * not marked as partially coded in the previous step */
590 if (decode_fully_flags
) {
592 current_superblock
= 0;
595 /* toggle the bit because as soon as the first run length is
596 * fetched the bit will be toggled again */
598 while (current_superblock
< s
->superblock_count
) {
600 /* skip any superblocks already marked as partially coded */
601 if (s
->superblock_coding
[current_superblock
] == SB_NOT_CODED
) {
603 if (current_run
-- == 0) {
605 current_run
= get_vlc2(gb
,
606 s
->superblock_run_length_vlc
.table
, 6, 2);
607 if (current_run
== 33)
608 current_run
+= get_bits(gb
, 12);
610 s
->superblock_coding
[current_superblock
] = 2*bit
;
612 current_superblock
++;
616 /* if there were partial blocks, initialize bitstream for
617 * unpacking fragment codings */
618 if (decode_partial_blocks
) {
622 /* toggle the bit because as soon as the first run length is
623 * fetched the bit will be toggled again */
628 /* figure out which fragments are coded; iterate through each
629 * superblock (all planes) */
630 s
->coded_fragment_list_index
= 0;
631 s
->next_coeff
= s
->coeffs
+ s
->fragment_count
;
632 s
->first_coded_y_fragment
= s
->first_coded_c_fragment
= 0;
633 s
->last_coded_y_fragment
= s
->last_coded_c_fragment
= -1;
634 first_c_fragment_seen
= 0;
635 memset(s
->macroblock_coding
, MODE_COPY
, s
->macroblock_count
);
636 for (i
= 0; i
< s
->superblock_count
; i
++) {
638 /* iterate through all 16 fragments in a superblock */
639 for (j
= 0; j
< 16; j
++) {
641 /* if the fragment is in bounds, check its coding status */
642 current_fragment
= s
->superblock_fragments
[i
* 16 + j
];
643 if (current_fragment
>= s
->fragment_count
) {
644 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
645 current_fragment
, s
->fragment_count
);
648 if (current_fragment
!= -1) {
649 if (s
->superblock_coding
[i
] == SB_NOT_CODED
) {
651 /* copy all the fragments from the prior frame */
652 s
->all_fragments
[current_fragment
].coding_method
=
655 } else if (s
->superblock_coding
[i
] == SB_PARTIALLY_CODED
) {
657 /* fragment may or may not be coded; this is the case
658 * that cares about the fragment coding runs */
659 if (current_run
-- == 0) {
661 current_run
= get_vlc2(gb
,
662 s
->fragment_run_length_vlc
.table
, 5, 2);
666 /* default mode; actual mode will be decoded in
668 s
->all_fragments
[current_fragment
].coding_method
=
670 s
->all_fragments
[current_fragment
].next_coeff
= s
->coeffs
+ current_fragment
;
671 s
->coded_fragment_list
[s
->coded_fragment_list_index
] =
673 if ((current_fragment
>= s
->fragment_start
[1]) &&
674 (s
->last_coded_y_fragment
== -1) &&
675 (!first_c_fragment_seen
)) {
676 s
->first_coded_c_fragment
= s
->coded_fragment_list_index
;
677 s
->last_coded_y_fragment
= s
->first_coded_c_fragment
- 1;
678 first_c_fragment_seen
= 1;
680 s
->coded_fragment_list_index
++;
681 s
->macroblock_coding
[s
->all_fragments
[current_fragment
].macroblock
] = MODE_INTER_NO_MV
;
683 /* not coded; copy this fragment from the prior frame */
684 s
->all_fragments
[current_fragment
].coding_method
=
690 /* fragments are fully coded in this superblock; actual
691 * coding will be determined in next step */
692 s
->all_fragments
[current_fragment
].coding_method
=
694 s
->all_fragments
[current_fragment
].next_coeff
= s
->coeffs
+ current_fragment
;
695 s
->coded_fragment_list
[s
->coded_fragment_list_index
] =
697 if ((current_fragment
>= s
->fragment_start
[1]) &&
698 (s
->last_coded_y_fragment
== -1) &&
699 (!first_c_fragment_seen
)) {
700 s
->first_coded_c_fragment
= s
->coded_fragment_list_index
;
701 s
->last_coded_y_fragment
= s
->first_coded_c_fragment
- 1;
702 first_c_fragment_seen
= 1;
704 s
->coded_fragment_list_index
++;
705 s
->macroblock_coding
[s
->all_fragments
[current_fragment
].macroblock
] = MODE_INTER_NO_MV
;
711 if (!first_c_fragment_seen
)
712 /* only Y fragments coded in this frame */
713 s
->last_coded_y_fragment
= s
->coded_fragment_list_index
- 1;
715 /* end the list of coded C fragments */
716 s
->last_coded_c_fragment
= s
->coded_fragment_list_index
- 1;
722 * This function unpacks all the coding mode data for individual macroblocks
723 * from the bitstream.
725 static int unpack_modes(Vp3DecodeContext
*s
, GetBitContext
*gb
)
729 int current_macroblock
;
730 int current_fragment
;
732 int custom_mode_alphabet
[CODING_MODE_COUNT
];
735 for (i
= 0; i
< s
->fragment_count
; i
++)
736 s
->all_fragments
[i
].coding_method
= MODE_INTRA
;
740 /* fetch the mode coding scheme for this frame */
741 scheme
= get_bits(gb
, 3);
743 /* is it a custom coding scheme? */
745 for (i
= 0; i
< 8; i
++)
746 custom_mode_alphabet
[i
] = MODE_INTER_NO_MV
;
747 for (i
= 0; i
< 8; i
++)
748 custom_mode_alphabet
[get_bits(gb
, 3)] = i
;
751 /* iterate through all of the macroblocks that contain 1 or more
753 for (i
= 0; i
< s
->u_superblock_start
; i
++) {
755 for (j
= 0; j
< 4; j
++) {
756 current_macroblock
= s
->superblock_macroblocks
[i
* 4 + j
];
757 if ((current_macroblock
== -1) ||
758 (s
->macroblock_coding
[current_macroblock
] == MODE_COPY
))
760 if (current_macroblock
>= s
->macroblock_count
) {
761 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_modes(): bad macroblock number (%d >= %d)\n",
762 current_macroblock
, s
->macroblock_count
);
766 /* mode 7 means get 3 bits for each coding mode */
768 coding_mode
= get_bits(gb
, 3);
770 coding_mode
= custom_mode_alphabet
771 [get_vlc2(gb
, s
->mode_code_vlc
.table
, 3, 3)];
773 coding_mode
= ModeAlphabet
[scheme
-1]
774 [get_vlc2(gb
, s
->mode_code_vlc
.table
, 3, 3)];
776 s
->macroblock_coding
[current_macroblock
] = coding_mode
;
777 for (k
= 0; k
< 6; k
++) {
779 s
->macroblock_fragments
[current_macroblock
* 6 + k
];
780 if (current_fragment
== -1)
782 if (current_fragment
>= s
->fragment_count
) {
783 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_modes(): bad fragment number (%d >= %d)\n",
784 current_fragment
, s
->fragment_count
);
787 if (s
->all_fragments
[current_fragment
].coding_method
!=
789 s
->all_fragments
[current_fragment
].coding_method
=
800 * This function unpacks all the motion vectors for the individual
801 * macroblocks from the bitstream.
803 static int unpack_vectors(Vp3DecodeContext
*s
, GetBitContext
*gb
)
809 int last_motion_x
= 0;
810 int last_motion_y
= 0;
811 int prior_last_motion_x
= 0;
812 int prior_last_motion_y
= 0;
813 int current_macroblock
;
814 int current_fragment
;
819 memset(motion_x
, 0, 6 * sizeof(int));
820 memset(motion_y
, 0, 6 * sizeof(int));
822 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
823 coding_mode
= get_bits1(gb
);
825 /* iterate through all of the macroblocks that contain 1 or more
827 for (i
= 0; i
< s
->u_superblock_start
; i
++) {
829 for (j
= 0; j
< 4; j
++) {
830 current_macroblock
= s
->superblock_macroblocks
[i
* 4 + j
];
831 if ((current_macroblock
== -1) ||
832 (s
->macroblock_coding
[current_macroblock
] == MODE_COPY
))
834 if (current_macroblock
>= s
->macroblock_count
) {
835 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_vectors(): bad macroblock number (%d >= %d)\n",
836 current_macroblock
, s
->macroblock_count
);
840 current_fragment
= s
->macroblock_fragments
[current_macroblock
* 6];
841 if (current_fragment
>= s
->fragment_count
) {
842 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_vectors(): bad fragment number (%d >= %d\n",
843 current_fragment
, s
->fragment_count
);
846 switch (s
->macroblock_coding
[current_macroblock
]) {
848 case MODE_INTER_PLUS_MV
:
850 /* all 6 fragments use the same motion vector */
851 if (coding_mode
== 0) {
852 motion_x
[0] = motion_vector_table
[get_vlc2(gb
, s
->motion_vector_vlc
.table
, 6, 2)];
853 motion_y
[0] = motion_vector_table
[get_vlc2(gb
, s
->motion_vector_vlc
.table
, 6, 2)];
855 motion_x
[0] = fixed_motion_vector_table
[get_bits(gb
, 6)];
856 motion_y
[0] = fixed_motion_vector_table
[get_bits(gb
, 6)];
859 for (k
= 1; k
< 6; k
++) {
860 motion_x
[k
] = motion_x
[0];
861 motion_y
[k
] = motion_y
[0];
864 /* vector maintenance, only on MODE_INTER_PLUS_MV */
865 if (s
->macroblock_coding
[current_macroblock
] ==
866 MODE_INTER_PLUS_MV
) {
867 prior_last_motion_x
= last_motion_x
;
868 prior_last_motion_y
= last_motion_y
;
869 last_motion_x
= motion_x
[0];
870 last_motion_y
= motion_y
[0];
874 case MODE_INTER_FOURMV
:
875 /* vector maintenance */
876 prior_last_motion_x
= last_motion_x
;
877 prior_last_motion_y
= last_motion_y
;
879 /* fetch 4 vectors from the bitstream, one for each
880 * Y fragment, then average for the C fragment vectors */
881 motion_x
[4] = motion_y
[4] = 0;
882 for (k
= 0; k
< 4; k
++) {
883 for (l
= 0; l
< s
->coded_fragment_list_index
; l
++)
884 if (s
->coded_fragment_list
[l
] == s
->macroblock_fragments
[6*current_macroblock
+ k
])
886 if (l
< s
->coded_fragment_list_index
) {
887 if (coding_mode
== 0) {
888 motion_x
[k
] = motion_vector_table
[get_vlc2(gb
, s
->motion_vector_vlc
.table
, 6, 2)];
889 motion_y
[k
] = motion_vector_table
[get_vlc2(gb
, s
->motion_vector_vlc
.table
, 6, 2)];
891 motion_x
[k
] = fixed_motion_vector_table
[get_bits(gb
, 6)];
892 motion_y
[k
] = fixed_motion_vector_table
[get_bits(gb
, 6)];
894 last_motion_x
= motion_x
[k
];
895 last_motion_y
= motion_y
[k
];
900 motion_x
[4] += motion_x
[k
];
901 motion_y
[4] += motion_y
[k
];
905 motion_x
[4]= RSHIFT(motion_x
[4], 2);
907 motion_y
[4]= RSHIFT(motion_y
[4], 2);
910 case MODE_INTER_LAST_MV
:
911 /* all 6 fragments use the last motion vector */
912 motion_x
[0] = last_motion_x
;
913 motion_y
[0] = last_motion_y
;
914 for (k
= 1; k
< 6; k
++) {
915 motion_x
[k
] = motion_x
[0];
916 motion_y
[k
] = motion_y
[0];
919 /* no vector maintenance (last vector remains the
923 case MODE_INTER_PRIOR_LAST
:
924 /* all 6 fragments use the motion vector prior to the
925 * last motion vector */
926 motion_x
[0] = prior_last_motion_x
;
927 motion_y
[0] = prior_last_motion_y
;
928 for (k
= 1; k
< 6; k
++) {
929 motion_x
[k
] = motion_x
[0];
930 motion_y
[k
] = motion_y
[0];
933 /* vector maintenance */
934 prior_last_motion_x
= last_motion_x
;
935 prior_last_motion_y
= last_motion_y
;
936 last_motion_x
= motion_x
[0];
937 last_motion_y
= motion_y
[0];
941 /* covers intra, inter without MV, golden without MV */
942 memset(motion_x
, 0, 6 * sizeof(int));
943 memset(motion_y
, 0, 6 * sizeof(int));
945 /* no vector maintenance */
949 /* assign the motion vectors to the correct fragments */
950 for (k
= 0; k
< 6; k
++) {
952 s
->macroblock_fragments
[current_macroblock
* 6 + k
];
953 if (current_fragment
== -1)
955 if (current_fragment
>= s
->fragment_count
) {
956 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_vectors(): bad fragment number (%d >= %d)\n",
957 current_fragment
, s
->fragment_count
);
960 s
->all_fragments
[current_fragment
].motion_x
= motion_x
[k
];
961 s
->all_fragments
[current_fragment
].motion_y
= motion_y
[k
];
969 static int unpack_block_qpis(Vp3DecodeContext
*s
, GetBitContext
*gb
)
971 int qpi
, i
, j
, bit
, run_length
, blocks_decoded
, num_blocks_at_qpi
;
972 int num_blocks
= s
->coded_fragment_list_index
;
974 for (qpi
= 0; qpi
< s
->nqps
-1 && num_blocks
> 0; qpi
++) {
975 i
= blocks_decoded
= num_blocks_at_qpi
= 0;
980 run_length
= get_vlc2(gb
, s
->superblock_run_length_vlc
.table
, 6, 2) + 1;
981 if (run_length
== 34)
982 run_length
+= get_bits(gb
, 12);
983 blocks_decoded
+= run_length
;
986 num_blocks_at_qpi
+= run_length
;
988 for (j
= 0; j
< run_length
; i
++) {
989 if (i
> s
->coded_fragment_list_index
)
992 if (s
->all_fragments
[s
->coded_fragment_list
[i
]].qpi
== qpi
) {
993 s
->all_fragments
[s
->coded_fragment_list
[i
]].qpi
+= bit
;
998 if (run_length
== 4129)
1002 } while (blocks_decoded
< num_blocks
);
1004 num_blocks
-= num_blocks_at_qpi
;
1011 * This function is called by unpack_dct_coeffs() to extract the VLCs from
1012 * the bitstream. The VLCs encode tokens which are used to unpack DCT
1013 * data. This function unpacks all the VLCs for either the Y plane or both
1014 * C planes, and is called for DC coefficients or different AC coefficient
1015 * levels (since different coefficient types require different VLC tables.
1017 * This function returns a residual eob run. E.g, if a particular token gave
1018 * instructions to EOB the next 5 fragments and there were only 2 fragments
1019 * left in the current fragment range, 3 would be returned so that it could
1020 * be passed into the next call to this same function.
1022 static int unpack_vlcs(Vp3DecodeContext
*s
, GetBitContext
*gb
,
1023 VLC
*table
, int coeff_index
,
1024 int first_fragment
, int last_fragment
,
1031 Vp3Fragment
*fragment
;
1032 uint8_t *perm
= s
->scantable
.permutated
;
1035 if ((first_fragment
>= s
->fragment_count
) ||
1036 (last_fragment
>= s
->fragment_count
)) {
1038 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_vlcs(): bad fragment number (%d -> %d ?)\n",
1039 first_fragment
, last_fragment
);
1043 for (i
= first_fragment
; i
<= last_fragment
; i
++) {
1044 int fragment_num
= s
->coded_fragment_list
[i
];
1046 if (s
->coeff_counts
[fragment_num
] > coeff_index
)
1048 fragment
= &s
->all_fragments
[fragment_num
];
1051 /* decode a VLC into a token */
1052 token
= get_vlc2(gb
, table
->table
, 5, 3);
1053 /* use the token to get a zero run, a coefficient, and an eob run */
1055 eob_run
= eob_run_base
[token
];
1056 if (eob_run_get_bits
[token
])
1057 eob_run
+= get_bits(gb
, eob_run_get_bits
[token
]);
1058 coeff
= zero_run
= 0;
1060 bits_to_get
= coeff_get_bits
[token
];
1062 coeff
= coeff_tables
[token
][0];
1064 coeff
= coeff_tables
[token
][get_bits(gb
, bits_to_get
)];
1066 zero_run
= zero_run_base
[token
];
1067 if (zero_run_get_bits
[token
])
1068 zero_run
+= get_bits(gb
, zero_run_get_bits
[token
]);
1073 s
->coeff_counts
[fragment_num
] += zero_run
;
1074 if (s
->coeff_counts
[fragment_num
] < 64){
1075 fragment
->next_coeff
->coeff
= coeff
;
1076 fragment
->next_coeff
->index
= perm
[s
->coeff_counts
[fragment_num
]++]; //FIXME perm here already?
1077 fragment
->next_coeff
->next
= s
->next_coeff
;
1078 s
->next_coeff
->next
=NULL
;
1079 fragment
->next_coeff
= s
->next_coeff
++;
1082 s
->coeff_counts
[fragment_num
] |= 128;
1091 * This function unpacks all of the DCT coefficient data from the
1094 static int unpack_dct_coeffs(Vp3DecodeContext
*s
, GetBitContext
*gb
)
1101 int residual_eob_run
= 0;
1103 /* fetch the DC table indexes */
1104 dc_y_table
= get_bits(gb
, 4);
1105 dc_c_table
= get_bits(gb
, 4);
1107 /* unpack the Y plane DC coefficients */
1108 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->dc_vlc
[dc_y_table
], 0,
1109 s
->first_coded_y_fragment
, s
->last_coded_y_fragment
, residual_eob_run
);
1111 /* unpack the C plane DC coefficients */
1112 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->dc_vlc
[dc_c_table
], 0,
1113 s
->first_coded_c_fragment
, s
->last_coded_c_fragment
, residual_eob_run
);
1115 /* fetch the AC table indexes */
1116 ac_y_table
= get_bits(gb
, 4);
1117 ac_c_table
= get_bits(gb
, 4);
1119 /* unpack the group 1 AC coefficients (coeffs 1-5) */
1120 for (i
= 1; i
<= 5; i
++) {
1121 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_1
[ac_y_table
], i
,
1122 s
->first_coded_y_fragment
, s
->last_coded_y_fragment
, residual_eob_run
);
1124 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_1
[ac_c_table
], i
,
1125 s
->first_coded_c_fragment
, s
->last_coded_c_fragment
, residual_eob_run
);
1128 /* unpack the group 2 AC coefficients (coeffs 6-14) */
1129 for (i
= 6; i
<= 14; i
++) {
1130 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_2
[ac_y_table
], i
,
1131 s
->first_coded_y_fragment
, s
->last_coded_y_fragment
, residual_eob_run
);
1133 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_2
[ac_c_table
], i
,
1134 s
->first_coded_c_fragment
, s
->last_coded_c_fragment
, residual_eob_run
);
1137 /* unpack the group 3 AC coefficients (coeffs 15-27) */
1138 for (i
= 15; i
<= 27; i
++) {
1139 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_3
[ac_y_table
], i
,
1140 s
->first_coded_y_fragment
, s
->last_coded_y_fragment
, residual_eob_run
);
1142 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_3
[ac_c_table
], i
,
1143 s
->first_coded_c_fragment
, s
->last_coded_c_fragment
, residual_eob_run
);
1146 /* unpack the group 4 AC coefficients (coeffs 28-63) */
1147 for (i
= 28; i
<= 63; i
++) {
1148 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_4
[ac_y_table
], i
,
1149 s
->first_coded_y_fragment
, s
->last_coded_y_fragment
, residual_eob_run
);
1151 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_4
[ac_c_table
], i
,
1152 s
->first_coded_c_fragment
, s
->last_coded_c_fragment
, residual_eob_run
);
1159 * This function reverses the DC prediction for each coded fragment in
1160 * the frame. Much of this function is adapted directly from the original
1163 #define COMPATIBLE_FRAME(x) \
1164 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1165 #define FRAME_CODED(x) (s->all_fragments[x].coding_method != MODE_COPY)
1166 #define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
1168 static void reverse_dc_prediction(Vp3DecodeContext
*s
,
1171 int fragment_height
)
1180 int i
= first_fragment
;
1184 /* DC values for the left, up-left, up, and up-right fragments */
1185 int vl
, vul
, vu
, vur
;
1187 /* indexes for the left, up-left, up, and up-right fragments */
1191 * The 6 fields mean:
1192 * 0: up-left multiplier
1194 * 2: up-right multiplier
1195 * 3: left multiplier
1197 int predictor_transform
[16][4] = {
1199 { 0, 0, 0,128}, // PL
1200 { 0, 0,128, 0}, // PUR
1201 { 0, 0, 53, 75}, // PUR|PL
1202 { 0,128, 0, 0}, // PU
1203 { 0, 64, 0, 64}, // PU|PL
1204 { 0,128, 0, 0}, // PU|PUR
1205 { 0, 0, 53, 75}, // PU|PUR|PL
1206 {128, 0, 0, 0}, // PUL
1207 { 0, 0, 0,128}, // PUL|PL
1208 { 64, 0, 64, 0}, // PUL|PUR
1209 { 0, 0, 53, 75}, // PUL|PUR|PL
1210 { 0,128, 0, 0}, // PUL|PU
1211 {-104,116, 0,116}, // PUL|PU|PL
1212 { 24, 80, 24, 0}, // PUL|PU|PUR
1213 {-104,116, 0,116} // PUL|PU|PUR|PL
1216 /* This table shows which types of blocks can use other blocks for
1217 * prediction. For example, INTRA is the only mode in this table to
1218 * have a frame number of 0. That means INTRA blocks can only predict
1219 * from other INTRA blocks. There are 2 golden frame coding types;
1220 * blocks encoding in these modes can only predict from other blocks
1221 * that were encoded with these 1 of these 2 modes. */
1222 unsigned char compatible_frame
[8] = {
1223 1, /* MODE_INTER_NO_MV */
1225 1, /* MODE_INTER_PLUS_MV */
1226 1, /* MODE_INTER_LAST_MV */
1227 1, /* MODE_INTER_PRIOR_MV */
1228 2, /* MODE_USING_GOLDEN */
1229 2, /* MODE_GOLDEN_MV */
1230 1 /* MODE_INTER_FOUR_MV */
1232 int current_frame_type
;
1234 /* there is a last DC predictor for each of the 3 frame types */
1239 vul
= vu
= vur
= vl
= 0;
1240 last_dc
[0] = last_dc
[1] = last_dc
[2] = 0;
1242 /* for each fragment row... */
1243 for (y
= 0; y
< fragment_height
; y
++) {
1245 /* for each fragment in a row... */
1246 for (x
= 0; x
< fragment_width
; x
++, i
++) {
1248 /* reverse prediction if this block was coded */
1249 if (s
->all_fragments
[i
].coding_method
!= MODE_COPY
) {
1251 current_frame_type
=
1252 compatible_frame
[s
->all_fragments
[i
].coding_method
];
1258 if(FRAME_CODED(l
) && COMPATIBLE_FRAME(l
))
1262 u
= i
-fragment_width
;
1264 if(FRAME_CODED(u
) && COMPATIBLE_FRAME(u
))
1267 ul
= i
-fragment_width
-1;
1269 if(FRAME_CODED(ul
) && COMPATIBLE_FRAME(ul
))
1272 if(x
+ 1 < fragment_width
){
1273 ur
= i
-fragment_width
+1;
1275 if(FRAME_CODED(ur
) && COMPATIBLE_FRAME(ur
))
1280 if (transform
== 0) {
1282 /* if there were no fragments to predict from, use last
1284 predicted_dc
= last_dc
[current_frame_type
];
1287 /* apply the appropriate predictor transform */
1289 (predictor_transform
[transform
][0] * vul
) +
1290 (predictor_transform
[transform
][1] * vu
) +
1291 (predictor_transform
[transform
][2] * vur
) +
1292 (predictor_transform
[transform
][3] * vl
);
1294 predicted_dc
/= 128;
1296 /* check for outranging on the [ul u l] and
1297 * [ul u ur l] predictors */
1298 if ((transform
== 13) || (transform
== 15)) {
1299 if (FFABS(predicted_dc
- vu
) > 128)
1301 else if (FFABS(predicted_dc
- vl
) > 128)
1303 else if (FFABS(predicted_dc
- vul
) > 128)
1308 /* at long last, apply the predictor */
1309 if(s
->coeffs
[i
].index
){
1310 *s
->next_coeff
= s
->coeffs
[i
];
1311 s
->coeffs
[i
].index
=0;
1312 s
->coeffs
[i
].coeff
=0;
1313 s
->coeffs
[i
].next
= s
->next_coeff
++;
1315 s
->coeffs
[i
].coeff
+= predicted_dc
;
1317 last_dc
[current_frame_type
] = DC_COEFF(i
);
1318 if(DC_COEFF(i
) && !(s
->coeff_counts
[i
]&127)){
1319 s
->coeff_counts
[i
]= 129;
1320 // s->all_fragments[i].next_coeff= s->next_coeff;
1321 s
->coeffs
[i
].next
= s
->next_coeff
;
1322 (s
->next_coeff
++)->next
=NULL
;
1330 * Perform the final rendering for a particular slice of data.
1331 * The slice number ranges from 0..(macroblock_height - 1).
1333 static void render_slice(Vp3DecodeContext
*s
, int slice
)
1336 int16_t *dequantizer
;
1337 DECLARE_ALIGNED_16(DCTELEM
, block
[64]);
1338 int motion_x
= 0xdeadbeef, motion_y
= 0xdeadbeef;
1339 int motion_halfpel_index
;
1340 uint8_t *motion_source
;
1342 int current_macroblock_entry
= slice
* s
->macroblock_width
* 6;
1344 if (slice
>= s
->macroblock_height
)
1347 for (plane
= 0; plane
< 3; plane
++) {
1348 uint8_t *output_plane
= s
->current_frame
.data
[plane
];
1349 uint8_t * last_plane
= s
-> last_frame
.data
[plane
];
1350 uint8_t *golden_plane
= s
-> golden_frame
.data
[plane
];
1351 int stride
= s
->current_frame
.linesize
[plane
];
1352 int plane_width
= s
->width
>> !!plane
;
1353 int plane_height
= s
->height
>> !!plane
;
1354 int y
= slice
* FRAGMENT_PIXELS
<< !plane
;
1355 int slice_height
= y
+ (FRAGMENT_PIXELS
<< !plane
);
1356 int i
= s
->macroblock_fragments
[current_macroblock_entry
+ plane
+ 3*!!plane
];
1358 if (!s
->flipped_image
) stride
= -stride
;
1361 if(FFABS(stride
) > 2048)
1362 return; //various tables are fixed size
1364 /* for each fragment row in the slice (both of them)... */
1365 for (; y
< slice_height
; y
+= 8) {
1367 /* for each fragment in a row... */
1368 for (x
= 0; x
< plane_width
; x
+= 8, i
++) {
1370 if ((i
< 0) || (i
>= s
->fragment_count
)) {
1371 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:render_slice(): bad fragment number (%d)\n", i
);
1375 /* transform if this block was coded */
1376 if ((s
->all_fragments
[i
].coding_method
!= MODE_COPY
) &&
1377 !((s
->avctx
->flags
& CODEC_FLAG_GRAY
) && plane
)) {
1379 if ((s
->all_fragments
[i
].coding_method
== MODE_USING_GOLDEN
) ||
1380 (s
->all_fragments
[i
].coding_method
== MODE_GOLDEN_MV
))
1381 motion_source
= golden_plane
;
1383 motion_source
= last_plane
;
1385 motion_source
+= s
->all_fragments
[i
].first_pixel
;
1386 motion_halfpel_index
= 0;
1388 /* sort out the motion vector if this fragment is coded
1389 * using a motion vector method */
1390 if ((s
->all_fragments
[i
].coding_method
> MODE_INTRA
) &&
1391 (s
->all_fragments
[i
].coding_method
!= MODE_USING_GOLDEN
)) {
1393 motion_x
= s
->all_fragments
[i
].motion_x
;
1394 motion_y
= s
->all_fragments
[i
].motion_y
;
1396 motion_x
= (motion_x
>>1) | (motion_x
&1);
1397 motion_y
= (motion_y
>>1) | (motion_y
&1);
1400 src_x
= (motion_x
>>1) + x
;
1401 src_y
= (motion_y
>>1) + y
;
1402 if ((motion_x
== 127) || (motion_y
== 127))
1403 av_log(s
->avctx
, AV_LOG_ERROR
, " help! got invalid motion vector! (%X, %X)\n", motion_x
, motion_y
);
1405 motion_halfpel_index
= motion_x
& 0x01;
1406 motion_source
+= (motion_x
>> 1);
1408 motion_halfpel_index
|= (motion_y
& 0x01) << 1;
1409 motion_source
+= ((motion_y
>> 1) * stride
);
1411 if(src_x
<0 || src_y
<0 || src_x
+ 9 >= plane_width
|| src_y
+ 9 >= plane_height
){
1412 uint8_t *temp
= s
->edge_emu_buffer
;
1413 if(stride
<0) temp
-= 9*stride
;
1414 else temp
+= 9*stride
;
1416 ff_emulated_edge_mc(temp
, motion_source
, stride
, 9, 9, src_x
, src_y
, plane_width
, plane_height
);
1417 motion_source
= temp
;
1422 /* first, take care of copying a block from either the
1423 * previous or the golden frame */
1424 if (s
->all_fragments
[i
].coding_method
!= MODE_INTRA
) {
1425 /* Note, it is possible to implement all MC cases with
1426 put_no_rnd_pixels_l2 which would look more like the
1427 VP3 source but this would be slower as
1428 put_no_rnd_pixels_tab is better optimzed */
1429 if(motion_halfpel_index
!= 3){
1430 s
->dsp
.put_no_rnd_pixels_tab
[1][motion_halfpel_index
](
1431 output_plane
+ s
->all_fragments
[i
].first_pixel
,
1432 motion_source
, stride
, 8);
1434 int d
= (motion_x
^ motion_y
)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1435 s
->dsp
.put_no_rnd_pixels_l2
[1](
1436 output_plane
+ s
->all_fragments
[i
].first_pixel
,
1438 motion_source
+ stride
+ 1 + d
,
1441 dequantizer
= s
->qmat
[s
->all_fragments
[i
].qpi
][1][plane
];
1443 dequantizer
= s
->qmat
[s
->all_fragments
[i
].qpi
][0][plane
];
1446 /* dequantize the DCT coefficients */
1447 if(s
->avctx
->idct_algo
==FF_IDCT_VP3
){
1448 Coeff
*coeff
= s
->coeffs
+ i
;
1449 s
->dsp
.clear_block(block
);
1451 block
[coeff
->index
]= coeff
->coeff
* dequantizer
[coeff
->index
];
1455 Coeff
*coeff
= s
->coeffs
+ i
;
1456 s
->dsp
.clear_block(block
);
1458 block
[coeff
->index
]= (coeff
->coeff
* dequantizer
[coeff
->index
] + 2)>>2;
1463 /* invert DCT and place (or add) in final output */
1465 if (s
->all_fragments
[i
].coding_method
== MODE_INTRA
) {
1466 if(s
->avctx
->idct_algo
!=FF_IDCT_VP3
)
1469 output_plane
+ s
->all_fragments
[i
].first_pixel
,
1474 output_plane
+ s
->all_fragments
[i
].first_pixel
,
1480 /* copy directly from the previous frame */
1481 s
->dsp
.put_pixels_tab
[1][0](
1482 output_plane
+ s
->all_fragments
[i
].first_pixel
,
1483 last_plane
+ s
->all_fragments
[i
].first_pixel
,
1488 /* perform the left edge filter if:
1489 * - the fragment is not on the left column
1490 * - the fragment is coded in this frame
1491 * - the fragment is not coded in this frame but the left
1492 * fragment is coded in this frame (this is done instead
1493 * of a right edge filter when rendering the left fragment
1494 * since this fragment is not available yet) */
1496 ((s
->all_fragments
[i
].coding_method
!= MODE_COPY
) ||
1497 ((s
->all_fragments
[i
].coding_method
== MODE_COPY
) &&
1498 (s
->all_fragments
[i
- 1].coding_method
!= MODE_COPY
)) )) {
1500 output_plane
+ s
->all_fragments
[i
].first_pixel
+ 7*stride
,
1501 -stride
, s
->bounding_values_array
+ 127);
1504 /* perform the top edge filter if:
1505 * - the fragment is not on the top row
1506 * - the fragment is coded in this frame
1507 * - the fragment is not coded in this frame but the above
1508 * fragment is coded in this frame (this is done instead
1509 * of a bottom edge filter when rendering the above
1510 * fragment since this fragment is not available yet) */
1512 ((s
->all_fragments
[i
].coding_method
!= MODE_COPY
) ||
1513 ((s
->all_fragments
[i
].coding_method
== MODE_COPY
) &&
1514 (s
->all_fragments
[i
- fragment_width
].coding_method
!= MODE_COPY
)) )) {
1516 output_plane
+ s
->all_fragments
[i
].first_pixel
- stride
,
1517 -stride
, s
->bounding_values_array
+ 127);
1524 /* this looks like a good place for slice dispatch... */
1526 * if (slice == s->macroblock_height - 1)
1527 * dispatch (both last slice & 2nd-to-last slice);
1528 * else if (slice > 0)
1529 * dispatch (slice - 1);
1535 static void apply_loop_filter(Vp3DecodeContext
*s
)
1539 int *bounding_values
= s
->bounding_values_array
+127;
1542 int bounding_values_array
[256];
1545 /* find the right loop limit value */
1546 for (x
= 63; x
>= 0; x
--) {
1547 if (vp31_ac_scale_factor
[x
] >= s
->quality_index
)
1550 filter_limit
= vp31_filter_limit_values
[s
->quality_index
];
1552 /* set up the bounding values */
1553 memset(bounding_values_array
, 0, 256 * sizeof(int));
1554 for (x
= 0; x
< filter_limit
; x
++) {
1555 bounding_values
[-x
- filter_limit
] = -filter_limit
+ x
;
1556 bounding_values
[-x
] = -x
;
1557 bounding_values
[x
] = x
;
1558 bounding_values
[x
+ filter_limit
] = filter_limit
- x
;
1562 for (plane
= 0; plane
< 3; plane
++) {
1563 int width
= s
->fragment_width
>> !!plane
;
1564 int height
= s
->fragment_height
>> !!plane
;
1565 int fragment
= s
->fragment_start
[plane
];
1566 int stride
= s
->current_frame
.linesize
[plane
];
1567 uint8_t *plane_data
= s
->current_frame
.data
[plane
];
1568 if (!s
->flipped_image
) stride
= -stride
;
1570 for (y
= 0; y
< height
; y
++) {
1572 for (x
= 0; x
< width
; x
++) {
1573 /* do not perform left edge filter for left columns frags */
1575 (s
->all_fragments
[fragment
].coding_method
!= MODE_COPY
)) {
1576 s
->dsp
.vp3_h_loop_filter(
1577 plane_data
+ s
->all_fragments
[fragment
].first_pixel
,
1578 stride
, bounding_values
);
1581 /* do not perform top edge filter for top row fragments */
1583 (s
->all_fragments
[fragment
].coding_method
!= MODE_COPY
)) {
1584 s
->dsp
.vp3_v_loop_filter(
1585 plane_data
+ s
->all_fragments
[fragment
].first_pixel
,
1586 stride
, bounding_values
);
1589 /* do not perform right edge filter for right column
1590 * fragments or if right fragment neighbor is also coded
1591 * in this frame (it will be filtered in next iteration) */
1592 if ((x
< width
- 1) &&
1593 (s
->all_fragments
[fragment
].coding_method
!= MODE_COPY
) &&
1594 (s
->all_fragments
[fragment
+ 1].coding_method
== MODE_COPY
)) {
1595 s
->dsp
.vp3_h_loop_filter(
1596 plane_data
+ s
->all_fragments
[fragment
+ 1].first_pixel
,
1597 stride
, bounding_values
);
1600 /* do not perform bottom edge filter for bottom row
1601 * fragments or if bottom fragment neighbor is also coded
1602 * in this frame (it will be filtered in the next row) */
1603 if ((y
< height
- 1) &&
1604 (s
->all_fragments
[fragment
].coding_method
!= MODE_COPY
) &&
1605 (s
->all_fragments
[fragment
+ width
].coding_method
== MODE_COPY
)) {
1606 s
->dsp
.vp3_v_loop_filter(
1607 plane_data
+ s
->all_fragments
[fragment
+ width
].first_pixel
,
1608 stride
, bounding_values
);
1618 * This function computes the first pixel addresses for each fragment.
1619 * This function needs to be invoked after the first frame is allocated
1620 * so that it has access to the plane strides.
1622 static void vp3_calculate_pixel_addresses(Vp3DecodeContext
*s
)
1624 #define Y_INITIAL(chroma_shift) s->flipped_image ? 1 : s->fragment_height >> chroma_shift
1625 #define Y_FINISHED(chroma_shift) s->flipped_image ? y <= s->fragment_height >> chroma_shift : y > 0
1628 const int y_inc
= s
->flipped_image
? 1 : -1;
1630 /* figure out the first pixel addresses for each of the fragments */
1633 for (y
= Y_INITIAL(0); Y_FINISHED(0); y
+= y_inc
) {
1634 for (x
= 0; x
< s
->fragment_width
; x
++) {
1635 s
->all_fragments
[i
++].first_pixel
=
1636 s
->golden_frame
.linesize
[0] * y
* FRAGMENT_PIXELS
-
1637 s
->golden_frame
.linesize
[0] +
1638 x
* FRAGMENT_PIXELS
;
1643 i
= s
->fragment_start
[1];
1644 for (y
= Y_INITIAL(1); Y_FINISHED(1); y
+= y_inc
) {
1645 for (x
= 0; x
< s
->fragment_width
/ 2; x
++) {
1646 s
->all_fragments
[i
++].first_pixel
=
1647 s
->golden_frame
.linesize
[1] * y
* FRAGMENT_PIXELS
-
1648 s
->golden_frame
.linesize
[1] +
1649 x
* FRAGMENT_PIXELS
;
1654 i
= s
->fragment_start
[2];
1655 for (y
= Y_INITIAL(1); Y_FINISHED(1); y
+= y_inc
) {
1656 for (x
= 0; x
< s
->fragment_width
/ 2; x
++) {
1657 s
->all_fragments
[i
++].first_pixel
=
1658 s
->golden_frame
.linesize
[2] * y
* FRAGMENT_PIXELS
-
1659 s
->golden_frame
.linesize
[2] +
1660 x
* FRAGMENT_PIXELS
;
1666 * This is the ffmpeg/libavcodec API init function.
1668 static av_cold
int vp3_decode_init(AVCodecContext
*avctx
)
1670 Vp3DecodeContext
*s
= avctx
->priv_data
;
1671 int i
, inter
, plane
;
1674 int y_superblock_count
;
1675 int c_superblock_count
;
1677 if (avctx
->codec_tag
== MKTAG('V','P','3','0'))
1683 s
->width
= FFALIGN(avctx
->width
, 16);
1684 s
->height
= FFALIGN(avctx
->height
, 16);
1685 avctx
->pix_fmt
= PIX_FMT_YUV420P
;
1686 avctx
->chroma_sample_location
= AVCHROMA_LOC_CENTER
;
1687 if(avctx
->idct_algo
==FF_IDCT_AUTO
)
1688 avctx
->idct_algo
=FF_IDCT_VP3
;
1689 dsputil_init(&s
->dsp
, avctx
);
1691 ff_init_scantable(s
->dsp
.idct_permutation
, &s
->scantable
, ff_zigzag_direct
);
1693 /* initialize to an impossible value which will force a recalculation
1694 * in the first frame decode */
1695 for (i
= 0; i
< 3; i
++)
1698 s
->y_superblock_width
= (s
->width
+ 31) / 32;
1699 s
->y_superblock_height
= (s
->height
+ 31) / 32;
1700 y_superblock_count
= s
->y_superblock_width
* s
->y_superblock_height
;
1702 /* work out the dimensions for the C planes */
1703 c_width
= s
->width
/ 2;
1704 c_height
= s
->height
/ 2;
1705 s
->c_superblock_width
= (c_width
+ 31) / 32;
1706 s
->c_superblock_height
= (c_height
+ 31) / 32;
1707 c_superblock_count
= s
->c_superblock_width
* s
->c_superblock_height
;
1709 s
->superblock_count
= y_superblock_count
+ (c_superblock_count
* 2);
1710 s
->u_superblock_start
= y_superblock_count
;
1711 s
->v_superblock_start
= s
->u_superblock_start
+ c_superblock_count
;
1712 s
->superblock_coding
= av_malloc(s
->superblock_count
);
1714 s
->macroblock_width
= (s
->width
+ 15) / 16;
1715 s
->macroblock_height
= (s
->height
+ 15) / 16;
1716 s
->macroblock_count
= s
->macroblock_width
* s
->macroblock_height
;
1718 s
->fragment_width
= s
->width
/ FRAGMENT_PIXELS
;
1719 s
->fragment_height
= s
->height
/ FRAGMENT_PIXELS
;
1721 /* fragment count covers all 8x8 blocks for all 3 planes */
1722 s
->fragment_count
= s
->fragment_width
* s
->fragment_height
* 3 / 2;
1723 s
->fragment_start
[1] = s
->fragment_width
* s
->fragment_height
;
1724 s
->fragment_start
[2] = s
->fragment_width
* s
->fragment_height
* 5 / 4;
1726 s
->all_fragments
= av_malloc(s
->fragment_count
* sizeof(Vp3Fragment
));
1727 s
->coeff_counts
= av_malloc(s
->fragment_count
* sizeof(*s
->coeff_counts
));
1728 s
->coeffs
= av_malloc(s
->fragment_count
* sizeof(Coeff
) * 65);
1729 s
->coded_fragment_list
= av_malloc(s
->fragment_count
* sizeof(int));
1730 s
->pixel_addresses_initialized
= 0;
1732 if (!s
->theora_tables
)
1734 for (i
= 0; i
< 64; i
++) {
1735 s
->coded_dc_scale_factor
[i
] = vp31_dc_scale_factor
[i
];
1736 s
->coded_ac_scale_factor
[i
] = vp31_ac_scale_factor
[i
];
1737 s
->base_matrix
[0][i
] = vp31_intra_y_dequant
[i
];
1738 s
->base_matrix
[1][i
] = vp31_intra_c_dequant
[i
];
1739 s
->base_matrix
[2][i
] = vp31_inter_dequant
[i
];
1740 s
->filter_limit_values
[i
] = vp31_filter_limit_values
[i
];
1743 for(inter
=0; inter
<2; inter
++){
1744 for(plane
=0; plane
<3; plane
++){
1745 s
->qr_count
[inter
][plane
]= 1;
1746 s
->qr_size
[inter
][plane
][0]= 63;
1747 s
->qr_base
[inter
][plane
][0]=
1748 s
->qr_base
[inter
][plane
][1]= 2*inter
+ (!!plane
)*!inter
;
1752 /* init VLC tables */
1753 for (i
= 0; i
< 16; i
++) {
1756 init_vlc(&s
->dc_vlc
[i
], 5, 32,
1757 &dc_bias
[i
][0][1], 4, 2,
1758 &dc_bias
[i
][0][0], 4, 2, 0);
1760 /* group 1 AC histograms */
1761 init_vlc(&s
->ac_vlc_1
[i
], 5, 32,
1762 &ac_bias_0
[i
][0][1], 4, 2,
1763 &ac_bias_0
[i
][0][0], 4, 2, 0);
1765 /* group 2 AC histograms */
1766 init_vlc(&s
->ac_vlc_2
[i
], 5, 32,
1767 &ac_bias_1
[i
][0][1], 4, 2,
1768 &ac_bias_1
[i
][0][0], 4, 2, 0);
1770 /* group 3 AC histograms */
1771 init_vlc(&s
->ac_vlc_3
[i
], 5, 32,
1772 &ac_bias_2
[i
][0][1], 4, 2,
1773 &ac_bias_2
[i
][0][0], 4, 2, 0);
1775 /* group 4 AC histograms */
1776 init_vlc(&s
->ac_vlc_4
[i
], 5, 32,
1777 &ac_bias_3
[i
][0][1], 4, 2,
1778 &ac_bias_3
[i
][0][0], 4, 2, 0);
1781 for (i
= 0; i
< 16; i
++) {
1784 init_vlc(&s
->dc_vlc
[i
], 5, 32,
1785 &s
->huffman_table
[i
][0][1], 4, 2,
1786 &s
->huffman_table
[i
][0][0], 4, 2, 0);
1788 /* group 1 AC histograms */
1789 init_vlc(&s
->ac_vlc_1
[i
], 5, 32,
1790 &s
->huffman_table
[i
+16][0][1], 4, 2,
1791 &s
->huffman_table
[i
+16][0][0], 4, 2, 0);
1793 /* group 2 AC histograms */
1794 init_vlc(&s
->ac_vlc_2
[i
], 5, 32,
1795 &s
->huffman_table
[i
+16*2][0][1], 4, 2,
1796 &s
->huffman_table
[i
+16*2][0][0], 4, 2, 0);
1798 /* group 3 AC histograms */
1799 init_vlc(&s
->ac_vlc_3
[i
], 5, 32,
1800 &s
->huffman_table
[i
+16*3][0][1], 4, 2,
1801 &s
->huffman_table
[i
+16*3][0][0], 4, 2, 0);
1803 /* group 4 AC histograms */
1804 init_vlc(&s
->ac_vlc_4
[i
], 5, 32,
1805 &s
->huffman_table
[i
+16*4][0][1], 4, 2,
1806 &s
->huffman_table
[i
+16*4][0][0], 4, 2, 0);
1810 init_vlc(&s
->superblock_run_length_vlc
, 6, 34,
1811 &superblock_run_length_vlc_table
[0][1], 4, 2,
1812 &superblock_run_length_vlc_table
[0][0], 4, 2, 0);
1814 init_vlc(&s
->fragment_run_length_vlc
, 5, 30,
1815 &fragment_run_length_vlc_table
[0][1], 4, 2,
1816 &fragment_run_length_vlc_table
[0][0], 4, 2, 0);
1818 init_vlc(&s
->mode_code_vlc
, 3, 8,
1819 &mode_code_vlc_table
[0][1], 2, 1,
1820 &mode_code_vlc_table
[0][0], 2, 1, 0);
1822 init_vlc(&s
->motion_vector_vlc
, 6, 63,
1823 &motion_vector_vlc_table
[0][1], 2, 1,
1824 &motion_vector_vlc_table
[0][0], 2, 1, 0);
1826 /* work out the block mapping tables */
1827 s
->superblock_fragments
= av_malloc(s
->superblock_count
* 16 * sizeof(int));
1828 s
->superblock_macroblocks
= av_malloc(s
->superblock_count
* 4 * sizeof(int));
1829 s
->macroblock_fragments
= av_malloc(s
->macroblock_count
* 6 * sizeof(int));
1830 s
->macroblock_coding
= av_malloc(s
->macroblock_count
+ 1);
1831 init_block_mapping(s
);
1833 for (i
= 0; i
< 3; i
++) {
1834 s
->current_frame
.data
[i
] = NULL
;
1835 s
->last_frame
.data
[i
] = NULL
;
1836 s
->golden_frame
.data
[i
] = NULL
;
1843 * This is the ffmpeg/libavcodec API frame decode function.
1845 static int vp3_decode_frame(AVCodecContext
*avctx
,
1846 void *data
, int *data_size
,
1849 const uint8_t *buf
= avpkt
->data
;
1850 int buf_size
= avpkt
->size
;
1851 Vp3DecodeContext
*s
= avctx
->priv_data
;
1853 static int counter
= 0;
1856 init_get_bits(&gb
, buf
, buf_size
* 8);
1858 if (s
->theora
&& get_bits1(&gb
))
1860 av_log(avctx
, AV_LOG_ERROR
, "Header packet passed to frame decoder, skipping\n");
1864 s
->keyframe
= !get_bits1(&gb
);
1867 for (i
= 0; i
< 3; i
++)
1868 s
->last_qps
[i
] = s
->qps
[i
];
1872 s
->qps
[s
->nqps
++]= get_bits(&gb
, 6);
1873 } while(s
->theora
>= 0x030200 && s
->nqps
<3 && get_bits1(&gb
));
1874 for (i
= s
->nqps
; i
< 3; i
++)
1877 if (s
->avctx
->debug
& FF_DEBUG_PICT_INFO
)
1878 av_log(s
->avctx
, AV_LOG_INFO
, " VP3 %sframe #%d: Q index = %d\n",
1879 s
->keyframe
?"key":"", counter
, s
->qps
[0]);
1882 if (s
->qps
[0] != s
->last_qps
[0])
1883 init_loop_filter(s
);
1885 for (i
= 0; i
< s
->nqps
; i
++)
1886 // reinit all dequantizers if the first one changed, because
1887 // the DC of the first quantizer must be used for all matrices
1888 if (s
->qps
[i
] != s
->last_qps
[i
] || s
->qps
[0] != s
->last_qps
[0])
1889 init_dequantizer(s
, i
);
1891 if (avctx
->skip_frame
>= AVDISCARD_NONKEY
&& !s
->keyframe
)
1897 skip_bits(&gb
, 4); /* width code */
1898 skip_bits(&gb
, 4); /* height code */
1901 s
->version
= get_bits(&gb
, 5);
1903 av_log(s
->avctx
, AV_LOG_DEBUG
, "VP version: %d\n", s
->version
);
1906 if (s
->version
|| s
->theora
)
1909 av_log(s
->avctx
, AV_LOG_ERROR
, "Warning, unsupported keyframe coding type?!\n");
1910 skip_bits(&gb
, 2); /* reserved? */
1913 if (s
->last_frame
.data
[0] == s
->golden_frame
.data
[0]) {
1914 if (s
->golden_frame
.data
[0])
1915 avctx
->release_buffer(avctx
, &s
->golden_frame
);
1916 s
->last_frame
= s
->golden_frame
; /* ensure that we catch any access to this released frame */
1918 if (s
->golden_frame
.data
[0])
1919 avctx
->release_buffer(avctx
, &s
->golden_frame
);
1920 if (s
->last_frame
.data
[0])
1921 avctx
->release_buffer(avctx
, &s
->last_frame
);
1924 s
->golden_frame
.reference
= 3;
1925 if(avctx
->get_buffer(avctx
, &s
->golden_frame
) < 0) {
1926 av_log(s
->avctx
, AV_LOG_ERROR
, "vp3: get_buffer() failed\n");
1930 /* golden frame is also the current frame */
1931 s
->current_frame
= s
->golden_frame
;
1933 /* time to figure out pixel addresses? */
1934 if (!s
->pixel_addresses_initialized
)
1936 vp3_calculate_pixel_addresses(s
);
1937 s
->pixel_addresses_initialized
= 1;
1940 /* allocate a new current frame */
1941 s
->current_frame
.reference
= 3;
1942 if (!s
->pixel_addresses_initialized
) {
1943 av_log(s
->avctx
, AV_LOG_ERROR
, "vp3: first frame not a keyframe\n");
1946 if(avctx
->get_buffer(avctx
, &s
->current_frame
) < 0) {
1947 av_log(s
->avctx
, AV_LOG_ERROR
, "vp3: get_buffer() failed\n");
1952 s
->current_frame
.qscale_table
= s
->qscale_table
; //FIXME allocate individual tables per AVFrame
1953 s
->current_frame
.qstride
= 0;
1957 if (unpack_superblocks(s
, &gb
)){
1958 av_log(s
->avctx
, AV_LOG_ERROR
, "error in unpack_superblocks\n");
1961 if (unpack_modes(s
, &gb
)){
1962 av_log(s
->avctx
, AV_LOG_ERROR
, "error in unpack_modes\n");
1965 if (unpack_vectors(s
, &gb
)){
1966 av_log(s
->avctx
, AV_LOG_ERROR
, "error in unpack_vectors\n");
1969 if (unpack_block_qpis(s
, &gb
)){
1970 av_log(s
->avctx
, AV_LOG_ERROR
, "error in unpack_block_qpis\n");
1973 if (unpack_dct_coeffs(s
, &gb
)){
1974 av_log(s
->avctx
, AV_LOG_ERROR
, "error in unpack_dct_coeffs\n");
1978 reverse_dc_prediction(s
, 0, s
->fragment_width
, s
->fragment_height
);
1979 if ((avctx
->flags
& CODEC_FLAG_GRAY
) == 0) {
1980 reverse_dc_prediction(s
, s
->fragment_start
[1],
1981 s
->fragment_width
/ 2, s
->fragment_height
/ 2);
1982 reverse_dc_prediction(s
, s
->fragment_start
[2],
1983 s
->fragment_width
/ 2, s
->fragment_height
/ 2);
1986 for (i
= 0; i
< s
->macroblock_height
; i
++)
1989 apply_loop_filter(s
);
1991 *data_size
=sizeof(AVFrame
);
1992 *(AVFrame
*)data
= s
->current_frame
;
1994 /* release the last frame, if it is allocated and if it is not the
1996 if ((s
->last_frame
.data
[0]) &&
1997 (s
->last_frame
.data
[0] != s
->golden_frame
.data
[0]))
1998 avctx
->release_buffer(avctx
, &s
->last_frame
);
2000 /* shuffle frames (last = current) */
2001 s
->last_frame
= s
->current_frame
;
2002 s
->current_frame
.data
[0]= NULL
; /* ensure that we catch any access to this released frame */
2008 * This is the ffmpeg/libavcodec API module cleanup function.
2010 static av_cold
int vp3_decode_end(AVCodecContext
*avctx
)
2012 Vp3DecodeContext
*s
= avctx
->priv_data
;
2015 av_free(s
->superblock_coding
);
2016 av_free(s
->all_fragments
);
2017 av_free(s
->coeff_counts
);
2019 av_free(s
->coded_fragment_list
);
2020 av_free(s
->superblock_fragments
);
2021 av_free(s
->superblock_macroblocks
);
2022 av_free(s
->macroblock_fragments
);
2023 av_free(s
->macroblock_coding
);
2025 for (i
= 0; i
< 16; i
++) {
2026 free_vlc(&s
->dc_vlc
[i
]);
2027 free_vlc(&s
->ac_vlc_1
[i
]);
2028 free_vlc(&s
->ac_vlc_2
[i
]);
2029 free_vlc(&s
->ac_vlc_3
[i
]);
2030 free_vlc(&s
->ac_vlc_4
[i
]);
2033 free_vlc(&s
->superblock_run_length_vlc
);
2034 free_vlc(&s
->fragment_run_length_vlc
);
2035 free_vlc(&s
->mode_code_vlc
);
2036 free_vlc(&s
->motion_vector_vlc
);
2038 /* release all frames */
2039 if (s
->golden_frame
.data
[0] && s
->golden_frame
.data
[0] != s
->last_frame
.data
[0])
2040 avctx
->release_buffer(avctx
, &s
->golden_frame
);
2041 if (s
->last_frame
.data
[0])
2042 avctx
->release_buffer(avctx
, &s
->last_frame
);
2043 /* no need to release the current_frame since it will always be pointing
2044 * to the same frame as either the golden or last frame */
2049 static int read_huffman_tree(AVCodecContext
*avctx
, GetBitContext
*gb
)
2051 Vp3DecodeContext
*s
= avctx
->priv_data
;
2053 if (get_bits1(gb
)) {
2055 if (s
->entries
>= 32) { /* overflow */
2056 av_log(avctx
, AV_LOG_ERROR
, "huffman tree overflow\n");
2059 token
= get_bits(gb
, 5);
2060 //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);
2061 s
->huffman_table
[s
->hti
][token
][0] = s
->hbits
;
2062 s
->huffman_table
[s
->hti
][token
][1] = s
->huff_code_size
;
2066 if (s
->huff_code_size
>= 32) {/* overflow */
2067 av_log(avctx
, AV_LOG_ERROR
, "huffman tree overflow\n");
2070 s
->huff_code_size
++;
2072 if (read_huffman_tree(avctx
, gb
))
2075 if (read_huffman_tree(avctx
, gb
))
2078 s
->huff_code_size
--;
2083 #if CONFIG_THEORA_DECODER
2084 static int theora_decode_header(AVCodecContext
*avctx
, GetBitContext
*gb
)
2086 Vp3DecodeContext
*s
= avctx
->priv_data
;
2087 int visible_width
, visible_height
;
2089 s
->theora
= get_bits_long(gb
, 24);
2090 av_log(avctx
, AV_LOG_DEBUG
, "Theora bitstream version %X\n", s
->theora
);
2092 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2093 /* but previous versions have the image flipped relative to vp3 */
2094 if (s
->theora
< 0x030200)
2096 s
->flipped_image
= 1;
2097 av_log(avctx
, AV_LOG_DEBUG
, "Old (<alpha3) Theora bitstream, flipped image\n");
2100 visible_width
= s
->width
= get_bits(gb
, 16) << 4;
2101 visible_height
= s
->height
= get_bits(gb
, 16) << 4;
2103 if(avcodec_check_dimensions(avctx
, s
->width
, s
->height
)){
2104 av_log(avctx
, AV_LOG_ERROR
, "Invalid dimensions (%dx%d)\n", s
->width
, s
->height
);
2105 s
->width
= s
->height
= 0;
2109 if (s
->theora
>= 0x030400)
2111 skip_bits(gb
, 32); /* total number of superblocks in a frame */
2112 // fixme, the next field is 36bits long
2113 skip_bits(gb
, 32); /* total number of blocks in a frame */
2114 skip_bits(gb
, 4); /* total number of blocks in a frame */
2115 skip_bits(gb
, 32); /* total number of macroblocks in a frame */
2118 if (s
->theora
>= 0x030200) {
2119 visible_width
= get_bits_long(gb
, 24);
2120 visible_height
= get_bits_long(gb
, 24);
2122 skip_bits(gb
, 8); /* offset x */
2123 skip_bits(gb
, 8); /* offset y */
2126 skip_bits(gb
, 32); /* fps numerator */
2127 skip_bits(gb
, 32); /* fps denumerator */
2128 skip_bits(gb
, 24); /* aspect numerator */
2129 skip_bits(gb
, 24); /* aspect denumerator */
2131 if (s
->theora
< 0x030200)
2132 skip_bits(gb
, 5); /* keyframe frequency force */
2133 skip_bits(gb
, 8); /* colorspace */
2134 if (s
->theora
>= 0x030400)
2135 skip_bits(gb
, 2); /* pixel format: 420,res,422,444 */
2136 skip_bits(gb
, 24); /* bitrate */
2138 skip_bits(gb
, 6); /* quality hint */
2140 if (s
->theora
>= 0x030200)
2142 skip_bits(gb
, 5); /* keyframe frequency force */
2144 if (s
->theora
< 0x030400)
2145 skip_bits(gb
, 5); /* spare bits */
2148 // align_get_bits(gb);
2150 if ( visible_width
<= s
->width
&& visible_width
> s
->width
-16
2151 && visible_height
<= s
->height
&& visible_height
> s
->height
-16)
2152 avcodec_set_dimensions(avctx
, visible_width
, visible_height
);
2154 avcodec_set_dimensions(avctx
, s
->width
, s
->height
);
2159 static int theora_decode_tables(AVCodecContext
*avctx
, GetBitContext
*gb
)
2161 Vp3DecodeContext
*s
= avctx
->priv_data
;
2162 int i
, n
, matrices
, inter
, plane
;
2164 if (s
->theora
>= 0x030200) {
2165 n
= get_bits(gb
, 3);
2166 /* loop filter limit values table */
2167 for (i
= 0; i
< 64; i
++)
2168 s
->filter_limit_values
[i
] = get_bits(gb
, n
);
2171 if (s
->theora
>= 0x030200)
2172 n
= get_bits(gb
, 4) + 1;
2175 /* quality threshold table */
2176 for (i
= 0; i
< 64; i
++)
2177 s
->coded_ac_scale_factor
[i
] = get_bits(gb
, n
);
2179 if (s
->theora
>= 0x030200)
2180 n
= get_bits(gb
, 4) + 1;
2183 /* dc scale factor table */
2184 for (i
= 0; i
< 64; i
++)
2185 s
->coded_dc_scale_factor
[i
] = get_bits(gb
, n
);
2187 if (s
->theora
>= 0x030200)
2188 matrices
= get_bits(gb
, 9) + 1;
2193 av_log(avctx
, AV_LOG_ERROR
, "invalid number of base matrixes\n");
2197 for(n
=0; n
<matrices
; n
++){
2198 for (i
= 0; i
< 64; i
++)
2199 s
->base_matrix
[n
][i
]= get_bits(gb
, 8);
2202 for (inter
= 0; inter
<= 1; inter
++) {
2203 for (plane
= 0; plane
<= 2; plane
++) {
2205 if (inter
|| plane
> 0)
2206 newqr
= get_bits1(gb
);
2209 if(inter
&& get_bits1(gb
)){
2213 qtj
= (3*inter
+ plane
- 1) / 3;
2214 plj
= (plane
+ 2) % 3;
2216 s
->qr_count
[inter
][plane
]= s
->qr_count
[qtj
][plj
];
2217 memcpy(s
->qr_size
[inter
][plane
], s
->qr_size
[qtj
][plj
], sizeof(s
->qr_size
[0][0]));
2218 memcpy(s
->qr_base
[inter
][plane
], s
->qr_base
[qtj
][plj
], sizeof(s
->qr_base
[0][0]));
2224 i
= get_bits(gb
, av_log2(matrices
-1)+1);
2226 av_log(avctx
, AV_LOG_ERROR
, "invalid base matrix index\n");
2229 s
->qr_base
[inter
][plane
][qri
]= i
;
2232 i
= get_bits(gb
, av_log2(63-qi
)+1) + 1;
2233 s
->qr_size
[inter
][plane
][qri
++]= i
;
2238 av_log(avctx
, AV_LOG_ERROR
, "invalid qi %d > 63\n", qi
);
2241 s
->qr_count
[inter
][plane
]= qri
;
2246 /* Huffman tables */
2247 for (s
->hti
= 0; s
->hti
< 80; s
->hti
++) {
2249 s
->huff_code_size
= 1;
2250 if (!get_bits1(gb
)) {
2252 if(read_huffman_tree(avctx
, gb
))
2255 if(read_huffman_tree(avctx
, gb
))
2260 s
->theora_tables
= 1;
2265 static av_cold
int theora_decode_init(AVCodecContext
*avctx
)
2267 Vp3DecodeContext
*s
= avctx
->priv_data
;
2270 uint8_t *header_start
[3];
2276 if (!avctx
->extradata_size
)
2278 av_log(avctx
, AV_LOG_ERROR
, "Missing extradata!\n");
2282 if (ff_split_xiph_headers(avctx
->extradata
, avctx
->extradata_size
,
2283 42, header_start
, header_len
) < 0) {
2284 av_log(avctx
, AV_LOG_ERROR
, "Corrupt extradata\n");
2289 init_get_bits(&gb
, header_start
[i
], header_len
[i
]);
2291 ptype
= get_bits(&gb
, 8);
2293 if (!(ptype
& 0x80))
2295 av_log(avctx
, AV_LOG_ERROR
, "Invalid extradata!\n");
2299 // FIXME: Check for this as well.
2300 skip_bits_long(&gb
, 6*8); /* "theora" */
2305 theora_decode_header(avctx
, &gb
);
2308 // FIXME: is this needed? it breaks sometimes
2309 // theora_decode_comments(avctx, gb);
2312 if (theora_decode_tables(avctx
, &gb
))
2316 av_log(avctx
, AV_LOG_ERROR
, "Unknown Theora config packet: %d\n", ptype
&~0x80);
2319 if(ptype
!= 0x81 && 8*header_len
[i
] != get_bits_count(&gb
))
2320 av_log(avctx
, AV_LOG_WARNING
, "%d bits left in packet %X\n", 8*header_len
[i
] - get_bits_count(&gb
), ptype
);
2321 if (s
->theora
< 0x030200)
2325 vp3_decode_init(avctx
);
2329 AVCodec theora_decoder
= {
2333 sizeof(Vp3DecodeContext
),
2340 .long_name
= NULL_IF_CONFIG_SMALL("Theora"),
2344 AVCodec vp3_decoder
= {
2348 sizeof(Vp3DecodeContext
),
2355 .long_name
= NULL_IF_CONFIG_SMALL("On2 VP3"),