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 typedef struct Coeff
{
51 //FIXME split things out into their own arrays
52 typedef struct Vp3Fragment
{
54 /* address of first pixel taking into account which plane the fragment
55 * lives on as well as the plane stride */
57 /* this is the macroblock that the fragment belongs to */
59 uint8_t coding_method
;
65 #define SB_NOT_CODED 0
66 #define SB_PARTIALLY_CODED 1
67 #define SB_FULLY_CODED 2
69 #define MODE_INTER_NO_MV 0
71 #define MODE_INTER_PLUS_MV 2
72 #define MODE_INTER_LAST_MV 3
73 #define MODE_INTER_PRIOR_LAST 4
74 #define MODE_USING_GOLDEN 5
75 #define MODE_GOLDEN_MV 6
76 #define MODE_INTER_FOURMV 7
77 #define CODING_MODE_COUNT 8
79 /* special internal mode */
82 /* There are 6 preset schemes, plus a free-form scheme */
83 static const int ModeAlphabet
[6][CODING_MODE_COUNT
] =
85 /* scheme 1: Last motion vector dominates */
86 { MODE_INTER_LAST_MV
, MODE_INTER_PRIOR_LAST
,
87 MODE_INTER_PLUS_MV
, MODE_INTER_NO_MV
,
88 MODE_INTRA
, MODE_USING_GOLDEN
,
89 MODE_GOLDEN_MV
, MODE_INTER_FOURMV
},
92 { MODE_INTER_LAST_MV
, MODE_INTER_PRIOR_LAST
,
93 MODE_INTER_NO_MV
, MODE_INTER_PLUS_MV
,
94 MODE_INTRA
, MODE_USING_GOLDEN
,
95 MODE_GOLDEN_MV
, MODE_INTER_FOURMV
},
98 { MODE_INTER_LAST_MV
, MODE_INTER_PLUS_MV
,
99 MODE_INTER_PRIOR_LAST
, MODE_INTER_NO_MV
,
100 MODE_INTRA
, MODE_USING_GOLDEN
,
101 MODE_GOLDEN_MV
, MODE_INTER_FOURMV
},
104 { MODE_INTER_LAST_MV
, MODE_INTER_PLUS_MV
,
105 MODE_INTER_NO_MV
, MODE_INTER_PRIOR_LAST
,
106 MODE_INTRA
, MODE_USING_GOLDEN
,
107 MODE_GOLDEN_MV
, MODE_INTER_FOURMV
},
109 /* scheme 5: No motion vector dominates */
110 { MODE_INTER_NO_MV
, MODE_INTER_LAST_MV
,
111 MODE_INTER_PRIOR_LAST
, MODE_INTER_PLUS_MV
,
112 MODE_INTRA
, MODE_USING_GOLDEN
,
113 MODE_GOLDEN_MV
, MODE_INTER_FOURMV
},
116 { MODE_INTER_NO_MV
, MODE_USING_GOLDEN
,
117 MODE_INTER_LAST_MV
, MODE_INTER_PRIOR_LAST
,
118 MODE_INTER_PLUS_MV
, MODE_INTRA
,
119 MODE_GOLDEN_MV
, MODE_INTER_FOURMV
},
123 #define MIN_DEQUANT_VAL 2
125 typedef struct Vp3DecodeContext
{
126 AVCodecContext
*avctx
;
127 int theora
, theora_tables
;
130 AVFrame golden_frame
;
132 AVFrame current_frame
;
141 int superblock_count
;
142 int y_superblock_width
;
143 int y_superblock_height
;
144 int c_superblock_width
;
145 int c_superblock_height
;
146 int u_superblock_start
;
147 int v_superblock_start
;
148 unsigned char *superblock_coding
;
150 int macroblock_count
;
151 int macroblock_width
;
152 int macroblock_height
;
158 Vp3Fragment
*all_fragments
;
159 uint8_t *coeff_counts
;
162 int fragment_start
[3];
167 uint16_t coded_dc_scale_factor
[64];
168 uint32_t coded_ac_scale_factor
[64];
169 uint8_t base_matrix
[384][64];
170 uint8_t qr_count
[2][3];
171 uint8_t qr_size
[2][3][64];
172 uint16_t qr_base
[2][3][64];
174 /* this is a list of indexes into the all_fragments array indicating
175 * which of the fragments are coded */
176 int *coded_fragment_list
;
177 int coded_fragment_list_index
;
178 int pixel_addresses_initialized
;
186 VLC superblock_run_length_vlc
;
187 VLC fragment_run_length_vlc
;
189 VLC motion_vector_vlc
;
191 /* these arrays need to be on 16-byte boundaries since SSE2 operations
193 DECLARE_ALIGNED_16(int16_t, qmat
[3][2][3][64]); //<qmat[qpi][is_inter][plane]
195 /* This table contains superblock_count * 16 entries. Each set of 16
196 * numbers corresponds to the fragment indexes 0..15 of the superblock.
197 * An entry will be -1 to indicate that no entry corresponds to that
199 int *superblock_fragments
;
201 /* This table contains superblock_count * 4 entries. Each set of 4
202 * numbers corresponds to the macroblock indexes 0..3 of the superblock.
203 * An entry will be -1 to indicate that no entry corresponds to that
205 int *superblock_macroblocks
;
207 /* This table contains macroblock_count * 6 entries. Each set of 6
208 * numbers corresponds to the fragment indexes 0..5 which comprise
209 * the macroblock (4 Y fragments and 2 C fragments). */
210 int *macroblock_fragments
;
211 /* This is an array that indicates how a particular macroblock
213 unsigned char *macroblock_coding
;
215 int first_coded_y_fragment
;
216 int first_coded_c_fragment
;
217 int last_coded_y_fragment
;
218 int last_coded_c_fragment
;
220 uint8_t edge_emu_buffer
[9*2048]; //FIXME dynamic alloc
221 int8_t qscale_table
[2048]; //FIXME dynamic alloc (width+15)/16
228 uint16_t huffman_table
[80][32][2];
230 uint8_t filter_limit_values
[64];
231 DECLARE_ALIGNED_8(int, bounding_values_array
[256+2]);
234 /************************************************************************
235 * VP3 specific functions
236 ************************************************************************/
239 * This function sets up all of the various blocks mappings:
240 * superblocks <-> fragments, macroblocks <-> fragments,
241 * superblocks <-> macroblocks
243 * Returns 0 is successful; returns 1 if *anything* went wrong.
245 static int init_block_mapping(Vp3DecodeContext
*s
)
248 signed int hilbert_walk_mb
[4];
250 int current_fragment
= 0;
251 int current_width
= 0;
252 int current_height
= 0;
255 int superblock_row_inc
= 0;
256 int mapping_index
= 0;
258 int current_macroblock
;
261 signed char travel_width
[16] = {
268 signed char travel_height
[16] = {
275 signed char travel_width_mb
[4] = {
279 signed char travel_height_mb
[4] = {
283 hilbert_walk_mb
[0] = 1;
284 hilbert_walk_mb
[1] = s
->macroblock_width
;
285 hilbert_walk_mb
[2] = 1;
286 hilbert_walk_mb
[3] = -s
->macroblock_width
;
288 /* iterate through each superblock (all planes) and map the fragments */
289 for (i
= 0; i
< s
->superblock_count
; i
++) {
290 /* time to re-assign the limits? */
293 /* start of Y superblocks */
294 right_edge
= s
->fragment_width
;
295 bottom_edge
= s
->fragment_height
;
298 superblock_row_inc
= 3 * s
->fragment_width
-
299 (s
->y_superblock_width
* 4 - s
->fragment_width
);
301 /* the first operation for this variable is to advance by 1 */
302 current_fragment
= -1;
304 } else if (i
== s
->u_superblock_start
) {
306 /* start of U superblocks */
307 right_edge
= s
->fragment_width
/ 2;
308 bottom_edge
= s
->fragment_height
/ 2;
311 superblock_row_inc
= 3 * (s
->fragment_width
/ 2) -
312 (s
->c_superblock_width
* 4 - s
->fragment_width
/ 2);
314 /* the first operation for this variable is to advance by 1 */
315 current_fragment
= s
->fragment_start
[1] - 1;
317 } else if (i
== s
->v_superblock_start
) {
319 /* start of V superblocks */
320 right_edge
= s
->fragment_width
/ 2;
321 bottom_edge
= s
->fragment_height
/ 2;
324 superblock_row_inc
= 3 * (s
->fragment_width
/ 2) -
325 (s
->c_superblock_width
* 4 - s
->fragment_width
/ 2);
327 /* the first operation for this variable is to advance by 1 */
328 current_fragment
= s
->fragment_start
[2] - 1;
332 if (current_width
>= right_edge
- 1) {
333 /* reset width and move to next superblock row */
337 /* fragment is now at the start of a new superblock row */
338 current_fragment
+= superblock_row_inc
;
341 /* iterate through all 16 fragments in a superblock */
342 for (j
= 0; j
< 16; j
++) {
343 current_fragment
+= travel_width
[j
] + right_edge
* travel_height
[j
];
344 current_width
+= travel_width
[j
];
345 current_height
+= travel_height
[j
];
347 /* check if the fragment is in bounds */
348 if ((current_width
< right_edge
) &&
349 (current_height
< bottom_edge
)) {
350 s
->superblock_fragments
[mapping_index
] = current_fragment
;
352 s
->superblock_fragments
[mapping_index
] = -1;
359 /* initialize the superblock <-> macroblock mapping; iterate through
360 * all of the Y plane superblocks to build this mapping */
361 right_edge
= s
->macroblock_width
;
362 bottom_edge
= s
->macroblock_height
;
365 superblock_row_inc
= s
->macroblock_width
-
366 (s
->y_superblock_width
* 2 - s
->macroblock_width
);
368 current_macroblock
= -1;
369 for (i
= 0; i
< s
->u_superblock_start
; i
++) {
371 if (current_width
>= right_edge
- 1) {
372 /* reset width and move to next superblock row */
376 /* macroblock is now at the start of a new superblock row */
377 current_macroblock
+= superblock_row_inc
;
380 /* iterate through each potential macroblock in the superblock */
381 for (j
= 0; j
< 4; j
++) {
382 current_macroblock
+= hilbert_walk_mb
[j
];
383 current_width
+= travel_width_mb
[j
];
384 current_height
+= travel_height_mb
[j
];
386 /* check if the macroblock is in bounds */
387 if ((current_width
< right_edge
) &&
388 (current_height
< bottom_edge
)) {
389 s
->superblock_macroblocks
[mapping_index
] = current_macroblock
;
391 s
->superblock_macroblocks
[mapping_index
] = -1;
398 /* initialize the macroblock <-> fragment mapping */
399 current_fragment
= 0;
400 current_macroblock
= 0;
402 for (i
= 0; i
< s
->fragment_height
; i
+= 2) {
404 for (j
= 0; j
< s
->fragment_width
; j
+= 2) {
406 s
->all_fragments
[current_fragment
].macroblock
= current_macroblock
;
407 s
->macroblock_fragments
[mapping_index
++] = current_fragment
;
409 if (j
+ 1 < s
->fragment_width
) {
410 s
->all_fragments
[current_fragment
+ 1].macroblock
= current_macroblock
;
411 s
->macroblock_fragments
[mapping_index
++] = current_fragment
+ 1;
413 s
->macroblock_fragments
[mapping_index
++] = -1;
415 if (i
+ 1 < s
->fragment_height
) {
416 s
->all_fragments
[current_fragment
+ s
->fragment_width
].macroblock
=
418 s
->macroblock_fragments
[mapping_index
++] =
419 current_fragment
+ s
->fragment_width
;
421 s
->macroblock_fragments
[mapping_index
++] = -1;
423 if ((j
+ 1 < s
->fragment_width
) && (i
+ 1 < s
->fragment_height
)) {
424 s
->all_fragments
[current_fragment
+ s
->fragment_width
+ 1].macroblock
=
426 s
->macroblock_fragments
[mapping_index
++] =
427 current_fragment
+ s
->fragment_width
+ 1;
429 s
->macroblock_fragments
[mapping_index
++] = -1;
432 c_fragment
= s
->fragment_start
[1] +
433 (i
* s
->fragment_width
/ 4) + (j
/ 2);
434 s
->all_fragments
[c_fragment
].macroblock
= s
->macroblock_count
;
435 s
->macroblock_fragments
[mapping_index
++] = c_fragment
;
437 c_fragment
= s
->fragment_start
[2] +
438 (i
* s
->fragment_width
/ 4) + (j
/ 2);
439 s
->all_fragments
[c_fragment
].macroblock
= s
->macroblock_count
;
440 s
->macroblock_fragments
[mapping_index
++] = c_fragment
;
442 if (j
+ 2 <= s
->fragment_width
)
443 current_fragment
+= 2;
446 current_macroblock
++;
449 current_fragment
+= s
->fragment_width
;
452 return 0; /* successful path out */
456 * This function wipes out all of the fragment data.
458 static void init_frame(Vp3DecodeContext
*s
, GetBitContext
*gb
)
462 /* zero out all of the fragment information */
463 s
->coded_fragment_list_index
= 0;
464 for (i
= 0; i
< s
->fragment_count
; i
++) {
465 s
->coeff_counts
[i
] = 0;
466 s
->all_fragments
[i
].motion_x
= 127;
467 s
->all_fragments
[i
].motion_y
= 127;
468 s
->all_fragments
[i
].next_coeff
= NULL
;
469 s
->all_fragments
[i
].qpi
= 0;
471 s
->coeffs
[i
].coeff
=0;
472 s
->coeffs
[i
].next
= NULL
;
477 * This function sets up the dequantization tables used for a particular
480 static void init_dequantizer(Vp3DecodeContext
*s
, int qpi
)
482 int ac_scale_factor
= s
->coded_ac_scale_factor
[s
->qps
[qpi
]];
483 int dc_scale_factor
= s
->coded_dc_scale_factor
[s
->qps
[qpi
]];
484 int i
, plane
, inter
, qri
, bmi
, bmj
, qistart
;
486 for(inter
=0; inter
<2; inter
++){
487 for(plane
=0; plane
<3; plane
++){
489 for(qri
=0; qri
<s
->qr_count
[inter
][plane
]; qri
++){
490 sum
+= s
->qr_size
[inter
][plane
][qri
];
491 if(s
->qps
[qpi
] <= sum
)
494 qistart
= sum
- s
->qr_size
[inter
][plane
][qri
];
495 bmi
= s
->qr_base
[inter
][plane
][qri
];
496 bmj
= s
->qr_base
[inter
][plane
][qri
+1];
498 int coeff
= ( 2*(sum
-s
->qps
[qpi
])*s
->base_matrix
[bmi
][i
]
499 - 2*(qistart
-s
->qps
[qpi
])*s
->base_matrix
[bmj
][i
]
500 + s
->qr_size
[inter
][plane
][qri
])
501 / (2*s
->qr_size
[inter
][plane
][qri
]);
503 int qmin
= 8<<(inter
+ !i
);
504 int qscale
= i
? ac_scale_factor
: dc_scale_factor
;
506 s
->qmat
[qpi
][inter
][plane
][s
->dsp
.idct_permutation
[i
]]= av_clip((qscale
* coeff
)/100 * 4, qmin
, 4096);
508 // all DC coefficients use the same quant so as not to interfere with DC prediction
509 s
->qmat
[qpi
][inter
][plane
][0] = s
->qmat
[0][inter
][plane
][0];
513 memset(s
->qscale_table
, (FFMAX(s
->qmat
[0][0][0][1], s
->qmat
[0][0][1][1])+8)/16, 512); //FIXME finetune
517 * This function initializes the loop filter boundary limits if the frame's
518 * quality index is different from the previous frame's.
520 * The filter_limit_values may not be larger than 127.
522 static void init_loop_filter(Vp3DecodeContext
*s
)
524 int *bounding_values
= s
->bounding_values_array
+127;
529 filter_limit
= s
->filter_limit_values
[s
->qps
[0]];
531 /* set up the bounding values */
532 memset(s
->bounding_values_array
, 0, 256 * sizeof(int));
533 for (x
= 0; x
< filter_limit
; x
++) {
534 bounding_values
[-x
] = -x
;
535 bounding_values
[x
] = x
;
537 for (x
= value
= filter_limit
; x
< 128 && value
; x
++, value
--) {
538 bounding_values
[ x
] = value
;
539 bounding_values
[-x
] = -value
;
542 bounding_values
[128] = value
;
543 bounding_values
[129] = bounding_values
[130] = filter_limit
* 0x02020202;
547 * This function unpacks all of the superblock/macroblock/fragment coding
548 * information from the bitstream.
550 static int unpack_superblocks(Vp3DecodeContext
*s
, GetBitContext
*gb
)
553 int current_superblock
= 0;
555 int decode_fully_flags
= 0;
556 int decode_partial_blocks
= 0;
557 int first_c_fragment_seen
;
560 int current_fragment
;
563 memset(s
->superblock_coding
, SB_FULLY_CODED
, s
->superblock_count
);
567 /* unpack the list of partially-coded superblocks */
569 /* toggle the bit because as soon as the first run length is
570 * fetched the bit will be toggled again */
572 while (current_superblock
< s
->superblock_count
) {
573 if (current_run
-- == 0) {
575 current_run
= get_vlc2(gb
,
576 s
->superblock_run_length_vlc
.table
, 6, 2);
577 if (current_run
== 33)
578 current_run
+= get_bits(gb
, 12);
580 /* if any of the superblocks are not partially coded, flag
581 * a boolean to decode the list of fully-coded superblocks */
583 decode_fully_flags
= 1;
586 /* make a note of the fact that there are partially coded
588 decode_partial_blocks
= 1;
591 s
->superblock_coding
[current_superblock
++] = bit
;
594 /* unpack the list of fully coded superblocks if any of the blocks were
595 * not marked as partially coded in the previous step */
596 if (decode_fully_flags
) {
598 current_superblock
= 0;
601 /* toggle the bit because as soon as the first run length is
602 * fetched the bit will be toggled again */
604 while (current_superblock
< s
->superblock_count
) {
606 /* skip any superblocks already marked as partially coded */
607 if (s
->superblock_coding
[current_superblock
] == SB_NOT_CODED
) {
609 if (current_run
-- == 0) {
611 current_run
= get_vlc2(gb
,
612 s
->superblock_run_length_vlc
.table
, 6, 2);
613 if (current_run
== 33)
614 current_run
+= get_bits(gb
, 12);
616 s
->superblock_coding
[current_superblock
] = 2*bit
;
618 current_superblock
++;
622 /* if there were partial blocks, initialize bitstream for
623 * unpacking fragment codings */
624 if (decode_partial_blocks
) {
628 /* toggle the bit because as soon as the first run length is
629 * fetched the bit will be toggled again */
634 /* figure out which fragments are coded; iterate through each
635 * superblock (all planes) */
636 s
->coded_fragment_list_index
= 0;
637 s
->next_coeff
= s
->coeffs
+ s
->fragment_count
;
638 s
->first_coded_y_fragment
= s
->first_coded_c_fragment
= 0;
639 s
->last_coded_y_fragment
= s
->last_coded_c_fragment
= -1;
640 first_c_fragment_seen
= 0;
641 memset(s
->macroblock_coding
, MODE_COPY
, s
->macroblock_count
);
642 for (i
= 0; i
< s
->superblock_count
; i
++) {
644 /* iterate through all 16 fragments in a superblock */
645 for (j
= 0; j
< 16; j
++) {
647 /* if the fragment is in bounds, check its coding status */
648 current_fragment
= s
->superblock_fragments
[i
* 16 + j
];
649 if (current_fragment
>= s
->fragment_count
) {
650 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
651 current_fragment
, s
->fragment_count
);
654 if (current_fragment
!= -1) {
655 if (s
->superblock_coding
[i
] == SB_NOT_CODED
) {
657 /* copy all the fragments from the prior frame */
658 s
->all_fragments
[current_fragment
].coding_method
=
661 } else if (s
->superblock_coding
[i
] == SB_PARTIALLY_CODED
) {
663 /* fragment may or may not be coded; this is the case
664 * that cares about the fragment coding runs */
665 if (current_run
-- == 0) {
667 current_run
= get_vlc2(gb
,
668 s
->fragment_run_length_vlc
.table
, 5, 2);
672 /* default mode; actual mode will be decoded in
674 s
->all_fragments
[current_fragment
].coding_method
=
676 s
->all_fragments
[current_fragment
].next_coeff
= s
->coeffs
+ current_fragment
;
677 s
->coded_fragment_list
[s
->coded_fragment_list_index
] =
679 if ((current_fragment
>= s
->fragment_start
[1]) &&
680 (s
->last_coded_y_fragment
== -1) &&
681 (!first_c_fragment_seen
)) {
682 s
->first_coded_c_fragment
= s
->coded_fragment_list_index
;
683 s
->last_coded_y_fragment
= s
->first_coded_c_fragment
- 1;
684 first_c_fragment_seen
= 1;
686 s
->coded_fragment_list_index
++;
687 s
->macroblock_coding
[s
->all_fragments
[current_fragment
].macroblock
] = MODE_INTER_NO_MV
;
689 /* not coded; copy this fragment from the prior frame */
690 s
->all_fragments
[current_fragment
].coding_method
=
696 /* fragments are fully coded in this superblock; actual
697 * coding will be determined in next step */
698 s
->all_fragments
[current_fragment
].coding_method
=
700 s
->all_fragments
[current_fragment
].next_coeff
= s
->coeffs
+ current_fragment
;
701 s
->coded_fragment_list
[s
->coded_fragment_list_index
] =
703 if ((current_fragment
>= s
->fragment_start
[1]) &&
704 (s
->last_coded_y_fragment
== -1) &&
705 (!first_c_fragment_seen
)) {
706 s
->first_coded_c_fragment
= s
->coded_fragment_list_index
;
707 s
->last_coded_y_fragment
= s
->first_coded_c_fragment
- 1;
708 first_c_fragment_seen
= 1;
710 s
->coded_fragment_list_index
++;
711 s
->macroblock_coding
[s
->all_fragments
[current_fragment
].macroblock
] = MODE_INTER_NO_MV
;
717 if (!first_c_fragment_seen
)
718 /* only Y fragments coded in this frame */
719 s
->last_coded_y_fragment
= s
->coded_fragment_list_index
- 1;
721 /* end the list of coded C fragments */
722 s
->last_coded_c_fragment
= s
->coded_fragment_list_index
- 1;
728 * This function unpacks all the coding mode data for individual macroblocks
729 * from the bitstream.
731 static int unpack_modes(Vp3DecodeContext
*s
, GetBitContext
*gb
)
735 int current_macroblock
;
736 int current_fragment
;
738 int custom_mode_alphabet
[CODING_MODE_COUNT
];
741 for (i
= 0; i
< s
->fragment_count
; i
++)
742 s
->all_fragments
[i
].coding_method
= MODE_INTRA
;
746 /* fetch the mode coding scheme for this frame */
747 scheme
= get_bits(gb
, 3);
749 /* is it a custom coding scheme? */
751 for (i
= 0; i
< 8; i
++)
752 custom_mode_alphabet
[i
] = MODE_INTER_NO_MV
;
753 for (i
= 0; i
< 8; i
++)
754 custom_mode_alphabet
[get_bits(gb
, 3)] = i
;
757 /* iterate through all of the macroblocks that contain 1 or more
759 for (i
= 0; i
< s
->u_superblock_start
; i
++) {
761 for (j
= 0; j
< 4; j
++) {
762 current_macroblock
= s
->superblock_macroblocks
[i
* 4 + j
];
763 if ((current_macroblock
== -1) ||
764 (s
->macroblock_coding
[current_macroblock
] == MODE_COPY
))
766 if (current_macroblock
>= s
->macroblock_count
) {
767 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_modes(): bad macroblock number (%d >= %d)\n",
768 current_macroblock
, s
->macroblock_count
);
772 /* mode 7 means get 3 bits for each coding mode */
774 coding_mode
= get_bits(gb
, 3);
776 coding_mode
= custom_mode_alphabet
777 [get_vlc2(gb
, s
->mode_code_vlc
.table
, 3, 3)];
779 coding_mode
= ModeAlphabet
[scheme
-1]
780 [get_vlc2(gb
, s
->mode_code_vlc
.table
, 3, 3)];
782 s
->macroblock_coding
[current_macroblock
] = coding_mode
;
783 for (k
= 0; k
< 6; k
++) {
785 s
->macroblock_fragments
[current_macroblock
* 6 + k
];
786 if (current_fragment
== -1)
788 if (current_fragment
>= s
->fragment_count
) {
789 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_modes(): bad fragment number (%d >= %d)\n",
790 current_fragment
, s
->fragment_count
);
793 if (s
->all_fragments
[current_fragment
].coding_method
!=
795 s
->all_fragments
[current_fragment
].coding_method
=
806 * This function unpacks all the motion vectors for the individual
807 * macroblocks from the bitstream.
809 static int unpack_vectors(Vp3DecodeContext
*s
, GetBitContext
*gb
)
815 int last_motion_x
= 0;
816 int last_motion_y
= 0;
817 int prior_last_motion_x
= 0;
818 int prior_last_motion_y
= 0;
819 int current_macroblock
;
820 int current_fragment
;
825 memset(motion_x
, 0, 6 * sizeof(int));
826 memset(motion_y
, 0, 6 * sizeof(int));
828 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
829 coding_mode
= get_bits1(gb
);
831 /* iterate through all of the macroblocks that contain 1 or more
833 for (i
= 0; i
< s
->u_superblock_start
; i
++) {
835 for (j
= 0; j
< 4; j
++) {
836 current_macroblock
= s
->superblock_macroblocks
[i
* 4 + j
];
837 if ((current_macroblock
== -1) ||
838 (s
->macroblock_coding
[current_macroblock
] == MODE_COPY
))
840 if (current_macroblock
>= s
->macroblock_count
) {
841 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_vectors(): bad macroblock number (%d >= %d)\n",
842 current_macroblock
, s
->macroblock_count
);
846 current_fragment
= s
->macroblock_fragments
[current_macroblock
* 6];
847 if (current_fragment
>= s
->fragment_count
) {
848 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_vectors(): bad fragment number (%d >= %d\n",
849 current_fragment
, s
->fragment_count
);
852 switch (s
->macroblock_coding
[current_macroblock
]) {
854 case MODE_INTER_PLUS_MV
:
856 /* all 6 fragments use the same motion vector */
857 if (coding_mode
== 0) {
858 motion_x
[0] = motion_vector_table
[get_vlc2(gb
, s
->motion_vector_vlc
.table
, 6, 2)];
859 motion_y
[0] = motion_vector_table
[get_vlc2(gb
, s
->motion_vector_vlc
.table
, 6, 2)];
861 motion_x
[0] = fixed_motion_vector_table
[get_bits(gb
, 6)];
862 motion_y
[0] = fixed_motion_vector_table
[get_bits(gb
, 6)];
865 for (k
= 1; k
< 6; k
++) {
866 motion_x
[k
] = motion_x
[0];
867 motion_y
[k
] = motion_y
[0];
870 /* vector maintenance, only on MODE_INTER_PLUS_MV */
871 if (s
->macroblock_coding
[current_macroblock
] ==
872 MODE_INTER_PLUS_MV
) {
873 prior_last_motion_x
= last_motion_x
;
874 prior_last_motion_y
= last_motion_y
;
875 last_motion_x
= motion_x
[0];
876 last_motion_y
= motion_y
[0];
880 case MODE_INTER_FOURMV
:
881 /* vector maintenance */
882 prior_last_motion_x
= last_motion_x
;
883 prior_last_motion_y
= last_motion_y
;
885 /* fetch 4 vectors from the bitstream, one for each
886 * Y fragment, then average for the C fragment vectors */
887 motion_x
[4] = motion_y
[4] = 0;
888 for (k
= 0; k
< 4; k
++) {
889 for (l
= 0; l
< s
->coded_fragment_list_index
; l
++)
890 if (s
->coded_fragment_list
[l
] == s
->macroblock_fragments
[6*current_macroblock
+ k
])
892 if (l
< s
->coded_fragment_list_index
) {
893 if (coding_mode
== 0) {
894 motion_x
[k
] = motion_vector_table
[get_vlc2(gb
, s
->motion_vector_vlc
.table
, 6, 2)];
895 motion_y
[k
] = motion_vector_table
[get_vlc2(gb
, s
->motion_vector_vlc
.table
, 6, 2)];
897 motion_x
[k
] = fixed_motion_vector_table
[get_bits(gb
, 6)];
898 motion_y
[k
] = fixed_motion_vector_table
[get_bits(gb
, 6)];
900 last_motion_x
= motion_x
[k
];
901 last_motion_y
= motion_y
[k
];
906 motion_x
[4] += motion_x
[k
];
907 motion_y
[4] += motion_y
[k
];
911 motion_x
[4]= RSHIFT(motion_x
[4], 2);
913 motion_y
[4]= RSHIFT(motion_y
[4], 2);
916 case MODE_INTER_LAST_MV
:
917 /* all 6 fragments use the last motion vector */
918 motion_x
[0] = last_motion_x
;
919 motion_y
[0] = last_motion_y
;
920 for (k
= 1; k
< 6; k
++) {
921 motion_x
[k
] = motion_x
[0];
922 motion_y
[k
] = motion_y
[0];
925 /* no vector maintenance (last vector remains the
929 case MODE_INTER_PRIOR_LAST
:
930 /* all 6 fragments use the motion vector prior to the
931 * last motion vector */
932 motion_x
[0] = prior_last_motion_x
;
933 motion_y
[0] = prior_last_motion_y
;
934 for (k
= 1; k
< 6; k
++) {
935 motion_x
[k
] = motion_x
[0];
936 motion_y
[k
] = motion_y
[0];
939 /* vector maintenance */
940 prior_last_motion_x
= last_motion_x
;
941 prior_last_motion_y
= last_motion_y
;
942 last_motion_x
= motion_x
[0];
943 last_motion_y
= motion_y
[0];
947 /* covers intra, inter without MV, golden without MV */
948 memset(motion_x
, 0, 6 * sizeof(int));
949 memset(motion_y
, 0, 6 * sizeof(int));
951 /* no vector maintenance */
955 /* assign the motion vectors to the correct fragments */
956 for (k
= 0; k
< 6; k
++) {
958 s
->macroblock_fragments
[current_macroblock
* 6 + k
];
959 if (current_fragment
== -1)
961 if (current_fragment
>= s
->fragment_count
) {
962 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_vectors(): bad fragment number (%d >= %d)\n",
963 current_fragment
, s
->fragment_count
);
966 s
->all_fragments
[current_fragment
].motion_x
= motion_x
[k
];
967 s
->all_fragments
[current_fragment
].motion_y
= motion_y
[k
];
975 static int unpack_block_qpis(Vp3DecodeContext
*s
, GetBitContext
*gb
)
977 int qpi
, i
, j
, bit
, run_length
, blocks_decoded
, num_blocks_at_qpi
;
978 int num_blocks
= s
->coded_fragment_list_index
;
980 for (qpi
= 0; qpi
< s
->nqps
-1 && num_blocks
> 0; qpi
++) {
981 i
= blocks_decoded
= num_blocks_at_qpi
= 0;
986 run_length
= get_vlc2(gb
, s
->superblock_run_length_vlc
.table
, 6, 2) + 1;
987 if (run_length
== 34)
988 run_length
+= get_bits(gb
, 12);
989 blocks_decoded
+= run_length
;
992 num_blocks_at_qpi
+= run_length
;
994 for (j
= 0; j
< run_length
; i
++) {
995 if (i
> s
->coded_fragment_list_index
)
998 if (s
->all_fragments
[s
->coded_fragment_list
[i
]].qpi
== qpi
) {
999 s
->all_fragments
[s
->coded_fragment_list
[i
]].qpi
+= bit
;
1004 if (run_length
== 4129)
1005 bit
= get_bits1(gb
);
1008 } while (blocks_decoded
< num_blocks
);
1010 num_blocks
-= num_blocks_at_qpi
;
1017 * This function is called by unpack_dct_coeffs() to extract the VLCs from
1018 * the bitstream. The VLCs encode tokens which are used to unpack DCT
1019 * data. This function unpacks all the VLCs for either the Y plane or both
1020 * C planes, and is called for DC coefficients or different AC coefficient
1021 * levels (since different coefficient types require different VLC tables.
1023 * This function returns a residual eob run. E.g, if a particular token gave
1024 * instructions to EOB the next 5 fragments and there were only 2 fragments
1025 * left in the current fragment range, 3 would be returned so that it could
1026 * be passed into the next call to this same function.
1028 static int unpack_vlcs(Vp3DecodeContext
*s
, GetBitContext
*gb
,
1029 VLC
*table
, int coeff_index
,
1030 int first_fragment
, int last_fragment
,
1037 Vp3Fragment
*fragment
;
1038 uint8_t *perm
= s
->scantable
.permutated
;
1041 if ((first_fragment
>= s
->fragment_count
) ||
1042 (last_fragment
>= s
->fragment_count
)) {
1044 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:unpack_vlcs(): bad fragment number (%d -> %d ?)\n",
1045 first_fragment
, last_fragment
);
1049 for (i
= first_fragment
; i
<= last_fragment
; i
++) {
1050 int fragment_num
= s
->coded_fragment_list
[i
];
1052 if (s
->coeff_counts
[fragment_num
] > coeff_index
)
1054 fragment
= &s
->all_fragments
[fragment_num
];
1057 /* decode a VLC into a token */
1058 token
= get_vlc2(gb
, table
->table
, 5, 3);
1059 /* use the token to get a zero run, a coefficient, and an eob run */
1061 eob_run
= eob_run_base
[token
];
1062 if (eob_run_get_bits
[token
])
1063 eob_run
+= get_bits(gb
, eob_run_get_bits
[token
]);
1064 coeff
= zero_run
= 0;
1066 bits_to_get
= coeff_get_bits
[token
];
1068 coeff
= coeff_tables
[token
][0];
1070 coeff
= coeff_tables
[token
][get_bits(gb
, bits_to_get
)];
1072 zero_run
= zero_run_base
[token
];
1073 if (zero_run_get_bits
[token
])
1074 zero_run
+= get_bits(gb
, zero_run_get_bits
[token
]);
1079 s
->coeff_counts
[fragment_num
] += zero_run
;
1080 if (s
->coeff_counts
[fragment_num
] < 64){
1081 fragment
->next_coeff
->coeff
= coeff
;
1082 fragment
->next_coeff
->index
= perm
[s
->coeff_counts
[fragment_num
]++]; //FIXME perm here already?
1083 fragment
->next_coeff
->next
= s
->next_coeff
;
1084 s
->next_coeff
->next
=NULL
;
1085 fragment
->next_coeff
= s
->next_coeff
++;
1088 s
->coeff_counts
[fragment_num
] |= 128;
1097 * This function unpacks all of the DCT coefficient data from the
1100 static int unpack_dct_coeffs(Vp3DecodeContext
*s
, GetBitContext
*gb
)
1107 int residual_eob_run
= 0;
1109 /* fetch the DC table indexes */
1110 dc_y_table
= get_bits(gb
, 4);
1111 dc_c_table
= get_bits(gb
, 4);
1113 /* unpack the Y plane DC coefficients */
1114 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->dc_vlc
[dc_y_table
], 0,
1115 s
->first_coded_y_fragment
, s
->last_coded_y_fragment
, residual_eob_run
);
1117 /* unpack the C plane DC coefficients */
1118 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->dc_vlc
[dc_c_table
], 0,
1119 s
->first_coded_c_fragment
, s
->last_coded_c_fragment
, residual_eob_run
);
1121 /* fetch the AC table indexes */
1122 ac_y_table
= get_bits(gb
, 4);
1123 ac_c_table
= get_bits(gb
, 4);
1125 /* unpack the group 1 AC coefficients (coeffs 1-5) */
1126 for (i
= 1; i
<= 5; i
++) {
1127 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_1
[ac_y_table
], i
,
1128 s
->first_coded_y_fragment
, s
->last_coded_y_fragment
, residual_eob_run
);
1130 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_1
[ac_c_table
], i
,
1131 s
->first_coded_c_fragment
, s
->last_coded_c_fragment
, residual_eob_run
);
1134 /* unpack the group 2 AC coefficients (coeffs 6-14) */
1135 for (i
= 6; i
<= 14; i
++) {
1136 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_2
[ac_y_table
], i
,
1137 s
->first_coded_y_fragment
, s
->last_coded_y_fragment
, residual_eob_run
);
1139 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_2
[ac_c_table
], i
,
1140 s
->first_coded_c_fragment
, s
->last_coded_c_fragment
, residual_eob_run
);
1143 /* unpack the group 3 AC coefficients (coeffs 15-27) */
1144 for (i
= 15; i
<= 27; i
++) {
1145 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_3
[ac_y_table
], i
,
1146 s
->first_coded_y_fragment
, s
->last_coded_y_fragment
, residual_eob_run
);
1148 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_3
[ac_c_table
], i
,
1149 s
->first_coded_c_fragment
, s
->last_coded_c_fragment
, residual_eob_run
);
1152 /* unpack the group 4 AC coefficients (coeffs 28-63) */
1153 for (i
= 28; i
<= 63; i
++) {
1154 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_4
[ac_y_table
], i
,
1155 s
->first_coded_y_fragment
, s
->last_coded_y_fragment
, residual_eob_run
);
1157 residual_eob_run
= unpack_vlcs(s
, gb
, &s
->ac_vlc_4
[ac_c_table
], i
,
1158 s
->first_coded_c_fragment
, s
->last_coded_c_fragment
, residual_eob_run
);
1165 * This function reverses the DC prediction for each coded fragment in
1166 * the frame. Much of this function is adapted directly from the original
1169 #define COMPATIBLE_FRAME(x) \
1170 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1171 #define FRAME_CODED(x) (s->all_fragments[x].coding_method != MODE_COPY)
1172 #define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
1174 static void reverse_dc_prediction(Vp3DecodeContext
*s
,
1177 int fragment_height
)
1186 int i
= first_fragment
;
1190 /* DC values for the left, up-left, up, and up-right fragments */
1191 int vl
, vul
, vu
, vur
;
1193 /* indexes for the left, up-left, up, and up-right fragments */
1197 * The 6 fields mean:
1198 * 0: up-left multiplier
1200 * 2: up-right multiplier
1201 * 3: left multiplier
1203 int predictor_transform
[16][4] = {
1205 { 0, 0, 0,128}, // PL
1206 { 0, 0,128, 0}, // PUR
1207 { 0, 0, 53, 75}, // PUR|PL
1208 { 0,128, 0, 0}, // PU
1209 { 0, 64, 0, 64}, // PU|PL
1210 { 0,128, 0, 0}, // PU|PUR
1211 { 0, 0, 53, 75}, // PU|PUR|PL
1212 {128, 0, 0, 0}, // PUL
1213 { 0, 0, 0,128}, // PUL|PL
1214 { 64, 0, 64, 0}, // PUL|PUR
1215 { 0, 0, 53, 75}, // PUL|PUR|PL
1216 { 0,128, 0, 0}, // PUL|PU
1217 {-104,116, 0,116}, // PUL|PU|PL
1218 { 24, 80, 24, 0}, // PUL|PU|PUR
1219 {-104,116, 0,116} // PUL|PU|PUR|PL
1222 /* This table shows which types of blocks can use other blocks for
1223 * prediction. For example, INTRA is the only mode in this table to
1224 * have a frame number of 0. That means INTRA blocks can only predict
1225 * from other INTRA blocks. There are 2 golden frame coding types;
1226 * blocks encoding in these modes can only predict from other blocks
1227 * that were encoded with these 1 of these 2 modes. */
1228 unsigned char compatible_frame
[8] = {
1229 1, /* MODE_INTER_NO_MV */
1231 1, /* MODE_INTER_PLUS_MV */
1232 1, /* MODE_INTER_LAST_MV */
1233 1, /* MODE_INTER_PRIOR_MV */
1234 2, /* MODE_USING_GOLDEN */
1235 2, /* MODE_GOLDEN_MV */
1236 1 /* MODE_INTER_FOUR_MV */
1238 int current_frame_type
;
1240 /* there is a last DC predictor for each of the 3 frame types */
1245 vul
= vu
= vur
= vl
= 0;
1246 last_dc
[0] = last_dc
[1] = last_dc
[2] = 0;
1248 /* for each fragment row... */
1249 for (y
= 0; y
< fragment_height
; y
++) {
1251 /* for each fragment in a row... */
1252 for (x
= 0; x
< fragment_width
; x
++, i
++) {
1254 /* reverse prediction if this block was coded */
1255 if (s
->all_fragments
[i
].coding_method
!= MODE_COPY
) {
1257 current_frame_type
=
1258 compatible_frame
[s
->all_fragments
[i
].coding_method
];
1264 if(FRAME_CODED(l
) && COMPATIBLE_FRAME(l
))
1268 u
= i
-fragment_width
;
1270 if(FRAME_CODED(u
) && COMPATIBLE_FRAME(u
))
1273 ul
= i
-fragment_width
-1;
1275 if(FRAME_CODED(ul
) && COMPATIBLE_FRAME(ul
))
1278 if(x
+ 1 < fragment_width
){
1279 ur
= i
-fragment_width
+1;
1281 if(FRAME_CODED(ur
) && COMPATIBLE_FRAME(ur
))
1286 if (transform
== 0) {
1288 /* if there were no fragments to predict from, use last
1290 predicted_dc
= last_dc
[current_frame_type
];
1293 /* apply the appropriate predictor transform */
1295 (predictor_transform
[transform
][0] * vul
) +
1296 (predictor_transform
[transform
][1] * vu
) +
1297 (predictor_transform
[transform
][2] * vur
) +
1298 (predictor_transform
[transform
][3] * vl
);
1300 predicted_dc
/= 128;
1302 /* check for outranging on the [ul u l] and
1303 * [ul u ur l] predictors */
1304 if ((transform
== 13) || (transform
== 15)) {
1305 if (FFABS(predicted_dc
- vu
) > 128)
1307 else if (FFABS(predicted_dc
- vl
) > 128)
1309 else if (FFABS(predicted_dc
- vul
) > 128)
1314 /* at long last, apply the predictor */
1315 if(s
->coeffs
[i
].index
){
1316 *s
->next_coeff
= s
->coeffs
[i
];
1317 s
->coeffs
[i
].index
=0;
1318 s
->coeffs
[i
].coeff
=0;
1319 s
->coeffs
[i
].next
= s
->next_coeff
++;
1321 s
->coeffs
[i
].coeff
+= predicted_dc
;
1323 last_dc
[current_frame_type
] = DC_COEFF(i
);
1324 if(DC_COEFF(i
) && !(s
->coeff_counts
[i
]&127)){
1325 s
->coeff_counts
[i
]= 129;
1326 // s->all_fragments[i].next_coeff= s->next_coeff;
1327 s
->coeffs
[i
].next
= s
->next_coeff
;
1328 (s
->next_coeff
++)->next
=NULL
;
1336 * Perform the final rendering for a particular slice of data.
1337 * The slice number ranges from 0..(macroblock_height - 1).
1339 static void render_slice(Vp3DecodeContext
*s
, int slice
)
1342 int16_t *dequantizer
;
1343 DECLARE_ALIGNED_16(DCTELEM
, block
[64]);
1344 int motion_x
= 0xdeadbeef, motion_y
= 0xdeadbeef;
1345 int motion_halfpel_index
;
1346 uint8_t *motion_source
;
1348 int current_macroblock_entry
= slice
* s
->macroblock_width
* 6;
1350 if (slice
>= s
->macroblock_height
)
1353 for (plane
= 0; plane
< 3; plane
++) {
1354 uint8_t *output_plane
= s
->current_frame
.data
[plane
];
1355 uint8_t * last_plane
= s
-> last_frame
.data
[plane
];
1356 uint8_t *golden_plane
= s
-> golden_frame
.data
[plane
];
1357 int stride
= s
->current_frame
.linesize
[plane
];
1358 int plane_width
= s
->width
>> !!plane
;
1359 int plane_height
= s
->height
>> !!plane
;
1360 int y
= slice
* FRAGMENT_PIXELS
<< !plane
;
1361 int slice_height
= y
+ (FRAGMENT_PIXELS
<< !plane
);
1362 int i
= s
->macroblock_fragments
[current_macroblock_entry
+ plane
+ 3*!!plane
];
1364 if (!s
->flipped_image
) stride
= -stride
;
1367 if(FFABS(stride
) > 2048)
1368 return; //various tables are fixed size
1370 /* for each fragment row in the slice (both of them)... */
1371 for (; y
< slice_height
; y
+= 8) {
1373 /* for each fragment in a row... */
1374 for (x
= 0; x
< plane_width
; x
+= 8, i
++) {
1376 if ((i
< 0) || (i
>= s
->fragment_count
)) {
1377 av_log(s
->avctx
, AV_LOG_ERROR
, " vp3:render_slice(): bad fragment number (%d)\n", i
);
1381 /* transform if this block was coded */
1382 if ((s
->all_fragments
[i
].coding_method
!= MODE_COPY
) &&
1383 !((s
->avctx
->flags
& CODEC_FLAG_GRAY
) && plane
)) {
1385 if ((s
->all_fragments
[i
].coding_method
== MODE_USING_GOLDEN
) ||
1386 (s
->all_fragments
[i
].coding_method
== MODE_GOLDEN_MV
))
1387 motion_source
= golden_plane
;
1389 motion_source
= last_plane
;
1391 motion_source
+= s
->all_fragments
[i
].first_pixel
;
1392 motion_halfpel_index
= 0;
1394 /* sort out the motion vector if this fragment is coded
1395 * using a motion vector method */
1396 if ((s
->all_fragments
[i
].coding_method
> MODE_INTRA
) &&
1397 (s
->all_fragments
[i
].coding_method
!= MODE_USING_GOLDEN
)) {
1399 motion_x
= s
->all_fragments
[i
].motion_x
;
1400 motion_y
= s
->all_fragments
[i
].motion_y
;
1402 motion_x
= (motion_x
>>1) | (motion_x
&1);
1403 motion_y
= (motion_y
>>1) | (motion_y
&1);
1406 src_x
= (motion_x
>>1) + x
;
1407 src_y
= (motion_y
>>1) + y
;
1408 if ((motion_x
== 127) || (motion_y
== 127))
1409 av_log(s
->avctx
, AV_LOG_ERROR
, " help! got invalid motion vector! (%X, %X)\n", motion_x
, motion_y
);
1411 motion_halfpel_index
= motion_x
& 0x01;
1412 motion_source
+= (motion_x
>> 1);
1414 motion_halfpel_index
|= (motion_y
& 0x01) << 1;
1415 motion_source
+= ((motion_y
>> 1) * stride
);
1417 if(src_x
<0 || src_y
<0 || src_x
+ 9 >= plane_width
|| src_y
+ 9 >= plane_height
){
1418 uint8_t *temp
= s
->edge_emu_buffer
;
1419 if(stride
<0) temp
-= 9*stride
;
1420 else temp
+= 9*stride
;
1422 ff_emulated_edge_mc(temp
, motion_source
, stride
, 9, 9, src_x
, src_y
, plane_width
, plane_height
);
1423 motion_source
= temp
;
1428 /* first, take care of copying a block from either the
1429 * previous or the golden frame */
1430 if (s
->all_fragments
[i
].coding_method
!= MODE_INTRA
) {
1431 /* Note, it is possible to implement all MC cases with
1432 put_no_rnd_pixels_l2 which would look more like the
1433 VP3 source but this would be slower as
1434 put_no_rnd_pixels_tab is better optimzed */
1435 if(motion_halfpel_index
!= 3){
1436 s
->dsp
.put_no_rnd_pixels_tab
[1][motion_halfpel_index
](
1437 output_plane
+ s
->all_fragments
[i
].first_pixel
,
1438 motion_source
, stride
, 8);
1440 int d
= (motion_x
^ motion_y
)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1441 s
->dsp
.put_no_rnd_pixels_l2
[1](
1442 output_plane
+ s
->all_fragments
[i
].first_pixel
,
1444 motion_source
+ stride
+ 1 + d
,
1447 dequantizer
= s
->qmat
[s
->all_fragments
[i
].qpi
][1][plane
];
1449 dequantizer
= s
->qmat
[s
->all_fragments
[i
].qpi
][0][plane
];
1452 /* dequantize the DCT coefficients */
1453 if(s
->avctx
->idct_algo
==FF_IDCT_VP3
){
1454 Coeff
*coeff
= s
->coeffs
+ i
;
1455 s
->dsp
.clear_block(block
);
1457 block
[coeff
->index
]= coeff
->coeff
* dequantizer
[coeff
->index
];
1461 Coeff
*coeff
= s
->coeffs
+ i
;
1462 s
->dsp
.clear_block(block
);
1464 block
[coeff
->index
]= (coeff
->coeff
* dequantizer
[coeff
->index
] + 2)>>2;
1469 /* invert DCT and place (or add) in final output */
1471 if (s
->all_fragments
[i
].coding_method
== MODE_INTRA
) {
1472 if(s
->avctx
->idct_algo
!=FF_IDCT_VP3
)
1475 output_plane
+ s
->all_fragments
[i
].first_pixel
,
1480 output_plane
+ s
->all_fragments
[i
].first_pixel
,
1486 /* copy directly from the previous frame */
1487 s
->dsp
.put_pixels_tab
[1][0](
1488 output_plane
+ s
->all_fragments
[i
].first_pixel
,
1489 last_plane
+ s
->all_fragments
[i
].first_pixel
,
1494 /* perform the left edge filter if:
1495 * - the fragment is not on the left column
1496 * - the fragment is coded in this frame
1497 * - the fragment is not coded in this frame but the left
1498 * fragment is coded in this frame (this is done instead
1499 * of a right edge filter when rendering the left fragment
1500 * since this fragment is not available yet) */
1502 ((s
->all_fragments
[i
].coding_method
!= MODE_COPY
) ||
1503 ((s
->all_fragments
[i
].coding_method
== MODE_COPY
) &&
1504 (s
->all_fragments
[i
- 1].coding_method
!= MODE_COPY
)) )) {
1506 output_plane
+ s
->all_fragments
[i
].first_pixel
+ 7*stride
,
1507 -stride
, s
->bounding_values_array
+ 127);
1510 /* perform the top edge filter if:
1511 * - the fragment is not on the top row
1512 * - the fragment is coded in this frame
1513 * - the fragment is not coded in this frame but the above
1514 * fragment is coded in this frame (this is done instead
1515 * of a bottom edge filter when rendering the above
1516 * fragment since this fragment is not available yet) */
1518 ((s
->all_fragments
[i
].coding_method
!= MODE_COPY
) ||
1519 ((s
->all_fragments
[i
].coding_method
== MODE_COPY
) &&
1520 (s
->all_fragments
[i
- fragment_width
].coding_method
!= MODE_COPY
)) )) {
1522 output_plane
+ s
->all_fragments
[i
].first_pixel
- stride
,
1523 -stride
, s
->bounding_values_array
+ 127);
1530 /* this looks like a good place for slice dispatch... */
1532 * if (slice == s->macroblock_height - 1)
1533 * dispatch (both last slice & 2nd-to-last slice);
1534 * else if (slice > 0)
1535 * dispatch (slice - 1);
1541 static void apply_loop_filter(Vp3DecodeContext
*s
)
1545 int *bounding_values
= s
->bounding_values_array
+127;
1548 int bounding_values_array
[256];
1551 /* find the right loop limit value */
1552 for (x
= 63; x
>= 0; x
--) {
1553 if (vp31_ac_scale_factor
[x
] >= s
->quality_index
)
1556 filter_limit
= vp31_filter_limit_values
[s
->quality_index
];
1558 /* set up the bounding values */
1559 memset(bounding_values_array
, 0, 256 * sizeof(int));
1560 for (x
= 0; x
< filter_limit
; x
++) {
1561 bounding_values
[-x
- filter_limit
] = -filter_limit
+ x
;
1562 bounding_values
[-x
] = -x
;
1563 bounding_values
[x
] = x
;
1564 bounding_values
[x
+ filter_limit
] = filter_limit
- x
;
1568 for (plane
= 0; plane
< 3; plane
++) {
1569 int width
= s
->fragment_width
>> !!plane
;
1570 int height
= s
->fragment_height
>> !!plane
;
1571 int fragment
= s
->fragment_start
[plane
];
1572 int stride
= s
->current_frame
.linesize
[plane
];
1573 uint8_t *plane_data
= s
->current_frame
.data
[plane
];
1574 if (!s
->flipped_image
) stride
= -stride
;
1576 for (y
= 0; y
< height
; y
++) {
1578 for (x
= 0; x
< width
; x
++) {
1579 /* do not perform left edge filter for left columns frags */
1581 (s
->all_fragments
[fragment
].coding_method
!= MODE_COPY
)) {
1582 s
->dsp
.vp3_h_loop_filter(
1583 plane_data
+ s
->all_fragments
[fragment
].first_pixel
,
1584 stride
, bounding_values
);
1587 /* do not perform top edge filter for top row fragments */
1589 (s
->all_fragments
[fragment
].coding_method
!= MODE_COPY
)) {
1590 s
->dsp
.vp3_v_loop_filter(
1591 plane_data
+ s
->all_fragments
[fragment
].first_pixel
,
1592 stride
, bounding_values
);
1595 /* do not perform right edge filter for right column
1596 * fragments or if right fragment neighbor is also coded
1597 * in this frame (it will be filtered in next iteration) */
1598 if ((x
< width
- 1) &&
1599 (s
->all_fragments
[fragment
].coding_method
!= MODE_COPY
) &&
1600 (s
->all_fragments
[fragment
+ 1].coding_method
== MODE_COPY
)) {
1601 s
->dsp
.vp3_h_loop_filter(
1602 plane_data
+ s
->all_fragments
[fragment
+ 1].first_pixel
,
1603 stride
, bounding_values
);
1606 /* do not perform bottom edge filter for bottom row
1607 * fragments or if bottom fragment neighbor is also coded
1608 * in this frame (it will be filtered in the next row) */
1609 if ((y
< height
- 1) &&
1610 (s
->all_fragments
[fragment
].coding_method
!= MODE_COPY
) &&
1611 (s
->all_fragments
[fragment
+ width
].coding_method
== MODE_COPY
)) {
1612 s
->dsp
.vp3_v_loop_filter(
1613 plane_data
+ s
->all_fragments
[fragment
+ width
].first_pixel
,
1614 stride
, bounding_values
);
1624 * This function computes the first pixel addresses for each fragment.
1625 * This function needs to be invoked after the first frame is allocated
1626 * so that it has access to the plane strides.
1628 static void vp3_calculate_pixel_addresses(Vp3DecodeContext
*s
)
1630 #define Y_INITIAL(chroma_shift) s->flipped_image ? 1 : s->fragment_height >> chroma_shift
1631 #define Y_FINISHED(chroma_shift) s->flipped_image ? y <= s->fragment_height >> chroma_shift : y > 0
1634 const int y_inc
= s
->flipped_image
? 1 : -1;
1636 /* figure out the first pixel addresses for each of the fragments */
1639 for (y
= Y_INITIAL(0); Y_FINISHED(0); y
+= y_inc
) {
1640 for (x
= 0; x
< s
->fragment_width
; x
++) {
1641 s
->all_fragments
[i
++].first_pixel
=
1642 s
->golden_frame
.linesize
[0] * y
* FRAGMENT_PIXELS
-
1643 s
->golden_frame
.linesize
[0] +
1644 x
* FRAGMENT_PIXELS
;
1649 i
= s
->fragment_start
[1];
1650 for (y
= Y_INITIAL(1); Y_FINISHED(1); y
+= y_inc
) {
1651 for (x
= 0; x
< s
->fragment_width
/ 2; x
++) {
1652 s
->all_fragments
[i
++].first_pixel
=
1653 s
->golden_frame
.linesize
[1] * y
* FRAGMENT_PIXELS
-
1654 s
->golden_frame
.linesize
[1] +
1655 x
* FRAGMENT_PIXELS
;
1660 i
= s
->fragment_start
[2];
1661 for (y
= Y_INITIAL(1); Y_FINISHED(1); y
+= y_inc
) {
1662 for (x
= 0; x
< s
->fragment_width
/ 2; x
++) {
1663 s
->all_fragments
[i
++].first_pixel
=
1664 s
->golden_frame
.linesize
[2] * y
* FRAGMENT_PIXELS
-
1665 s
->golden_frame
.linesize
[2] +
1666 x
* FRAGMENT_PIXELS
;
1672 * This is the ffmpeg/libavcodec API init function.
1674 static av_cold
int vp3_decode_init(AVCodecContext
*avctx
)
1676 Vp3DecodeContext
*s
= avctx
->priv_data
;
1677 int i
, inter
, plane
;
1680 int y_superblock_count
;
1681 int c_superblock_count
;
1683 if (avctx
->codec_tag
== MKTAG('V','P','3','0'))
1689 s
->width
= FFALIGN(avctx
->width
, 16);
1690 s
->height
= FFALIGN(avctx
->height
, 16);
1691 avctx
->pix_fmt
= PIX_FMT_YUV420P
;
1692 avctx
->chroma_sample_location
= AVCHROMA_LOC_CENTER
;
1693 if(avctx
->idct_algo
==FF_IDCT_AUTO
)
1694 avctx
->idct_algo
=FF_IDCT_VP3
;
1695 dsputil_init(&s
->dsp
, avctx
);
1697 ff_init_scantable(s
->dsp
.idct_permutation
, &s
->scantable
, ff_zigzag_direct
);
1699 /* initialize to an impossible value which will force a recalculation
1700 * in the first frame decode */
1701 for (i
= 0; i
< 3; i
++)
1704 s
->y_superblock_width
= (s
->width
+ 31) / 32;
1705 s
->y_superblock_height
= (s
->height
+ 31) / 32;
1706 y_superblock_count
= s
->y_superblock_width
* s
->y_superblock_height
;
1708 /* work out the dimensions for the C planes */
1709 c_width
= s
->width
/ 2;
1710 c_height
= s
->height
/ 2;
1711 s
->c_superblock_width
= (c_width
+ 31) / 32;
1712 s
->c_superblock_height
= (c_height
+ 31) / 32;
1713 c_superblock_count
= s
->c_superblock_width
* s
->c_superblock_height
;
1715 s
->superblock_count
= y_superblock_count
+ (c_superblock_count
* 2);
1716 s
->u_superblock_start
= y_superblock_count
;
1717 s
->v_superblock_start
= s
->u_superblock_start
+ c_superblock_count
;
1718 s
->superblock_coding
= av_malloc(s
->superblock_count
);
1720 s
->macroblock_width
= (s
->width
+ 15) / 16;
1721 s
->macroblock_height
= (s
->height
+ 15) / 16;
1722 s
->macroblock_count
= s
->macroblock_width
* s
->macroblock_height
;
1724 s
->fragment_width
= s
->width
/ FRAGMENT_PIXELS
;
1725 s
->fragment_height
= s
->height
/ FRAGMENT_PIXELS
;
1727 /* fragment count covers all 8x8 blocks for all 3 planes */
1728 s
->fragment_count
= s
->fragment_width
* s
->fragment_height
* 3 / 2;
1729 s
->fragment_start
[1] = s
->fragment_width
* s
->fragment_height
;
1730 s
->fragment_start
[2] = s
->fragment_width
* s
->fragment_height
* 5 / 4;
1732 s
->all_fragments
= av_malloc(s
->fragment_count
* sizeof(Vp3Fragment
));
1733 s
->coeff_counts
= av_malloc(s
->fragment_count
* sizeof(*s
->coeff_counts
));
1734 s
->coeffs
= av_malloc(s
->fragment_count
* sizeof(Coeff
) * 65);
1735 s
->coded_fragment_list
= av_malloc(s
->fragment_count
* sizeof(int));
1736 s
->pixel_addresses_initialized
= 0;
1738 if (!s
->theora_tables
)
1740 for (i
= 0; i
< 64; i
++) {
1741 s
->coded_dc_scale_factor
[i
] = vp31_dc_scale_factor
[i
];
1742 s
->coded_ac_scale_factor
[i
] = vp31_ac_scale_factor
[i
];
1743 s
->base_matrix
[0][i
] = vp31_intra_y_dequant
[i
];
1744 s
->base_matrix
[1][i
] = vp31_intra_c_dequant
[i
];
1745 s
->base_matrix
[2][i
] = vp31_inter_dequant
[i
];
1746 s
->filter_limit_values
[i
] = vp31_filter_limit_values
[i
];
1749 for(inter
=0; inter
<2; inter
++){
1750 for(plane
=0; plane
<3; plane
++){
1751 s
->qr_count
[inter
][plane
]= 1;
1752 s
->qr_size
[inter
][plane
][0]= 63;
1753 s
->qr_base
[inter
][plane
][0]=
1754 s
->qr_base
[inter
][plane
][1]= 2*inter
+ (!!plane
)*!inter
;
1758 /* init VLC tables */
1759 for (i
= 0; i
< 16; i
++) {
1762 init_vlc(&s
->dc_vlc
[i
], 5, 32,
1763 &dc_bias
[i
][0][1], 4, 2,
1764 &dc_bias
[i
][0][0], 4, 2, 0);
1766 /* group 1 AC histograms */
1767 init_vlc(&s
->ac_vlc_1
[i
], 5, 32,
1768 &ac_bias_0
[i
][0][1], 4, 2,
1769 &ac_bias_0
[i
][0][0], 4, 2, 0);
1771 /* group 2 AC histograms */
1772 init_vlc(&s
->ac_vlc_2
[i
], 5, 32,
1773 &ac_bias_1
[i
][0][1], 4, 2,
1774 &ac_bias_1
[i
][0][0], 4, 2, 0);
1776 /* group 3 AC histograms */
1777 init_vlc(&s
->ac_vlc_3
[i
], 5, 32,
1778 &ac_bias_2
[i
][0][1], 4, 2,
1779 &ac_bias_2
[i
][0][0], 4, 2, 0);
1781 /* group 4 AC histograms */
1782 init_vlc(&s
->ac_vlc_4
[i
], 5, 32,
1783 &ac_bias_3
[i
][0][1], 4, 2,
1784 &ac_bias_3
[i
][0][0], 4, 2, 0);
1787 for (i
= 0; i
< 16; i
++) {
1790 if (init_vlc(&s
->dc_vlc
[i
], 5, 32,
1791 &s
->huffman_table
[i
][0][1], 4, 2,
1792 &s
->huffman_table
[i
][0][0], 4, 2, 0) < 0)
1795 /* group 1 AC histograms */
1796 if (init_vlc(&s
->ac_vlc_1
[i
], 5, 32,
1797 &s
->huffman_table
[i
+16][0][1], 4, 2,
1798 &s
->huffman_table
[i
+16][0][0], 4, 2, 0) < 0)
1801 /* group 2 AC histograms */
1802 if (init_vlc(&s
->ac_vlc_2
[i
], 5, 32,
1803 &s
->huffman_table
[i
+16*2][0][1], 4, 2,
1804 &s
->huffman_table
[i
+16*2][0][0], 4, 2, 0) < 0)
1807 /* group 3 AC histograms */
1808 if (init_vlc(&s
->ac_vlc_3
[i
], 5, 32,
1809 &s
->huffman_table
[i
+16*3][0][1], 4, 2,
1810 &s
->huffman_table
[i
+16*3][0][0], 4, 2, 0) < 0)
1813 /* group 4 AC histograms */
1814 if (init_vlc(&s
->ac_vlc_4
[i
], 5, 32,
1815 &s
->huffman_table
[i
+16*4][0][1], 4, 2,
1816 &s
->huffman_table
[i
+16*4][0][0], 4, 2, 0) < 0)
1821 init_vlc(&s
->superblock_run_length_vlc
, 6, 34,
1822 &superblock_run_length_vlc_table
[0][1], 4, 2,
1823 &superblock_run_length_vlc_table
[0][0], 4, 2, 0);
1825 init_vlc(&s
->fragment_run_length_vlc
, 5, 30,
1826 &fragment_run_length_vlc_table
[0][1], 4, 2,
1827 &fragment_run_length_vlc_table
[0][0], 4, 2, 0);
1829 init_vlc(&s
->mode_code_vlc
, 3, 8,
1830 &mode_code_vlc_table
[0][1], 2, 1,
1831 &mode_code_vlc_table
[0][0], 2, 1, 0);
1833 init_vlc(&s
->motion_vector_vlc
, 6, 63,
1834 &motion_vector_vlc_table
[0][1], 2, 1,
1835 &motion_vector_vlc_table
[0][0], 2, 1, 0);
1837 /* work out the block mapping tables */
1838 s
->superblock_fragments
= av_malloc(s
->superblock_count
* 16 * sizeof(int));
1839 s
->superblock_macroblocks
= av_malloc(s
->superblock_count
* 4 * sizeof(int));
1840 s
->macroblock_fragments
= av_malloc(s
->macroblock_count
* 6 * sizeof(int));
1841 s
->macroblock_coding
= av_malloc(s
->macroblock_count
+ 1);
1842 init_block_mapping(s
);
1844 for (i
= 0; i
< 3; i
++) {
1845 s
->current_frame
.data
[i
] = NULL
;
1846 s
->last_frame
.data
[i
] = NULL
;
1847 s
->golden_frame
.data
[i
] = NULL
;
1853 av_log(avctx
, AV_LOG_FATAL
, "Invalid huffman table\n");
1858 * This is the ffmpeg/libavcodec API frame decode function.
1860 static int vp3_decode_frame(AVCodecContext
*avctx
,
1861 void *data
, int *data_size
,
1864 const uint8_t *buf
= avpkt
->data
;
1865 int buf_size
= avpkt
->size
;
1866 Vp3DecodeContext
*s
= avctx
->priv_data
;
1868 static int counter
= 0;
1871 init_get_bits(&gb
, buf
, buf_size
* 8);
1873 if (s
->theora
&& get_bits1(&gb
))
1875 av_log(avctx
, AV_LOG_ERROR
, "Header packet passed to frame decoder, skipping\n");
1879 s
->keyframe
= !get_bits1(&gb
);
1882 for (i
= 0; i
< 3; i
++)
1883 s
->last_qps
[i
] = s
->qps
[i
];
1887 s
->qps
[s
->nqps
++]= get_bits(&gb
, 6);
1888 } while(s
->theora
>= 0x030200 && s
->nqps
<3 && get_bits1(&gb
));
1889 for (i
= s
->nqps
; i
< 3; i
++)
1892 if (s
->avctx
->debug
& FF_DEBUG_PICT_INFO
)
1893 av_log(s
->avctx
, AV_LOG_INFO
, " VP3 %sframe #%d: Q index = %d\n",
1894 s
->keyframe
?"key":"", counter
, s
->qps
[0]);
1897 if (s
->qps
[0] != s
->last_qps
[0])
1898 init_loop_filter(s
);
1900 for (i
= 0; i
< s
->nqps
; i
++)
1901 // reinit all dequantizers if the first one changed, because
1902 // the DC of the first quantizer must be used for all matrices
1903 if (s
->qps
[i
] != s
->last_qps
[i
] || s
->qps
[0] != s
->last_qps
[0])
1904 init_dequantizer(s
, i
);
1906 if (avctx
->skip_frame
>= AVDISCARD_NONKEY
&& !s
->keyframe
)
1912 skip_bits(&gb
, 4); /* width code */
1913 skip_bits(&gb
, 4); /* height code */
1916 s
->version
= get_bits(&gb
, 5);
1918 av_log(s
->avctx
, AV_LOG_DEBUG
, "VP version: %d\n", s
->version
);
1921 if (s
->version
|| s
->theora
)
1924 av_log(s
->avctx
, AV_LOG_ERROR
, "Warning, unsupported keyframe coding type?!\n");
1925 skip_bits(&gb
, 2); /* reserved? */
1928 if (s
->last_frame
.data
[0] == s
->golden_frame
.data
[0]) {
1929 if (s
->golden_frame
.data
[0])
1930 avctx
->release_buffer(avctx
, &s
->golden_frame
);
1931 s
->last_frame
= s
->golden_frame
; /* ensure that we catch any access to this released frame */
1933 if (s
->golden_frame
.data
[0])
1934 avctx
->release_buffer(avctx
, &s
->golden_frame
);
1935 if (s
->last_frame
.data
[0])
1936 avctx
->release_buffer(avctx
, &s
->last_frame
);
1939 s
->golden_frame
.reference
= 3;
1940 if(avctx
->get_buffer(avctx
, &s
->golden_frame
) < 0) {
1941 av_log(s
->avctx
, AV_LOG_ERROR
, "vp3: get_buffer() failed\n");
1945 /* golden frame is also the current frame */
1946 s
->current_frame
= s
->golden_frame
;
1948 /* time to figure out pixel addresses? */
1949 if (!s
->pixel_addresses_initialized
)
1951 vp3_calculate_pixel_addresses(s
);
1952 s
->pixel_addresses_initialized
= 1;
1955 /* allocate a new current frame */
1956 s
->current_frame
.reference
= 3;
1957 if (!s
->pixel_addresses_initialized
) {
1958 av_log(s
->avctx
, AV_LOG_ERROR
, "vp3: first frame not a keyframe\n");
1961 if(avctx
->get_buffer(avctx
, &s
->current_frame
) < 0) {
1962 av_log(s
->avctx
, AV_LOG_ERROR
, "vp3: get_buffer() failed\n");
1967 s
->current_frame
.qscale_table
= s
->qscale_table
; //FIXME allocate individual tables per AVFrame
1968 s
->current_frame
.qstride
= 0;
1972 if (unpack_superblocks(s
, &gb
)){
1973 av_log(s
->avctx
, AV_LOG_ERROR
, "error in unpack_superblocks\n");
1976 if (unpack_modes(s
, &gb
)){
1977 av_log(s
->avctx
, AV_LOG_ERROR
, "error in unpack_modes\n");
1980 if (unpack_vectors(s
, &gb
)){
1981 av_log(s
->avctx
, AV_LOG_ERROR
, "error in unpack_vectors\n");
1984 if (unpack_block_qpis(s
, &gb
)){
1985 av_log(s
->avctx
, AV_LOG_ERROR
, "error in unpack_block_qpis\n");
1988 if (unpack_dct_coeffs(s
, &gb
)){
1989 av_log(s
->avctx
, AV_LOG_ERROR
, "error in unpack_dct_coeffs\n");
1993 reverse_dc_prediction(s
, 0, s
->fragment_width
, s
->fragment_height
);
1994 if ((avctx
->flags
& CODEC_FLAG_GRAY
) == 0) {
1995 reverse_dc_prediction(s
, s
->fragment_start
[1],
1996 s
->fragment_width
/ 2, s
->fragment_height
/ 2);
1997 reverse_dc_prediction(s
, s
->fragment_start
[2],
1998 s
->fragment_width
/ 2, s
->fragment_height
/ 2);
2001 for (i
= 0; i
< s
->macroblock_height
; i
++)
2004 apply_loop_filter(s
);
2006 *data_size
=sizeof(AVFrame
);
2007 *(AVFrame
*)data
= s
->current_frame
;
2009 /* release the last frame, if it is allocated and if it is not the
2011 if ((s
->last_frame
.data
[0]) &&
2012 (s
->last_frame
.data
[0] != s
->golden_frame
.data
[0]))
2013 avctx
->release_buffer(avctx
, &s
->last_frame
);
2015 /* shuffle frames (last = current) */
2016 s
->last_frame
= s
->current_frame
;
2017 s
->current_frame
.data
[0]= NULL
; /* ensure that we catch any access to this released frame */
2023 * This is the ffmpeg/libavcodec API module cleanup function.
2025 static av_cold
int vp3_decode_end(AVCodecContext
*avctx
)
2027 Vp3DecodeContext
*s
= avctx
->priv_data
;
2030 av_free(s
->superblock_coding
);
2031 av_free(s
->all_fragments
);
2032 av_free(s
->coeff_counts
);
2034 av_free(s
->coded_fragment_list
);
2035 av_free(s
->superblock_fragments
);
2036 av_free(s
->superblock_macroblocks
);
2037 av_free(s
->macroblock_fragments
);
2038 av_free(s
->macroblock_coding
);
2040 for (i
= 0; i
< 16; i
++) {
2041 free_vlc(&s
->dc_vlc
[i
]);
2042 free_vlc(&s
->ac_vlc_1
[i
]);
2043 free_vlc(&s
->ac_vlc_2
[i
]);
2044 free_vlc(&s
->ac_vlc_3
[i
]);
2045 free_vlc(&s
->ac_vlc_4
[i
]);
2048 free_vlc(&s
->superblock_run_length_vlc
);
2049 free_vlc(&s
->fragment_run_length_vlc
);
2050 free_vlc(&s
->mode_code_vlc
);
2051 free_vlc(&s
->motion_vector_vlc
);
2053 /* release all frames */
2054 if (s
->golden_frame
.data
[0] && s
->golden_frame
.data
[0] != s
->last_frame
.data
[0])
2055 avctx
->release_buffer(avctx
, &s
->golden_frame
);
2056 if (s
->last_frame
.data
[0])
2057 avctx
->release_buffer(avctx
, &s
->last_frame
);
2058 /* no need to release the current_frame since it will always be pointing
2059 * to the same frame as either the golden or last frame */
2064 static int read_huffman_tree(AVCodecContext
*avctx
, GetBitContext
*gb
)
2066 Vp3DecodeContext
*s
= avctx
->priv_data
;
2068 if (get_bits1(gb
)) {
2070 if (s
->entries
>= 32) { /* overflow */
2071 av_log(avctx
, AV_LOG_ERROR
, "huffman tree overflow\n");
2074 token
= get_bits(gb
, 5);
2075 //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);
2076 s
->huffman_table
[s
->hti
][token
][0] = s
->hbits
;
2077 s
->huffman_table
[s
->hti
][token
][1] = s
->huff_code_size
;
2081 if (s
->huff_code_size
>= 32) {/* overflow */
2082 av_log(avctx
, AV_LOG_ERROR
, "huffman tree overflow\n");
2085 s
->huff_code_size
++;
2087 if (read_huffman_tree(avctx
, gb
))
2090 if (read_huffman_tree(avctx
, gb
))
2093 s
->huff_code_size
--;
2098 #if CONFIG_THEORA_DECODER
2099 static int theora_decode_header(AVCodecContext
*avctx
, GetBitContext
*gb
)
2101 Vp3DecodeContext
*s
= avctx
->priv_data
;
2102 int visible_width
, visible_height
;
2104 s
->theora
= get_bits_long(gb
, 24);
2105 av_log(avctx
, AV_LOG_DEBUG
, "Theora bitstream version %X\n", s
->theora
);
2107 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2108 /* but previous versions have the image flipped relative to vp3 */
2109 if (s
->theora
< 0x030200)
2111 s
->flipped_image
= 1;
2112 av_log(avctx
, AV_LOG_DEBUG
, "Old (<alpha3) Theora bitstream, flipped image\n");
2115 visible_width
= s
->width
= get_bits(gb
, 16) << 4;
2116 visible_height
= s
->height
= get_bits(gb
, 16) << 4;
2118 if(avcodec_check_dimensions(avctx
, s
->width
, s
->height
)){
2119 av_log(avctx
, AV_LOG_ERROR
, "Invalid dimensions (%dx%d)\n", s
->width
, s
->height
);
2120 s
->width
= s
->height
= 0;
2124 if (s
->theora
>= 0x030400)
2126 skip_bits(gb
, 32); /* total number of superblocks in a frame */
2127 // fixme, the next field is 36bits long
2128 skip_bits(gb
, 32); /* total number of blocks in a frame */
2129 skip_bits(gb
, 4); /* total number of blocks in a frame */
2130 skip_bits(gb
, 32); /* total number of macroblocks in a frame */
2133 if (s
->theora
>= 0x030200) {
2134 visible_width
= get_bits_long(gb
, 24);
2135 visible_height
= get_bits_long(gb
, 24);
2137 skip_bits(gb
, 8); /* offset x */
2138 skip_bits(gb
, 8); /* offset y */
2141 skip_bits(gb
, 32); /* fps numerator */
2142 skip_bits(gb
, 32); /* fps denumerator */
2143 skip_bits(gb
, 24); /* aspect numerator */
2144 skip_bits(gb
, 24); /* aspect denumerator */
2146 if (s
->theora
< 0x030200)
2147 skip_bits(gb
, 5); /* keyframe frequency force */
2148 skip_bits(gb
, 8); /* colorspace */
2149 if (s
->theora
>= 0x030400)
2150 skip_bits(gb
, 2); /* pixel format: 420,res,422,444 */
2151 skip_bits(gb
, 24); /* bitrate */
2153 skip_bits(gb
, 6); /* quality hint */
2155 if (s
->theora
>= 0x030200)
2157 skip_bits(gb
, 5); /* keyframe frequency force */
2159 if (s
->theora
< 0x030400)
2160 skip_bits(gb
, 5); /* spare bits */
2163 // align_get_bits(gb);
2165 if ( visible_width
<= s
->width
&& visible_width
> s
->width
-16
2166 && visible_height
<= s
->height
&& visible_height
> s
->height
-16)
2167 avcodec_set_dimensions(avctx
, visible_width
, visible_height
);
2169 avcodec_set_dimensions(avctx
, s
->width
, s
->height
);
2174 static int theora_decode_tables(AVCodecContext
*avctx
, GetBitContext
*gb
)
2176 Vp3DecodeContext
*s
= avctx
->priv_data
;
2177 int i
, n
, matrices
, inter
, plane
;
2179 if (s
->theora
>= 0x030200) {
2180 n
= get_bits(gb
, 3);
2181 /* loop filter limit values table */
2182 for (i
= 0; i
< 64; i
++) {
2183 s
->filter_limit_values
[i
] = get_bits(gb
, n
);
2184 if (s
->filter_limit_values
[i
] > 127) {
2185 av_log(avctx
, AV_LOG_ERROR
, "filter limit value too large (%i > 127), clamping\n", s
->filter_limit_values
[i
]);
2186 s
->filter_limit_values
[i
] = 127;
2191 if (s
->theora
>= 0x030200)
2192 n
= get_bits(gb
, 4) + 1;
2195 /* quality threshold table */
2196 for (i
= 0; i
< 64; i
++)
2197 s
->coded_ac_scale_factor
[i
] = get_bits(gb
, n
);
2199 if (s
->theora
>= 0x030200)
2200 n
= get_bits(gb
, 4) + 1;
2203 /* dc scale factor table */
2204 for (i
= 0; i
< 64; i
++)
2205 s
->coded_dc_scale_factor
[i
] = get_bits(gb
, n
);
2207 if (s
->theora
>= 0x030200)
2208 matrices
= get_bits(gb
, 9) + 1;
2213 av_log(avctx
, AV_LOG_ERROR
, "invalid number of base matrixes\n");
2217 for(n
=0; n
<matrices
; n
++){
2218 for (i
= 0; i
< 64; i
++)
2219 s
->base_matrix
[n
][i
]= get_bits(gb
, 8);
2222 for (inter
= 0; inter
<= 1; inter
++) {
2223 for (plane
= 0; plane
<= 2; plane
++) {
2225 if (inter
|| plane
> 0)
2226 newqr
= get_bits1(gb
);
2229 if(inter
&& get_bits1(gb
)){
2233 qtj
= (3*inter
+ plane
- 1) / 3;
2234 plj
= (plane
+ 2) % 3;
2236 s
->qr_count
[inter
][plane
]= s
->qr_count
[qtj
][plj
];
2237 memcpy(s
->qr_size
[inter
][plane
], s
->qr_size
[qtj
][plj
], sizeof(s
->qr_size
[0][0]));
2238 memcpy(s
->qr_base
[inter
][plane
], s
->qr_base
[qtj
][plj
], sizeof(s
->qr_base
[0][0]));
2244 i
= get_bits(gb
, av_log2(matrices
-1)+1);
2246 av_log(avctx
, AV_LOG_ERROR
, "invalid base matrix index\n");
2249 s
->qr_base
[inter
][plane
][qri
]= i
;
2252 i
= get_bits(gb
, av_log2(63-qi
)+1) + 1;
2253 s
->qr_size
[inter
][plane
][qri
++]= i
;
2258 av_log(avctx
, AV_LOG_ERROR
, "invalid qi %d > 63\n", qi
);
2261 s
->qr_count
[inter
][plane
]= qri
;
2266 /* Huffman tables */
2267 for (s
->hti
= 0; s
->hti
< 80; s
->hti
++) {
2269 s
->huff_code_size
= 1;
2270 if (!get_bits1(gb
)) {
2272 if(read_huffman_tree(avctx
, gb
))
2275 if(read_huffman_tree(avctx
, gb
))
2280 s
->theora_tables
= 1;
2285 static av_cold
int theora_decode_init(AVCodecContext
*avctx
)
2287 Vp3DecodeContext
*s
= avctx
->priv_data
;
2290 uint8_t *header_start
[3];
2296 if (!avctx
->extradata_size
)
2298 av_log(avctx
, AV_LOG_ERROR
, "Missing extradata!\n");
2302 if (ff_split_xiph_headers(avctx
->extradata
, avctx
->extradata_size
,
2303 42, header_start
, header_len
) < 0) {
2304 av_log(avctx
, AV_LOG_ERROR
, "Corrupt extradata\n");
2309 init_get_bits(&gb
, header_start
[i
], header_len
[i
]);
2311 ptype
= get_bits(&gb
, 8);
2313 if (!(ptype
& 0x80))
2315 av_log(avctx
, AV_LOG_ERROR
, "Invalid extradata!\n");
2319 // FIXME: Check for this as well.
2320 skip_bits_long(&gb
, 6*8); /* "theora" */
2325 theora_decode_header(avctx
, &gb
);
2328 // FIXME: is this needed? it breaks sometimes
2329 // theora_decode_comments(avctx, gb);
2332 if (theora_decode_tables(avctx
, &gb
))
2336 av_log(avctx
, AV_LOG_ERROR
, "Unknown Theora config packet: %d\n", ptype
&~0x80);
2339 if(ptype
!= 0x81 && 8*header_len
[i
] != get_bits_count(&gb
))
2340 av_log(avctx
, AV_LOG_WARNING
, "%d bits left in packet %X\n", 8*header_len
[i
] - get_bits_count(&gb
), ptype
);
2341 if (s
->theora
< 0x030200)
2345 return vp3_decode_init(avctx
);
2348 AVCodec theora_decoder
= {
2352 sizeof(Vp3DecodeContext
),
2359 .long_name
= NULL_IF_CONFIG_SMALL("Theora"),
2363 AVCodec vp3_decoder
= {
2367 sizeof(Vp3DecodeContext
),
2374 .long_name
= NULL_IF_CONFIG_SMALL("On2 VP3"),