3 * Copyright (c) 2002 Fabrice Bellard.
4 * Copyright (c) 2004 Roman Shaposhnik.
7 * Copyright (c) 2003 Roman Shaposhnik.
9 * 50 Mbps (DVCPRO50) support
10 * Copyright (c) 2006 Daniel Maas <dmaas@maasdigital.com>
12 * Many thanks to Dan Dennedy <dan@dennedy.org> for providing wealth
13 * of DV technical info.
15 * This file is part of FFmpeg.
17 * FFmpeg is free software; you can redistribute it and/or
18 * modify it under the terms of the GNU Lesser General Public
19 * License as published by the Free Software Foundation; either
20 * version 2.1 of the License, or (at your option) any later version.
22 * FFmpeg is distributed in the hope that it will be useful,
23 * but WITHOUT ANY WARRANTY; without even the implied warranty of
24 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
25 * Lesser General Public License for more details.
27 * You should have received a copy of the GNU Lesser General Public
28 * License along with FFmpeg; if not, write to the Free Software
29 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
36 #define ALT_BITSTREAM_READER
39 #include "mpegvideo.h"
40 #include "simple_idct.h"
46 typedef struct DVVideoContext
{
49 AVCodecContext
*avctx
;
52 uint8_t dv_zigzag
[2][64];
53 uint8_t dv_idct_shift
[2][2][22][64];
55 void (*get_pixels
)(DCTELEM
*block
, const uint8_t *pixels
, int line_size
);
56 void (*fdct
[2])(DCTELEM
*block
);
57 void (*idct_put
[2])(uint8_t *dest
, int line_size
, DCTELEM
*block
);
60 /* MultiThreading - dv_anchor applies to entire DV codec, not just the avcontext */
61 /* one element is needed for each video segment in a DV frame */
62 /* at most there are 2 DIF channels * 12 DIF sequences * 27 video segments (PAL 50Mbps) */
63 #define DV_ANCHOR_SIZE (2*12*27)
65 static void* dv_anchor
[DV_ANCHOR_SIZE
];
67 #define TEX_VLC_BITS 9
69 #ifdef DV_CODEC_TINY_TARGET
70 #define DV_VLC_MAP_RUN_SIZE 15
71 #define DV_VLC_MAP_LEV_SIZE 23
73 #define DV_VLC_MAP_RUN_SIZE 64
74 #define DV_VLC_MAP_LEV_SIZE 512 //FIXME sign was removed so this should be /2 but needs check
77 /* XXX: also include quantization */
78 static RL_VLC_ELEM dv_rl_vlc
[1184];
79 /* VLC encoding lookup table */
80 static struct dv_vlc_pair
{
83 } dv_vlc_map
[DV_VLC_MAP_RUN_SIZE
][DV_VLC_MAP_LEV_SIZE
];
85 static void dv_build_unquantize_tables(DVVideoContext
*s
, uint8_t* perm
)
89 /* NOTE: max left shift is 6 */
90 for(q
= 0; q
< 22; q
++) {
92 for(i
= 1; i
< 64; i
++) {
95 s
->dv_idct_shift
[0][0][q
][j
] =
96 dv_quant_shifts
[q
][dv_88_areas
[i
]] + 1;
97 s
->dv_idct_shift
[1][0][q
][j
] = s
->dv_idct_shift
[0][0][q
][j
] + 1;
101 for(i
= 1; i
< 64; i
++) {
103 s
->dv_idct_shift
[0][1][q
][i
] =
104 dv_quant_shifts
[q
][dv_248_areas
[i
]] + 1;
105 s
->dv_idct_shift
[1][1][q
][i
] = s
->dv_idct_shift
[0][1][q
][i
] + 1;
110 static int dvvideo_init(AVCodecContext
*avctx
)
112 DVVideoContext
*s
= avctx
->priv_data
;
119 uint16_t new_dv_vlc_bits
[NB_DV_VLC
*2];
120 uint8_t new_dv_vlc_len
[NB_DV_VLC
*2];
121 uint8_t new_dv_vlc_run
[NB_DV_VLC
*2];
122 int16_t new_dv_vlc_level
[NB_DV_VLC
*2];
126 /* dv_anchor lets each thread know its Id */
127 for (i
=0; i
<DV_ANCHOR_SIZE
; i
++)
128 dv_anchor
[i
] = (void*)(size_t)i
;
130 /* it's faster to include sign bit in a generic VLC parsing scheme */
131 for (i
=0, j
=0; i
<NB_DV_VLC
; i
++, j
++) {
132 new_dv_vlc_bits
[j
] = dv_vlc_bits
[i
];
133 new_dv_vlc_len
[j
] = dv_vlc_len
[i
];
134 new_dv_vlc_run
[j
] = dv_vlc_run
[i
];
135 new_dv_vlc_level
[j
] = dv_vlc_level
[i
];
137 if (dv_vlc_level
[i
]) {
138 new_dv_vlc_bits
[j
] <<= 1;
142 new_dv_vlc_bits
[j
] = (dv_vlc_bits
[i
] << 1) | 1;
143 new_dv_vlc_len
[j
] = dv_vlc_len
[i
] + 1;
144 new_dv_vlc_run
[j
] = dv_vlc_run
[i
];
145 new_dv_vlc_level
[j
] = -dv_vlc_level
[i
];
149 /* NOTE: as a trick, we use the fact the no codes are unused
150 to accelerate the parsing of partial codes */
151 init_vlc(&dv_vlc
, TEX_VLC_BITS
, j
,
152 new_dv_vlc_len
, 1, 1, new_dv_vlc_bits
, 2, 2, 0);
153 assert(dv_vlc
.table_size
== 1184);
155 for(i
= 0; i
< dv_vlc
.table_size
; i
++){
156 int code
= dv_vlc
.table
[i
][0];
157 int len
= dv_vlc
.table
[i
][1];
160 if(len
<0){ //more bits needed
164 run
= new_dv_vlc_run
[code
] + 1;
165 level
= new_dv_vlc_level
[code
];
167 dv_rl_vlc
[i
].len
= len
;
168 dv_rl_vlc
[i
].level
= level
;
169 dv_rl_vlc
[i
].run
= run
;
173 for (i
= 0; i
< NB_DV_VLC
- 1; i
++) {
174 if (dv_vlc_run
[i
] >= DV_VLC_MAP_RUN_SIZE
)
176 #ifdef DV_CODEC_TINY_TARGET
177 if (dv_vlc_level
[i
] >= DV_VLC_MAP_LEV_SIZE
)
181 if (dv_vlc_map
[dv_vlc_run
[i
]][dv_vlc_level
[i
]].size
!= 0)
184 dv_vlc_map
[dv_vlc_run
[i
]][dv_vlc_level
[i
]].vlc
= dv_vlc_bits
[i
] <<
186 dv_vlc_map
[dv_vlc_run
[i
]][dv_vlc_level
[i
]].size
= dv_vlc_len
[i
] +
189 for (i
= 0; i
< DV_VLC_MAP_RUN_SIZE
; i
++) {
190 #ifdef DV_CODEC_TINY_TARGET
191 for (j
= 1; j
< DV_VLC_MAP_LEV_SIZE
; j
++) {
192 if (dv_vlc_map
[i
][j
].size
== 0) {
193 dv_vlc_map
[i
][j
].vlc
= dv_vlc_map
[0][j
].vlc
|
194 (dv_vlc_map
[i
-1][0].vlc
<< (dv_vlc_map
[0][j
].size
));
195 dv_vlc_map
[i
][j
].size
= dv_vlc_map
[i
-1][0].size
+
196 dv_vlc_map
[0][j
].size
;
200 for (j
= 1; j
< DV_VLC_MAP_LEV_SIZE
/2; j
++) {
201 if (dv_vlc_map
[i
][j
].size
== 0) {
202 dv_vlc_map
[i
][j
].vlc
= dv_vlc_map
[0][j
].vlc
|
203 (dv_vlc_map
[i
-1][0].vlc
<< (dv_vlc_map
[0][j
].size
));
204 dv_vlc_map
[i
][j
].size
= dv_vlc_map
[i
-1][0].size
+
205 dv_vlc_map
[0][j
].size
;
207 dv_vlc_map
[i
][((uint16_t)(-j
))&0x1ff].vlc
=
208 dv_vlc_map
[i
][j
].vlc
| 1;
209 dv_vlc_map
[i
][((uint16_t)(-j
))&0x1ff].size
=
210 dv_vlc_map
[i
][j
].size
;
216 /* Generic DSP setup */
217 dsputil_init(&dsp
, avctx
);
218 s
->get_pixels
= dsp
.get_pixels
;
221 s
->fdct
[0] = dsp
.fdct
;
222 s
->idct_put
[0] = dsp
.idct_put
;
224 s
->dv_zigzag
[0][i
] = dsp
.idct_permutation
[ff_zigzag_direct
[i
]];
227 s
->fdct
[1] = dsp
.fdct248
;
228 s
->idct_put
[1] = ff_simple_idct248_put
; // FIXME: need to add it to DSP
230 for (i
=0; i
<64; i
++){
231 int j
= ff_zigzag248_direct
[i
];
232 s
->dv_zigzag
[1][i
] = dsp
.idct_permutation
[(j
&7) + (j
&8)*4 + (j
&48)/2];
235 memcpy(s
->dv_zigzag
[1], ff_zigzag248_direct
, 64);
237 /* XXX: do it only for constant case */
238 dv_build_unquantize_tables(s
, dsp
.idct_permutation
);
240 avctx
->coded_frame
= &s
->picture
;
247 // #define printf(...) av_log(NULL, AV_LOG_ERROR, __VA_ARGS__)
249 typedef struct BlockInfo
{
250 const uint8_t *shift_table
;
251 const uint8_t *scan_table
;
252 const int *iweight_table
;
253 uint8_t pos
; /* position in block */
255 uint8_t partial_bit_count
;
256 uint16_t partial_bit_buffer
;
260 /* block size in bits */
261 static const uint16_t block_sizes
[6] = {
262 112, 112, 112, 112, 80, 80
264 /* bit budget for AC only in 5 MBs */
265 static const int vs_total_ac_bits
= (100 * 4 + 68*2) * 5;
266 /* see dv_88_areas and dv_248_areas for details */
267 static const int mb_area_start
[5] = { 1, 6, 21, 43, 64 };
269 static inline int get_bits_left(GetBitContext
*s
)
271 return s
->size_in_bits
- get_bits_count(s
);
274 static inline int get_bits_size(GetBitContext
*s
)
276 return s
->size_in_bits
;
279 static inline int put_bits_left(PutBitContext
* s
)
281 return (s
->buf_end
- s
->buf
) * 8 - put_bits_count(s
);
284 /* decode ac coefs */
285 static void dv_decode_ac(GetBitContext
*gb
, BlockInfo
*mb
, DCTELEM
*block
)
287 int last_index
= get_bits_size(gb
);
288 const uint8_t *scan_table
= mb
->scan_table
;
289 const uint8_t *shift_table
= mb
->shift_table
;
290 const int *iweight_table
= mb
->iweight_table
;
292 int partial_bit_count
= mb
->partial_bit_count
;
293 int level
, pos1
, run
, vlc_len
, index
;
296 UPDATE_CACHE(re
, gb
);
298 /* if we must parse a partial vlc, we do it here */
299 if (partial_bit_count
> 0) {
300 re_cache
= ((unsigned)re_cache
>> partial_bit_count
) |
301 (mb
->partial_bit_buffer
<< (sizeof(re_cache
)*8 - partial_bit_count
));
302 re_index
-= partial_bit_count
;
303 mb
->partial_bit_count
= 0;
306 /* get the AC coefficients until last_index is reached */
309 printf("%2d: bits=%04x index=%d\n", pos
, SHOW_UBITS(re
, gb
, 16), re_index
);
311 /* our own optimized GET_RL_VLC */
312 index
= NEG_USR32(re_cache
, TEX_VLC_BITS
);
313 vlc_len
= dv_rl_vlc
[index
].len
;
315 index
= NEG_USR32((unsigned)re_cache
<< TEX_VLC_BITS
, -vlc_len
) + dv_rl_vlc
[index
].level
;
316 vlc_len
= TEX_VLC_BITS
- vlc_len
;
318 level
= dv_rl_vlc
[index
].level
;
319 run
= dv_rl_vlc
[index
].run
;
321 /* gotta check if we're still within gb boundaries */
322 if (re_index
+ vlc_len
> last_index
) {
323 /* should be < 16 bits otherwise a codeword could have been parsed */
324 mb
->partial_bit_count
= last_index
- re_index
;
325 mb
->partial_bit_buffer
= NEG_USR32(re_cache
, mb
->partial_bit_count
);
326 re_index
= last_index
;
332 printf("run=%d level=%d\n", run
, level
);
338 pos1
= scan_table
[pos
];
339 level
<<= shift_table
[pos1
];
341 /* unweigh, round, and shift down */
342 level
= (level
*iweight_table
[pos
] + (1 << (dv_iweight_bits
-1))) >> dv_iweight_bits
;
346 UPDATE_CACHE(re
, gb
);
348 CLOSE_READER(re
, gb
);
352 static inline void bit_copy(PutBitContext
*pb
, GetBitContext
*gb
)
354 int bits_left
= get_bits_left(gb
);
355 while (bits_left
>= MIN_CACHE_BITS
) {
356 put_bits(pb
, MIN_CACHE_BITS
, get_bits(gb
, MIN_CACHE_BITS
));
357 bits_left
-= MIN_CACHE_BITS
;
360 put_bits(pb
, bits_left
, get_bits(gb
, bits_left
));
364 /* mb_x and mb_y are in units of 8 pixels */
365 static inline void dv_decode_video_segment(DVVideoContext
*s
,
366 const uint8_t *buf_ptr1
,
367 const uint16_t *mb_pos_ptr
)
369 int quant
, dc
, dct_mode
, class1
, j
;
370 int mb_index
, mb_x
, mb_y
, v
, last_index
;
371 DCTELEM
*block
, *block1
;
374 void (*idct_put
)(uint8_t *dest
, int line_size
, DCTELEM
*block
);
375 const uint8_t *buf_ptr
;
376 PutBitContext pb
, vs_pb
;
378 BlockInfo mb_data
[5 * 6], *mb
, *mb1
;
379 DECLARE_ALIGNED_8(DCTELEM
, sblock
[5*6][64]);
380 DECLARE_ALIGNED_8(uint8_t, mb_bit_buffer
[80 + 4]); /* allow some slack */
381 DECLARE_ALIGNED_8(uint8_t, vs_bit_buffer
[5 * 80 + 4]); /* allow some slack */
382 const int log2_blocksize
= 3-s
->avctx
->lowres
;
384 assert((((int)mb_bit_buffer
)&7)==0);
385 assert((((int)vs_bit_buffer
)&7)==0);
387 memset(sblock
, 0, sizeof(sblock
));
389 /* pass 1 : read DC and AC coefficients in blocks */
391 block1
= &sblock
[0][0];
393 init_put_bits(&vs_pb
, vs_bit_buffer
, 5 * 80);
394 for(mb_index
= 0; mb_index
< 5; mb_index
++, mb1
+= 6, block1
+= 6 * 64) {
396 quant
= buf_ptr
[3] & 0x0f;
398 init_put_bits(&pb
, mb_bit_buffer
, 80);
401 for(j
= 0;j
< 6; j
++) {
402 last_index
= block_sizes
[j
];
403 init_get_bits(&gb
, buf_ptr
, last_index
);
406 dc
= get_sbits(&gb
, 9);
407 dct_mode
= get_bits1(&gb
);
408 mb
->dct_mode
= dct_mode
;
409 mb
->scan_table
= s
->dv_zigzag
[dct_mode
];
410 mb
->iweight_table
= dct_mode
? dv_iweight_248
: dv_iweight_88
;
411 class1
= get_bits(&gb
, 2);
412 mb
->shift_table
= s
->dv_idct_shift
[class1
== 3][dct_mode
]
413 [quant
+ dv_quant_offset
[class1
]];
415 /* convert to unsigned because 128 is not added in the
419 buf_ptr
+= last_index
>> 3;
421 mb
->partial_bit_count
= 0;
424 printf("MB block: %d, %d ", mb_index
, j
);
426 dv_decode_ac(&gb
, mb
, block
);
428 /* write the remaining bits in a new buffer only if the
437 /* pass 2 : we can do it just after */
439 printf("***pass 2 size=%d MB#=%d\n", put_bits_count(&pb
), mb_index
);
443 init_get_bits(&gb
, mb_bit_buffer
, put_bits_count(&pb
));
445 for(j
= 0;j
< 6; j
++, block
+= 64, mb
++) {
446 if (mb
->pos
< 64 && get_bits_left(&gb
) > 0) {
447 dv_decode_ac(&gb
, mb
, block
);
448 /* if still not finished, no need to parse other blocks */
453 /* all blocks are finished, so the extra bytes can be used at
454 the video segment level */
456 bit_copy(&vs_pb
, &gb
);
459 /* we need a pass other the whole video segment */
461 printf("***pass 3 size=%d\n", put_bits_count(&vs_pb
));
463 block
= &sblock
[0][0];
465 init_get_bits(&gb
, vs_bit_buffer
, put_bits_count(&vs_pb
));
466 flush_put_bits(&vs_pb
);
467 for(mb_index
= 0; mb_index
< 5; mb_index
++) {
468 for(j
= 0;j
< 6; j
++) {
471 printf("start %d:%d\n", mb_index
, j
);
473 dv_decode_ac(&gb
, mb
, block
);
475 if (mb
->pos
>= 64 && mb
->pos
< 127)
476 av_log(NULL
, AV_LOG_ERROR
, "AC EOB marker is absent pos=%d\n", mb
->pos
);
482 /* compute idct and place blocks */
483 block
= &sblock
[0][0];
485 for(mb_index
= 0; mb_index
< 5; mb_index
++) {
489 if (s
->sys
->pix_fmt
== PIX_FMT_YUV422P
) {
490 y_ptr
= s
->picture
.data
[0] + ((mb_y
* s
->picture
.linesize
[0] + (mb_x
>>1))<<log2_blocksize
);
491 c_offset
= ((mb_y
* s
->picture
.linesize
[1] + (mb_x
>> 2))<<log2_blocksize
);
492 } else { /* 4:1:1 or 4:2:0 */
493 y_ptr
= s
->picture
.data
[0] + ((mb_y
* s
->picture
.linesize
[0] + mb_x
)<<log2_blocksize
);
494 if (s
->sys
->pix_fmt
== PIX_FMT_YUV411P
)
495 c_offset
= ((mb_y
* s
->picture
.linesize
[1] + (mb_x
>> 2))<<log2_blocksize
);
497 c_offset
= (((mb_y
>> 1) * s
->picture
.linesize
[1] + (mb_x
>> 1))<<log2_blocksize
);
499 for(j
= 0;j
< 6; j
++) {
500 idct_put
= s
->idct_put
[mb
->dct_mode
&& log2_blocksize
==3];
501 if (s
->sys
->pix_fmt
== PIX_FMT_YUV422P
) { /* 4:2:2 */
502 if (j
== 0 || j
== 2) {
504 idct_put(y_ptr
+ ((j
>> 1)<<log2_blocksize
),
505 s
->picture
.linesize
[0], block
);
508 idct_put(s
->picture
.data
[6 - j
] + c_offset
,
509 s
->picture
.linesize
[6 - j
], block
);
511 /* note: j=1 and j=3 are "dummy" blocks in 4:2:2 */
512 } else { /* 4:1:1 or 4:2:0 */
514 if (s
->sys
->pix_fmt
== PIX_FMT_YUV411P
&& mb_x
< (704 / 8)) {
515 /* NOTE: at end of line, the macroblock is handled as 420 */
516 idct_put(y_ptr
+ (j
<<log2_blocksize
), s
->picture
.linesize
[0], block
);
518 idct_put(y_ptr
+ (((j
& 1) + (j
>> 1) * s
->picture
.linesize
[0])<<log2_blocksize
),
519 s
->picture
.linesize
[0], block
);
522 if (s
->sys
->pix_fmt
== PIX_FMT_YUV411P
&& mb_x
>= (704 / 8)) {
523 uint64_t aligned_pixels
[64/8];
524 uint8_t *pixels
= (uint8_t*)aligned_pixels
;
525 uint8_t *c_ptr
, *c_ptr1
, *ptr
, *ptr1
;
527 /* NOTE: at end of line, the macroblock is handled as 420 */
528 idct_put(pixels
, 8, block
);
529 linesize
= s
->picture
.linesize
[6 - j
];
530 c_ptr
= s
->picture
.data
[6 - j
] + c_offset
;
532 for(y
= 0;y
< (1<<log2_blocksize
); y
++) {
533 ptr1
= ptr
+ (1<<(log2_blocksize
-1));
534 c_ptr1
= c_ptr
+ (linesize
<<log2_blocksize
);
535 for(x
=0; x
< (1<<(log2_blocksize
-1)); x
++){
536 c_ptr
[x
]= ptr
[x
]; c_ptr1
[x
]= ptr1
[x
];
542 /* don't ask me why they inverted Cb and Cr ! */
543 idct_put(s
->picture
.data
[6 - j
] + c_offset
,
544 s
->picture
.linesize
[6 - j
], block
);
554 #ifdef DV_CODEC_TINY_TARGET
555 /* Converts run and level (where level != 0) pair into vlc, returning bit size */
556 static av_always_inline
int dv_rl2vlc(int run
, int level
, int sign
, uint32_t* vlc
)
559 if (run
< DV_VLC_MAP_RUN_SIZE
&& level
< DV_VLC_MAP_LEV_SIZE
) {
560 *vlc
= dv_vlc_map
[run
][level
].vlc
| sign
;
561 size
= dv_vlc_map
[run
][level
].size
;
564 if (level
< DV_VLC_MAP_LEV_SIZE
) {
565 *vlc
= dv_vlc_map
[0][level
].vlc
| sign
;
566 size
= dv_vlc_map
[0][level
].size
;
568 *vlc
= 0xfe00 | (level
<< 1) | sign
;
572 *vlc
|= ((run
< 16) ? dv_vlc_map
[run
-1][0].vlc
:
573 (0x1f80 | (run
- 1))) << size
;
574 size
+= (run
< 16) ? dv_vlc_map
[run
-1][0].size
: 13;
581 static av_always_inline
int dv_rl2vlc_size(int run
, int level
)
585 if (run
< DV_VLC_MAP_RUN_SIZE
&& level
< DV_VLC_MAP_LEV_SIZE
) {
586 size
= dv_vlc_map
[run
][level
].size
;
589 size
= (level
< DV_VLC_MAP_LEV_SIZE
) ? dv_vlc_map
[0][level
].size
: 16;
591 size
+= (run
< 16) ? dv_vlc_map
[run
-1][0].size
: 13;
597 static av_always_inline
int dv_rl2vlc(int run
, int l
, int sign
, uint32_t* vlc
)
599 *vlc
= dv_vlc_map
[run
][l
].vlc
| sign
;
600 return dv_vlc_map
[run
][l
].size
;
603 static av_always_inline
int dv_rl2vlc_size(int run
, int l
)
605 return dv_vlc_map
[run
][l
].size
;
609 typedef struct EncBlockInfo
{
619 uint8_t partial_bit_count
;
620 uint32_t partial_bit_buffer
; /* we can't use uint16_t here */
623 static av_always_inline PutBitContext
* dv_encode_ac(EncBlockInfo
* bi
, PutBitContext
* pb_pool
,
624 PutBitContext
* pb_end
)
628 PutBitContext
* pb
= pb_pool
;
629 int size
= bi
->partial_bit_count
;
630 uint32_t vlc
= bi
->partial_bit_buffer
;
632 bi
->partial_bit_count
= bi
->partial_bit_buffer
= 0;
634 /* Find suitable storage space */
635 for (; size
> (bits_left
= put_bits_left(pb
)); pb
++) {
638 put_bits(pb
, bits_left
, vlc
>> size
);
639 vlc
= vlc
& ((1<<size
)-1);
641 if (pb
+ 1 >= pb_end
) {
642 bi
->partial_bit_count
= size
;
643 bi
->partial_bit_buffer
= vlc
;
649 put_bits(pb
, size
, vlc
);
654 /* Construct the next VLC */
656 bi
->cur_ac
= bi
->next
[prev
];
658 size
= dv_rl2vlc(bi
->cur_ac
- prev
- 1, bi
->mb
[bi
->cur_ac
], bi
->sign
[bi
->cur_ac
], &vlc
);
660 size
= 4; vlc
= 6; /* End Of Block stamp */
666 static av_always_inline
void dv_set_class_number(DCTELEM
* blk
, EncBlockInfo
* bi
,
667 const uint8_t* zigzag_scan
, const int *weight
, int bias
)
670 /* We offer two different methods for class number assignment: the
671 method suggested in SMPTE 314M Table 22, and an improved
672 method. The SMPTE method is very conservative; it assigns class
673 3 (i.e. severe quantization) to any block where the largest AC
674 component is greater than 36. ffmpeg's DV encoder tracks AC bit
675 consumption precisely, so there is no need to bias most blocks
676 towards strongly lossy compression. Instead, we assign class 2
677 to most blocks, and use class 3 only when strictly necessary
678 (for blocks whose largest AC component exceeds 255). */
680 #if 0 /* SMPTE spec method */
681 static const int classes
[] = {12, 24, 36, 0xffff};
682 #else /* improved ffmpeg method */
683 static const int classes
[] = {-1, -1, 255, 0xffff};
690 for (area
= 0; area
< 4; area
++) {
691 bi
->prev
[area
] = prev
;
692 bi
->bit_size
[area
] = 1; // 4 areas 4 bits for EOB :)
693 for (i
=mb_area_start
[area
]; i
<mb_area_start
[area
+1]; i
++) {
694 int level
= blk
[zigzag_scan
[i
]];
696 if (level
+15 > 30U) {
697 bi
->sign
[i
] = (level
>>31)&1;
698 /* weigh it and and shift down into range, adding for rounding */
699 /* the extra division by a factor of 2^4 reverses the 8x expansion of the DCT
700 AND the 2x doubling of the weights */
701 level
= (FFABS(level
) * weight
[i
] + (1<<(dv_weight_bits
+3))) >> (dv_weight_bits
+4);
703 if(level
>max
) max
= level
;
704 bi
->bit_size
[area
] += dv_rl2vlc_size(i
- prev
- 1, level
);
711 for(bi
->cno
= 0; max
> classes
[bi
->cno
]; bi
->cno
++);
719 for (area
= 0; area
< 4; area
++) {
720 bi
->prev
[area
] = prev
;
721 bi
->bit_size
[area
] = 1; // 4 areas 4 bits for EOB :)
722 for (; i
<mb_area_start
[area
+1]; i
= bi
->next
[i
]) {
726 bi
->bit_size
[area
] += dv_rl2vlc_size(i
- prev
- 1, bi
->mb
[i
]);
736 //FIXME replace this by dsputil
737 #define SC(x, y) ((s[x] - s[y]) ^ ((s[x] - s[y]) >> 7))
738 static av_always_inline
int dv_guess_dct_mode(DCTELEM
*blk
) {
744 /* Compute 8-8 score (small values give a better chance for 8-8 DCT) */
747 score88
+= SC(0, 8) + SC(1, 9) + SC(2, 10) + SC(3, 11) +
748 SC(4, 12) + SC(5,13) + SC(6, 14) + SC(7, 15);
751 /* Compute 2-4-8 score (small values give a better chance for 2-4-8 DCT) */
754 score248
+= SC(0, 16) + SC(1,17) + SC(2, 18) + SC(3, 19) +
755 SC(4, 20) + SC(5,21) + SC(6, 22) + SC(7, 23);
759 return (score88
- score248
> -10);
762 static inline void dv_guess_qnos(EncBlockInfo
* blks
, int* qnos
)
765 int i
, j
, k
, a
, prev
, a2
;
768 size
[0] = size
[1] = size
[2] = size
[3] = size
[4] = 1<<24;
771 for (i
=0; i
<5; i
++) {
777 for (j
=0; j
<6; j
++, b
++) {
778 for (a
=0; a
<4; a
++) {
779 if (b
->area_q
[a
] != dv_quant_shifts
[qnos
[i
] + dv_quant_offset
[b
->cno
]][a
]) {
780 b
->bit_size
[a
] = 1; // 4 areas 4 bits for EOB :)
783 assert(b
->next
[prev
] >= mb_area_start
[a
+1] || b
->mb
[prev
]);
784 for (k
= b
->next
[prev
] ; k
<mb_area_start
[a
+1]; k
= b
->next
[k
]) {
787 b
->bit_size
[a
] += dv_rl2vlc_size(k
- prev
- 1, b
->mb
[k
]);
790 if(b
->next
[k
] >= mb_area_start
[a
+1] && b
->next
[k
]<64){
791 for(a2
=a
+1; b
->next
[k
] >= mb_area_start
[a2
+1]; a2
++)
794 assert(b
->mb
[b
->next
[k
]]);
795 b
->bit_size
[a2
] += dv_rl2vlc_size(b
->next
[k
] - prev
- 1, b
->mb
[b
->next
[k
]])
796 -dv_rl2vlc_size(b
->next
[k
] - k
- 1, b
->mb
[b
->next
[k
]]);
797 assert(b
->prev
[a2
]==k
&& (a2
+1 >= 4 || b
->prev
[a2
+1]!=k
));
800 b
->next
[prev
] = b
->next
[k
];
805 size
[i
] += b
->bit_size
[a
];
808 if(vs_total_ac_bits
>= size
[0] + size
[1] + size
[2] + size
[3] + size
[4])
811 } while (qnos
[0]|qnos
[1]|qnos
[2]|qnos
[3]|qnos
[4]);
814 for(a
=2; a
==2 || vs_total_ac_bits
< size
[0]; a
+=a
){
816 size
[0] = 5*6*4; //EOB
817 for (j
=0; j
<6*5; j
++, b
++) {
819 for (k
= b
->next
[prev
]; k
<64; k
= b
->next
[k
]) {
820 if(b
->mb
[k
] < a
&& b
->mb
[k
] > -a
){
821 b
->next
[prev
] = b
->next
[k
];
823 size
[0] += dv_rl2vlc_size(k
- prev
- 1, b
->mb
[k
]);
831 static inline void dv_encode_video_segment(DVVideoContext
*s
,
833 const uint16_t *mb_pos_ptr
)
835 int mb_index
, i
, j
, v
;
836 int mb_x
, mb_y
, c_offset
, linesize
;
841 DECLARE_ALIGNED_16(DCTELEM
, block
[64]);
842 EncBlockInfo enc_blks
[5*6];
843 PutBitContext pbs
[5*6];
845 EncBlockInfo
* enc_blk
;
849 assert((((int)block
) & 15) == 0);
851 enc_blk
= &enc_blks
[0];
853 for(mb_index
= 0; mb_index
< 5; mb_index
++) {
857 if (s
->sys
->pix_fmt
== PIX_FMT_YUV422P
) {
858 y_ptr
= s
->picture
.data
[0] + (mb_y
* s
->picture
.linesize
[0] * 8) + (mb_x
* 4);
860 y_ptr
= s
->picture
.data
[0] + (mb_y
* s
->picture
.linesize
[0] * 8) + (mb_x
* 8);
862 if (s
->sys
->pix_fmt
== PIX_FMT_YUV420P
) {
863 c_offset
= (((mb_y
>> 1) * s
->picture
.linesize
[1] * 8) + ((mb_x
>> 1) * 8));
864 } else { /* 4:2:2 or 4:1:1 */
865 c_offset
= ((mb_y
* s
->picture
.linesize
[1] * 8) + ((mb_x
>> 2) * 8));
868 qnos
[mb_index
] = 15; /* No quantization */
869 ptr
= dif
+ mb_index
*80 + 4;
870 for(j
= 0;j
< 6; j
++) {
872 if (s
->sys
->pix_fmt
== PIX_FMT_YUV422P
) { /* 4:2:2 */
873 if (j
== 0 || j
== 2) {
875 data
= y_ptr
+ ((j
>>1) * 8);
876 linesize
= s
->picture
.linesize
[0];
879 data
= s
->picture
.data
[6 - j
] + c_offset
;
880 linesize
= s
->picture
.linesize
[6 - j
];
882 /* j=1 and j=3 are "dummy" blocks, used for AC data only */
887 } else { /* 4:1:1 or 4:2:0 */
888 if (j
< 4) { /* Four Y blocks */
889 /* NOTE: at end of line, the macroblock is handled as 420 */
890 if (s
->sys
->pix_fmt
== PIX_FMT_YUV411P
&& mb_x
< (704 / 8)) {
891 data
= y_ptr
+ (j
* 8);
893 data
= y_ptr
+ ((j
& 1) * 8) + ((j
>> 1) * 8 * s
->picture
.linesize
[0]);
895 linesize
= s
->picture
.linesize
[0];
896 } else { /* Cr and Cb blocks */
897 /* don't ask Fabrice why they inverted Cb and Cr ! */
898 data
= s
->picture
.data
[6 - j
] + c_offset
;
899 linesize
= s
->picture
.linesize
[6 - j
];
900 if (s
->sys
->pix_fmt
== PIX_FMT_YUV411P
&& mb_x
>= (704 / 8))
905 /* Everything is set up -- now just copy data -> DCT block */
906 if (do_edge_wrap
) { /* Edge wrap copy: 4x16 -> 8x8 */
910 d
= data
+ 8 * linesize
;
911 b
[0] = data
[0]; b
[1] = data
[1]; b
[2] = data
[2]; b
[3] = data
[3];
912 b
[4] = d
[0]; b
[5] = d
[1]; b
[6] = d
[2]; b
[7] = d
[3];
916 } else { /* Simple copy: 8x8 -> 8x8 */
918 s
->get_pixels(block
, data
, linesize
);
921 if(s
->avctx
->flags
& CODEC_FLAG_INTERLACED_DCT
)
922 enc_blk
->dct_mode
= dv_guess_dct_mode(block
);
924 enc_blk
->dct_mode
= 0;
925 enc_blk
->area_q
[0] = enc_blk
->area_q
[1] = enc_blk
->area_q
[2] = enc_blk
->area_q
[3] = 0;
926 enc_blk
->partial_bit_count
= 0;
927 enc_blk
->partial_bit_buffer
= 0;
931 /* We rely on the fact that encoding all zeros leads to an immediate EOB,
932 which is precisely what the spec calls for in the "dummy" blocks. */
933 memset(block
, 0, sizeof(block
));
935 s
->fdct
[enc_blk
->dct_mode
](block
);
938 dv_set_class_number(block
, enc_blk
,
939 enc_blk
->dct_mode
? ff_zigzag248_direct
: ff_zigzag_direct
,
940 enc_blk
->dct_mode
? dv_weight_248
: dv_weight_88
,
943 init_put_bits(pb
, ptr
, block_sizes
[j
]/8);
944 put_bits(pb
, 9, (uint16_t)(((enc_blk
->mb
[0] >> 3) - 1024 + 2) >> 2));
945 put_bits(pb
, 1, enc_blk
->dct_mode
);
946 put_bits(pb
, 2, enc_blk
->cno
);
948 vs_bit_size
+= enc_blk
->bit_size
[0] + enc_blk
->bit_size
[1] +
949 enc_blk
->bit_size
[2] + enc_blk
->bit_size
[3];
952 ptr
+= block_sizes
[j
]/8;
956 if (vs_total_ac_bits
< vs_bit_size
)
957 dv_guess_qnos(&enc_blks
[0], &qnos
[0]);
959 for (i
=0; i
<5; i
++) {
960 dif
[i
*80 + 3] = qnos
[i
];
963 /* First pass over individual cells only */
964 for (j
=0; j
<5*6; j
++)
965 dv_encode_ac(&enc_blks
[j
], &pbs
[j
], &pbs
[j
+1]);
967 /* Second pass over each MB space */
968 for (j
=0; j
<5*6; j
+=6) {
970 for (i
=0; i
<6; i
++) {
971 if (enc_blks
[i
+j
].partial_bit_count
)
972 pb
=dv_encode_ac(&enc_blks
[i
+j
], pb
, &pbs
[j
+6]);
976 /* Third and final pass over the whole vides segment space */
978 for (j
=0; j
<5*6; j
++) {
979 if (enc_blks
[j
].partial_bit_count
)
980 pb
=dv_encode_ac(&enc_blks
[j
], pb
, &pbs
[6*5]);
981 if (enc_blks
[j
].partial_bit_count
)
982 av_log(NULL
, AV_LOG_ERROR
, "ac bitstream overflow\n");
985 for (j
=0; j
<5*6; j
++)
986 flush_put_bits(&pbs
[j
]);
989 static int dv_decode_mt(AVCodecContext
*avctx
, void* sl
)
991 DVVideoContext
*s
= avctx
->priv_data
;
992 int slice
= (size_t)sl
;
994 /* which DIF channel is this? */
995 int chan
= slice
/ (s
->sys
->difseg_size
* 27);
997 /* slice within the DIF channel */
998 int chan_slice
= slice
% (s
->sys
->difseg_size
* 27);
1000 /* byte offset of this channel's data */
1001 int chan_offset
= chan
* s
->sys
->difseg_size
* 150 * 80;
1003 dv_decode_video_segment(s
, &s
->buf
[((chan_slice
/27)*6+(chan_slice
/3)+chan_slice
*5+7)*80 + chan_offset
],
1004 &s
->sys
->video_place
[slice
*5]);
1008 #ifdef CONFIG_ENCODERS
1009 static int dv_encode_mt(AVCodecContext
*avctx
, void* sl
)
1011 DVVideoContext
*s
= avctx
->priv_data
;
1012 int slice
= (size_t)sl
;
1014 /* which DIF channel is this? */
1015 int chan
= slice
/ (s
->sys
->difseg_size
* 27);
1017 /* slice within the DIF channel */
1018 int chan_slice
= slice
% (s
->sys
->difseg_size
* 27);
1020 /* byte offset of this channel's data */
1021 int chan_offset
= chan
* s
->sys
->difseg_size
* 150 * 80;
1023 dv_encode_video_segment(s
, &s
->buf
[((chan_slice
/27)*6+(chan_slice
/3)+chan_slice
*5+7)*80 + chan_offset
],
1024 &s
->sys
->video_place
[slice
*5]);
1029 #ifdef CONFIG_DECODERS
1030 /* NOTE: exactly one frame must be given (120000 bytes for NTSC,
1031 144000 bytes for PAL - or twice those for 50Mbps) */
1032 static int dvvideo_decode_frame(AVCodecContext
*avctx
,
1033 void *data
, int *data_size
,
1034 const uint8_t *buf
, int buf_size
)
1036 DVVideoContext
*s
= avctx
->priv_data
;
1038 s
->sys
= dv_frame_profile(buf
);
1039 if (!s
->sys
|| buf_size
< s
->sys
->frame_size
)
1040 return -1; /* NOTE: we only accept several full frames */
1042 if(s
->picture
.data
[0])
1043 avctx
->release_buffer(avctx
, &s
->picture
);
1045 s
->picture
.reference
= 0;
1046 s
->picture
.key_frame
= 1;
1047 s
->picture
.pict_type
= FF_I_TYPE
;
1048 avctx
->pix_fmt
= s
->sys
->pix_fmt
;
1049 avcodec_set_dimensions(avctx
, s
->sys
->width
, s
->sys
->height
);
1050 if(avctx
->get_buffer(avctx
, &s
->picture
) < 0) {
1051 av_log(avctx
, AV_LOG_ERROR
, "get_buffer() failed\n");
1054 s
->picture
.interlaced_frame
= 1;
1055 s
->picture
.top_field_first
= 0;
1058 avctx
->execute(avctx
, dv_decode_mt
, (void**)&dv_anchor
[0], NULL
,
1059 s
->sys
->n_difchan
* s
->sys
->difseg_size
* 27);
1064 *data_size
= sizeof(AVFrame
);
1065 *(AVFrame
*)data
= s
->picture
;
1067 return s
->sys
->frame_size
;
1072 static inline int dv_write_pack(enum dv_pack_type pack_id
, DVVideoContext
*c
, uint8_t* buf
)
1075 * Here's what SMPTE314M says about these two:
1076 * (page 6) APTn, AP1n, AP2n, AP3n: These data shall be identical
1077 * as track application IDs (APTn = 001, AP1n =
1078 * 001, AP2n = 001, AP3n = 001), if the source signal
1079 * comes from a digital VCR. If the signal source is
1080 * unknown, all bits for these data shall be set to 1.
1081 * (page 12) STYPE: STYPE defines a signal type of video signal
1082 * 00000b = 4:1:1 compression
1083 * 00100b = 4:2:2 compression
1085 * Now, I've got two problems with these statements:
1086 * 1. it looks like APT == 111b should be a safe bet, but it isn't.
1087 * It seems that for PAL as defined in IEC 61834 we have to set
1088 * APT to 000 and for SMPTE314M to 001.
1089 * 2. It is not at all clear what STYPE is used for 4:2:0 PAL
1090 * compression scheme (if any).
1092 int apt
= (c
->sys
->pix_fmt
== PIX_FMT_YUV420P
? 0 : 1);
1093 int stype
= (c
->sys
->pix_fmt
== PIX_FMT_YUV422P
? 4 : 0);
1096 if((int)(av_q2d(c
->avctx
->sample_aspect_ratio
) * c
->avctx
->width
/ c
->avctx
->height
* 10) == 17) /* 16:9 */
1099 buf
[0] = (uint8_t)pack_id
;
1101 case dv_header525
: /* I can't imagine why these two weren't defined as real */
1102 case dv_header625
: /* packs in SMPTE314M -- they definitely look like ones */
1103 buf
[1] = 0xf8 | /* reserved -- always 1 */
1104 (apt
& 0x07); /* APT: Track application ID */
1105 buf
[2] = (0 << 7) | /* TF1: audio data is 0 - valid; 1 - invalid */
1106 (0x0f << 3) | /* reserved -- always 1 */
1107 (apt
& 0x07); /* AP1: Audio application ID */
1108 buf
[3] = (0 << 7) | /* TF2: video data is 0 - valid; 1 - invalid */
1109 (0x0f << 3) | /* reserved -- always 1 */
1110 (apt
& 0x07); /* AP2: Video application ID */
1111 buf
[4] = (0 << 7) | /* TF3: subcode(SSYB) is 0 - valid; 1 - invalid */
1112 (0x0f << 3) | /* reserved -- always 1 */
1113 (apt
& 0x07); /* AP3: Subcode application ID */
1115 case dv_video_source
:
1116 buf
[1] = 0xff; /* reserved -- always 1 */
1117 buf
[2] = (1 << 7) | /* B/W: 0 - b/w, 1 - color */
1118 (1 << 6) | /* following CLF is valid - 0, invalid - 1 */
1119 (3 << 4) | /* CLF: color frames id (see ITU-R BT.470-4) */
1120 0xf; /* reserved -- always 1 */
1121 buf
[3] = (3 << 6) | /* reserved -- always 1 */
1122 (c
->sys
->dsf
<< 5) | /* system: 60fields/50fields */
1123 stype
; /* signal type video compression */
1124 buf
[4] = 0xff; /* VISC: 0xff -- no information */
1126 case dv_video_control
:
1127 buf
[1] = (0 << 6) | /* Copy generation management (CGMS) 0 -- free */
1128 0x3f; /* reserved -- always 1 */
1129 buf
[2] = 0xc8 | /* reserved -- always b11001xxx */
1131 buf
[3] = (1 << 7) | /* Frame/field flag 1 -- frame, 0 -- field */
1132 (1 << 6) | /* First/second field flag 0 -- field 2, 1 -- field 1 */
1133 (1 << 5) | /* Frame change flag 0 -- same picture as before, 1 -- different */
1134 (1 << 4) | /* 1 - interlaced, 0 - noninterlaced */
1135 0xc; /* reserved -- always b1100 */
1136 buf
[4] = 0xff; /* reserved -- always 1 */
1139 buf
[1] = buf
[2] = buf
[3] = buf
[4] = 0xff;
1144 static void dv_format_frame(DVVideoContext
* c
, uint8_t* buf
)
1148 for (chan
= 0; chan
< c
->sys
->n_difchan
; chan
++) {
1149 for (i
= 0; i
< c
->sys
->difseg_size
; i
++) {
1150 memset(buf
, 0xff, 80 * 6); /* First 6 DIF blocks are for control data */
1152 /* DV header: 1DIF */
1153 buf
+= dv_write_dif_id(dv_sect_header
, chan
, i
, 0, buf
);
1154 buf
+= dv_write_pack((c
->sys
->dsf
? dv_header625
: dv_header525
), c
, buf
);
1155 buf
+= 72; /* unused bytes */
1157 /* DV subcode: 2DIFs */
1158 for (j
= 0; j
< 2; j
++) {
1159 buf
+= dv_write_dif_id(dv_sect_subcode
, chan
, i
, j
, buf
);
1160 for (k
= 0; k
< 6; k
++)
1161 buf
+= dv_write_ssyb_id(k
, (i
< c
->sys
->difseg_size
/2), buf
) + 5;
1162 buf
+= 29; /* unused bytes */
1165 /* DV VAUX: 3DIFS */
1166 for (j
= 0; j
< 3; j
++) {
1167 buf
+= dv_write_dif_id(dv_sect_vaux
, chan
, i
, j
, buf
);
1168 buf
+= dv_write_pack(dv_video_source
, c
, buf
);
1169 buf
+= dv_write_pack(dv_video_control
, c
, buf
);
1171 buf
+= dv_write_pack(dv_video_source
, c
, buf
);
1172 buf
+= dv_write_pack(dv_video_control
, c
, buf
);
1173 buf
+= 4*5 + 2; /* unused bytes */
1176 /* DV Audio/Video: 135 Video DIFs + 9 Audio DIFs */
1177 for (j
= 0; j
< 135; j
++) {
1179 memset(buf
, 0xff, 80);
1180 buf
+= dv_write_dif_id(dv_sect_audio
, chan
, i
, j
/15, buf
);
1181 buf
+= 77; /* audio control & shuffled PCM audio */
1183 buf
+= dv_write_dif_id(dv_sect_video
, chan
, i
, j
, buf
);
1184 buf
+= 77; /* 1 video macro block: 1 bytes control
1185 4 * 14 bytes Y 8x8 data
1186 10 bytes Cr 8x8 data
1187 10 bytes Cb 8x8 data */
1194 #ifdef CONFIG_ENCODERS
1195 static int dvvideo_encode_frame(AVCodecContext
*c
, uint8_t *buf
, int buf_size
,
1198 DVVideoContext
*s
= c
->priv_data
;
1200 s
->sys
= dv_codec_profile(c
);
1203 if(buf_size
< s
->sys
->frame_size
)
1206 c
->pix_fmt
= s
->sys
->pix_fmt
;
1207 s
->picture
= *((AVFrame
*)data
);
1208 s
->picture
.key_frame
= 1;
1209 s
->picture
.pict_type
= FF_I_TYPE
;
1212 c
->execute(c
, dv_encode_mt
, (void**)&dv_anchor
[0], NULL
,
1213 s
->sys
->n_difchan
* s
->sys
->difseg_size
* 27);
1217 dv_format_frame(s
, buf
);
1219 return s
->sys
->frame_size
;
1223 static int dvvideo_close(AVCodecContext
*c
)
1225 DVVideoContext
*s
= c
->priv_data
;
1227 if(s
->picture
.data
[0])
1228 c
->release_buffer(c
, &s
->picture
);
1234 #ifdef CONFIG_DVVIDEO_ENCODER
1235 AVCodec dvvideo_encoder
= {
1239 sizeof(DVVideoContext
),
1241 dvvideo_encode_frame
,
1242 .pix_fmts
= (enum PixelFormat
[]) {PIX_FMT_YUV411P
, PIX_FMT_YUV422P
, PIX_FMT_YUV420P
, -1},
1244 #endif // CONFIG_DVVIDEO_ENCODER
1246 #ifdef CONFIG_DVVIDEO_DECODER
1247 AVCodec dvvideo_decoder
= {
1251 sizeof(DVVideoContext
),
1255 dvvideo_decode_frame
,