ffmpeg / libavcodec / dv.c @ ce83d2bd
History | View | Annotate | Download (47.6 KB)
1 |
/*
|
---|---|
2 |
* DV decoder
|
3 |
* Copyright (c) 2002 Fabrice Bellard
|
4 |
* Copyright (c) 2004 Roman Shaposhnik
|
5 |
*
|
6 |
* DV encoder
|
7 |
* Copyright (c) 2003 Roman Shaposhnik
|
8 |
*
|
9 |
* 50 Mbps (DVCPRO50) support
|
10 |
* Copyright (c) 2006 Daniel Maas <dmaas@maasdigital.com>
|
11 |
*
|
12 |
* 100 Mbps (DVCPRO HD) support
|
13 |
* Initial code by Daniel Maas <dmaas@maasdigital.com> (funded by BBC R&D)
|
14 |
* Final code by Roman Shaposhnik
|
15 |
*
|
16 |
* Many thanks to Dan Dennedy <dan@dennedy.org> for providing wealth
|
17 |
* of DV technical info.
|
18 |
*
|
19 |
* This file is part of FFmpeg.
|
20 |
*
|
21 |
* FFmpeg is free software; you can redistribute it and/or
|
22 |
* modify it under the terms of the GNU Lesser General Public
|
23 |
* License as published by the Free Software Foundation; either
|
24 |
* version 2.1 of the License, or (at your option) any later version.
|
25 |
*
|
26 |
* FFmpeg is distributed in the hope that it will be useful,
|
27 |
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
28 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
29 |
* Lesser General Public License for more details.
|
30 |
*
|
31 |
* You should have received a copy of the GNU Lesser General Public
|
32 |
* License along with FFmpeg; if not, write to the Free Software
|
33 |
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
34 |
*/
|
35 |
|
36 |
/**
|
37 |
* @file libavcodec/dv.c
|
38 |
* DV codec.
|
39 |
*/
|
40 |
#define ALT_BITSTREAM_READER
|
41 |
#include "avcodec.h" |
42 |
#include "dsputil.h" |
43 |
#include "bitstream.h" |
44 |
#include "simple_idct.h" |
45 |
#include "dvdata.h" |
46 |
|
47 |
//#undef NDEBUG
|
48 |
//#include <assert.h>
|
49 |
|
50 |
typedef struct DVVideoContext { |
51 |
const DVprofile *sys;
|
52 |
AVFrame picture; |
53 |
AVCodecContext *avctx; |
54 |
uint8_t *buf; |
55 |
|
56 |
uint8_t dv_zigzag[2][64]; |
57 |
|
58 |
void (*get_pixels)(DCTELEM *block, const uint8_t *pixels, int line_size); |
59 |
void (*fdct[2])(DCTELEM *block); |
60 |
void (*idct_put[2])(uint8_t *dest, int line_size, DCTELEM *block); |
61 |
me_cmp_func ildct_cmp; |
62 |
} DVVideoContext; |
63 |
|
64 |
#define TEX_VLC_BITS 9 |
65 |
|
66 |
#if CONFIG_SMALL
|
67 |
#define DV_VLC_MAP_RUN_SIZE 15 |
68 |
#define DV_VLC_MAP_LEV_SIZE 23 |
69 |
#else
|
70 |
#define DV_VLC_MAP_RUN_SIZE 64 |
71 |
#define DV_VLC_MAP_LEV_SIZE 512 //FIXME sign was removed so this should be /2 but needs check |
72 |
#endif
|
73 |
|
74 |
/* XXX: also include quantization */
|
75 |
static RL_VLC_ELEM dv_rl_vlc[1184]; |
76 |
/* VLC encoding lookup table */
|
77 |
static struct dv_vlc_pair { |
78 |
uint32_t vlc; |
79 |
uint8_t size; |
80 |
} dv_vlc_map[DV_VLC_MAP_RUN_SIZE][DV_VLC_MAP_LEV_SIZE]; |
81 |
|
82 |
static inline int dv_work_pool_size(const DVprofile *d) |
83 |
{ |
84 |
int size = d->n_difchan*d->difseg_size*27; |
85 |
if (DV_PROFILE_IS_1080i50(d))
|
86 |
size -= 3*27; |
87 |
if (DV_PROFILE_IS_720p50(d))
|
88 |
size -= 4*27; |
89 |
return size;
|
90 |
} |
91 |
|
92 |
static inline void dv_calc_mb_coordinates(const DVprofile *d, int chan, int seq, int slot, |
93 |
uint16_t *tbl) |
94 |
{ |
95 |
static const uint8_t off[] = { 2, 6, 8, 0, 4 }; |
96 |
static const uint8_t shuf1[] = { 36, 18, 54, 0, 72 }; |
97 |
static const uint8_t shuf2[] = { 24, 12, 36, 0, 48 }; |
98 |
static const uint8_t shuf3[] = { 18, 9, 27, 0, 36 }; |
99 |
|
100 |
static const uint8_t l_start[] = {0, 4, 9, 13, 18, 22, 27, 31, 36, 40}; |
101 |
static const uint8_t l_start_shuffled[] = { 9, 4, 13, 0, 18 }; |
102 |
|
103 |
static const uint8_t serpent1[] = {0, 1, 2, 2, 1, 0, |
104 |
0, 1, 2, 2, 1, 0, |
105 |
0, 1, 2, 2, 1, 0, |
106 |
0, 1, 2, 2, 1, 0, |
107 |
0, 1, 2}; |
108 |
static const uint8_t serpent2[] = {0, 1, 2, 3, 4, 5, 5, 4, 3, 2, 1, 0, |
109 |
0, 1, 2, 3, 4, 5, 5, 4, 3, 2, 1, 0, |
110 |
0, 1, 2, 3, 4, 5}; |
111 |
|
112 |
static const uint8_t remap[][2] = {{ 0, 0}, { 0, 0}, { 0, 0}, { 0, 0}, /* dummy */ |
113 |
{ 0, 0}, { 0, 1}, { 0, 2}, { 0, 3}, {10, 0}, |
114 |
{10, 1}, {10, 2}, {10, 3}, {20, 0}, {20, 1}, |
115 |
{20, 2}, {20, 3}, {30, 0}, {30, 1}, {30, 2}, |
116 |
{30, 3}, {40, 0}, {40, 1}, {40, 2}, {40, 3}, |
117 |
{50, 0}, {50, 1}, {50, 2}, {50, 3}, {60, 0}, |
118 |
{60, 1}, {60, 2}, {60, 3}, {70, 0}, {70, 1}, |
119 |
{70, 2}, {70, 3}, { 0,64}, { 0,65}, { 0,66}, |
120 |
{10,64}, {10,65}, {10,66}, {20,64}, {20,65}, |
121 |
{20,66}, {30,64}, {30,65}, {30,66}, {40,64}, |
122 |
{40,65}, {40,66}, {50,64}, {50,65}, {50,66}, |
123 |
{60,64}, {60,65}, {60,66}, {70,64}, {70,65}, |
124 |
{70,66}, { 0,67}, {20,67}, {40,67}, {60,67}}; |
125 |
|
126 |
int i, k, m;
|
127 |
int x, y, blk;
|
128 |
|
129 |
for (m=0; m<5; m++) { |
130 |
switch (d->width) {
|
131 |
case 1440: |
132 |
blk = (chan*11+seq)*27+slot; |
133 |
|
134 |
if (chan == 0 && seq == 11) { |
135 |
x = m*27+slot;
|
136 |
if (x<90) { |
137 |
y = 0;
|
138 |
} else {
|
139 |
x = (x - 90)*2; |
140 |
y = 67;
|
141 |
} |
142 |
} else {
|
143 |
i = (4*chan + blk + off[m])%11; |
144 |
k = (blk/11)%27; |
145 |
|
146 |
x = shuf1[m] + (chan&1)*9 + k%9; |
147 |
y = (i*3+k/9)*2 + (chan>>1) + 1; |
148 |
} |
149 |
tbl[m] = (x<<1)|(y<<9); |
150 |
break;
|
151 |
case 1280: |
152 |
blk = (chan*10+seq)*27+slot; |
153 |
|
154 |
i = (4*chan + (seq/5) + 2*blk + off[m])%10; |
155 |
k = (blk/5)%27; |
156 |
|
157 |
x = shuf1[m]+(chan&1)*9 + k%9; |
158 |
y = (i*3+k/9)*2 + (chan>>1) + 4; |
159 |
|
160 |
if (x >= 80) { |
161 |
x = remap[y][0]+((x-80)<<(y>59)); |
162 |
y = remap[y][1];
|
163 |
} |
164 |
tbl[m] = (x<<1)|(y<<9); |
165 |
break;
|
166 |
case 960: |
167 |
blk = (chan*10+seq)*27+slot; |
168 |
|
169 |
i = (4*chan + (seq/5) + 2*blk + off[m])%10; |
170 |
k = (blk/5)%27 + (i&1)*3; |
171 |
|
172 |
x = shuf2[m] + k%6 + 6*(chan&1); |
173 |
y = l_start[i] + k/6 + 45*(chan>>1); |
174 |
tbl[m] = (x<<1)|(y<<9); |
175 |
break;
|
176 |
case 720: |
177 |
switch (d->pix_fmt) {
|
178 |
case PIX_FMT_YUV422P:
|
179 |
x = shuf3[m] + slot/3;
|
180 |
y = serpent1[slot] + |
181 |
((((seq + off[m]) % d->difseg_size)<<1) + chan)*3; |
182 |
tbl[m] = (x<<1)|(y<<8); |
183 |
break;
|
184 |
case PIX_FMT_YUV420P:
|
185 |
x = shuf3[m] + slot/3;
|
186 |
y = serpent1[slot] + |
187 |
((seq + off[m]) % d->difseg_size)*3;
|
188 |
tbl[m] = (x<<1)|(y<<9); |
189 |
break;
|
190 |
case PIX_FMT_YUV411P:
|
191 |
i = (seq + off[m]) % d->difseg_size; |
192 |
k = slot + ((m==1||m==2)?3:0); |
193 |
|
194 |
x = l_start_shuffled[m] + k/6;
|
195 |
y = serpent2[k] + i*6;
|
196 |
if (x>21) |
197 |
y = y*2 - i*6; |
198 |
tbl[m] = (x<<2)|(y<<8); |
199 |
break;
|
200 |
} |
201 |
default:
|
202 |
break;
|
203 |
} |
204 |
} |
205 |
} |
206 |
|
207 |
static int dv_init_dynamic_tables(const DVprofile *d) |
208 |
{ |
209 |
int j,i,c,s,p;
|
210 |
uint32_t *factor1, *factor2; |
211 |
const int *iweight1, *iweight2; |
212 |
|
213 |
if (!d->work_chunks[dv_work_pool_size(d)-1].buf_offset) { |
214 |
p = i = 0;
|
215 |
for (c=0; c<d->n_difchan; c++) { |
216 |
for (s=0; s<d->difseg_size; s++) { |
217 |
p += 6;
|
218 |
for (j=0; j<27; j++) { |
219 |
p += !(j%3);
|
220 |
if (!(DV_PROFILE_IS_1080i50(d) && c != 0 && s == 11) && |
221 |
!(DV_PROFILE_IS_720p50(d) && s > 9)) {
|
222 |
dv_calc_mb_coordinates(d, c, s, j, &d->work_chunks[i].mb_coordinates[0]);
|
223 |
d->work_chunks[i++].buf_offset = p; |
224 |
} |
225 |
p += 5;
|
226 |
} |
227 |
} |
228 |
} |
229 |
} |
230 |
|
231 |
if (!d->idct_factor[DV_PROFILE_IS_HD(d)?8191:5631]) { |
232 |
factor1 = &d->idct_factor[0];
|
233 |
factor2 = &d->idct_factor[DV_PROFILE_IS_HD(d)?4096:2816]; |
234 |
if (d->height == 720) { |
235 |
iweight1 = &dv_iweight_720_y[0];
|
236 |
iweight2 = &dv_iweight_720_c[0];
|
237 |
} else {
|
238 |
iweight1 = &dv_iweight_1080_y[0];
|
239 |
iweight2 = &dv_iweight_1080_c[0];
|
240 |
} |
241 |
if (DV_PROFILE_IS_HD(d)) {
|
242 |
for (c = 0; c < 4; c++) { |
243 |
for (s = 0; s < 16; s++) { |
244 |
for (i = 0; i < 64; i++) { |
245 |
*factor1++ = (dv100_qstep[s] << (c + 9)) * iweight1[i];
|
246 |
*factor2++ = (dv100_qstep[s] << (c + 9)) * iweight2[i];
|
247 |
} |
248 |
} |
249 |
} |
250 |
} else {
|
251 |
iweight1 = &dv_iweight_88[0];
|
252 |
for (j = 0; j < 2; j++, iweight1 = &dv_iweight_248[0]) { |
253 |
for (s = 0; s < 22; s++) { |
254 |
for (i = c = 0; c < 4; c++) { |
255 |
for (; i < dv_quant_areas[c]; i++) {
|
256 |
*factor1 = iweight1[i] << (dv_quant_shifts[s][c] + 1);
|
257 |
*factor2++ = (*factor1++) << 1;
|
258 |
} |
259 |
} |
260 |
} |
261 |
} |
262 |
} |
263 |
} |
264 |
|
265 |
return 0; |
266 |
} |
267 |
|
268 |
static av_cold int dvvideo_init(AVCodecContext *avctx) |
269 |
{ |
270 |
DVVideoContext *s = avctx->priv_data; |
271 |
DSPContext dsp; |
272 |
static int done = 0; |
273 |
int i, j;
|
274 |
|
275 |
if (!done) {
|
276 |
VLC dv_vlc; |
277 |
uint16_t new_dv_vlc_bits[NB_DV_VLC*2];
|
278 |
uint8_t new_dv_vlc_len[NB_DV_VLC*2];
|
279 |
uint8_t new_dv_vlc_run[NB_DV_VLC*2];
|
280 |
int16_t new_dv_vlc_level[NB_DV_VLC*2];
|
281 |
|
282 |
done = 1;
|
283 |
|
284 |
/* it's faster to include sign bit in a generic VLC parsing scheme */
|
285 |
for (i = 0, j = 0; i < NB_DV_VLC; i++, j++) { |
286 |
new_dv_vlc_bits[j] = dv_vlc_bits[i]; |
287 |
new_dv_vlc_len[j] = dv_vlc_len[i]; |
288 |
new_dv_vlc_run[j] = dv_vlc_run[i]; |
289 |
new_dv_vlc_level[j] = dv_vlc_level[i]; |
290 |
|
291 |
if (dv_vlc_level[i]) {
|
292 |
new_dv_vlc_bits[j] <<= 1;
|
293 |
new_dv_vlc_len[j]++; |
294 |
|
295 |
j++; |
296 |
new_dv_vlc_bits[j] = (dv_vlc_bits[i] << 1) | 1; |
297 |
new_dv_vlc_len[j] = dv_vlc_len[i] + 1;
|
298 |
new_dv_vlc_run[j] = dv_vlc_run[i]; |
299 |
new_dv_vlc_level[j] = -dv_vlc_level[i]; |
300 |
} |
301 |
} |
302 |
|
303 |
/* NOTE: as a trick, we use the fact the no codes are unused
|
304 |
to accelerate the parsing of partial codes */
|
305 |
init_vlc(&dv_vlc, TEX_VLC_BITS, j, |
306 |
new_dv_vlc_len, 1, 1, new_dv_vlc_bits, 2, 2, 0); |
307 |
assert(dv_vlc.table_size == 1184);
|
308 |
|
309 |
for (i = 0; i < dv_vlc.table_size; i++){ |
310 |
int code = dv_vlc.table[i][0]; |
311 |
int len = dv_vlc.table[i][1]; |
312 |
int level, run;
|
313 |
|
314 |
if (len < 0){ //more bits needed |
315 |
run = 0;
|
316 |
level = code; |
317 |
} else {
|
318 |
run = new_dv_vlc_run [code] + 1;
|
319 |
level = new_dv_vlc_level[code]; |
320 |
} |
321 |
dv_rl_vlc[i].len = len; |
322 |
dv_rl_vlc[i].level = level; |
323 |
dv_rl_vlc[i].run = run; |
324 |
} |
325 |
free_vlc(&dv_vlc); |
326 |
|
327 |
for (i = 0; i < NB_DV_VLC - 1; i++) { |
328 |
if (dv_vlc_run[i] >= DV_VLC_MAP_RUN_SIZE)
|
329 |
continue;
|
330 |
#if CONFIG_SMALL
|
331 |
if (dv_vlc_level[i] >= DV_VLC_MAP_LEV_SIZE)
|
332 |
continue;
|
333 |
#endif
|
334 |
|
335 |
if (dv_vlc_map[dv_vlc_run[i]][dv_vlc_level[i]].size != 0) |
336 |
continue;
|
337 |
|
338 |
dv_vlc_map[dv_vlc_run[i]][dv_vlc_level[i]].vlc = |
339 |
dv_vlc_bits[i] << (!!dv_vlc_level[i]); |
340 |
dv_vlc_map[dv_vlc_run[i]][dv_vlc_level[i]].size = |
341 |
dv_vlc_len[i] + (!!dv_vlc_level[i]); |
342 |
} |
343 |
for (i = 0; i < DV_VLC_MAP_RUN_SIZE; i++) { |
344 |
#if CONFIG_SMALL
|
345 |
for (j = 1; j < DV_VLC_MAP_LEV_SIZE; j++) { |
346 |
if (dv_vlc_map[i][j].size == 0) { |
347 |
dv_vlc_map[i][j].vlc = dv_vlc_map[0][j].vlc |
|
348 |
(dv_vlc_map[i-1][0].vlc << (dv_vlc_map[0][j].size)); |
349 |
dv_vlc_map[i][j].size = dv_vlc_map[i-1][0].size + |
350 |
dv_vlc_map[0][j].size;
|
351 |
} |
352 |
} |
353 |
#else
|
354 |
for (j = 1; j < DV_VLC_MAP_LEV_SIZE/2; j++) { |
355 |
if (dv_vlc_map[i][j].size == 0) { |
356 |
dv_vlc_map[i][j].vlc = dv_vlc_map[0][j].vlc |
|
357 |
(dv_vlc_map[i-1][0].vlc << (dv_vlc_map[0][j].size)); |
358 |
dv_vlc_map[i][j].size = dv_vlc_map[i-1][0].size + |
359 |
dv_vlc_map[0][j].size;
|
360 |
} |
361 |
dv_vlc_map[i][((uint16_t)(-j))&0x1ff].vlc =
|
362 |
dv_vlc_map[i][j].vlc | 1;
|
363 |
dv_vlc_map[i][((uint16_t)(-j))&0x1ff].size =
|
364 |
dv_vlc_map[i][j].size; |
365 |
} |
366 |
#endif
|
367 |
} |
368 |
} |
369 |
|
370 |
/* Generic DSP setup */
|
371 |
dsputil_init(&dsp, avctx); |
372 |
ff_set_cmp(&dsp, dsp.ildct_cmp, avctx->ildct_cmp); |
373 |
s->get_pixels = dsp.get_pixels; |
374 |
s->ildct_cmp = dsp.ildct_cmp[5];
|
375 |
|
376 |
/* 88DCT setup */
|
377 |
s->fdct[0] = dsp.fdct;
|
378 |
s->idct_put[0] = dsp.idct_put;
|
379 |
for (i = 0; i < 64; i++) |
380 |
s->dv_zigzag[0][i] = dsp.idct_permutation[ff_zigzag_direct[i]];
|
381 |
|
382 |
/* 248DCT setup */
|
383 |
s->fdct[1] = dsp.fdct248;
|
384 |
s->idct_put[1] = ff_simple_idct248_put; // FIXME: need to add it to DSP |
385 |
if (avctx->lowres){
|
386 |
for (i = 0; i < 64; i++){ |
387 |
int j = ff_zigzag248_direct[i];
|
388 |
s->dv_zigzag[1][i] = dsp.idct_permutation[(j & 7) + (j & 8) * 4 + (j & 48) / 2]; |
389 |
} |
390 |
}else
|
391 |
memcpy(s->dv_zigzag[1], ff_zigzag248_direct, 64); |
392 |
|
393 |
avctx->coded_frame = &s->picture; |
394 |
s->avctx = avctx; |
395 |
|
396 |
return 0; |
397 |
} |
398 |
|
399 |
// #define VLC_DEBUG
|
400 |
// #define printf(...) av_log(NULL, AV_LOG_ERROR, __VA_ARGS__)
|
401 |
|
402 |
typedef struct BlockInfo { |
403 |
const uint32_t *factor_table;
|
404 |
const uint8_t *scan_table;
|
405 |
uint8_t pos; /* position in block */
|
406 |
void (*idct_put)(uint8_t *dest, int line_size, DCTELEM *block); |
407 |
uint8_t partial_bit_count; |
408 |
uint16_t partial_bit_buffer; |
409 |
int shift_offset;
|
410 |
} BlockInfo; |
411 |
|
412 |
/* bit budget for AC only in 5 MBs */
|
413 |
static const int vs_total_ac_bits = (100 * 4 + 68*2) * 5; |
414 |
/* see dv_88_areas and dv_248_areas for details */
|
415 |
static const int mb_area_start[5] = { 1, 6, 21, 43, 64 }; |
416 |
|
417 |
static inline int get_bits_left(GetBitContext *s) |
418 |
{ |
419 |
return s->size_in_bits - get_bits_count(s);
|
420 |
} |
421 |
|
422 |
static inline int put_bits_left(PutBitContext* s) |
423 |
{ |
424 |
return (s->buf_end - s->buf) * 8 - put_bits_count(s); |
425 |
} |
426 |
|
427 |
/* decode ac coefficients */
|
428 |
static void dv_decode_ac(GetBitContext *gb, BlockInfo *mb, DCTELEM *block) |
429 |
{ |
430 |
int last_index = gb->size_in_bits;
|
431 |
const uint8_t *scan_table = mb->scan_table;
|
432 |
const uint32_t *factor_table = mb->factor_table;
|
433 |
int pos = mb->pos;
|
434 |
int partial_bit_count = mb->partial_bit_count;
|
435 |
int level, run, vlc_len, index;
|
436 |
|
437 |
OPEN_READER(re, gb); |
438 |
UPDATE_CACHE(re, gb); |
439 |
|
440 |
/* if we must parse a partial vlc, we do it here */
|
441 |
if (partial_bit_count > 0) { |
442 |
re_cache = ((unsigned)re_cache >> partial_bit_count) |
|
443 |
(mb->partial_bit_buffer << (sizeof(re_cache) * 8 - partial_bit_count)); |
444 |
re_index -= partial_bit_count; |
445 |
mb->partial_bit_count = 0;
|
446 |
} |
447 |
|
448 |
/* get the AC coefficients until last_index is reached */
|
449 |
for (;;) {
|
450 |
#ifdef VLC_DEBUG
|
451 |
printf("%2d: bits=%04x index=%d\n", pos, SHOW_UBITS(re, gb, 16), re_index); |
452 |
#endif
|
453 |
/* our own optimized GET_RL_VLC */
|
454 |
index = NEG_USR32(re_cache, TEX_VLC_BITS); |
455 |
vlc_len = dv_rl_vlc[index].len; |
456 |
if (vlc_len < 0) { |
457 |
index = NEG_USR32((unsigned)re_cache << TEX_VLC_BITS, -vlc_len) + dv_rl_vlc[index].level;
|
458 |
vlc_len = TEX_VLC_BITS - vlc_len; |
459 |
} |
460 |
level = dv_rl_vlc[index].level; |
461 |
run = dv_rl_vlc[index].run; |
462 |
|
463 |
/* gotta check if we're still within gb boundaries */
|
464 |
if (re_index + vlc_len > last_index) {
|
465 |
/* should be < 16 bits otherwise a codeword could have been parsed */
|
466 |
mb->partial_bit_count = last_index - re_index; |
467 |
mb->partial_bit_buffer = NEG_USR32(re_cache, mb->partial_bit_count); |
468 |
re_index = last_index; |
469 |
break;
|
470 |
} |
471 |
re_index += vlc_len; |
472 |
|
473 |
#ifdef VLC_DEBUG
|
474 |
printf("run=%d level=%d\n", run, level);
|
475 |
#endif
|
476 |
pos += run; |
477 |
if (pos >= 64) |
478 |
break;
|
479 |
|
480 |
level = (level * factor_table[pos] + (1 << (dv_iweight_bits - 1))) >> dv_iweight_bits; |
481 |
block[scan_table[pos]] = level; |
482 |
|
483 |
UPDATE_CACHE(re, gb); |
484 |
} |
485 |
CLOSE_READER(re, gb); |
486 |
mb->pos = pos; |
487 |
} |
488 |
|
489 |
static inline void bit_copy(PutBitContext *pb, GetBitContext *gb) |
490 |
{ |
491 |
int bits_left = get_bits_left(gb);
|
492 |
while (bits_left >= MIN_CACHE_BITS) {
|
493 |
put_bits(pb, MIN_CACHE_BITS, get_bits(gb, MIN_CACHE_BITS)); |
494 |
bits_left -= MIN_CACHE_BITS; |
495 |
} |
496 |
if (bits_left > 0) { |
497 |
put_bits(pb, bits_left, get_bits(gb, bits_left)); |
498 |
} |
499 |
} |
500 |
|
501 |
static inline void dv_calculate_mb_xy(DVVideoContext *s, DVwork_chunk *work_chunk, int m, int *mb_x, int *mb_y) |
502 |
{ |
503 |
*mb_x = work_chunk->mb_coordinates[m] & 0xff;
|
504 |
*mb_y = work_chunk->mb_coordinates[m] >> 8;
|
505 |
|
506 |
/* We work with 720p frames split in half. The odd half-frame (chan==2,3) is displaced :-( */
|
507 |
if (s->sys->height == 720 && !(s->buf[1]&0x0C)) { |
508 |
*mb_y -= (*mb_y>17)?18:-72; /* shifting the Y coordinate down by 72/2 macro blocks */ |
509 |
} |
510 |
} |
511 |
|
512 |
/* mb_x and mb_y are in units of 8 pixels */
|
513 |
static int dv_decode_video_segment(AVCodecContext *avctx, void *arg) |
514 |
{ |
515 |
DVVideoContext *s = avctx->priv_data; |
516 |
DVwork_chunk *work_chunk = arg; |
517 |
int quant, dc, dct_mode, class1, j;
|
518 |
int mb_index, mb_x, mb_y, last_index;
|
519 |
int y_stride, linesize;
|
520 |
DCTELEM *block, *block1; |
521 |
int c_offset;
|
522 |
uint8_t *y_ptr; |
523 |
const uint8_t *buf_ptr;
|
524 |
PutBitContext pb, vs_pb; |
525 |
GetBitContext gb; |
526 |
BlockInfo mb_data[5 * DV_MAX_BPM], *mb, *mb1;
|
527 |
DECLARE_ALIGNED_16(DCTELEM, sblock[5*DV_MAX_BPM][64]); |
528 |
DECLARE_ALIGNED_8(uint8_t, mb_bit_buffer[80 + 4]); /* allow some slack */ |
529 |
DECLARE_ALIGNED_8(uint8_t, vs_bit_buffer[5 * 80 + 4]); /* allow some slack */ |
530 |
const int log2_blocksize = 3-s->avctx->lowres; |
531 |
int is_field_mode[5]; |
532 |
|
533 |
assert((((int)mb_bit_buffer) & 7) == 0); |
534 |
assert((((int)vs_bit_buffer) & 7) == 0); |
535 |
|
536 |
memset(sblock, 0, sizeof(sblock)); |
537 |
|
538 |
/* pass 1 : read DC and AC coefficients in blocks */
|
539 |
buf_ptr = &s->buf[work_chunk->buf_offset*80];
|
540 |
block1 = &sblock[0][0]; |
541 |
mb1 = mb_data; |
542 |
init_put_bits(&vs_pb, vs_bit_buffer, 5 * 80); |
543 |
for (mb_index = 0; mb_index < 5; mb_index++, mb1 += s->sys->bpm, block1 += s->sys->bpm * 64) { |
544 |
/* skip header */
|
545 |
quant = buf_ptr[3] & 0x0f; |
546 |
buf_ptr += 4;
|
547 |
init_put_bits(&pb, mb_bit_buffer, 80);
|
548 |
mb = mb1; |
549 |
block = block1; |
550 |
is_field_mode[mb_index] = 0;
|
551 |
for (j = 0; j < s->sys->bpm; j++) { |
552 |
last_index = s->sys->block_sizes[j]; |
553 |
init_get_bits(&gb, buf_ptr, last_index); |
554 |
|
555 |
/* get the dc */
|
556 |
dc = get_sbits(&gb, 9);
|
557 |
dct_mode = get_bits1(&gb); |
558 |
class1 = get_bits(&gb, 2);
|
559 |
if (DV_PROFILE_IS_HD(s->sys)) {
|
560 |
mb->idct_put = s->idct_put[0];
|
561 |
mb->scan_table = s->dv_zigzag[0];
|
562 |
mb->factor_table = &s->sys->idct_factor[(j >= 4)*4*16*64 + class1*16*64 + quant*64]; |
563 |
is_field_mode[mb_index] |= !j && dct_mode; |
564 |
} else {
|
565 |
mb->idct_put = s->idct_put[dct_mode && log2_blocksize == 3];
|
566 |
mb->scan_table = s->dv_zigzag[dct_mode]; |
567 |
mb->factor_table = &s->sys->idct_factor[(class1 == 3)*2*22*64 + dct_mode*22*64 + |
568 |
(quant + dv_quant_offset[class1])*64];
|
569 |
} |
570 |
dc = dc << 2;
|
571 |
/* convert to unsigned because 128 is not added in the
|
572 |
standard IDCT */
|
573 |
dc += 1024;
|
574 |
block[0] = dc;
|
575 |
buf_ptr += last_index >> 3;
|
576 |
mb->pos = 0;
|
577 |
mb->partial_bit_count = 0;
|
578 |
|
579 |
#ifdef VLC_DEBUG
|
580 |
printf("MB block: %d, %d ", mb_index, j);
|
581 |
#endif
|
582 |
dv_decode_ac(&gb, mb, block); |
583 |
|
584 |
/* write the remaining bits in a new buffer only if the
|
585 |
block is finished */
|
586 |
if (mb->pos >= 64) |
587 |
bit_copy(&pb, &gb); |
588 |
|
589 |
block += 64;
|
590 |
mb++; |
591 |
} |
592 |
|
593 |
/* pass 2 : we can do it just after */
|
594 |
#ifdef VLC_DEBUG
|
595 |
printf("***pass 2 size=%d MB#=%d\n", put_bits_count(&pb), mb_index);
|
596 |
#endif
|
597 |
block = block1; |
598 |
mb = mb1; |
599 |
init_get_bits(&gb, mb_bit_buffer, put_bits_count(&pb)); |
600 |
flush_put_bits(&pb); |
601 |
for (j = 0; j < s->sys->bpm; j++, block += 64, mb++) { |
602 |
if (mb->pos < 64 && get_bits_left(&gb) > 0) { |
603 |
dv_decode_ac(&gb, mb, block); |
604 |
/* if still not finished, no need to parse other blocks */
|
605 |
if (mb->pos < 64) |
606 |
break;
|
607 |
} |
608 |
} |
609 |
/* all blocks are finished, so the extra bytes can be used at
|
610 |
the video segment level */
|
611 |
if (j >= s->sys->bpm)
|
612 |
bit_copy(&vs_pb, &gb); |
613 |
} |
614 |
|
615 |
/* we need a pass other the whole video segment */
|
616 |
#ifdef VLC_DEBUG
|
617 |
printf("***pass 3 size=%d\n", put_bits_count(&vs_pb));
|
618 |
#endif
|
619 |
block = &sblock[0][0]; |
620 |
mb = mb_data; |
621 |
init_get_bits(&gb, vs_bit_buffer, put_bits_count(&vs_pb)); |
622 |
flush_put_bits(&vs_pb); |
623 |
for (mb_index = 0; mb_index < 5; mb_index++) { |
624 |
for (j = 0; j < s->sys->bpm; j++) { |
625 |
if (mb->pos < 64) { |
626 |
#ifdef VLC_DEBUG
|
627 |
printf("start %d:%d\n", mb_index, j);
|
628 |
#endif
|
629 |
dv_decode_ac(&gb, mb, block); |
630 |
} |
631 |
if (mb->pos >= 64 && mb->pos < 127) |
632 |
av_log(NULL, AV_LOG_ERROR, "AC EOB marker is absent pos=%d\n", mb->pos); |
633 |
block += 64;
|
634 |
mb++; |
635 |
} |
636 |
} |
637 |
|
638 |
/* compute idct and place blocks */
|
639 |
block = &sblock[0][0]; |
640 |
mb = mb_data; |
641 |
for (mb_index = 0; mb_index < 5; mb_index++) { |
642 |
dv_calculate_mb_xy(s, work_chunk, mb_index, &mb_x, &mb_y); |
643 |
|
644 |
/* idct_put'ting luminance */
|
645 |
if ((s->sys->pix_fmt == PIX_FMT_YUV420P) ||
|
646 |
(s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x >= (704 / 8)) || |
647 |
(s->sys->height >= 720 && mb_y != 134)) { |
648 |
y_stride = (s->picture.linesize[0] << ((!is_field_mode[mb_index]) * log2_blocksize));
|
649 |
} else {
|
650 |
y_stride = (2 << log2_blocksize);
|
651 |
} |
652 |
y_ptr = s->picture.data[0] + ((mb_y * s->picture.linesize[0] + mb_x) << log2_blocksize); |
653 |
linesize = s->picture.linesize[0] << is_field_mode[mb_index];
|
654 |
mb[0] .idct_put(y_ptr , linesize, block + 0*64); |
655 |
if (s->sys->video_stype == 4) { /* SD 422 */ |
656 |
mb[2].idct_put(y_ptr + (1 << log2_blocksize) , linesize, block + 2*64); |
657 |
} else {
|
658 |
mb[1].idct_put(y_ptr + (1 << log2_blocksize) , linesize, block + 1*64); |
659 |
mb[2].idct_put(y_ptr + y_stride, linesize, block + 2*64); |
660 |
mb[3].idct_put(y_ptr + (1 << log2_blocksize) + y_stride, linesize, block + 3*64); |
661 |
} |
662 |
mb += 4;
|
663 |
block += 4*64; |
664 |
|
665 |
/* idct_put'ting chrominance */
|
666 |
c_offset = (((mb_y >> (s->sys->pix_fmt == PIX_FMT_YUV420P)) * s->picture.linesize[1] +
|
667 |
(mb_x >> ((s->sys->pix_fmt == PIX_FMT_YUV411P) ? 2 : 1))) << log2_blocksize); |
668 |
for (j = 2; j; j--) { |
669 |
uint8_t *c_ptr = s->picture.data[j] + c_offset; |
670 |
if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x >= (704 / 8)) { |
671 |
uint64_t aligned_pixels[64/8]; |
672 |
uint8_t *pixels = (uint8_t*)aligned_pixels; |
673 |
uint8_t *c_ptr1, *ptr1; |
674 |
int x, y;
|
675 |
mb->idct_put(pixels, 8, block);
|
676 |
for (y = 0; y < (1 << log2_blocksize); y++, c_ptr += s->picture.linesize[j], pixels += 8) { |
677 |
ptr1 = pixels + (1 << (log2_blocksize - 1)); |
678 |
c_ptr1 = c_ptr + (s->picture.linesize[j] << log2_blocksize); |
679 |
for (x = 0; x < (1 << (log2_blocksize - 1)); x++) { |
680 |
c_ptr[x] = pixels[x]; |
681 |
c_ptr1[x] = ptr1[x]; |
682 |
} |
683 |
} |
684 |
block += 64; mb++;
|
685 |
} else {
|
686 |
y_stride = (mb_y == 134) ? (1 << log2_blocksize) : |
687 |
s->picture.linesize[j] << ((!is_field_mode[mb_index]) * log2_blocksize); |
688 |
linesize = s->picture.linesize[j] << is_field_mode[mb_index]; |
689 |
(mb++)-> idct_put(c_ptr , linesize, block); block += 64;
|
690 |
if (s->sys->bpm == 8) { |
691 |
(mb++)->idct_put(c_ptr + y_stride, linesize, block); block += 64;
|
692 |
} |
693 |
} |
694 |
} |
695 |
} |
696 |
return 0; |
697 |
} |
698 |
|
699 |
#if CONFIG_SMALL
|
700 |
/* Converts run and level (where level != 0) pair into vlc, returning bit size */
|
701 |
static av_always_inline int dv_rl2vlc(int run, int level, int sign, uint32_t* vlc) |
702 |
{ |
703 |
int size;
|
704 |
if (run < DV_VLC_MAP_RUN_SIZE && level < DV_VLC_MAP_LEV_SIZE) {
|
705 |
*vlc = dv_vlc_map[run][level].vlc | sign; |
706 |
size = dv_vlc_map[run][level].size; |
707 |
} |
708 |
else {
|
709 |
if (level < DV_VLC_MAP_LEV_SIZE) {
|
710 |
*vlc = dv_vlc_map[0][level].vlc | sign;
|
711 |
size = dv_vlc_map[0][level].size;
|
712 |
} else {
|
713 |
*vlc = 0xfe00 | (level << 1) | sign; |
714 |
size = 16;
|
715 |
} |
716 |
if (run) {
|
717 |
*vlc |= ((run < 16) ? dv_vlc_map[run-1][0].vlc : |
718 |
(0x1f80 | (run - 1))) << size; |
719 |
size += (run < 16) ? dv_vlc_map[run-1][0].size : 13; |
720 |
} |
721 |
} |
722 |
|
723 |
return size;
|
724 |
} |
725 |
|
726 |
static av_always_inline int dv_rl2vlc_size(int run, int level) |
727 |
{ |
728 |
int size;
|
729 |
|
730 |
if (run < DV_VLC_MAP_RUN_SIZE && level < DV_VLC_MAP_LEV_SIZE) {
|
731 |
size = dv_vlc_map[run][level].size; |
732 |
} |
733 |
else {
|
734 |
size = (level < DV_VLC_MAP_LEV_SIZE) ? dv_vlc_map[0][level].size : 16; |
735 |
if (run) {
|
736 |
size += (run < 16) ? dv_vlc_map[run-1][0].size : 13; |
737 |
} |
738 |
} |
739 |
return size;
|
740 |
} |
741 |
#else
|
742 |
static av_always_inline int dv_rl2vlc(int run, int l, int sign, uint32_t* vlc) |
743 |
{ |
744 |
*vlc = dv_vlc_map[run][l].vlc | sign; |
745 |
return dv_vlc_map[run][l].size;
|
746 |
} |
747 |
|
748 |
static av_always_inline int dv_rl2vlc_size(int run, int l) |
749 |
{ |
750 |
return dv_vlc_map[run][l].size;
|
751 |
} |
752 |
#endif
|
753 |
|
754 |
typedef struct EncBlockInfo { |
755 |
int area_q[4]; |
756 |
int bit_size[4]; |
757 |
int prev[5]; |
758 |
int cur_ac;
|
759 |
int cno;
|
760 |
int dct_mode;
|
761 |
DCTELEM mb[64];
|
762 |
uint8_t next[64];
|
763 |
uint8_t sign[64];
|
764 |
uint8_t partial_bit_count; |
765 |
uint32_t partial_bit_buffer; /* we can't use uint16_t here */
|
766 |
} EncBlockInfo; |
767 |
|
768 |
static av_always_inline PutBitContext* dv_encode_ac(EncBlockInfo* bi,
|
769 |
PutBitContext* pb_pool, |
770 |
PutBitContext* pb_end) |
771 |
{ |
772 |
int prev, bits_left;
|
773 |
PutBitContext* pb = pb_pool; |
774 |
int size = bi->partial_bit_count;
|
775 |
uint32_t vlc = bi->partial_bit_buffer; |
776 |
|
777 |
bi->partial_bit_count = bi->partial_bit_buffer = 0;
|
778 |
for (;;){
|
779 |
/* Find suitable storage space */
|
780 |
for (; size > (bits_left = put_bits_left(pb)); pb++) {
|
781 |
if (bits_left) {
|
782 |
size -= bits_left; |
783 |
put_bits(pb, bits_left, vlc >> size); |
784 |
vlc = vlc & ((1 << size) - 1); |
785 |
} |
786 |
if (pb + 1 >= pb_end) { |
787 |
bi->partial_bit_count = size; |
788 |
bi->partial_bit_buffer = vlc; |
789 |
return pb;
|
790 |
} |
791 |
} |
792 |
|
793 |
/* Store VLC */
|
794 |
put_bits(pb, size, vlc); |
795 |
|
796 |
if (bi->cur_ac >= 64) |
797 |
break;
|
798 |
|
799 |
/* Construct the next VLC */
|
800 |
prev = bi->cur_ac; |
801 |
bi->cur_ac = bi->next[prev]; |
802 |
if (bi->cur_ac < 64){ |
803 |
size = dv_rl2vlc(bi->cur_ac - prev - 1, bi->mb[bi->cur_ac], bi->sign[bi->cur_ac], &vlc);
|
804 |
} else {
|
805 |
size = 4; vlc = 6; /* End Of Block stamp */ |
806 |
} |
807 |
} |
808 |
return pb;
|
809 |
} |
810 |
|
811 |
static av_always_inline int dv_guess_dct_mode(DVVideoContext *s, uint8_t *data, int linesize) { |
812 |
if (s->avctx->flags & CODEC_FLAG_INTERLACED_DCT) {
|
813 |
int ps = s->ildct_cmp(NULL, data, NULL, linesize, 8) - 400; |
814 |
if (ps > 0) { |
815 |
int is = s->ildct_cmp(NULL, data , NULL, linesize<<1, 4) + |
816 |
s->ildct_cmp(NULL, data + linesize, NULL, linesize<<1, 4); |
817 |
return (ps > is);
|
818 |
} |
819 |
} |
820 |
|
821 |
return 0; |
822 |
} |
823 |
|
824 |
static av_always_inline int dv_init_enc_block(EncBlockInfo* bi, uint8_t *data, int linesize, DVVideoContext *s, int bias) |
825 |
{ |
826 |
const int *weight; |
827 |
const uint8_t* zigzag_scan;
|
828 |
DECLARE_ALIGNED_16(DCTELEM, blk[64]);
|
829 |
int i, area;
|
830 |
/* We offer two different methods for class number assignment: the
|
831 |
method suggested in SMPTE 314M Table 22, and an improved
|
832 |
method. The SMPTE method is very conservative; it assigns class
|
833 |
3 (i.e. severe quantization) to any block where the largest AC
|
834 |
component is greater than 36. FFmpeg's DV encoder tracks AC bit
|
835 |
consumption precisely, so there is no need to bias most blocks
|
836 |
towards strongly lossy compression. Instead, we assign class 2
|
837 |
to most blocks, and use class 3 only when strictly necessary
|
838 |
(for blocks whose largest AC component exceeds 255). */
|
839 |
|
840 |
#if 0 /* SMPTE spec method */
|
841 |
static const int classes[] = {12, 24, 36, 0xffff};
|
842 |
#else /* improved FFmpeg method */
|
843 |
static const int classes[] = {-1, -1, 255, 0xffff}; |
844 |
#endif
|
845 |
int max = classes[0]; |
846 |
int prev = 0; |
847 |
|
848 |
assert((((int)blk) & 15) == 0); |
849 |
|
850 |
bi->area_q[0] = bi->area_q[1] = bi->area_q[2] = bi->area_q[3] = 0; |
851 |
bi->partial_bit_count = 0;
|
852 |
bi->partial_bit_buffer = 0;
|
853 |
bi->cur_ac = 0;
|
854 |
if (data) {
|
855 |
bi->dct_mode = dv_guess_dct_mode(s, data, linesize); |
856 |
s->get_pixels(blk, data, linesize); |
857 |
s->fdct[bi->dct_mode](blk); |
858 |
} else {
|
859 |
/* We rely on the fact that encoding all zeros leads to an immediate EOB,
|
860 |
which is precisely what the spec calls for in the "dummy" blocks. */
|
861 |
memset(blk, 0, sizeof(blk)); |
862 |
bi->dct_mode = 0;
|
863 |
} |
864 |
bi->mb[0] = blk[0]; |
865 |
|
866 |
zigzag_scan = bi->dct_mode ? ff_zigzag248_direct : ff_zigzag_direct; |
867 |
weight = bi->dct_mode ? dv_weight_248 : dv_weight_88; |
868 |
|
869 |
for (area = 0; area < 4; area++) { |
870 |
bi->prev[area] = prev; |
871 |
bi->bit_size[area] = 1; // 4 areas 4 bits for EOB :) |
872 |
for (i = mb_area_start[area]; i < mb_area_start[area+1]; i++) { |
873 |
int level = blk[zigzag_scan[i]];
|
874 |
|
875 |
if (level + 15 > 30U) { |
876 |
bi->sign[i] = (level >> 31) & 1; |
877 |
/* weigh it and and shift down into range, adding for rounding */
|
878 |
/* the extra division by a factor of 2^4 reverses the 8x expansion of the DCT
|
879 |
AND the 2x doubling of the weights */
|
880 |
level = (FFABS(level) * weight[i] + (1 << (dv_weight_bits+3))) >> (dv_weight_bits+4); |
881 |
bi->mb[i] = level; |
882 |
if (level > max)
|
883 |
max = level; |
884 |
bi->bit_size[area] += dv_rl2vlc_size(i - prev - 1, level);
|
885 |
bi->next[prev]= i; |
886 |
prev = i; |
887 |
} |
888 |
} |
889 |
} |
890 |
bi->next[prev]= i; |
891 |
for (bi->cno = 0; max > classes[bi->cno]; bi->cno++); |
892 |
|
893 |
bi->cno += bias; |
894 |
|
895 |
if (bi->cno >= 3) { |
896 |
bi->cno = 3;
|
897 |
prev = 0;
|
898 |
i = bi->next[prev]; |
899 |
for (area = 0; area < 4; area++) { |
900 |
bi->prev[area] = prev; |
901 |
bi->bit_size[area] = 1; // 4 areas 4 bits for EOB :) |
902 |
for (; i < mb_area_start[area+1]; i = bi->next[i]) { |
903 |
bi->mb[i] >>= 1;
|
904 |
|
905 |
if (bi->mb[i]) {
|
906 |
bi->bit_size[area] += dv_rl2vlc_size(i - prev - 1, bi->mb[i]);
|
907 |
bi->next[prev]= i; |
908 |
prev = i; |
909 |
} |
910 |
} |
911 |
} |
912 |
bi->next[prev]= i; |
913 |
} |
914 |
|
915 |
return bi->bit_size[0] + bi->bit_size[1] + bi->bit_size[2] + bi->bit_size[3]; |
916 |
} |
917 |
|
918 |
static inline void dv_guess_qnos(EncBlockInfo* blks, int* qnos) |
919 |
{ |
920 |
int size[5]; |
921 |
int i, j, k, a, prev, a2;
|
922 |
EncBlockInfo* b; |
923 |
|
924 |
size[0] = size[1] = size[2] = size[3] = size[4] = 1 << 24; |
925 |
do {
|
926 |
b = blks; |
927 |
for (i = 0; i < 5; i++) { |
928 |
if (!qnos[i])
|
929 |
continue;
|
930 |
|
931 |
qnos[i]--; |
932 |
size[i] = 0;
|
933 |
for (j = 0; j < 6; j++, b++) { |
934 |
for (a = 0; a < 4; a++) { |
935 |
if (b->area_q[a] != dv_quant_shifts[qnos[i] + dv_quant_offset[b->cno]][a]) {
|
936 |
b->bit_size[a] = 1; // 4 areas 4 bits for EOB :) |
937 |
b->area_q[a]++; |
938 |
prev = b->prev[a]; |
939 |
assert(b->next[prev] >= mb_area_start[a+1] || b->mb[prev]);
|
940 |
for (k = b->next[prev] ; k < mb_area_start[a+1]; k = b->next[k]) { |
941 |
b->mb[k] >>= 1;
|
942 |
if (b->mb[k]) {
|
943 |
b->bit_size[a] += dv_rl2vlc_size(k - prev - 1, b->mb[k]);
|
944 |
prev = k; |
945 |
} else {
|
946 |
if (b->next[k] >= mb_area_start[a+1] && b->next[k]<64){ |
947 |
for (a2 = a + 1; b->next[k] >= mb_area_start[a2+1]; a2++) |
948 |
b->prev[a2] = prev; |
949 |
assert(a2 < 4);
|
950 |
assert(b->mb[b->next[k]]); |
951 |
b->bit_size[a2] += dv_rl2vlc_size(b->next[k] - prev - 1, b->mb[b->next[k]])
|
952 |
-dv_rl2vlc_size(b->next[k] - k - 1, b->mb[b->next[k]]);
|
953 |
assert(b->prev[a2] == k && (a2 + 1 >= 4 || b->prev[a2+1] != k)); |
954 |
b->prev[a2] = prev; |
955 |
} |
956 |
b->next[prev] = b->next[k]; |
957 |
} |
958 |
} |
959 |
b->prev[a+1]= prev;
|
960 |
} |
961 |
size[i] += b->bit_size[a]; |
962 |
} |
963 |
} |
964 |
if (vs_total_ac_bits >= size[0] + size[1] + size[2] + size[3] + size[4]) |
965 |
return;
|
966 |
} |
967 |
} while (qnos[0]|qnos[1]|qnos[2]|qnos[3]|qnos[4]); |
968 |
|
969 |
|
970 |
for (a = 2; a == 2 || vs_total_ac_bits < size[0]; a += a){ |
971 |
b = blks; |
972 |
size[0] = 5 * 6 * 4; //EOB |
973 |
for (j = 0; j < 6 *5; j++, b++) { |
974 |
prev = b->prev[0];
|
975 |
for (k = b->next[prev]; k < 64; k = b->next[k]) { |
976 |
if (b->mb[k] < a && b->mb[k] > -a){
|
977 |
b->next[prev] = b->next[k]; |
978 |
}else{
|
979 |
size[0] += dv_rl2vlc_size(k - prev - 1, b->mb[k]); |
980 |
prev = k; |
981 |
} |
982 |
} |
983 |
} |
984 |
} |
985 |
} |
986 |
|
987 |
static int dv_encode_video_segment(AVCodecContext *avctx, void *arg) |
988 |
{ |
989 |
DVVideoContext *s = avctx->priv_data; |
990 |
DVwork_chunk *work_chunk = arg; |
991 |
int mb_index, i, j;
|
992 |
int mb_x, mb_y, c_offset, linesize, y_stride;
|
993 |
uint8_t* y_ptr; |
994 |
uint8_t* dif; |
995 |
uint8_t scratch[64];
|
996 |
EncBlockInfo enc_blks[5*DV_MAX_BPM];
|
997 |
PutBitContext pbs[5*DV_MAX_BPM];
|
998 |
PutBitContext* pb; |
999 |
EncBlockInfo* enc_blk; |
1000 |
int vs_bit_size = 0; |
1001 |
int qnos[5] = {15, 15, 15, 15, 15}; /* No quantization */ |
1002 |
int* qnosp = &qnos[0]; |
1003 |
|
1004 |
dif = &s->buf[work_chunk->buf_offset*80];
|
1005 |
enc_blk = &enc_blks[0];
|
1006 |
for (mb_index = 0; mb_index < 5; mb_index++) { |
1007 |
dv_calculate_mb_xy(s, work_chunk, mb_index, &mb_x, &mb_y); |
1008 |
|
1009 |
/* initializing luminance blocks */
|
1010 |
if ((s->sys->pix_fmt == PIX_FMT_YUV420P) ||
|
1011 |
(s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x >= (704 / 8)) || |
1012 |
(s->sys->height >= 720 && mb_y != 134)) { |
1013 |
y_stride = s->picture.linesize[0] << 3; |
1014 |
} else {
|
1015 |
y_stride = 16;
|
1016 |
} |
1017 |
y_ptr = s->picture.data[0] + ((mb_y * s->picture.linesize[0] + mb_x) << 3); |
1018 |
linesize = s->picture.linesize[0];
|
1019 |
|
1020 |
if (s->sys->video_stype == 4) { /* SD 422 */ |
1021 |
vs_bit_size += |
1022 |
dv_init_enc_block(enc_blk+0, y_ptr , linesize, s, 0) + |
1023 |
dv_init_enc_block(enc_blk+1, NULL , linesize, s, 0) + |
1024 |
dv_init_enc_block(enc_blk+2, y_ptr + 8 , linesize, s, 0) + |
1025 |
dv_init_enc_block(enc_blk+3, NULL , linesize, s, 0); |
1026 |
} else {
|
1027 |
vs_bit_size += |
1028 |
dv_init_enc_block(enc_blk+0, y_ptr , linesize, s, 0) + |
1029 |
dv_init_enc_block(enc_blk+1, y_ptr + 8 , linesize, s, 0) + |
1030 |
dv_init_enc_block(enc_blk+2, y_ptr + y_stride, linesize, s, 0) + |
1031 |
dv_init_enc_block(enc_blk+3, y_ptr + 8 + y_stride, linesize, s, 0); |
1032 |
} |
1033 |
enc_blk += 4;
|
1034 |
|
1035 |
/* initializing chrominance blocks */
|
1036 |
c_offset = (((mb_y >> (s->sys->pix_fmt == PIX_FMT_YUV420P)) * s->picture.linesize[1] +
|
1037 |
(mb_x >> ((s->sys->pix_fmt == PIX_FMT_YUV411P) ? 2 : 1))) << 3); |
1038 |
for (j = 2; j; j--) { |
1039 |
uint8_t *c_ptr = s->picture.data[j] + c_offset; |
1040 |
linesize = s->picture.linesize[j]; |
1041 |
y_stride = (mb_y == 134) ? 8 : (s->picture.linesize[j] << 3); |
1042 |
if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x >= (704 / 8)) { |
1043 |
uint8_t* d; |
1044 |
uint8_t* b = scratch; |
1045 |
for (i = 0; i < 8; i++) { |
1046 |
d = c_ptr + (linesize << 3);
|
1047 |
b[0] = c_ptr[0]; b[1] = c_ptr[1]; b[2] = c_ptr[2]; b[3] = c_ptr[3]; |
1048 |
b[4] = d[0]; b[5] = d[1]; b[6] = d[2]; b[7] = d[3]; |
1049 |
c_ptr += linesize; |
1050 |
b += 8;
|
1051 |
} |
1052 |
c_ptr = scratch; |
1053 |
linesize = 8;
|
1054 |
} |
1055 |
|
1056 |
vs_bit_size += dv_init_enc_block( enc_blk++, c_ptr , linesize, s, 1);
|
1057 |
if (s->sys->bpm == 8) { |
1058 |
vs_bit_size += dv_init_enc_block(enc_blk++, c_ptr + y_stride, linesize, s, 1);
|
1059 |
} |
1060 |
} |
1061 |
} |
1062 |
|
1063 |
if (vs_total_ac_bits < vs_bit_size)
|
1064 |
dv_guess_qnos(&enc_blks[0], qnosp);
|
1065 |
|
1066 |
/* DIF encoding process */
|
1067 |
for (j=0; j<5*s->sys->bpm;) { |
1068 |
int start_mb = j;
|
1069 |
|
1070 |
dif[3] = *qnosp++;
|
1071 |
dif += 4;
|
1072 |
|
1073 |
/* First pass over individual cells only */
|
1074 |
for (i=0; i<s->sys->bpm; i++, j++) { |
1075 |
int sz = s->sys->block_sizes[i]>>3; |
1076 |
|
1077 |
init_put_bits(&pbs[j], dif, sz); |
1078 |
put_bits(&pbs[j], 9, (uint16_t)(((enc_blks[j].mb[0] >> 3) - 1024 + 2) >> 2)); |
1079 |
put_bits(&pbs[j], 1, enc_blks[j].dct_mode);
|
1080 |
put_bits(&pbs[j], 2, enc_blks[j].cno);
|
1081 |
|
1082 |
dv_encode_ac(&enc_blks[j], &pbs[j], &pbs[j+1]);
|
1083 |
dif += sz; |
1084 |
} |
1085 |
|
1086 |
/* Second pass over each MB space */
|
1087 |
pb = &pbs[start_mb]; |
1088 |
for (i=0; i<s->sys->bpm; i++) { |
1089 |
if (enc_blks[start_mb+i].partial_bit_count)
|
1090 |
pb = dv_encode_ac(&enc_blks[start_mb+i], pb, &pbs[start_mb+s->sys->bpm]); |
1091 |
} |
1092 |
} |
1093 |
|
1094 |
/* Third and final pass over the whole video segment space */
|
1095 |
pb = &pbs[0];
|
1096 |
for (j=0; j<5*s->sys->bpm; j++) { |
1097 |
if (enc_blks[j].partial_bit_count)
|
1098 |
pb = dv_encode_ac(&enc_blks[j], pb, &pbs[s->sys->bpm*5]);
|
1099 |
if (enc_blks[j].partial_bit_count)
|
1100 |
av_log(NULL, AV_LOG_ERROR, "ac bitstream overflow\n"); |
1101 |
} |
1102 |
|
1103 |
for (j=0; j<5*s->sys->bpm; j++) |
1104 |
flush_put_bits(&pbs[j]); |
1105 |
|
1106 |
return 0; |
1107 |
} |
1108 |
|
1109 |
#if CONFIG_DVVIDEO_DECODER
|
1110 |
/* NOTE: exactly one frame must be given (120000 bytes for NTSC,
|
1111 |
144000 bytes for PAL - or twice those for 50Mbps) */
|
1112 |
static int dvvideo_decode_frame(AVCodecContext *avctx, |
1113 |
void *data, int *data_size, |
1114 |
const uint8_t *buf, int buf_size) |
1115 |
{ |
1116 |
DVVideoContext *s = avctx->priv_data; |
1117 |
|
1118 |
s->sys = dv_frame_profile(buf); |
1119 |
if (!s->sys || buf_size < s->sys->frame_size || dv_init_dynamic_tables(s->sys))
|
1120 |
return -1; /* NOTE: we only accept several full frames */ |
1121 |
|
1122 |
if (s->picture.data[0]) |
1123 |
avctx->release_buffer(avctx, &s->picture); |
1124 |
|
1125 |
s->picture.reference = 0;
|
1126 |
s->picture.key_frame = 1;
|
1127 |
s->picture.pict_type = FF_I_TYPE; |
1128 |
avctx->pix_fmt = s->sys->pix_fmt; |
1129 |
avctx->time_base = s->sys->time_base; |
1130 |
avcodec_set_dimensions(avctx, s->sys->width, s->sys->height); |
1131 |
if (avctx->get_buffer(avctx, &s->picture) < 0) { |
1132 |
av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
|
1133 |
return -1; |
1134 |
} |
1135 |
s->picture.interlaced_frame = 1;
|
1136 |
s->picture.top_field_first = 0;
|
1137 |
|
1138 |
s->buf = buf; |
1139 |
avctx->execute(avctx, dv_decode_video_segment, s->sys->work_chunks, NULL,
|
1140 |
dv_work_pool_size(s->sys), sizeof(DVwork_chunk));
|
1141 |
|
1142 |
emms_c(); |
1143 |
|
1144 |
/* return image */
|
1145 |
*data_size = sizeof(AVFrame);
|
1146 |
*(AVFrame*)data = s->picture; |
1147 |
|
1148 |
return s->sys->frame_size;
|
1149 |
} |
1150 |
#endif /* CONFIG_DVVIDEO_DECODER */ |
1151 |
|
1152 |
|
1153 |
static inline int dv_write_pack(enum dv_pack_type pack_id, DVVideoContext *c, |
1154 |
uint8_t* buf) |
1155 |
{ |
1156 |
/*
|
1157 |
* Here's what SMPTE314M says about these two:
|
1158 |
* (page 6) APTn, AP1n, AP2n, AP3n: These data shall be identical
|
1159 |
* as track application IDs (APTn = 001, AP1n =
|
1160 |
* 001, AP2n = 001, AP3n = 001), if the source signal
|
1161 |
* comes from a digital VCR. If the signal source is
|
1162 |
* unknown, all bits for these data shall be set to 1.
|
1163 |
* (page 12) STYPE: STYPE defines a signal type of video signal
|
1164 |
* 00000b = 4:1:1 compression
|
1165 |
* 00100b = 4:2:2 compression
|
1166 |
* XXXXXX = Reserved
|
1167 |
* Now, I've got two problems with these statements:
|
1168 |
* 1. it looks like APT == 111b should be a safe bet, but it isn't.
|
1169 |
* It seems that for PAL as defined in IEC 61834 we have to set
|
1170 |
* APT to 000 and for SMPTE314M to 001.
|
1171 |
* 2. It is not at all clear what STYPE is used for 4:2:0 PAL
|
1172 |
* compression scheme (if any).
|
1173 |
*/
|
1174 |
int apt = (c->sys->pix_fmt == PIX_FMT_YUV420P ? 0 : 1); |
1175 |
|
1176 |
uint8_t aspect = 0;
|
1177 |
if ((int)(av_q2d(c->avctx->sample_aspect_ratio) * c->avctx->width / c->avctx->height * 10) >= 17) /* 16:9 */ |
1178 |
aspect = 0x02;
|
1179 |
|
1180 |
buf[0] = (uint8_t)pack_id;
|
1181 |
switch (pack_id) {
|
1182 |
case dv_header525: /* I can't imagine why these two weren't defined as real */ |
1183 |
case dv_header625: /* packs in SMPTE314M -- they definitely look like ones */ |
1184 |
buf[1] = 0xf8 | /* reserved -- always 1 */ |
1185 |
(apt & 0x07); /* APT: Track application ID */ |
1186 |
buf[2] = (0 << 7) | /* TF1: audio data is 0 - valid; 1 - invalid */ |
1187 |
(0x0f << 3) | /* reserved -- always 1 */ |
1188 |
(apt & 0x07); /* AP1: Audio application ID */ |
1189 |
buf[3] = (0 << 7) | /* TF2: video data is 0 - valid; 1 - invalid */ |
1190 |
(0x0f << 3) | /* reserved -- always 1 */ |
1191 |
(apt & 0x07); /* AP2: Video application ID */ |
1192 |
buf[4] = (0 << 7) | /* TF3: subcode(SSYB) is 0 - valid; 1 - invalid */ |
1193 |
(0x0f << 3) | /* reserved -- always 1 */ |
1194 |
(apt & 0x07); /* AP3: Subcode application ID */ |
1195 |
break;
|
1196 |
case dv_video_source:
|
1197 |
buf[1] = 0xff; /* reserved -- always 1 */ |
1198 |
buf[2] = (1 << 7) | /* B/W: 0 - b/w, 1 - color */ |
1199 |
(1 << 6) | /* following CLF is valid - 0, invalid - 1 */ |
1200 |
(3 << 4) | /* CLF: color frames ID (see ITU-R BT.470-4) */ |
1201 |
0xf; /* reserved -- always 1 */ |
1202 |
buf[3] = (3 << 6) | /* reserved -- always 1 */ |
1203 |
(c->sys->dsf << 5) | /* system: 60fields/50fields */ |
1204 |
c->sys->video_stype; /* signal type video compression */
|
1205 |
buf[4] = 0xff; /* VISC: 0xff -- no information */ |
1206 |
break;
|
1207 |
case dv_video_control:
|
1208 |
buf[1] = (0 << 6) | /* Copy generation management (CGMS) 0 -- free */ |
1209 |
0x3f; /* reserved -- always 1 */ |
1210 |
buf[2] = 0xc8 | /* reserved -- always b11001xxx */ |
1211 |
aspect; |
1212 |
buf[3] = (1 << 7) | /* frame/field flag 1 -- frame, 0 -- field */ |
1213 |
(1 << 6) | /* first/second field flag 0 -- field 2, 1 -- field 1 */ |
1214 |
(1 << 5) | /* frame change flag 0 -- same picture as before, 1 -- different */ |
1215 |
(1 << 4) | /* 1 - interlaced, 0 - noninterlaced */ |
1216 |
0xc; /* reserved -- always b1100 */ |
1217 |
buf[4] = 0xff; /* reserved -- always 1 */ |
1218 |
break;
|
1219 |
default:
|
1220 |
buf[1] = buf[2] = buf[3] = buf[4] = 0xff; |
1221 |
} |
1222 |
return 5; |
1223 |
} |
1224 |
|
1225 |
#if CONFIG_DVVIDEO_ENCODER
|
1226 |
static void dv_format_frame(DVVideoContext* c, uint8_t* buf) |
1227 |
{ |
1228 |
int chan, i, j, k;
|
1229 |
|
1230 |
for (chan = 0; chan < c->sys->n_difchan; chan++) { |
1231 |
for (i = 0; i < c->sys->difseg_size; i++) { |
1232 |
memset(buf, 0xff, 80 * 6); /* first 6 DIF blocks are for control data */ |
1233 |
|
1234 |
/* DV header: 1DIF */
|
1235 |
buf += dv_write_dif_id(dv_sect_header, chan, i, 0, buf);
|
1236 |
buf += dv_write_pack((c->sys->dsf ? dv_header625 : dv_header525), c, buf); |
1237 |
buf += 72; /* unused bytes */ |
1238 |
|
1239 |
/* DV subcode: 2DIFs */
|
1240 |
for (j = 0; j < 2; j++) { |
1241 |
buf += dv_write_dif_id(dv_sect_subcode, chan, i, j, buf); |
1242 |
for (k = 0; k < 6; k++) |
1243 |
buf += dv_write_ssyb_id(k, (i < c->sys->difseg_size/2), buf) + 5; |
1244 |
buf += 29; /* unused bytes */ |
1245 |
} |
1246 |
|
1247 |
/* DV VAUX: 3DIFS */
|
1248 |
for (j = 0; j < 3; j++) { |
1249 |
buf += dv_write_dif_id(dv_sect_vaux, chan, i, j, buf); |
1250 |
buf += dv_write_pack(dv_video_source, c, buf); |
1251 |
buf += dv_write_pack(dv_video_control, c, buf); |
1252 |
buf += 7*5; |
1253 |
buf += dv_write_pack(dv_video_source, c, buf); |
1254 |
buf += dv_write_pack(dv_video_control, c, buf); |
1255 |
buf += 4*5 + 2; /* unused bytes */ |
1256 |
} |
1257 |
|
1258 |
/* DV Audio/Video: 135 Video DIFs + 9 Audio DIFs */
|
1259 |
for (j = 0; j < 135; j++) { |
1260 |
if (j%15 == 0) { |
1261 |
memset(buf, 0xff, 80); |
1262 |
buf += dv_write_dif_id(dv_sect_audio, chan, i, j/15, buf);
|
1263 |
buf += 77; /* audio control & shuffled PCM audio */ |
1264 |
} |
1265 |
buf += dv_write_dif_id(dv_sect_video, chan, i, j, buf); |
1266 |
buf += 77; /* 1 video macroblock: 1 bytes control |
1267 |
4 * 14 bytes Y 8x8 data
|
1268 |
10 bytes Cr 8x8 data
|
1269 |
10 bytes Cb 8x8 data */
|
1270 |
} |
1271 |
} |
1272 |
} |
1273 |
} |
1274 |
|
1275 |
|
1276 |
static int dvvideo_encode_frame(AVCodecContext *c, uint8_t *buf, int buf_size, |
1277 |
void *data)
|
1278 |
{ |
1279 |
DVVideoContext *s = c->priv_data; |
1280 |
|
1281 |
s->sys = dv_codec_profile(c); |
1282 |
if (!s->sys || buf_size < s->sys->frame_size || dv_init_dynamic_tables(s->sys))
|
1283 |
return -1; |
1284 |
|
1285 |
c->pix_fmt = s->sys->pix_fmt; |
1286 |
s->picture = *((AVFrame *)data); |
1287 |
s->picture.key_frame = 1;
|
1288 |
s->picture.pict_type = FF_I_TYPE; |
1289 |
|
1290 |
s->buf = buf; |
1291 |
c->execute(c, dv_encode_video_segment, s->sys->work_chunks, NULL,
|
1292 |
dv_work_pool_size(s->sys), sizeof(DVwork_chunk));
|
1293 |
|
1294 |
emms_c(); |
1295 |
|
1296 |
dv_format_frame(s, buf); |
1297 |
|
1298 |
return s->sys->frame_size;
|
1299 |
} |
1300 |
#endif
|
1301 |
|
1302 |
static int dvvideo_close(AVCodecContext *c) |
1303 |
{ |
1304 |
DVVideoContext *s = c->priv_data; |
1305 |
|
1306 |
if (s->picture.data[0]) |
1307 |
c->release_buffer(c, &s->picture); |
1308 |
|
1309 |
return 0; |
1310 |
} |
1311 |
|
1312 |
|
1313 |
#if CONFIG_DVVIDEO_ENCODER
|
1314 |
AVCodec dvvideo_encoder = { |
1315 |
"dvvideo",
|
1316 |
CODEC_TYPE_VIDEO, |
1317 |
CODEC_ID_DVVIDEO, |
1318 |
sizeof(DVVideoContext),
|
1319 |
dvvideo_init, |
1320 |
dvvideo_encode_frame, |
1321 |
.pix_fmts = (enum PixelFormat[]) {PIX_FMT_YUV411P, PIX_FMT_YUV422P, PIX_FMT_YUV420P, PIX_FMT_NONE},
|
1322 |
.long_name = NULL_IF_CONFIG_SMALL("DV (Digital Video)"),
|
1323 |
}; |
1324 |
#endif // CONFIG_DVVIDEO_ENCODER |
1325 |
|
1326 |
#if CONFIG_DVVIDEO_DECODER
|
1327 |
AVCodec dvvideo_decoder = { |
1328 |
"dvvideo",
|
1329 |
CODEC_TYPE_VIDEO, |
1330 |
CODEC_ID_DVVIDEO, |
1331 |
sizeof(DVVideoContext),
|
1332 |
dvvideo_init, |
1333 |
NULL,
|
1334 |
dvvideo_close, |
1335 |
dvvideo_decode_frame, |
1336 |
CODEC_CAP_DR1, |
1337 |
NULL,
|
1338 |
.long_name = NULL_IF_CONFIG_SMALL("DV (Digital Video)"),
|
1339 |
}; |
1340 |
#endif
|