VirtualBox

source: vbox/trunk/src/libs/ffmpeg-20060710/libavcodec/dv.c@ 9441

Last change on this file since 9441 was 5776, checked in by vboxsync, 17 years ago

ffmpeg: exported to OSE

File size: 38.1 KB
Line 
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 <[email protected]>
11 *
12 * Many thanks to Dan Dennedy <[email protected]> for providing wealth
13 * of DV technical info.
14 *
15 * This library is free software; you can redistribute it and/or
16 * modify it under the terms of the GNU Lesser General Public
17 * License as published by the Free Software Foundation; either
18 * version 2 of the License, or (at your option) any later version.
19 *
20 * This library is distributed in the hope that it will be useful,
21 * but WITHOUT ANY WARRANTY; without even the implied warranty of
22 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
23 * Lesser General Public License for more details.
24 *
25 * You should have received a copy of the GNU Lesser General Public
26 * License along with this library; if not, write to the Free Software
27 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
28 */
29
30/**
31 * @file dv.c
32 * DV codec.
33 */
34#include "avcodec.h"
35#include "dsputil.h"
36#include "mpegvideo.h"
37#include "simple_idct.h"
38#include "dvdata.h"
39
40//#undef NDEBUG
41//#include <assert.h>
42
43typedef struct DVVideoContext {
44 const DVprofile* sys;
45 AVFrame picture;
46 AVCodecContext *avctx;
47 uint8_t *buf;
48
49 uint8_t dv_zigzag[2][64];
50 uint8_t dv_idct_shift[2][2][22][64];
51
52 void (*get_pixels)(DCTELEM *block, const uint8_t *pixels, int line_size);
53 void (*fdct[2])(DCTELEM *block);
54 void (*idct_put[2])(uint8_t *dest, int line_size, DCTELEM *block);
55} DVVideoContext;
56
57/* MultiThreading - dv_anchor applies to entire DV codec, not just the avcontext */
58/* one element is needed for each video segment in a DV frame */
59/* at most there are 2 DIF channels * 12 DIF sequences * 27 video segments (PAL 50Mbps) */
60#define DV_ANCHOR_SIZE (2*12*27)
61
62static void* dv_anchor[DV_ANCHOR_SIZE];
63
64#define TEX_VLC_BITS 9
65
66#ifdef DV_CODEC_TINY_TARGET
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 */
75static RL_VLC_ELEM *dv_rl_vlc;
76/* VLC encoding lookup table */
77static struct dv_vlc_pair {
78 uint32_t vlc;
79 uint8_t size;
80} (*dv_vlc_map)[DV_VLC_MAP_LEV_SIZE] = NULL;
81
82static void dv_build_unquantize_tables(DVVideoContext *s, uint8_t* perm)
83{
84 int i, q, j;
85
86 /* NOTE: max left shift is 6 */
87 for(q = 0; q < 22; q++) {
88 /* 88DCT */
89 for(i = 1; i < 64; i++) {
90 /* 88 table */
91 j = perm[i];
92 s->dv_idct_shift[0][0][q][j] =
93 dv_quant_shifts[q][dv_88_areas[i]] + 1;
94 s->dv_idct_shift[1][0][q][j] = s->dv_idct_shift[0][0][q][j] + 1;
95 }
96
97 /* 248DCT */
98 for(i = 1; i < 64; i++) {
99 /* 248 table */
100 s->dv_idct_shift[0][1][q][i] =
101 dv_quant_shifts[q][dv_248_areas[i]] + 1;
102 s->dv_idct_shift[1][1][q][i] = s->dv_idct_shift[0][1][q][i] + 1;
103 }
104 }
105}
106
107static int dvvideo_init(AVCodecContext *avctx)
108{
109 DVVideoContext *s = avctx->priv_data;
110 DSPContext dsp;
111 static int done=0;
112 int i, j;
113
114 if (!done) {
115 VLC dv_vlc;
116 uint16_t new_dv_vlc_bits[NB_DV_VLC*2];
117 uint8_t new_dv_vlc_len[NB_DV_VLC*2];
118 uint8_t new_dv_vlc_run[NB_DV_VLC*2];
119 int16_t new_dv_vlc_level[NB_DV_VLC*2];
120
121 done = 1;
122
123 dv_vlc_map = av_mallocz_static(DV_VLC_MAP_LEV_SIZE*DV_VLC_MAP_RUN_SIZE*sizeof(struct dv_vlc_pair));
124 if (!dv_vlc_map)
125 return -ENOMEM;
126
127 /* dv_anchor lets each thread know its Id */
128 for (i=0; i<DV_ANCHOR_SIZE; i++)
129 dv_anchor[i] = (void*)(size_t)i;
130
131 /* it's faster to include sign bit in a generic VLC parsing scheme */
132 for (i=0, j=0; i<NB_DV_VLC; i++, j++) {
133 new_dv_vlc_bits[j] = dv_vlc_bits[i];
134 new_dv_vlc_len[j] = dv_vlc_len[i];
135 new_dv_vlc_run[j] = dv_vlc_run[i];
136 new_dv_vlc_level[j] = dv_vlc_level[i];
137
138 if (dv_vlc_level[i]) {
139 new_dv_vlc_bits[j] <<= 1;
140 new_dv_vlc_len[j]++;
141
142 j++;
143 new_dv_vlc_bits[j] = (dv_vlc_bits[i] << 1) | 1;
144 new_dv_vlc_len[j] = dv_vlc_len[i] + 1;
145 new_dv_vlc_run[j] = dv_vlc_run[i];
146 new_dv_vlc_level[j] = -dv_vlc_level[i];
147 }
148 }
149
150 /* NOTE: as a trick, we use the fact the no codes are unused
151 to accelerate the parsing of partial codes */
152 init_vlc(&dv_vlc, TEX_VLC_BITS, j,
153 new_dv_vlc_len, 1, 1, new_dv_vlc_bits, 2, 2, 0);
154
155 dv_rl_vlc = av_mallocz_static(dv_vlc.table_size * sizeof(RL_VLC_ELEM));
156 if (!dv_rl_vlc)
157 return -ENOMEM;
158
159 for(i = 0; i < dv_vlc.table_size; i++){
160 int code= dv_vlc.table[i][0];
161 int len = dv_vlc.table[i][1];
162 int level, run;
163
164 if(len<0){ //more bits needed
165 run= 0;
166 level= code;
167 } else {
168 run= new_dv_vlc_run[code] + 1;
169 level= new_dv_vlc_level[code];
170 }
171 dv_rl_vlc[i].len = len;
172 dv_rl_vlc[i].level = level;
173 dv_rl_vlc[i].run = run;
174 }
175 free_vlc(&dv_vlc);
176
177 for (i = 0; i < NB_DV_VLC - 1; i++) {
178 if (dv_vlc_run[i] >= DV_VLC_MAP_RUN_SIZE)
179 continue;
180#ifdef DV_CODEC_TINY_TARGET
181 if (dv_vlc_level[i] >= DV_VLC_MAP_LEV_SIZE)
182 continue;
183#endif
184
185 if (dv_vlc_map[dv_vlc_run[i]][dv_vlc_level[i]].size != 0)
186 continue;
187
188 dv_vlc_map[dv_vlc_run[i]][dv_vlc_level[i]].vlc = dv_vlc_bits[i] <<
189 (!!dv_vlc_level[i]);
190 dv_vlc_map[dv_vlc_run[i]][dv_vlc_level[i]].size = dv_vlc_len[i] +
191 (!!dv_vlc_level[i]);
192 }
193 for (i = 0; i < DV_VLC_MAP_RUN_SIZE; i++) {
194#ifdef DV_CODEC_TINY_TARGET
195 for (j = 1; j < DV_VLC_MAP_LEV_SIZE; j++) {
196 if (dv_vlc_map[i][j].size == 0) {
197 dv_vlc_map[i][j].vlc = dv_vlc_map[0][j].vlc |
198 (dv_vlc_map[i-1][0].vlc << (dv_vlc_map[0][j].size));
199 dv_vlc_map[i][j].size = dv_vlc_map[i-1][0].size +
200 dv_vlc_map[0][j].size;
201 }
202 }
203#else
204 for (j = 1; j < DV_VLC_MAP_LEV_SIZE/2; j++) {
205 if (dv_vlc_map[i][j].size == 0) {
206 dv_vlc_map[i][j].vlc = dv_vlc_map[0][j].vlc |
207 (dv_vlc_map[i-1][0].vlc << (dv_vlc_map[0][j].size));
208 dv_vlc_map[i][j].size = dv_vlc_map[i-1][0].size +
209 dv_vlc_map[0][j].size;
210 }
211 dv_vlc_map[i][((uint16_t)(-j))&0x1ff].vlc =
212 dv_vlc_map[i][j].vlc | 1;
213 dv_vlc_map[i][((uint16_t)(-j))&0x1ff].size =
214 dv_vlc_map[i][j].size;
215 }
216#endif
217 }
218 }
219
220 /* Generic DSP setup */
221 dsputil_init(&dsp, avctx);
222 s->get_pixels = dsp.get_pixels;
223
224 /* 88DCT setup */
225 s->fdct[0] = dsp.fdct;
226 s->idct_put[0] = dsp.idct_put;
227 for (i=0; i<64; i++)
228 s->dv_zigzag[0][i] = dsp.idct_permutation[ff_zigzag_direct[i]];
229
230 /* 248DCT setup */
231 s->fdct[1] = dsp.fdct248;
232 s->idct_put[1] = simple_idct248_put; // FIXME: need to add it to DSP
233 if(avctx->lowres){
234 for (i=0; i<64; i++){
235 int j= ff_zigzag248_direct[i];
236 s->dv_zigzag[1][i] = dsp.idct_permutation[(j&7) + (j&8)*4 + (j&48)/2];
237 }
238 }else
239 memcpy(s->dv_zigzag[1], ff_zigzag248_direct, 64);
240
241 /* XXX: do it only for constant case */
242 dv_build_unquantize_tables(s, dsp.idct_permutation);
243
244 avctx->coded_frame = &s->picture;
245 s->avctx= avctx;
246
247 return 0;
248}
249
250// #define VLC_DEBUG
251// #define printf(...) av_log(NULL, AV_LOG_ERROR, __VA_ARGS__)
252
253typedef struct BlockInfo {
254 const uint8_t *shift_table;
255 const uint8_t *scan_table;
256 const int *iweight_table;
257 uint8_t pos; /* position in block */
258 uint8_t dct_mode;
259 uint8_t partial_bit_count;
260 uint16_t partial_bit_buffer;
261 int shift_offset;
262} BlockInfo;
263
264/* block size in bits */
265static const uint16_t block_sizes[6] = {
266 112, 112, 112, 112, 80, 80
267};
268/* bit budget for AC only in 5 MBs */
269static const int vs_total_ac_bits = (100 * 4 + 68*2) * 5;
270/* see dv_88_areas and dv_248_areas for details */
271static const int mb_area_start[5] = { 1, 6, 21, 43, 64 };
272
273#ifndef ALT_BITSTREAM_READER
274#warning only works with ALT_BITSTREAM_READER
275static int re_index; //Hack to make it compile
276#endif
277
278static inline int get_bits_left(GetBitContext *s)
279{
280 return s->size_in_bits - get_bits_count(s);
281}
282
283static inline int get_bits_size(GetBitContext *s)
284{
285 return s->size_in_bits;
286}
287
288static inline int put_bits_left(PutBitContext* s)
289{
290 return (s->buf_end - s->buf) * 8 - put_bits_count(s);
291}
292
293/* decode ac coefs */
294static void dv_decode_ac(GetBitContext *gb, BlockInfo *mb, DCTELEM *block)
295{
296 int last_index = get_bits_size(gb);
297 const uint8_t *scan_table = mb->scan_table;
298 const uint8_t *shift_table = mb->shift_table;
299 const int *iweight_table = mb->iweight_table;
300 int pos = mb->pos;
301 int partial_bit_count = mb->partial_bit_count;
302 int level, pos1, run, vlc_len, index;
303
304 OPEN_READER(re, gb);
305 UPDATE_CACHE(re, gb);
306
307 /* if we must parse a partial vlc, we do it here */
308 if (partial_bit_count > 0) {
309 re_cache = ((unsigned)re_cache >> partial_bit_count) |
310 (mb->partial_bit_buffer << (sizeof(re_cache)*8 - partial_bit_count));
311 re_index -= partial_bit_count;
312 mb->partial_bit_count = 0;
313 }
314
315 /* get the AC coefficients until last_index is reached */
316 for(;;) {
317#ifdef VLC_DEBUG
318 printf("%2d: bits=%04x index=%d\n", pos, SHOW_UBITS(re, gb, 16), re_index);
319#endif
320 /* our own optimized GET_RL_VLC */
321 index = NEG_USR32(re_cache, TEX_VLC_BITS);
322 vlc_len = dv_rl_vlc[index].len;
323 if (vlc_len < 0) {
324 index = NEG_USR32((unsigned)re_cache << TEX_VLC_BITS, -vlc_len) + dv_rl_vlc[index].level;
325 vlc_len = TEX_VLC_BITS - vlc_len;
326 }
327 level = dv_rl_vlc[index].level;
328 run = dv_rl_vlc[index].run;
329
330 /* gotta check if we're still within gb boundaries */
331 if (re_index + vlc_len > last_index) {
332 /* should be < 16 bits otherwise a codeword could have been parsed */
333 mb->partial_bit_count = last_index - re_index;
334 mb->partial_bit_buffer = NEG_USR32(re_cache, mb->partial_bit_count);
335 re_index = last_index;
336 break;
337 }
338 re_index += vlc_len;
339
340#ifdef VLC_DEBUG
341 printf("run=%d level=%d\n", run, level);
342#endif
343 pos += run;
344 if (pos >= 64)
345 break;
346
347 pos1 = scan_table[pos];
348 level <<= shift_table[pos1];
349
350 /* unweigh, round, and shift down */
351 level = (level*iweight_table[pos] + (1 << (dv_iweight_bits-1))) >> dv_iweight_bits;
352
353 block[pos1] = level;
354
355 UPDATE_CACHE(re, gb);
356 }
357 CLOSE_READER(re, gb);
358 mb->pos = pos;
359}
360
361static inline void bit_copy(PutBitContext *pb, GetBitContext *gb)
362{
363 int bits_left = get_bits_left(gb);
364 while (bits_left >= MIN_CACHE_BITS) {
365 put_bits(pb, MIN_CACHE_BITS, get_bits(gb, MIN_CACHE_BITS));
366 bits_left -= MIN_CACHE_BITS;
367 }
368 if (bits_left > 0) {
369 put_bits(pb, bits_left, get_bits(gb, bits_left));
370 }
371}
372
373/* mb_x and mb_y are in units of 8 pixels */
374static inline void dv_decode_video_segment(DVVideoContext *s,
375 uint8_t *buf_ptr1,
376 const uint16_t *mb_pos_ptr)
377{
378 int quant, dc, dct_mode, class1, j;
379 int mb_index, mb_x, mb_y, v, last_index;
380 DCTELEM *block, *block1;
381 int c_offset;
382 uint8_t *y_ptr;
383 void (*idct_put)(uint8_t *dest, int line_size, DCTELEM *block);
384 uint8_t *buf_ptr;
385 PutBitContext pb, vs_pb;
386 GetBitContext gb;
387 BlockInfo mb_data[5 * 6], *mb, *mb1;
388 DECLARE_ALIGNED_8(DCTELEM, sblock[5*6][64]);
389 DECLARE_ALIGNED_8(uint8_t, mb_bit_buffer[80 + 4]); /* allow some slack */
390 DECLARE_ALIGNED_8(uint8_t, vs_bit_buffer[5 * 80 + 4]); /* allow some slack */
391 const int log2_blocksize= 3-s->avctx->lowres;
392
393 assert((((int)mb_bit_buffer)&7)==0);
394 assert((((int)vs_bit_buffer)&7)==0);
395
396 memset(sblock, 0, sizeof(sblock));
397
398 /* pass 1 : read DC and AC coefficients in blocks */
399 buf_ptr = buf_ptr1;
400 block1 = &sblock[0][0];
401 mb1 = mb_data;
402 init_put_bits(&vs_pb, vs_bit_buffer, 5 * 80);
403 for(mb_index = 0; mb_index < 5; mb_index++, mb1 += 6, block1 += 6 * 64) {
404 /* skip header */
405 quant = buf_ptr[3] & 0x0f;
406 buf_ptr += 4;
407 init_put_bits(&pb, mb_bit_buffer, 80);
408 mb = mb1;
409 block = block1;
410 for(j = 0;j < 6; j++) {
411 last_index = block_sizes[j];
412 init_get_bits(&gb, buf_ptr, last_index);
413
414 /* get the dc */
415 dc = get_sbits(&gb, 9);
416 dct_mode = get_bits1(&gb);
417 mb->dct_mode = dct_mode;
418 mb->scan_table = s->dv_zigzag[dct_mode];
419 mb->iweight_table = dct_mode ? dv_iweight_248 : dv_iweight_88;
420 class1 = get_bits(&gb, 2);
421 mb->shift_table = s->dv_idct_shift[class1 == 3][dct_mode]
422 [quant + dv_quant_offset[class1]];
423 dc = dc << 2;
424 /* convert to unsigned because 128 is not added in the
425 standard IDCT */
426 dc += 1024;
427 block[0] = dc;
428 buf_ptr += last_index >> 3;
429 mb->pos = 0;
430 mb->partial_bit_count = 0;
431
432#ifdef VLC_DEBUG
433 printf("MB block: %d, %d ", mb_index, j);
434#endif
435 dv_decode_ac(&gb, mb, block);
436
437 /* write the remaining bits in a new buffer only if the
438 block is finished */
439 if (mb->pos >= 64)
440 bit_copy(&pb, &gb);
441
442 block += 64;
443 mb++;
444 }
445
446 /* pass 2 : we can do it just after */
447#ifdef VLC_DEBUG
448 printf("***pass 2 size=%d MB#=%d\n", put_bits_count(&pb), mb_index);
449#endif
450 block = block1;
451 mb = mb1;
452 init_get_bits(&gb, mb_bit_buffer, put_bits_count(&pb));
453 flush_put_bits(&pb);
454 for(j = 0;j < 6; j++, block += 64, mb++) {
455 if (mb->pos < 64 && get_bits_left(&gb) > 0) {
456 dv_decode_ac(&gb, mb, block);
457 /* if still not finished, no need to parse other blocks */
458 if (mb->pos < 64)
459 break;
460 }
461 }
462 /* all blocks are finished, so the extra bytes can be used at
463 the video segment level */
464 if (j >= 6)
465 bit_copy(&vs_pb, &gb);
466 }
467
468 /* we need a pass other the whole video segment */
469#ifdef VLC_DEBUG
470 printf("***pass 3 size=%d\n", put_bits_count(&vs_pb));
471#endif
472 block = &sblock[0][0];
473 mb = mb_data;
474 init_get_bits(&gb, vs_bit_buffer, put_bits_count(&vs_pb));
475 flush_put_bits(&vs_pb);
476 for(mb_index = 0; mb_index < 5; mb_index++) {
477 for(j = 0;j < 6; j++) {
478 if (mb->pos < 64) {
479#ifdef VLC_DEBUG
480 printf("start %d:%d\n", mb_index, j);
481#endif
482 dv_decode_ac(&gb, mb, block);
483 }
484 if (mb->pos >= 64 && mb->pos < 127)
485 av_log(NULL, AV_LOG_ERROR, "AC EOB marker is absent pos=%d\n", mb->pos);
486 block += 64;
487 mb++;
488 }
489 }
490
491 /* compute idct and place blocks */
492 block = &sblock[0][0];
493 mb = mb_data;
494 for(mb_index = 0; mb_index < 5; mb_index++) {
495 v = *mb_pos_ptr++;
496 mb_x = v & 0xff;
497 mb_y = v >> 8;
498 if (s->sys->pix_fmt == PIX_FMT_YUV422P) {
499 y_ptr = s->picture.data[0] + ((mb_y * s->picture.linesize[0] + (mb_x>>1))<<log2_blocksize);
500 c_offset = ((mb_y * s->picture.linesize[1] + (mb_x >> 2))<<log2_blocksize);
501 } else { /* 4:1:1 or 4:2:0 */
502 y_ptr = s->picture.data[0] + ((mb_y * s->picture.linesize[0] + mb_x)<<log2_blocksize);
503 if (s->sys->pix_fmt == PIX_FMT_YUV411P)
504 c_offset = ((mb_y * s->picture.linesize[1] + (mb_x >> 2))<<log2_blocksize);
505 else /* 4:2:0 */
506 c_offset = (((mb_y >> 1) * s->picture.linesize[1] + (mb_x >> 1))<<log2_blocksize);
507 }
508 for(j = 0;j < 6; j++) {
509 idct_put = s->idct_put[mb->dct_mode && log2_blocksize==3];
510 if (s->sys->pix_fmt == PIX_FMT_YUV422P) { /* 4:2:2 */
511 if (j == 0 || j == 2) {
512 /* Y0 Y1 */
513 idct_put(y_ptr + ((j >> 1)<<log2_blocksize),
514 s->picture.linesize[0], block);
515 } else if(j > 3) {
516 /* Cr Cb */
517 idct_put(s->picture.data[6 - j] + c_offset,
518 s->picture.linesize[6 - j], block);
519 }
520 /* note: j=1 and j=3 are "dummy" blocks in 4:2:2 */
521 } else { /* 4:1:1 or 4:2:0 */
522 if (j < 4) {
523 if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x < (704 / 8)) {
524 /* NOTE: at end of line, the macroblock is handled as 420 */
525 idct_put(y_ptr + (j<<log2_blocksize), s->picture.linesize[0], block);
526 } else {
527 idct_put(y_ptr + (((j & 1) + (j >> 1) * s->picture.linesize[0])<<log2_blocksize),
528 s->picture.linesize[0], block);
529 }
530 } else {
531 if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x >= (704 / 8)) {
532 uint64_t aligned_pixels[64/8];
533 uint8_t *pixels= (uint8_t*)aligned_pixels;
534 uint8_t *c_ptr, *c_ptr1, *ptr, *ptr1;
535 int x, y, linesize;
536 /* NOTE: at end of line, the macroblock is handled as 420 */
537 idct_put(pixels, 8, block);
538 linesize = s->picture.linesize[6 - j];
539 c_ptr = s->picture.data[6 - j] + c_offset;
540 ptr = pixels;
541 for(y = 0;y < (1<<log2_blocksize); y++) {
542 ptr1= ptr + (1<<(log2_blocksize-1));
543 c_ptr1 = c_ptr + (linesize<<log2_blocksize);
544 for(x=0; x < (1<<(log2_blocksize-1)); x++){
545 c_ptr[x]= ptr[x]; c_ptr1[x]= ptr1[x];
546 }
547 c_ptr += linesize;
548 ptr += 8;
549 }
550 } else {
551 /* don't ask me why they inverted Cb and Cr ! */
552 idct_put(s->picture.data[6 - j] + c_offset,
553 s->picture.linesize[6 - j], block);
554 }
555 }
556 }
557 block += 64;
558 mb++;
559 }
560 }
561}
562
563#ifdef DV_CODEC_TINY_TARGET
564/* Converts run and level (where level != 0) pair into vlc, returning bit size */
565static always_inline int dv_rl2vlc(int run, int level, int sign, uint32_t* vlc)
566{
567 int size;
568 if (run < DV_VLC_MAP_RUN_SIZE && level < DV_VLC_MAP_LEV_SIZE) {
569 *vlc = dv_vlc_map[run][level].vlc | sign;
570 size = dv_vlc_map[run][level].size;
571 }
572 else {
573 if (level < DV_VLC_MAP_LEV_SIZE) {
574 *vlc = dv_vlc_map[0][level].vlc | sign;
575 size = dv_vlc_map[0][level].size;
576 } else {
577 *vlc = 0xfe00 | (level << 1) | sign;
578 size = 16;
579 }
580 if (run) {
581 *vlc |= ((run < 16) ? dv_vlc_map[run-1][0].vlc :
582 (0x1f80 | (run - 1))) << size;
583 size += (run < 16) ? dv_vlc_map[run-1][0].size : 13;
584 }
585 }
586
587 return size;
588}
589
590static always_inline int dv_rl2vlc_size(int run, int level)
591{
592 int size;
593
594 if (run < DV_VLC_MAP_RUN_SIZE && level < DV_VLC_MAP_LEV_SIZE) {
595 size = dv_vlc_map[run][level].size;
596 }
597 else {
598 size = (level < DV_VLC_MAP_LEV_SIZE) ? dv_vlc_map[0][level].size : 16;
599 if (run) {
600 size += (run < 16) ? dv_vlc_map[run-1][0].size : 13;
601 }
602 }
603 return size;
604}
605#else
606static always_inline int dv_rl2vlc(int run, int l, int sign, uint32_t* vlc)
607{
608 *vlc = dv_vlc_map[run][l].vlc | sign;
609 return dv_vlc_map[run][l].size;
610}
611
612static always_inline int dv_rl2vlc_size(int run, int l)
613{
614 return dv_vlc_map[run][l].size;
615}
616#endif
617
618typedef struct EncBlockInfo {
619 int area_q[4];
620 int bit_size[4];
621 int prev[5];
622 int cur_ac;
623 int cno;
624 int dct_mode;
625 DCTELEM mb[64];
626 uint8_t next[64];
627 uint8_t sign[64];
628 uint8_t partial_bit_count;
629 uint32_t partial_bit_buffer; /* we can't use uint16_t here */
630} EncBlockInfo;
631
632static always_inline PutBitContext* dv_encode_ac(EncBlockInfo* bi, PutBitContext* pb_pool,
633 PutBitContext* pb_end)
634{
635 int prev;
636 int bits_left;
637 PutBitContext* pb = pb_pool;
638 int size = bi->partial_bit_count;
639 uint32_t vlc = bi->partial_bit_buffer;
640
641 bi->partial_bit_count = bi->partial_bit_buffer = 0;
642 for(;;){
643 /* Find suitable storage space */
644 for (; size > (bits_left = put_bits_left(pb)); pb++) {
645 if (bits_left) {
646 size -= bits_left;
647 put_bits(pb, bits_left, vlc >> size);
648 vlc = vlc & ((1<<size)-1);
649 }
650 if (pb + 1 >= pb_end) {
651 bi->partial_bit_count = size;
652 bi->partial_bit_buffer = vlc;
653 return pb;
654 }
655 }
656
657 /* Store VLC */
658 put_bits(pb, size, vlc);
659
660 if(bi->cur_ac>=64)
661 break;
662
663 /* Construct the next VLC */
664 prev= bi->cur_ac;
665 bi->cur_ac = bi->next[prev];
666 if(bi->cur_ac < 64){
667 size = dv_rl2vlc(bi->cur_ac - prev - 1, bi->mb[bi->cur_ac], bi->sign[bi->cur_ac], &vlc);
668 } else {
669 size = 4; vlc = 6; /* End Of Block stamp */
670 }
671 }
672 return pb;
673}
674
675static always_inline void dv_set_class_number(DCTELEM* blk, EncBlockInfo* bi,
676 const uint8_t* zigzag_scan, const int *weight, int bias)
677{
678 int i, area;
679 /* We offer two different methods for class number assignment: the
680 method suggested in SMPTE 314M Table 22, and an improved
681 method. The SMPTE method is very conservative; it assigns class
682 3 (i.e. severe quantization) to any block where the largest AC
683 component is greater than 36. ffmpeg's DV encoder tracks AC bit
684 consumption precisely, so there is no need to bias most blocks
685 towards strongly lossy compression. Instead, we assign class 2
686 to most blocks, and use class 3 only when strictly necessary
687 (for blocks whose largest AC component exceeds 255). */
688
689#if 0 /* SMPTE spec method */
690 static const int classes[] = {12, 24, 36, 0xffff};
691#else /* improved ffmpeg method */
692 static const int classes[] = {-1, -1, 255, 0xffff};
693#endif
694 int max=classes[0];
695 int prev=0;
696
697 bi->mb[0] = blk[0];
698
699 for (area = 0; area < 4; area++) {
700 bi->prev[area] = prev;
701 bi->bit_size[area] = 1; // 4 areas 4 bits for EOB :)
702 for (i=mb_area_start[area]; i<mb_area_start[area+1]; i++) {
703 int level = blk[zigzag_scan[i]];
704
705 if (level+15 > 30U) {
706 bi->sign[i] = (level>>31)&1;
707 /* weigh it and and shift down into range, adding for rounding */
708 /* the extra division by a factor of 2^4 reverses the 8x expansion of the DCT
709 AND the 2x doubling of the weights */
710 level = (ABS(level) * weight[i] + (1<<(dv_weight_bits+3))) >> (dv_weight_bits+4);
711 bi->mb[i] = level;
712 if(level>max) max= level;
713 bi->bit_size[area] += dv_rl2vlc_size(i - prev - 1, level);
714 bi->next[prev]= i;
715 prev= i;
716 }
717 }
718 }
719 bi->next[prev]= i;
720 for(bi->cno = 0; max > classes[bi->cno]; bi->cno++);
721
722 bi->cno += bias;
723
724 if (bi->cno >= 3) {
725 bi->cno = 3;
726 prev=0;
727 i= bi->next[prev];
728 for (area = 0; area < 4; area++) {
729 bi->prev[area] = prev;
730 bi->bit_size[area] = 1; // 4 areas 4 bits for EOB :)
731 for (; i<mb_area_start[area+1]; i= bi->next[i]) {
732 bi->mb[i] >>=1;
733
734 if (bi->mb[i]) {
735 bi->bit_size[area] += dv_rl2vlc_size(i - prev - 1, bi->mb[i]);
736 bi->next[prev]= i;
737 prev= i;
738 }
739 }
740 }
741 bi->next[prev]= i;
742 }
743}
744
745//FIXME replace this by dsputil
746#define SC(x, y) ((s[x] - s[y]) ^ ((s[x] - s[y]) >> 7))
747static always_inline int dv_guess_dct_mode(DCTELEM *blk) {
748 DCTELEM *s;
749 int score88 = 0;
750 int score248 = 0;
751 int i;
752
753 /* Compute 8-8 score (small values give a better chance for 8-8 DCT) */
754 s = blk;
755 for(i=0; i<7; i++) {
756 score88 += SC(0, 8) + SC(1, 9) + SC(2, 10) + SC(3, 11) +
757 SC(4, 12) + SC(5,13) + SC(6, 14) + SC(7, 15);
758 s += 8;
759 }
760 /* Compute 2-4-8 score (small values give a better chance for 2-4-8 DCT) */
761 s = blk;
762 for(i=0; i<6; i++) {
763 score248 += SC(0, 16) + SC(1,17) + SC(2, 18) + SC(3, 19) +
764 SC(4, 20) + SC(5,21) + SC(6, 22) + SC(7, 23);
765 s += 8;
766 }
767
768 return (score88 - score248 > -10);
769}
770
771static inline void dv_guess_qnos(EncBlockInfo* blks, int* qnos)
772{
773 int size[5];
774 int i, j, k, a, prev, a2;
775 EncBlockInfo* b;
776
777 size[0] = size[1] = size[2] = size[3] = size[4] = 1<<24;
778 do {
779 b = blks;
780 for (i=0; i<5; i++) {
781 if (!qnos[i])
782 continue;
783
784 qnos[i]--;
785 size[i] = 0;
786 for (j=0; j<6; j++, b++) {
787 for (a=0; a<4; a++) {
788 if (b->area_q[a] != dv_quant_shifts[qnos[i] + dv_quant_offset[b->cno]][a]) {
789 b->bit_size[a] = 1; // 4 areas 4 bits for EOB :)
790 b->area_q[a]++;
791 prev= b->prev[a];
792 assert(b->next[prev] >= mb_area_start[a+1] || b->mb[prev]);
793 for (k= b->next[prev] ; k<mb_area_start[a+1]; k= b->next[k]) {
794 b->mb[k] >>= 1;
795 if (b->mb[k]) {
796 b->bit_size[a] += dv_rl2vlc_size(k - prev - 1, b->mb[k]);
797 prev= k;
798 } else {
799 if(b->next[k] >= mb_area_start[a+1] && b->next[k]<64){
800 for(a2=a+1; b->next[k] >= mb_area_start[a2+1]; a2++)
801 b->prev[a2] = prev;
802 assert(a2<4);
803 assert(b->mb[b->next[k]]);
804 b->bit_size[a2] += dv_rl2vlc_size(b->next[k] - prev - 1, b->mb[b->next[k]])
805 -dv_rl2vlc_size(b->next[k] - k - 1, b->mb[b->next[k]]);
806 assert(b->prev[a2]==k && (a2+1 >= 4 || b->prev[a2+1]!=k));
807 b->prev[a2] = prev;
808 }
809 b->next[prev] = b->next[k];
810 }
811 }
812 b->prev[a+1]= prev;
813 }
814 size[i] += b->bit_size[a];
815 }
816 }
817 if(vs_total_ac_bits >= size[0] + size[1] + size[2] + size[3] + size[4])
818 return;
819 }
820 } while (qnos[0]|qnos[1]|qnos[2]|qnos[3]|qnos[4]);
821
822
823 for(a=2; a==2 || vs_total_ac_bits < size[0]; a+=a){
824 b = blks;
825 size[0] = 5*6*4; //EOB
826 for (j=0; j<6*5; j++, b++) {
827 prev= b->prev[0];
828 for (k= b->next[prev]; k<64; k= b->next[k]) {
829 if(b->mb[k] < a && b->mb[k] > -a){
830 b->next[prev] = b->next[k];
831 }else{
832 size[0] += dv_rl2vlc_size(k - prev - 1, b->mb[k]);
833 prev= k;
834 }
835 }
836 }
837 }
838}
839
840static inline void dv_encode_video_segment(DVVideoContext *s,
841 uint8_t *dif,
842 const uint16_t *mb_pos_ptr)
843{
844 int mb_index, i, j, v;
845 int mb_x, mb_y, c_offset, linesize;
846 uint8_t* y_ptr;
847 uint8_t* data;
848 uint8_t* ptr;
849 int do_edge_wrap;
850 DECLARE_ALIGNED_8(DCTELEM, block[64]);
851 EncBlockInfo enc_blks[5*6];
852 PutBitContext pbs[5*6];
853 PutBitContext* pb;
854 EncBlockInfo* enc_blk;
855 int vs_bit_size = 0;
856 int qnos[5];
857
858 assert((((int)block) & 7) == 0);
859
860 enc_blk = &enc_blks[0];
861 pb = &pbs[0];
862 for(mb_index = 0; mb_index < 5; mb_index++) {
863 v = *mb_pos_ptr++;
864 mb_x = v & 0xff;
865 mb_y = v >> 8;
866 if (s->sys->pix_fmt == PIX_FMT_YUV422P) {
867 y_ptr = s->picture.data[0] + (mb_y * s->picture.linesize[0] * 8) + (mb_x * 4);
868 } else { /* 4:1:1 */
869 y_ptr = s->picture.data[0] + (mb_y * s->picture.linesize[0] * 8) + (mb_x * 8);
870 }
871 if (s->sys->pix_fmt == PIX_FMT_YUV420P) {
872 c_offset = (((mb_y >> 1) * s->picture.linesize[1] * 8) + ((mb_x >> 1) * 8));
873 } else { /* 4:2:2 or 4:1:1 */
874 c_offset = ((mb_y * s->picture.linesize[1] * 8) + ((mb_x >> 2) * 8));
875 }
876 do_edge_wrap = 0;
877 qnos[mb_index] = 15; /* No quantization */
878 ptr = dif + mb_index*80 + 4;
879 for(j = 0;j < 6; j++) {
880 int dummy = 0;
881 if (s->sys->pix_fmt == PIX_FMT_YUV422P) { /* 4:2:2 */
882 if (j == 0 || j == 2) {
883 /* Y0 Y1 */
884 data = y_ptr + ((j>>1) * 8);
885 linesize = s->picture.linesize[0];
886 } else if (j > 3) {
887 /* Cr Cb */
888 data = s->picture.data[6 - j] + c_offset;
889 linesize = s->picture.linesize[6 - j];
890 } else {
891 /* j=1 and j=3 are "dummy" blocks, used for AC data only */
892 data = 0;
893 linesize = 0;
894 dummy = 1;
895 }
896 } else { /* 4:1:1 or 4:2:0 */
897 if (j < 4) { /* Four Y blocks */
898 /* NOTE: at end of line, the macroblock is handled as 420 */
899 if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x < (704 / 8)) {
900 data = y_ptr + (j * 8);
901 } else {
902 data = y_ptr + ((j & 1) * 8) + ((j >> 1) * 8 * s->picture.linesize[0]);
903 }
904 linesize = s->picture.linesize[0];
905 } else { /* Cr and Cb blocks */
906 /* don't ask Fabrice why they inverted Cb and Cr ! */
907 data = s->picture.data[6 - j] + c_offset;
908 linesize = s->picture.linesize[6 - j];
909 if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x >= (704 / 8))
910 do_edge_wrap = 1;
911 }
912 }
913
914 /* Everything is set up -- now just copy data -> DCT block */
915 if (do_edge_wrap) { /* Edge wrap copy: 4x16 -> 8x8 */
916 uint8_t* d;
917 DCTELEM *b = block;
918 for (i=0;i<8;i++) {
919 d = data + 8 * linesize;
920 b[0] = data[0]; b[1] = data[1]; b[2] = data[2]; b[3] = data[3];
921 b[4] = d[0]; b[5] = d[1]; b[6] = d[2]; b[7] = d[3];
922 data += linesize;
923 b += 8;
924 }
925 } else { /* Simple copy: 8x8 -> 8x8 */
926 if (!dummy)
927 s->get_pixels(block, data, linesize);
928 }
929
930 if(s->avctx->flags & CODEC_FLAG_INTERLACED_DCT)
931 enc_blk->dct_mode = dv_guess_dct_mode(block);
932 else
933 enc_blk->dct_mode = 0;
934 enc_blk->area_q[0] = enc_blk->area_q[1] = enc_blk->area_q[2] = enc_blk->area_q[3] = 0;
935 enc_blk->partial_bit_count = 0;
936 enc_blk->partial_bit_buffer = 0;
937 enc_blk->cur_ac = 0;
938
939 if (dummy) {
940 /* We rely on the fact that encoding all zeros leads to an immediate EOB,
941 which is precisely what the spec calls for in the "dummy" blocks. */
942 memset(block, 0, sizeof(block));
943 } else {
944 s->fdct[enc_blk->dct_mode](block);
945 }
946
947 dv_set_class_number(block, enc_blk,
948 enc_blk->dct_mode ? ff_zigzag248_direct : ff_zigzag_direct,
949 enc_blk->dct_mode ? dv_weight_248 : dv_weight_88,
950 j/4);
951
952 init_put_bits(pb, ptr, block_sizes[j]/8);
953 put_bits(pb, 9, (uint16_t)(((enc_blk->mb[0] >> 3) - 1024 + 2) >> 2));
954 put_bits(pb, 1, enc_blk->dct_mode);
955 put_bits(pb, 2, enc_blk->cno);
956
957 vs_bit_size += enc_blk->bit_size[0] + enc_blk->bit_size[1] +
958 enc_blk->bit_size[2] + enc_blk->bit_size[3];
959 ++enc_blk;
960 ++pb;
961 ptr += block_sizes[j]/8;
962 }
963 }
964
965 if (vs_total_ac_bits < vs_bit_size)
966 dv_guess_qnos(&enc_blks[0], &qnos[0]);
967
968 for (i=0; i<5; i++) {
969 dif[i*80 + 3] = qnos[i];
970 }
971
972 /* First pass over individual cells only */
973 for (j=0; j<5*6; j++)
974 dv_encode_ac(&enc_blks[j], &pbs[j], &pbs[j+1]);
975
976 /* Second pass over each MB space */
977 for (j=0; j<5*6; j+=6) {
978 pb= &pbs[j];
979 for (i=0; i<6; i++) {
980 if (enc_blks[i+j].partial_bit_count)
981 pb=dv_encode_ac(&enc_blks[i+j], pb, &pbs[j+6]);
982 }
983 }
984
985 /* Third and final pass over the whole vides segment space */
986 pb= &pbs[0];
987 for (j=0; j<5*6; j++) {
988 if (enc_blks[j].partial_bit_count)
989 pb=dv_encode_ac(&enc_blks[j], pb, &pbs[6*5]);
990 if (enc_blks[j].partial_bit_count)
991 av_log(NULL, AV_LOG_ERROR, "ac bitstream overflow\n");
992 }
993
994 for (j=0; j<5*6; j++)
995 flush_put_bits(&pbs[j]);
996}
997
998static int dv_decode_mt(AVCodecContext *avctx, void* sl)
999{
1000 DVVideoContext *s = avctx->priv_data;
1001 int slice = (size_t)sl;
1002
1003 /* which DIF channel is this? */
1004 int chan = slice / (s->sys->difseg_size * 27);
1005
1006 /* slice within the DIF channel */
1007 int chan_slice = slice % (s->sys->difseg_size * 27);
1008
1009 /* byte offset of this channel's data */
1010 int chan_offset = chan * s->sys->difseg_size * 150 * 80;
1011
1012 dv_decode_video_segment(s, &s->buf[((chan_slice/27)*6+(chan_slice/3)+chan_slice*5+7)*80 + chan_offset],
1013 &s->sys->video_place[slice*5]);
1014 return 0;
1015}
1016
1017static int dv_encode_mt(AVCodecContext *avctx, void* sl)
1018{
1019 DVVideoContext *s = avctx->priv_data;
1020 int slice = (size_t)sl;
1021
1022 /* which DIF channel is this? */
1023 int chan = slice / (s->sys->difseg_size * 27);
1024
1025 /* slice within the DIF channel */
1026 int chan_slice = slice % (s->sys->difseg_size * 27);
1027
1028 /* byte offset of this channel's data */
1029 int chan_offset = chan * s->sys->difseg_size * 150 * 80;
1030
1031 dv_encode_video_segment(s, &s->buf[((chan_slice/27)*6+(chan_slice/3)+chan_slice*5+7)*80 + chan_offset],
1032 &s->sys->video_place[slice*5]);
1033 return 0;
1034}
1035
1036/* NOTE: exactly one frame must be given (120000 bytes for NTSC,
1037 144000 bytes for PAL - or twice those for 50Mbps) */
1038static int dvvideo_decode_frame(AVCodecContext *avctx,
1039 void *data, int *data_size,
1040 uint8_t *buf, int buf_size)
1041{
1042 DVVideoContext *s = avctx->priv_data;
1043
1044 s->sys = dv_frame_profile(buf);
1045 if (!s->sys || buf_size < s->sys->frame_size)
1046 return -1; /* NOTE: we only accept several full frames */
1047
1048 if(s->picture.data[0])
1049 avctx->release_buffer(avctx, &s->picture);
1050
1051 s->picture.reference = 0;
1052 s->picture.key_frame = 1;
1053 s->picture.pict_type = FF_I_TYPE;
1054 avctx->pix_fmt = s->sys->pix_fmt;
1055 avcodec_set_dimensions(avctx, s->sys->width, s->sys->height);
1056 if(avctx->get_buffer(avctx, &s->picture) < 0) {
1057 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1058 return -1;
1059 }
1060 s->picture.interlaced_frame = 1;
1061 s->picture.top_field_first = 0;
1062
1063 s->buf = buf;
1064 avctx->execute(avctx, dv_decode_mt, (void**)&dv_anchor[0], NULL,
1065 s->sys->n_difchan * s->sys->difseg_size * 27);
1066
1067 emms_c();
1068
1069 /* return image */
1070 *data_size = sizeof(AVFrame);
1071 *(AVFrame*)data= s->picture;
1072
1073 return s->sys->frame_size;
1074}
1075
1076static int dvvideo_encode_frame(AVCodecContext *c, uint8_t *buf, int buf_size,
1077 void *data)
1078{
1079 DVVideoContext *s = c->priv_data;
1080
1081 s->sys = dv_codec_profile(c);
1082 if (!s->sys)
1083 return -1;
1084 if(buf_size < s->sys->frame_size)
1085 return -1;
1086
1087 c->pix_fmt = s->sys->pix_fmt;
1088 s->picture = *((AVFrame *)data);
1089 s->picture.key_frame = 1;
1090 s->picture.pict_type = FF_I_TYPE;
1091
1092 s->buf = buf;
1093 c->execute(c, dv_encode_mt, (void**)&dv_anchor[0], NULL,
1094 s->sys->n_difchan * s->sys->difseg_size * 27);
1095
1096 emms_c();
1097
1098 /* Fill in just enough of the header for dv_frame_profile() to
1099 return the correct result, so that the frame can be decoded
1100 correctly. The rest of the metadata is filled in by the dvvideo
1101 avformat. (this should probably change so that encode_frame()
1102 fills in ALL of the metadata - e.g. for Quicktime-wrapped DV
1103 streams) */
1104
1105 /* NTSC/PAL format */
1106 buf[3] = s->sys->dsf ? 0x80 : 0x00;
1107
1108 /* 25Mbps or 50Mbps */
1109 buf[80*5 + 48 + 3] = (s->sys->pix_fmt == PIX_FMT_YUV422P) ? 0x4 : 0x0;
1110
1111 return s->sys->frame_size;
1112}
1113
1114static int dvvideo_close(AVCodecContext *c)
1115{
1116
1117 return 0;
1118}
1119
1120
1121#ifdef CONFIG_DVVIDEO_ENCODER
1122AVCodec dvvideo_encoder = {
1123 "dvvideo",
1124 CODEC_TYPE_VIDEO,
1125 CODEC_ID_DVVIDEO,
1126 sizeof(DVVideoContext),
1127 dvvideo_init,
1128 dvvideo_encode_frame,
1129 dvvideo_close,
1130 NULL,
1131 CODEC_CAP_DR1,
1132 NULL
1133};
1134#endif // CONFIG_DVVIDEO_ENCODER
1135
1136AVCodec dvvideo_decoder = {
1137 "dvvideo",
1138 CODEC_TYPE_VIDEO,
1139 CODEC_ID_DVVIDEO,
1140 sizeof(DVVideoContext),
1141 dvvideo_init,
1142 NULL,
1143 dvvideo_close,
1144 dvvideo_decode_frame,
1145 CODEC_CAP_DR1,
1146 NULL
1147};
Note: See TracBrowser for help on using the repository browser.

© 2024 Oracle Support Privacy / Do Not Sell My Info Terms of Use Trademark Policy Automated Access Etiquette