* introducing dct248 into the DSP context.
[libav.git] / libavcodec / dv.c
1 /*
2 * DV decoder
3 * Copyright (c) 2002 Fabrice Bellard.
4 *
5 * DV encoder
6 * Copyright (c) 2003 Roman Shaposhnik.
7 *
8 * Many thanks to Dan Dennedy <dan@dennedy.org> for providing wealth
9 * of DV technical info.
10 *
11 * This library is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU Lesser General Public
13 * License as published by the Free Software Foundation; either
14 * version 2 of the License, or (at your option) any later version.
15 *
16 * This library is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * Lesser General Public License for more details.
20 *
21 * You should have received a copy of the GNU Lesser General Public
22 * License along with this library; if not, write to the Free Software
23 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
24 */
25
26 /**
27 * @file dv.c
28 * DV codec.
29 */
30 #include "avcodec.h"
31 #include "dsputil.h"
32 #include "mpegvideo.h"
33 #include "simple_idct.h"
34 #include "dvdata.h"
35
36 typedef struct DVVideoDecodeContext {
37 const DVprofile* sys;
38 AVFrame picture;
39
40 uint8_t dv_zigzag[2][64];
41 uint8_t dv_idct_shift[2][22][64];
42 uint8_t dv_dct_shift[2][22][64];
43
44 void (*get_pixels)(DCTELEM *block, const uint8_t *pixels, int line_size);
45 void (*fdct[2])(DCTELEM *block);
46 void (*idct_put[2])(uint8_t *dest, int line_size, DCTELEM *block);
47
48 GetBitContext gb;
49 DCTELEM block[5*6][64] __align8;
50 } DVVideoDecodeContext;
51
52 #define TEX_VLC_BITS 9
53 /* XXX: also include quantization */
54 static RL_VLC_ELEM *dv_rl_vlc[1];
55 static VLC_TYPE dv_vlc_codes[15][23];
56
57 static void dv_build_unquantize_tables(DVVideoDecodeContext *s, uint8_t* perm)
58 {
59 int i, q, j;
60
61 /* NOTE: max left shift is 6 */
62 for(q = 0; q < 22; q++) {
63 /* 88DCT */
64 for(i = 1; i < 64; i++) {
65 /* 88 table */
66 j = perm[i];
67 s->dv_idct_shift[0][q][j] =
68 dv_quant_shifts[q][dv_88_areas[i]] + 1;
69 s->dv_dct_shift[0][q][i] =
70 dv_quant_shifts[q][dv_88_areas[ff_zigzag_direct[i]]] + 4;
71 }
72
73 /* 248DCT */
74 for(i = 1; i < 64; i++) {
75 /* 248 table */
76 s->dv_idct_shift[1][q][i] =
77 dv_quant_shifts[q][dv_248_areas[i]] + 1;
78 s->dv_dct_shift[1][q][i] =
79 dv_quant_shifts[q][dv_248_areas[ff_zigzag248_direct[i]]] + 4;
80 }
81 }
82 }
83
84 static int dvvideo_init(AVCodecContext *avctx)
85 {
86 DVVideoDecodeContext *s = avctx->priv_data;
87 DSPContext dsp;
88 static int done=0;
89 int i;
90
91 if (!done) {
92 int i;
93 VLC dv_vlc;
94
95 done = 1;
96
97 /* NOTE: as a trick, we use the fact the no codes are unused
98 to accelerate the parsing of partial codes */
99 init_vlc(&dv_vlc, TEX_VLC_BITS, NB_DV_VLC,
100 dv_vlc_len, 1, 1, dv_vlc_bits, 2, 2);
101
102 dv_rl_vlc[0] = av_malloc(dv_vlc.table_size * sizeof(RL_VLC_ELEM));
103 for(i = 0; i < dv_vlc.table_size; i++){
104 int code= dv_vlc.table[i][0];
105 int len = dv_vlc.table[i][1];
106 int level, run;
107
108 if(len<0){ //more bits needed
109 run= 0;
110 level= code;
111 } else if (code == (NB_DV_VLC - 1)) {
112 /* EOB */
113 run = 0;
114 level = 256;
115 } else {
116 run= dv_vlc_run[code] + 1;
117 level= dv_vlc_level[code];
118 }
119 dv_rl_vlc[0][i].len = len;
120 dv_rl_vlc[0][i].level = level;
121 dv_rl_vlc[0][i].run = run;
122 }
123
124 memset(dv_vlc_codes, 0xff, sizeof(dv_vlc_codes));
125 for (i = 0; i < NB_DV_VLC - 1; i++) {
126 if (dv_vlc_run[i] < 15 && dv_vlc_level[i] < 23 && dv_vlc_len[i] < 15)
127 dv_vlc_codes[dv_vlc_run[i]][dv_vlc_level[i]] = i;
128 }
129 }
130
131 /* Generic DSP setup */
132 dsputil_init(&dsp, avctx);
133 s->get_pixels = dsp.get_pixels;
134
135 /* 88DCT setup */
136 s->fdct[0] = dsp.fdct;
137 s->idct_put[0] = dsp.idct_put;
138 for (i=0; i<64; i++)
139 s->dv_zigzag[0][i] = dsp.idct_permutation[ff_zigzag_direct[i]];
140
141 /* 248DCT setup */
142 s->fdct[1] = dsp.fdct248;
143 s->idct_put[1] = simple_idct248_put; // FIXME: need to add it to DSP
144 memcpy(s->dv_zigzag[1], ff_zigzag248_direct, 64);
145
146 /* XXX: do it only for constant case */
147 dv_build_unquantize_tables(s, dsp.idct_permutation);
148
149 /* FIXME: I really don't think this should be here */
150 if (dv_codec_profile(avctx))
151 avctx->pix_fmt = dv_codec_profile(avctx)->pix_fmt;
152 avctx->coded_frame = &s->picture;
153
154 return 0;
155 }
156
157 // #define VLC_DEBUG
158
159 typedef struct BlockInfo {
160 const uint8_t *shift_table;
161 const uint8_t *scan_table;
162 uint8_t pos; /* position in block */
163 uint8_t eob_reached; /* true if EOB has been reached */
164 uint8_t dct_mode;
165 uint8_t partial_bit_count;
166 uint16_t partial_bit_buffer;
167 int shift_offset;
168 } BlockInfo;
169
170 /* block size in bits */
171 static const uint16_t block_sizes[6] = {
172 112, 112, 112, 112, 80, 80
173 };
174
175 #ifndef ALT_BITSTREAM_READER
176 #warning only works with ALT_BITSTREAM_READER
177 #endif
178
179 /* decode ac coefs */
180 static void dv_decode_ac(DVVideoDecodeContext *s,
181 BlockInfo *mb, DCTELEM *block, int last_index)
182 {
183 int last_re_index;
184 int shift_offset = mb->shift_offset;
185 const uint8_t *scan_table = mb->scan_table;
186 const uint8_t *shift_table = mb->shift_table;
187 int pos = mb->pos;
188 int level, pos1, sign, run;
189 int partial_bit_count;
190 #ifndef ALT_BITSTREAM_READER //FIXME
191 int re_index=0;
192 int re1_index=0;
193 #endif
194 OPEN_READER(re, &s->gb);
195
196 #ifdef VLC_DEBUG
197 printf("start\n");
198 #endif
199
200 /* if we must parse a partial vlc, we do it here */
201 partial_bit_count = mb->partial_bit_count;
202 if (partial_bit_count > 0) {
203 uint8_t buf[4];
204 uint32_t v;
205 int l, l1;
206 GetBitContext gb1;
207
208 /* build the dummy bit buffer */
209 l = 16 - partial_bit_count;
210 UPDATE_CACHE(re, &s->gb);
211 #ifdef VLC_DEBUG
212 printf("show=%04x\n", SHOW_UBITS(re, &s->gb, 16));
213 #endif
214 v = (mb->partial_bit_buffer << l) | SHOW_UBITS(re, &s->gb, l);
215 buf[0] = v >> 8;
216 buf[1] = v;
217 #ifdef VLC_DEBUG
218 printf("v=%04x cnt=%d %04x\n",
219 v, partial_bit_count, (mb->partial_bit_buffer << l));
220 #endif
221 /* try to read the codeword */
222 init_get_bits(&gb1, buf, 4*8);
223 {
224 OPEN_READER(re1, &gb1);
225 UPDATE_CACHE(re1, &gb1);
226 GET_RL_VLC(level, run, re1, &gb1, dv_rl_vlc[0],
227 TEX_VLC_BITS, 2);
228 l = re1_index;
229 CLOSE_READER(re1, &gb1);
230 }
231 #ifdef VLC_DEBUG
232 printf("****run=%d level=%d size=%d\n", run, level, l);
233 #endif
234 /* compute codeword length */
235 l1 = (level != 256 && level != 0);
236 /* if too long, we cannot parse */
237 l -= partial_bit_count;
238 if ((re_index + l + l1) > last_index)
239 return;
240 /* skip read bits */
241 last_re_index = 0; /* avoid warning */
242 re_index += l;
243 /* by definition, if we can read the vlc, all partial bits
244 will be read (otherwise we could have read the vlc before) */
245 mb->partial_bit_count = 0;
246 UPDATE_CACHE(re, &s->gb);
247 goto handle_vlc;
248 }
249
250 /* get the AC coefficients until last_index is reached */
251 for(;;) {
252 UPDATE_CACHE(re, &s->gb);
253 #ifdef VLC_DEBUG
254 printf("%2d: bits=%04x index=%d\n",
255 pos, SHOW_UBITS(re, &s->gb, 16), re_index);
256 #endif
257 last_re_index = re_index;
258 GET_RL_VLC(level, run, re, &s->gb, dv_rl_vlc[0],
259 TEX_VLC_BITS, 2);
260 handle_vlc:
261 #ifdef VLC_DEBUG
262 printf("run=%d level=%d\n", run, level);
263 #endif
264 if (level == 256) {
265 if (re_index > last_index) {
266 cannot_read:
267 /* put position before read code */
268 re_index = last_re_index;
269 mb->eob_reached = 0;
270 break;
271 }
272 /* EOB */
273 mb->eob_reached = 1;
274 break;
275 } else if (level != 0) {
276 if ((re_index + 1) > last_index)
277 goto cannot_read;
278 sign = SHOW_SBITS(re, &s->gb, 1);
279 level = (level ^ sign) - sign;
280 LAST_SKIP_BITS(re, &s->gb, 1);
281 pos += run;
282 /* error */
283 if (pos >= 64) {
284 goto read_error;
285 }
286 pos1 = scan_table[pos];
287 level = level << (shift_table[pos1] + shift_offset);
288 block[pos1] = level;
289 // printf("run=%d level=%d shift=%d\n", run, level, shift_table[pos1]);
290 } else {
291 if (re_index > last_index)
292 goto cannot_read;
293 /* level is zero: means run without coding. No
294 sign is coded */
295 pos += run;
296 /* error */
297 if (pos >= 64) {
298 read_error:
299 #if defined(VLC_DEBUG) || 1
300 fprintf(stderr, "error pos=%d\n", pos);
301 #endif
302 /* for errors, we consider the eob is reached */
303 mb->eob_reached = 1;
304 break;
305 }
306 }
307 }
308 CLOSE_READER(re, &s->gb);
309 mb->pos = pos;
310 }
311
312 static inline void bit_copy(PutBitContext *pb, GetBitContext *gb, int bits_left)
313 {
314 while (bits_left >= 16) {
315 put_bits(pb, 16, get_bits(gb, 16));
316 bits_left -= 16;
317 }
318 if (bits_left > 0) {
319 put_bits(pb, bits_left, get_bits(gb, bits_left));
320 }
321 }
322
323 /* mb_x and mb_y are in units of 8 pixels */
324 static inline void dv_decode_video_segment(DVVideoDecodeContext *s,
325 uint8_t *buf_ptr1,
326 const uint16_t *mb_pos_ptr)
327 {
328 int quant, dc, dct_mode, class1, j;
329 int mb_index, mb_x, mb_y, v, last_index;
330 DCTELEM *block, *block1;
331 int c_offset, bits_left;
332 uint8_t *y_ptr;
333 BlockInfo mb_data[5 * 6], *mb, *mb1;
334 void (*idct_put)(uint8_t *dest, int line_size, DCTELEM *block);
335 uint8_t *buf_ptr;
336 PutBitContext pb, vs_pb;
337 uint8_t mb_bit_buffer[80 + 4]; /* allow some slack */
338 int mb_bit_count;
339 uint8_t vs_bit_buffer[5 * 80 + 4]; /* allow some slack */
340 int vs_bit_count;
341
342 memset(s->block, 0, sizeof(s->block));
343
344 /* pass 1 : read DC and AC coefficients in blocks */
345 buf_ptr = buf_ptr1;
346 block1 = &s->block[0][0];
347 mb1 = mb_data;
348 init_put_bits(&vs_pb, vs_bit_buffer, 5 * 80);
349 vs_bit_count = 0;
350 for(mb_index = 0; mb_index < 5; mb_index++) {
351 /* skip header */
352 quant = buf_ptr[3] & 0x0f;
353 buf_ptr += 4;
354 init_put_bits(&pb, mb_bit_buffer, 80);
355 mb_bit_count = 0;
356 mb = mb1;
357 block = block1;
358 for(j = 0;j < 6; j++) {
359 /* NOTE: size is not important here */
360 init_get_bits(&s->gb, buf_ptr, 14*8);
361
362 /* get the dc */
363 dc = get_bits(&s->gb, 9);
364 dc = (dc << (32 - 9)) >> (32 - 9);
365 dct_mode = get_bits1(&s->gb);
366 mb->dct_mode = dct_mode;
367 mb->scan_table = s->dv_zigzag[dct_mode];
368 class1 = get_bits(&s->gb, 2);
369 mb->shift_offset = (class1 == 3);
370 mb->shift_table = s->dv_idct_shift[dct_mode]
371 [quant + dv_quant_offset[class1]];
372 dc = dc << 2;
373 /* convert to unsigned because 128 is not added in the
374 standard IDCT */
375 dc += 1024;
376 block[0] = dc;
377 last_index = block_sizes[j];
378 buf_ptr += last_index >> 3;
379 mb->pos = 0;
380 mb->partial_bit_count = 0;
381
382 #ifdef VLC_DEBUG
383 printf("MB block: %d, %d ", mb_index, j);
384 #endif
385 dv_decode_ac(s, mb, block, last_index);
386
387 /* write the remaining bits in a new buffer only if the
388 block is finished */
389 bits_left = last_index - get_bits_count(&s->gb);
390 if (mb->eob_reached) {
391 mb->partial_bit_count = 0;
392 mb_bit_count += bits_left;
393 bit_copy(&pb, &s->gb, bits_left);
394 } else {
395 /* should be < 16 bits otherwise a codeword could have
396 been parsed */
397 mb->partial_bit_count = bits_left;
398 mb->partial_bit_buffer = get_bits(&s->gb, bits_left);
399 }
400 block += 64;
401 mb++;
402 }
403
404 flush_put_bits(&pb);
405
406 /* pass 2 : we can do it just after */
407 #ifdef VLC_DEBUG
408 printf("***pass 2 size=%d MB#=%d\n", mb_bit_count, mb_index);
409 #endif
410 block = block1;
411 mb = mb1;
412 init_get_bits(&s->gb, mb_bit_buffer, 80*8);
413 for(j = 0;j < 6; j++) {
414 if (!mb->eob_reached && get_bits_count(&s->gb) < mb_bit_count) {
415 dv_decode_ac(s, mb, block, mb_bit_count);
416 /* if still not finished, no need to parse other blocks */
417 if (!mb->eob_reached) {
418 /* we could not parse the current AC coefficient,
419 so we add the remaining bytes */
420 bits_left = mb_bit_count - get_bits_count(&s->gb);
421 if (bits_left > 0) {
422 mb->partial_bit_count += bits_left;
423 mb->partial_bit_buffer =
424 (mb->partial_bit_buffer << bits_left) |
425 get_bits(&s->gb, bits_left);
426 }
427 goto next_mb;
428 }
429 }
430 block += 64;
431 mb++;
432 }
433 /* all blocks are finished, so the extra bytes can be used at
434 the video segment level */
435 bits_left = mb_bit_count - get_bits_count(&s->gb);
436 vs_bit_count += bits_left;
437 bit_copy(&vs_pb, &s->gb, bits_left);
438 next_mb:
439 mb1 += 6;
440 block1 += 6 * 64;
441 }
442
443 /* we need a pass other the whole video segment */
444 flush_put_bits(&vs_pb);
445
446 #ifdef VLC_DEBUG
447 printf("***pass 3 size=%d\n", vs_bit_count);
448 #endif
449 block = &s->block[0][0];
450 mb = mb_data;
451 init_get_bits(&s->gb, vs_bit_buffer, 5 * 80*8);
452 for(mb_index = 0; mb_index < 5; mb_index++) {
453 for(j = 0;j < 6; j++) {
454 if (!mb->eob_reached) {
455 #ifdef VLC_DEBUG
456 printf("start %d:%d\n", mb_index, j);
457 #endif
458 dv_decode_ac(s, mb, block, vs_bit_count);
459 }
460 block += 64;
461 mb++;
462 }
463 }
464
465 /* compute idct and place blocks */
466 block = &s->block[0][0];
467 mb = mb_data;
468 for(mb_index = 0; mb_index < 5; mb_index++) {
469 v = *mb_pos_ptr++;
470 mb_x = v & 0xff;
471 mb_y = v >> 8;
472 y_ptr = s->picture.data[0] + (mb_y * s->picture.linesize[0] * 8) + (mb_x * 8);
473 if (s->sys->pix_fmt == PIX_FMT_YUV411P)
474 c_offset = (mb_y * s->picture.linesize[1] * 8) + ((mb_x >> 2) * 8);
475 else
476 c_offset = ((mb_y >> 1) * s->picture.linesize[1] * 8) + ((mb_x >> 1) * 8);
477 for(j = 0;j < 6; j++) {
478 idct_put = s->idct_put[mb->dct_mode];
479 if (j < 4) {
480 if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x < (704 / 8)) {
481 /* NOTE: at end of line, the macroblock is handled as 420 */
482 idct_put(y_ptr + (j * 8), s->picture.linesize[0], block);
483 } else {
484 idct_put(y_ptr + ((j & 1) * 8) + ((j >> 1) * 8 * s->picture.linesize[0]),
485 s->picture.linesize[0], block);
486 }
487 } else {
488 if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x >= (704 / 8)) {
489 uint64_t aligned_pixels[64/8];
490 uint8_t *pixels= (uint8_t*)aligned_pixels;
491 uint8_t *c_ptr, *c_ptr1, *ptr;
492 int y, linesize;
493 /* NOTE: at end of line, the macroblock is handled as 420 */
494 idct_put(pixels, 8, block);
495 linesize = s->picture.linesize[6 - j];
496 c_ptr = s->picture.data[6 - j] + c_offset;
497 ptr = pixels;
498 for(y = 0;y < 8; y++) {
499 /* convert to 411P */
500 c_ptr1 = c_ptr + 8*linesize;
501 c_ptr[0]= ptr[0]; c_ptr1[0]= ptr[4];
502 c_ptr[1]= ptr[1]; c_ptr1[1]= ptr[5];
503 c_ptr[2]= ptr[2]; c_ptr1[2]= ptr[6];
504 c_ptr[3]= ptr[3]; c_ptr1[3]= ptr[7];
505 c_ptr += linesize;
506 ptr += 8;
507 }
508 } else {
509 /* don't ask me why they inverted Cb and Cr ! */
510 idct_put(s->picture.data[6 - j] + c_offset,
511 s->picture.linesize[6 - j], block);
512 }
513 }
514 block += 64;
515 mb++;
516 }
517 }
518 }
519
520 /* Converts run and level (where level != 0) pair into vlc, returning bit size */
521 static inline int dv_rl2vlc(int run, int l, uint32_t* vlc)
522 {
523 int sign = l >> 8;
524 int level = (l ^ sign) - sign;
525 int size;
526
527 sign = (sign & 1);
528
529 if (run < 15 && level < 23 && dv_vlc_codes[run][level] != -1) {
530 *vlc = (dv_vlc_bits[dv_vlc_codes[run][level]] << 1) | sign;
531 size = dv_vlc_len[dv_vlc_codes[run][level]] + 1;
532 }
533 else {
534 if (level < 23) {
535 *vlc = (dv_vlc_bits[dv_vlc_codes[0][level]] << 1) | sign;
536 size = dv_vlc_len[dv_vlc_codes[0][level]] + 1;
537 } else {
538 *vlc = 0xfe00 | (level << 1) | sign;
539 size = 16;
540 }
541
542 switch(run) {
543 case 0:
544 break;
545 case 1:
546 case 2:
547 *vlc |= ((0x7ce | (run - 1)) << size);
548 size += 11;
549 break;
550 case 3:
551 case 4:
552 case 5:
553 case 6:
554 *vlc |= ((0xfac | (run - 3)) << size);
555 size += 12;
556 break;
557 default:
558 *vlc |= ((0x1f80 | (run - 1)) << size);
559 size += 13;
560 break;
561 }
562 }
563
564 return size;
565 }
566
567 typedef struct EncBlockInfo {
568 int qno;
569 int cno;
570 int dct_mode;
571 int block_size;
572 DCTELEM *mb;
573 PutBitContext pb;
574 const uint8_t* zigzag_scan;
575 uint8_t *dv_shift;
576 } EncBlockInfo;
577
578 static inline int dv_bits_left(EncBlockInfo* bi)
579 {
580 return (bi->block_size - get_bit_count(&bi->pb));
581 }
582
583 static inline void dv_encode_ac(EncBlockInfo* bi, PutBitContext* heap)
584 {
585 int i, level, size, run = 0;
586 uint32_t vlc;
587 PutBitContext* cpb = &bi->pb;
588 int bias = (bi->cno == 3);
589
590 for (i=1; i<64; i++) {
591 level = bi->mb[bi->zigzag_scan[i]] / (1<<(bi->dv_shift[i] + bias));
592 if (level != 0) {
593 size = dv_rl2vlc(run, level, &vlc);
594 put_vlc:
595
596 #ifdef VLC_DEBUG
597 printf(" %3d:%3d", run, level);
598 #endif
599 if (cpb == &bi->pb && size > dv_bits_left(bi)) {
600 size -= dv_bits_left(bi);
601 put_bits(cpb, dv_bits_left(bi), vlc >> size);
602 vlc = vlc & ((1<<size)-1);
603 cpb = heap;
604 }
605 put_bits(cpb, size, vlc);
606 run = 0;
607 } else
608 run++;
609 }
610
611 if (i == 64) {
612 size = 4; vlc = 6; /* End Of Block stamp */
613 goto put_vlc;
614 }
615 }
616
617 static inline void dv_redistr_bits(EncBlockInfo* bi, int count, uint8_t* extra_data, int extra_bits, PutBitContext* heap)
618 {
619 int i;
620 GetBitContext gb;
621
622 init_get_bits(&gb, extra_data, extra_bits);
623
624 for (i=0; i<count; i++) {
625 int bits_left = dv_bits_left(bi);
626 #ifdef VLC_DEBUG
627 if (bits_left)
628 printf("------------> inserting %d bytes in %d:%d\n", bits_left, i/6, i%6);
629 #endif
630 if (bits_left > extra_bits) {
631 bit_copy(&bi->pb, &gb, extra_bits);
632 extra_bits = 0;
633 break;
634 } else
635 bit_copy(&bi->pb, &gb, bits_left);
636
637 extra_bits -= bits_left;
638 bi++;
639 }
640
641 if (extra_bits > 0 && heap)
642 bit_copy(heap, &gb, extra_bits);
643 }
644
645 static inline void dv_set_class_number(EncBlockInfo* bi, int j)
646 {
647 int i, max_ac = 0;
648
649 for (i=1; i<64; i++) {
650 int ac = abs(bi->mb[ff_zigzag_direct[i]]) / 4;
651 if (max_ac < ac)
652 max_ac = ac;
653 }
654 if (max_ac < 12)
655 bi->cno = j;
656 else if (max_ac < 24)
657 bi->cno = j + 1;
658 else if (max_ac < 36)
659 bi->cno = j + 2;
660 else
661 bi->cno = j + 3;
662
663 if (bi->cno > 3)
664 bi->cno = 3;
665 }
666
667 #define SQ(a) ((a)*(a))
668 static int dv_score_lines(DCTELEM *s, int stride) {
669 int score=0;
670 int x, y;
671
672 for(y=0; y<4; y++) {
673 for(x=0; x<8; x+=4){
674 score+= SQ(s[x ] - s[x +stride]) + SQ(s[x+1] - s[x+1+stride])
675 +SQ(s[x+2] - s[x+2+stride]) + SQ(s[x+3] - s[x+3+stride]);
676 }
677 s+= stride;
678 }
679
680 return score;
681 }
682
683 /*
684 * This is a very rough initial implementaion. The performance is
685 * horrible and the weighting is missing. But it's missing from the
686 * decoding step also -- so at least we're on the same page with decoder ;-)
687 */
688 static inline void dv_encode_video_segment(DVVideoDecodeContext *s,
689 uint8_t *dif,
690 const uint16_t *mb_pos_ptr)
691 {
692 int mb_index, i, j, v;
693 int mb_x, mb_y, c_offset, linesize;
694 uint8_t* y_ptr;
695 uint8_t* data;
696 int do_edge_wrap;
697 DCTELEM *block;
698 EncBlockInfo enc_blks[5*6];
699 EncBlockInfo* enc_blk;
700 int free_vs_bits;
701 int extra_bits;
702 PutBitContext extra_vs;
703 uint8_t extra_vs_data[5*6*128];
704 uint8_t extra_mb_data[6*128];
705
706 int QNO = 15;
707
708 /* Stage 1 -- doing DCT on 5 MBs */
709 block = &s->block[0][0];
710 enc_blk = &enc_blks[0];
711 for(mb_index = 0; mb_index < 5; mb_index++) {
712 v = *mb_pos_ptr++;
713 mb_x = v & 0xff;
714 mb_y = v >> 8;
715 y_ptr = s->picture.data[0] + (mb_y * s->picture.linesize[0] * 8) + (mb_x * 8);
716 c_offset = (s->sys->pix_fmt == PIX_FMT_YUV411P) ?
717 ((mb_y * s->picture.linesize[1] * 8) + ((mb_x >> 2) * 8)) :
718 (((mb_y >> 1) * s->picture.linesize[1] * 8) + ((mb_x >> 1) * 8));
719 do_edge_wrap = 0;
720 for(j = 0;j < 6; j++) {
721 if (j < 4) { /* Four Y blocks */
722 /* NOTE: at end of line, the macroblock is handled as 420 */
723 if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x < (704 / 8)) {
724 data = y_ptr + (j * 8);
725 } else {
726 data = y_ptr + ((j & 1) * 8) + ((j >> 1) * 8 * s->picture.linesize[0]);
727 }
728 linesize = s->picture.linesize[0];
729 } else { /* Cr and Cb blocks */
730 /* don't ask Fabrice why they inverted Cb and Cr ! */
731 data = s->picture.data[6 - j] + c_offset;
732 linesize = s->picture.linesize[6 - j];
733 if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x >= (704 / 8))
734 do_edge_wrap = 1;
735 }
736
737 /* Everything is set up -- now just copy data -> DCT block */
738 if (do_edge_wrap) { /* Edge wrap copy: 4x16 -> 8x8 */
739 uint8_t* d;
740 DCTELEM *b = block;
741 for (i=0;i<8;i++) {
742 d = data + 8 * linesize;
743 b[0] = data[0]; b[1] = data[1]; b[2] = data[2]; b[3] = data[3];
744 b[4] = d[0]; b[5] = d[1]; b[6] = d[2]; b[7] = d[3];
745 data += linesize;
746 b += 8;
747 }
748 } else { /* Simple copy: 8x8 -> 8x8 */
749 s->get_pixels(block, data, linesize);
750 }
751
752 if (dv_score_lines(block, 8) + dv_score_lines(block+8*4, 8) - 100 >
753 dv_score_lines(block, 16) + dv_score_lines(block+8, 16)) {
754 enc_blk->dct_mode = 1;
755 enc_blk->zigzag_scan = ff_zigzag248_direct;
756 } else {
757 enc_blk->dct_mode = 0;
758 enc_blk->zigzag_scan = ff_zigzag_direct;
759 }
760 enc_blk->mb = block;
761 enc_blk->block_size = block_sizes[j];
762
763 s->fdct[enc_blk->dct_mode](block);
764
765 dv_set_class_number(enc_blk, j/4*(j%2));
766
767 block += 64;
768 enc_blk++;
769 }
770 }
771
772 /* Stage 2 -- encoding by trial-and-error */
773 encode_vs:
774 enc_blk = &enc_blks[0];
775 for (i=0; i<5; i++) {
776 uint8_t* p = dif + i*80 + 4;
777 for (j=0; j<6; j++) {
778 enc_blk->qno = QNO;
779 enc_blk->dv_shift = &(s->dv_dct_shift[0]
780 [QNO + dv_quant_offset[enc_blk->cno]][0]);
781 init_put_bits(&enc_blk->pb, p, block_sizes[j]/8);
782 enc_blk++;
783 p += block_sizes[j]/8;
784 }
785 }
786
787 init_put_bits(&extra_vs, extra_vs_data, sizeof(extra_vs_data));
788 free_vs_bits = 0;
789 enc_blk = &enc_blks[0];
790 for (i=0; i<5; i++) {
791 PutBitContext extra_mb;
792 EncBlockInfo* enc_blk2 = enc_blk;
793 int free_mb_bits = 0;
794
795 init_put_bits(&extra_mb, extra_mb_data, sizeof(extra_mb_data));
796 dif[i*80 + 3] = enc_blk->qno;
797
798 for (j=0; j<6; j++) {
799 uint16_t dc = ((enc_blk->mb[0] >> 3) - 1024) >> 2;
800
801 put_bits(&enc_blk->pb, 9, dc);
802 put_bits(&enc_blk->pb, 1, enc_blk->dct_mode);
803 put_bits(&enc_blk->pb, 2, enc_blk->cno);
804
805 #ifdef VLC_DEBUG
806 printf("[%d, %d]: ", i, j);
807 #endif
808 dv_encode_ac(enc_blk, &extra_mb);
809 #ifdef VLC_DEBUG
810 printf("\n");
811 #endif
812
813 free_mb_bits += dv_bits_left(enc_blk);
814 enc_blk++;
815 }
816
817 /* We can't flush extra_mb just yet -- since it'll round up bit number */
818 extra_bits = get_bit_count(&extra_mb);
819 if (free_mb_bits > extra_bits)
820 free_vs_bits += free_mb_bits - extra_bits;
821
822 if (extra_bits) { /* FIXME: speed up things when free_mb_bits == 0 */
823 flush_put_bits(&extra_mb);
824 dv_redistr_bits(enc_blk2, 6, extra_mb_data, extra_bits, &extra_vs);
825 }
826 }
827
828 /* We can't flush extra_mb just yet -- since it'll round up bit number */
829 extra_bits = get_bit_count(&extra_vs);
830 if (extra_bits > free_vs_bits && QNO) { /* FIXME: very crude trial-and-error */
831 QNO--;
832 goto encode_vs;
833 }
834
835 if (extra_bits) {
836 flush_put_bits(&extra_vs);
837 dv_redistr_bits(&enc_blks[0], 5*6, extra_vs_data, extra_bits, NULL);
838 }
839
840 for (i=0; i<6*5; i++) {
841 flush_put_bits(&enc_blks[i].pb);
842 #ifdef VLC_DEBUG
843 printf("[%d:%d] qno=%d cno=%d\n", i/6, i%6, enc_blks[i].qno, enc_blks[i].cno);
844 #endif
845 }
846 }
847
848 /* NOTE: exactly one frame must be given (120000 bytes for NTSC,
849 144000 bytes for PAL) */
850 static int dvvideo_decode_frame(AVCodecContext *avctx,
851 void *data, int *data_size,
852 uint8_t *buf, int buf_size)
853 {
854 DVVideoDecodeContext *s = avctx->priv_data;
855 int ds, vs;
856 const uint16_t *mb_pos_ptr;
857
858 s->sys = dv_frame_profile(buf);
859 if (!s->sys || buf_size < s->sys->frame_size)
860 return -1; /* NOTE: we only accept several full frames */
861
862
863 if(s->picture.data[0])
864 avctx->release_buffer(avctx, &s->picture);
865
866 s->picture.reference = 0;
867 avctx->pix_fmt = s->sys->pix_fmt;
868 avctx->width = s->sys->width;
869 avctx->height = s->sys->height;
870 if(avctx->get_buffer(avctx, &s->picture) < 0) {
871 fprintf(stderr, "get_buffer() failed\n");
872 return -1;
873 }
874 s->picture.interlaced_frame = 1;
875 s->picture.top_field_first = 0;
876
877 /* for each DIF segment */
878 mb_pos_ptr = s->sys->video_place;
879 for (ds = 0; ds < s->sys->difseg_size; ds++) {
880 buf += 6 * 80; /* skip DIF segment header */
881
882 for(vs = 0; vs < 27; vs++) {
883 if ((vs % 3) == 0)
884 buf += 80; /* skip audio block */
885
886 #ifdef VLC_DEBUG
887 printf("********************* %d, %d **********************\n", ds, vs);
888 #endif
889 dv_decode_video_segment(s, buf, mb_pos_ptr);
890 buf += 5 * 80;
891 mb_pos_ptr += 5;
892 }
893 }
894
895 emms_c();
896
897 /* return image */
898 *data_size = sizeof(AVFrame);
899 *(AVFrame*)data= s->picture;
900
901 return s->sys->frame_size;
902 }
903
904 static int dvvideo_encode_frame(AVCodecContext *c, uint8_t *buf, int buf_size,
905 void *data)
906 {
907 DVVideoDecodeContext *s = c->priv_data;
908 const uint16_t *mb_pos_ptr;
909 int ds, vs;
910
911 s->sys = dv_codec_profile(c);
912 if (!s->sys)
913 return -1;
914
915 c->pix_fmt = s->sys->pix_fmt;
916 s->picture = *((AVFrame *)data);
917
918 /* for each DIF segment */
919 mb_pos_ptr = s->sys->video_place;
920 for (ds = 0; ds < s->sys->difseg_size; ds++) {
921 buf += 6 * 80; /* skip DIF segment header */
922
923 for(vs = 0; vs < 27; vs++) {
924 if ((vs % 3) == 0)
925 buf += 80; /* skip audio block */
926
927 #ifdef VLC_DEBUG
928 printf("********************* %d, %d **********************\n", ds, vs);
929 #endif
930 dv_encode_video_segment(s, buf, mb_pos_ptr);
931 buf += 5 * 80;
932 mb_pos_ptr += 5;
933 }
934 }
935
936 emms_c();
937 return s->sys->frame_size;
938 }
939
940 static int dvvideo_end(AVCodecContext *avctx)
941 {
942 avcodec_default_free_buffers(avctx);
943 return 0;
944 }
945
946 AVCodec dvvideo_decoder = {
947 "dvvideo",
948 CODEC_TYPE_VIDEO,
949 CODEC_ID_DVVIDEO,
950 sizeof(DVVideoDecodeContext),
951 dvvideo_init,
952 dvvideo_encode_frame,
953 dvvideo_end,
954 dvvideo_decode_frame,
955 CODEC_CAP_DR1,
956 NULL
957 };