* simplification patch by Jeffrey Ryan Muizelaar
[libav.git] / libavcodec / dv.c
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 * Many thanks to Dan Dennedy <dan@dennedy.org> for providing wealth
10 * of DV technical info.
11 *
12 * This library is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU Lesser General Public
14 * License as published by the Free Software Foundation; either
15 * version 2 of the License, or (at your option) any later version.
16 *
17 * This library is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * Lesser General Public License for more details.
21 *
22 * You should have received a copy of the GNU Lesser General Public
23 * License along with this library; if not, write to the Free Software
24 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
25 */
26
27 /**
28 * @file dv.c
29 * DV codec.
30 */
31 #include "avcodec.h"
32 #include "dsputil.h"
33 #include "mpegvideo.h"
34 #include "simple_idct.h"
35 #include "dvdata.h"
36
37 typedef struct DVVideoContext {
38 const DVprofile* sys;
39 AVFrame picture;
40 uint8_t *buf;
41
42 uint8_t dv_zigzag[2][64];
43 uint8_t dv_idct_shift[2][2][22][64];
44
45 void (*get_pixels)(DCTELEM *block, const uint8_t *pixels, int line_size);
46 void (*fdct[2])(DCTELEM *block);
47 void (*idct_put[2])(uint8_t *dest, int line_size, DCTELEM *block);
48 } DVVideoContext;
49
50 #define TEX_VLC_BITS 9
51
52 #ifdef DV_CODEC_TINY_TARGET
53 #define DV_VLC_MAP_RUN_SIZE 15
54 #define DV_VLC_MAP_LEV_SIZE 23
55 #else
56 #define DV_VLC_MAP_RUN_SIZE 64
57 #define DV_VLC_MAP_LEV_SIZE 512
58 #endif
59
60 /* MultiThreading */
61 static uint8_t** dv_anchor;
62
63 /* XXX: also include quantization */
64 static RL_VLC_ELEM *dv_rl_vlc;
65 /* VLC encoding lookup table */
66 static struct dv_vlc_pair {
67 uint32_t vlc;
68 uint8_t size;
69 } (*dv_vlc_map)[DV_VLC_MAP_LEV_SIZE] = NULL;
70
71 static void dv_build_unquantize_tables(DVVideoContext *s, uint8_t* perm)
72 {
73 int i, q, j;
74
75 /* NOTE: max left shift is 6 */
76 for(q = 0; q < 22; q++) {
77 /* 88DCT */
78 for(i = 1; i < 64; i++) {
79 /* 88 table */
80 j = perm[i];
81 s->dv_idct_shift[0][0][q][j] =
82 dv_quant_shifts[q][dv_88_areas[i]] + 1;
83 s->dv_idct_shift[1][0][q][j] = s->dv_idct_shift[0][0][q][j] + 1;
84 }
85
86 /* 248DCT */
87 for(i = 1; i < 64; i++) {
88 /* 248 table */
89 s->dv_idct_shift[0][1][q][i] =
90 dv_quant_shifts[q][dv_248_areas[i]] + 1;
91 s->dv_idct_shift[1][1][q][i] = s->dv_idct_shift[0][1][q][i] + 1;
92 }
93 }
94 }
95
96 static int dvvideo_init(AVCodecContext *avctx)
97 {
98 DVVideoContext *s = avctx->priv_data;
99 DSPContext dsp;
100 static int done=0;
101 int i, j;
102
103 if (!done) {
104 VLC dv_vlc;
105 uint16_t new_dv_vlc_bits[NB_DV_VLC*2];
106 uint8_t new_dv_vlc_len[NB_DV_VLC*2];
107 uint8_t new_dv_vlc_run[NB_DV_VLC*2];
108 int16_t new_dv_vlc_level[NB_DV_VLC*2];
109
110 done = 1;
111
112 dv_vlc_map = av_mallocz(DV_VLC_MAP_LEV_SIZE*DV_VLC_MAP_RUN_SIZE*sizeof(struct dv_vlc_pair));
113 if (!dv_vlc_map)
114 return -ENOMEM;
115
116 /* dv_anchor lets each thread know its Id */
117 dv_anchor = av_malloc(12*27*sizeof(void*));
118 if (!dv_anchor) {
119 av_free(dv_vlc_map);
120 return -ENOMEM;
121 }
122 for (i=0; i<12*27; i++)
123 dv_anchor[i] = (void*)(size_t)i;
124
125 /* it's faster to include sign bit in a generic VLC parsing scheme */
126 for (i=0, j=0; i<NB_DV_VLC; i++, j++) {
127 new_dv_vlc_bits[j] = dv_vlc_bits[i];
128 new_dv_vlc_len[j] = dv_vlc_len[i];
129 new_dv_vlc_run[j] = dv_vlc_run[i];
130 new_dv_vlc_level[j] = dv_vlc_level[i];
131
132 if (dv_vlc_level[i]) {
133 new_dv_vlc_bits[j] <<= 1;
134 new_dv_vlc_len[j]++;
135
136 j++;
137 new_dv_vlc_bits[j] = (dv_vlc_bits[i] << 1) | 1;
138 new_dv_vlc_len[j] = dv_vlc_len[i] + 1;
139 new_dv_vlc_run[j] = dv_vlc_run[i];
140 new_dv_vlc_level[j] = -dv_vlc_level[i];
141 }
142 }
143
144 /* NOTE: as a trick, we use the fact the no codes are unused
145 to accelerate the parsing of partial codes */
146 init_vlc(&dv_vlc, TEX_VLC_BITS, j,
147 new_dv_vlc_len, 1, 1, new_dv_vlc_bits, 2, 2);
148
149 dv_rl_vlc = av_malloc(dv_vlc.table_size * sizeof(RL_VLC_ELEM));
150 if (!dv_rl_vlc) {
151 av_free(dv_anchor);
152 av_free(dv_vlc_map);
153 return -ENOMEM;
154 }
155 for(i = 0; i < dv_vlc.table_size; i++){
156 int code= dv_vlc.table[i][0];
157 int len = dv_vlc.table[i][1];
158 int level, run;
159
160 if(len<0){ //more bits needed
161 run= 0;
162 level= code;
163 } else {
164 run= new_dv_vlc_run[code] + 1;
165 level= new_dv_vlc_level[code];
166 }
167 dv_rl_vlc[i].len = len;
168 dv_rl_vlc[i].level = level;
169 dv_rl_vlc[i].run = run;
170 }
171 free_vlc(&dv_vlc);
172
173 for (i = 0; i < NB_DV_VLC - 1; i++) {
174 if (dv_vlc_run[i] >= DV_VLC_MAP_RUN_SIZE)
175 continue;
176 #ifdef DV_CODEC_TINY_TARGET
177 if (dv_vlc_level[i] >= DV_VLC_MAP_LEV_SIZE)
178 continue;
179 #endif
180
181 if (dv_vlc_map[dv_vlc_run[i]][dv_vlc_level[i]].size != 0)
182 continue;
183
184 dv_vlc_map[dv_vlc_run[i]][dv_vlc_level[i]].vlc = dv_vlc_bits[i] <<
185 (!!dv_vlc_level[i]);
186 dv_vlc_map[dv_vlc_run[i]][dv_vlc_level[i]].size = dv_vlc_len[i] +
187 (!!dv_vlc_level[i]);
188 }
189 for (i = 0; i < DV_VLC_MAP_RUN_SIZE; i++) {
190 #ifdef DV_CODEC_TINY_TARGET
191 for (j = 1; j < DV_VLC_MAP_LEV_SIZE; j++) {
192 if (dv_vlc_map[i][j].size == 0) {
193 dv_vlc_map[i][j].vlc = dv_vlc_map[0][j].vlc |
194 (dv_vlc_map[i-1][0].vlc << (dv_vlc_map[0][j].size));
195 dv_vlc_map[i][j].size = dv_vlc_map[i-1][0].size +
196 dv_vlc_map[0][j].size;
197 }
198 }
199 #else
200 for (j = 1; j < DV_VLC_MAP_LEV_SIZE/2; j++) {
201 if (dv_vlc_map[i][j].size == 0) {
202 dv_vlc_map[i][j].vlc = dv_vlc_map[0][j].vlc |
203 (dv_vlc_map[i-1][0].vlc << (dv_vlc_map[0][j].size));
204 dv_vlc_map[i][j].size = dv_vlc_map[i-1][0].size +
205 dv_vlc_map[0][j].size;
206 }
207 dv_vlc_map[i][((uint16_t)(-j))&0x1ff].vlc =
208 dv_vlc_map[i][j].vlc | 1;
209 dv_vlc_map[i][((uint16_t)(-j))&0x1ff].size =
210 dv_vlc_map[i][j].size;
211 }
212 #endif
213 }
214 }
215
216 /* Generic DSP setup */
217 dsputil_init(&dsp, avctx);
218 s->get_pixels = dsp.get_pixels;
219
220 /* 88DCT setup */
221 s->fdct[0] = dsp.fdct;
222 s->idct_put[0] = dsp.idct_put;
223 for (i=0; i<64; i++)
224 s->dv_zigzag[0][i] = dsp.idct_permutation[ff_zigzag_direct[i]];
225
226 /* 248DCT setup */
227 s->fdct[1] = dsp.fdct248;
228 s->idct_put[1] = simple_idct248_put; // FIXME: need to add it to DSP
229 memcpy(s->dv_zigzag[1], ff_zigzag248_direct, 64);
230
231 /* XXX: do it only for constant case */
232 dv_build_unquantize_tables(s, dsp.idct_permutation);
233
234 /* FIXME: I really don't think this should be here */
235 if (dv_codec_profile(avctx))
236 avctx->pix_fmt = dv_codec_profile(avctx)->pix_fmt;
237 avctx->coded_frame = &s->picture;
238
239 return 0;
240 }
241
242 // #define VLC_DEBUG
243 // #define printf(...) av_log(NULL, AV_LOG_ERROR, __VA_ARGS__)
244
245 typedef struct BlockInfo {
246 const uint8_t *shift_table;
247 const uint8_t *scan_table;
248 uint8_t pos; /* position in block */
249 uint8_t dct_mode;
250 uint8_t partial_bit_count;
251 uint16_t partial_bit_buffer;
252 int shift_offset;
253 } BlockInfo;
254
255 /* block size in bits */
256 static const uint16_t block_sizes[6] = {
257 112, 112, 112, 112, 80, 80
258 };
259 /* bit budget for AC only in 5 MBs */
260 static const int vs_total_ac_bits = (100 * 4 + 68*2) * 5;
261 /* see dv_88_areas and dv_248_areas for details */
262 static const int mb_area_start[5] = { 1, 6, 21, 43, 64 };
263
264 #ifndef ALT_BITSTREAM_READER
265 #warning only works with ALT_BITSTREAM_READER
266 #endif
267
268 static inline int get_bits_left(GetBitContext *s)
269 {
270 return s->size_in_bits - get_bits_count(s);
271 }
272
273 static inline int get_bits_size(GetBitContext *s)
274 {
275 return s->size_in_bits;
276 }
277
278 static inline int put_bits_left(PutBitContext* s)
279 {
280 return (s->buf_end - s->buf) * 8 - put_bits_count(s);
281 }
282
283 /* decode ac coefs */
284 static void dv_decode_ac(GetBitContext *gb, BlockInfo *mb, DCTELEM *block)
285 {
286 int last_index = get_bits_size(gb);
287 const uint8_t *scan_table = mb->scan_table;
288 const uint8_t *shift_table = mb->shift_table;
289 int pos = mb->pos;
290 int partial_bit_count = mb->partial_bit_count;
291 int level, pos1, run, vlc_len, index;
292
293 OPEN_READER(re, gb);
294 UPDATE_CACHE(re, gb);
295
296 /* if we must parse a partial vlc, we do it here */
297 if (partial_bit_count > 0) {
298 re_cache = ((unsigned)re_cache >> partial_bit_count) |
299 (mb->partial_bit_buffer << (sizeof(re_cache)*8 - partial_bit_count));
300 re_index -= partial_bit_count;
301 mb->partial_bit_count = 0;
302 }
303
304 /* get the AC coefficients until last_index is reached */
305 for(;;) {
306 #ifdef VLC_DEBUG
307 printf("%2d: bits=%04x index=%d\n", pos, SHOW_UBITS(re, gb, 16), re_index);
308 #endif
309 /* our own optimized GET_RL_VLC */
310 index = NEG_USR32(re_cache, TEX_VLC_BITS);
311 vlc_len = dv_rl_vlc[index].len;
312 if (vlc_len < 0) {
313 index = NEG_USR32((unsigned)re_cache << TEX_VLC_BITS, -vlc_len) + dv_rl_vlc[index].level;
314 vlc_len = TEX_VLC_BITS - vlc_len;
315 }
316 level = dv_rl_vlc[index].level;
317 run = dv_rl_vlc[index].run;
318
319 /* gotta check if we're still within gb boundaries */
320 if (re_index + vlc_len > last_index) {
321 /* should be < 16 bits otherwise a codeword could have been parsed */
322 mb->partial_bit_count = last_index - re_index;
323 mb->partial_bit_buffer = NEG_USR32(re_cache, mb->partial_bit_count);
324 re_index = last_index;
325 break;
326 }
327 re_index += vlc_len;
328
329 #ifdef VLC_DEBUG
330 printf("run=%d level=%d\n", run, level);
331 #endif
332 pos += run;
333 if (pos >= 64)
334 break;
335
336 if (level) {
337 pos1 = scan_table[pos];
338 block[pos1] = level << shift_table[pos1];
339 }
340
341 UPDATE_CACHE(re, gb);
342 }
343 CLOSE_READER(re, gb);
344 mb->pos = pos;
345 }
346
347 static inline void bit_copy(PutBitContext *pb, GetBitContext *gb)
348 {
349 int bits_left = get_bits_left(gb);
350 while (bits_left >= 16) {
351 put_bits(pb, 16, get_bits(gb, 16));
352 bits_left -= 16;
353 }
354 if (bits_left > 0) {
355 put_bits(pb, bits_left, get_bits(gb, bits_left));
356 }
357 }
358
359 /* mb_x and mb_y are in units of 8 pixels */
360 static inline void dv_decode_video_segment(DVVideoContext *s,
361 uint8_t *buf_ptr1,
362 const uint16_t *mb_pos_ptr)
363 {
364 int quant, dc, dct_mode, class1, j;
365 int mb_index, mb_x, mb_y, v, last_index;
366 DCTELEM *block, *block1;
367 int c_offset;
368 uint8_t *y_ptr;
369 void (*idct_put)(uint8_t *dest, int line_size, DCTELEM *block);
370 uint8_t *buf_ptr;
371 PutBitContext pb, vs_pb;
372 GetBitContext gb;
373 BlockInfo mb_data[5 * 6], *mb, *mb1;
374 DCTELEM sblock[5*6][64] __align8;
375 uint8_t mb_bit_buffer[80 + 4]; /* allow some slack */
376 uint8_t vs_bit_buffer[5 * 80 + 4]; /* allow some slack */
377
378 memset(sblock, 0, sizeof(sblock));
379
380 /* pass 1 : read DC and AC coefficients in blocks */
381 buf_ptr = buf_ptr1;
382 block1 = &sblock[0][0];
383 mb1 = mb_data;
384 init_put_bits(&vs_pb, vs_bit_buffer, 5 * 80);
385 for(mb_index = 0; mb_index < 5; mb_index++, mb1 += 6, block1 += 6 * 64) {
386 /* skip header */
387 quant = buf_ptr[3] & 0x0f;
388 buf_ptr += 4;
389 init_put_bits(&pb, mb_bit_buffer, 80);
390 mb = mb1;
391 block = block1;
392 for(j = 0;j < 6; j++) {
393 last_index = block_sizes[j];
394 init_get_bits(&gb, buf_ptr, last_index);
395
396 /* get the dc */
397 dc = get_sbits(&gb, 9);
398 dct_mode = get_bits1(&gb);
399 mb->dct_mode = dct_mode;
400 mb->scan_table = s->dv_zigzag[dct_mode];
401 class1 = get_bits(&gb, 2);
402 mb->shift_table = s->dv_idct_shift[class1 == 3][dct_mode]
403 [quant + dv_quant_offset[class1]];
404 dc = dc << 2;
405 /* convert to unsigned because 128 is not added in the
406 standard IDCT */
407 dc += 1024;
408 block[0] = dc;
409 buf_ptr += last_index >> 3;
410 mb->pos = 0;
411 mb->partial_bit_count = 0;
412
413 #ifdef VLC_DEBUG
414 printf("MB block: %d, %d ", mb_index, j);
415 #endif
416 dv_decode_ac(&gb, mb, block);
417
418 /* write the remaining bits in a new buffer only if the
419 block is finished */
420 if (mb->pos >= 64)
421 bit_copy(&pb, &gb);
422
423 block += 64;
424 mb++;
425 }
426
427 /* pass 2 : we can do it just after */
428 #ifdef VLC_DEBUG
429 printf("***pass 2 size=%d MB#=%d\n", put_bits_count(&pb), mb_index);
430 #endif
431 block = block1;
432 mb = mb1;
433 init_get_bits(&gb, mb_bit_buffer, put_bits_count(&pb));
434 flush_put_bits(&pb);
435 for(j = 0;j < 6; j++, block += 64, mb++) {
436 if (mb->pos < 64 && get_bits_left(&gb) > 0) {
437 dv_decode_ac(&gb, mb, block);
438 /* if still not finished, no need to parse other blocks */
439 if (mb->pos < 64)
440 break;
441 }
442 }
443 /* all blocks are finished, so the extra bytes can be used at
444 the video segment level */
445 if (j >= 6)
446 bit_copy(&vs_pb, &gb);
447 }
448
449 /* we need a pass other the whole video segment */
450 #ifdef VLC_DEBUG
451 printf("***pass 3 size=%d\n", put_bits_count(&vs_pb));
452 #endif
453 block = &sblock[0][0];
454 mb = mb_data;
455 init_get_bits(&gb, vs_bit_buffer, put_bits_count(&vs_pb));
456 flush_put_bits(&vs_pb);
457 for(mb_index = 0; mb_index < 5; mb_index++) {
458 for(j = 0;j < 6; j++) {
459 if (mb->pos < 64) {
460 #ifdef VLC_DEBUG
461 printf("start %d:%d\n", mb_index, j);
462 #endif
463 dv_decode_ac(&gb, mb, block);
464 }
465 if (mb->pos >= 64 && mb->pos < 127)
466 av_log(NULL, AV_LOG_ERROR, "AC EOB marker is absent pos=%d\n", mb->pos);
467 block += 64;
468 mb++;
469 }
470 }
471
472 /* compute idct and place blocks */
473 block = &sblock[0][0];
474 mb = mb_data;
475 for(mb_index = 0; mb_index < 5; mb_index++) {
476 v = *mb_pos_ptr++;
477 mb_x = v & 0xff;
478 mb_y = v >> 8;
479 y_ptr = s->picture.data[0] + (mb_y * s->picture.linesize[0] * 8) + (mb_x * 8);
480 if (s->sys->pix_fmt == PIX_FMT_YUV411P)
481 c_offset = (mb_y * s->picture.linesize[1] * 8) + ((mb_x >> 2) * 8);
482 else
483 c_offset = ((mb_y >> 1) * s->picture.linesize[1] * 8) + ((mb_x >> 1) * 8);
484 for(j = 0;j < 6; j++) {
485 idct_put = s->idct_put[mb->dct_mode];
486 if (j < 4) {
487 if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x < (704 / 8)) {
488 /* NOTE: at end of line, the macroblock is handled as 420 */
489 idct_put(y_ptr + (j * 8), s->picture.linesize[0], block);
490 } else {
491 idct_put(y_ptr + ((j & 1) * 8) + ((j >> 1) * 8 * s->picture.linesize[0]),
492 s->picture.linesize[0], block);
493 }
494 } else {
495 if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x >= (704 / 8)) {
496 uint64_t aligned_pixels[64/8];
497 uint8_t *pixels= (uint8_t*)aligned_pixels;
498 uint8_t *c_ptr, *c_ptr1, *ptr;
499 int y, linesize;
500 /* NOTE: at end of line, the macroblock is handled as 420 */
501 idct_put(pixels, 8, block);
502 linesize = s->picture.linesize[6 - j];
503 c_ptr = s->picture.data[6 - j] + c_offset;
504 ptr = pixels;
505 for(y = 0;y < 8; y++) {
506 /* convert to 411P */
507 c_ptr1 = c_ptr + 8*linesize;
508 c_ptr[0]= ptr[0]; c_ptr1[0]= ptr[4];
509 c_ptr[1]= ptr[1]; c_ptr1[1]= ptr[5];
510 c_ptr[2]= ptr[2]; c_ptr1[2]= ptr[6];
511 c_ptr[3]= ptr[3]; c_ptr1[3]= ptr[7];
512 c_ptr += linesize;
513 ptr += 8;
514 }
515 } else {
516 /* don't ask me why they inverted Cb and Cr ! */
517 idct_put(s->picture.data[6 - j] + c_offset,
518 s->picture.linesize[6 - j], block);
519 }
520 }
521 block += 64;
522 mb++;
523 }
524 }
525 }
526
527 #ifdef DV_CODEC_TINY_TARGET
528 /* Converts run and level (where level != 0) pair into vlc, returning bit size */
529 static always_inline int dv_rl2vlc(int run, int l, uint32_t* vlc)
530 {
531 int sign = l >> 8;
532 int level = (l ^ sign) - sign;
533 int size;
534
535 sign = (sign & 1);
536
537 if (run < DV_VLC_MAP_RUN_SIZE && level < DV_VLC_MAP_LEV_SIZE) {
538 *vlc = dv_vlc_map[run][level].vlc | sign;
539 size = dv_vlc_map[run][level].size;
540 }
541 else {
542 if (level < DV_VLC_MAP_LEV_SIZE) {
543 *vlc = dv_vlc_map[0][level].vlc | sign;
544 size = dv_vlc_map[0][level].size;
545 } else {
546 *vlc = 0xfe00 | (level << 1) | sign;
547 size = 16;
548 }
549 if (run) {
550 *vlc |= ((run < 16) ? dv_vlc_map[run-1][0].vlc :
551 (0x1f80 | (run - 1))) << size;
552 size += (run < 16) ? dv_vlc_map[run-1][0].size : 13;
553 }
554 }
555
556 return size;
557 }
558
559 static always_inline int dv_rl2vlc_size(int run, int l)
560 {
561 int level = (l ^ (l >> 8)) - (l >> 8);
562 int size;
563
564 if (run < DV_VLC_MAP_RUN_SIZE && level < DV_VLC_MAP_LEV_SIZE) {
565 size = dv_vlc_map[run][level].size;
566 }
567 else {
568 size = (level < DV_VLC_MAP_LEV_SIZE) ? dv_vlc_map[0][level].size : 16;
569 if (run) {
570 size += (run < 16) ? dv_vlc_map[run-1][0].size : 13;
571 }
572 }
573 return size;
574 }
575 #else
576 static always_inline int dv_rl2vlc(int run, int l, uint32_t* vlc)
577 {
578 *vlc = dv_vlc_map[run][((uint16_t)l)&0x1ff].vlc;
579 return dv_vlc_map[run][((uint16_t)l)&0x1ff].size;
580 }
581
582 static always_inline int dv_rl2vlc_size(int run, int l)
583 {
584 return dv_vlc_map[run][((uint16_t)l)&0x1ff].size;
585 }
586 #endif
587
588 typedef struct EncBlockInfo {
589 int area_q[4];
590 int bit_size[4];
591 int prev_run[4];
592 int cur_ac;
593 int cno;
594 int dct_mode;
595 DCTELEM *mb;
596 uint8_t partial_bit_count;
597 uint32_t partial_bit_buffer; /* we can't use uint16_t here */
598 } EncBlockInfo;
599
600 static always_inline void dv_encode_ac(EncBlockInfo* bi, PutBitContext* pb_pool,
601 int pb_size)
602 {
603 int run;
604 int bits_left;
605 PutBitContext* pb = pb_pool;
606 int size = bi->partial_bit_count;
607 uint32_t vlc = bi->partial_bit_buffer;
608
609 bi->partial_bit_count = bi->partial_bit_buffer = 0;
610 vlc_loop:
611 /* Find suitable storage space */
612 for (; size > (bits_left = put_bits_left(pb)); pb++) {
613 if (bits_left) {
614 size -= bits_left;
615 put_bits(pb, bits_left, vlc >> size);
616 vlc = vlc & ((1<<size)-1);
617 }
618 if (pb_size == 1) {
619 bi->partial_bit_count = size;
620 bi->partial_bit_buffer = vlc;
621 return;
622 }
623 --pb_size;
624 }
625
626 /* Store VLC */
627 put_bits(pb, size, vlc);
628
629 /* Construct the next VLC */
630 run = 0;
631 for (; bi->cur_ac < 64; bi->cur_ac++, run++) {
632 if (bi->mb[bi->cur_ac]) {
633 size = dv_rl2vlc(run, bi->mb[bi->cur_ac], &vlc);
634 bi->cur_ac++;
635 goto vlc_loop;
636 }
637 }
638
639 if (bi->cur_ac == 64) {
640 size = 4; vlc = 6; /* End Of Block stamp */
641 bi->cur_ac++;
642 goto vlc_loop;
643 }
644 }
645
646 static always_inline void dv_set_class_number(DCTELEM* blk, EncBlockInfo* bi,
647 const uint8_t* zigzag_scan, int bias)
648 {
649 int i, area;
650 int run;
651 int classes[] = {12, 24, 36, 0xffff};
652
653 run = 0;
654 bi->mb[0] = blk[0];
655 bi->cno = 0;
656 for (area = 0; area < 4; area++) {
657 bi->prev_run[area] = run;
658 bi->bit_size[area] = 0;
659 for (i=mb_area_start[area]; i<mb_area_start[area+1]; i++) {
660 bi->mb[i] = (blk[zigzag_scan[i]] / 16);
661 while ((bi->mb[i] ^ (bi->mb[i] >> 8)) > classes[bi->cno])
662 bi->cno++;
663
664 if (bi->mb[i]) {
665 bi->bit_size[area] += dv_rl2vlc_size(run, bi->mb[i]);
666 run = 0;
667 } else
668 ++run;
669 }
670 }
671 bi->bit_size[3] += 4; /* EOB marker */
672 bi->cno += bias;
673
674 if (bi->cno >= 3) { /* FIXME: we have to recreate bit_size[], prev_run[] */
675 bi->cno = 3;
676 for (i=1; i<64; i++)
677 bi->mb[i] /= 2;
678 }
679 }
680
681 #define SC(x, y) ((s[x] - s[y]) ^ ((s[x] - s[y]) >> 7))
682 static always_inline int dv_guess_dct_mode(DCTELEM *blk) {
683 DCTELEM *s;
684 int score88 = 0;
685 int score248 = 0;
686 int i;
687
688 /* Compute 8-8 score (small values give a better chance for 8-8 DCT) */
689 s = blk;
690 for(i=0; i<7; i++) {
691 score88 += SC(0, 8) + SC(1, 9) + SC(2, 10) + SC(3, 11) +
692 SC(4, 12) + SC(5,13) + SC(6, 14) + SC(7, 15);
693 s += 8;
694 }
695 /* Compute 2-4-8 score (small values give a better chance for 2-4-8 DCT) */
696 s = blk;
697 for(i=0; i<6; i++) {
698 score248 += SC(0, 16) + SC(1,17) + SC(2, 18) + SC(3, 19) +
699 SC(4, 20) + SC(5,21) + SC(6, 22) + SC(7, 23);
700 s += 8;
701 }
702
703 return (score88 - score248 > -10);
704 }
705
706 static inline void dv_guess_qnos(EncBlockInfo* blks, int* qnos)
707 {
708 int size[5];
709 int i, j, k, a, run;
710 EncBlockInfo* b;
711
712 do {
713 b = blks;
714 for (i=0; i<5; i++) {
715 if (!qnos[i])
716 continue;
717
718 qnos[i]--;
719 size[i] = 0;
720 for (j=0; j<6; j++, b++) {
721 for (a=0; a<4; a++) {
722 if (b->area_q[a] != dv_quant_shifts[qnos[i] + dv_quant_offset[b->cno]][a]) {
723 b->bit_size[a] = (a==3)?4:0;
724 b->area_q[a]++;
725 run = b->prev_run[a];
726 for (k=mb_area_start[a]; k<mb_area_start[a+1]; k++) {
727 b->mb[k] /= 2;
728 if (b->mb[k]) {
729 b->bit_size[a] += dv_rl2vlc_size(run, b->mb[k]);
730 run = 0;
731 } else
732 ++run;
733 }
734 }
735 size[i] += b->bit_size[a];
736 }
737 }
738 }
739 } while ((vs_total_ac_bits < size[0] + size[1] + size[2] + size[3] + size[4]) &&
740 (qnos[0]|qnos[1]|qnos[2]|qnos[3]|qnos[4]));
741 }
742
743 /*
744 * This is a very rough initial implementaion. The performance is
745 * horrible and the weighting is missing. But it's missing from the
746 * decoding step also -- so at least we're on the same page with decoder ;-)
747 */
748 static inline void dv_encode_video_segment(DVVideoContext *s,
749 uint8_t *dif,
750 const uint16_t *mb_pos_ptr)
751 {
752 int mb_index, i, j, v;
753 int mb_x, mb_y, c_offset, linesize;
754 uint8_t* y_ptr;
755 uint8_t* data;
756 uint8_t* ptr;
757 int do_edge_wrap;
758 DCTELEM block[64] __align8;
759 DCTELEM sblock[5*6][64] __align8;
760 EncBlockInfo enc_blks[5*6];
761 PutBitContext pbs[5*6];
762 PutBitContext* pb;
763 EncBlockInfo* enc_blk;
764 int vs_bit_size = 0;
765 int qnos[5];
766
767 enc_blk = &enc_blks[0];
768 pb = &pbs[0];
769 for(mb_index = 0; mb_index < 5; mb_index++) {
770 v = *mb_pos_ptr++;
771 mb_x = v & 0xff;
772 mb_y = v >> 8;
773 y_ptr = s->picture.data[0] + (mb_y * s->picture.linesize[0] * 8) + (mb_x * 8);
774 c_offset = (s->sys->pix_fmt == PIX_FMT_YUV411P) ?
775 ((mb_y * s->picture.linesize[1] * 8) + ((mb_x >> 2) * 8)) :
776 (((mb_y >> 1) * s->picture.linesize[1] * 8) + ((mb_x >> 1) * 8));
777 do_edge_wrap = 0;
778 qnos[mb_index] = 15; /* No quantization */
779 ptr = dif + mb_index*80 + 4;
780 for(j = 0;j < 6; j++) {
781 if (j < 4) { /* Four Y blocks */
782 /* NOTE: at end of line, the macroblock is handled as 420 */
783 if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x < (704 / 8)) {
784 data = y_ptr + (j * 8);
785 } else {
786 data = y_ptr + ((j & 1) * 8) + ((j >> 1) * 8 * s->picture.linesize[0]);
787 }
788 linesize = s->picture.linesize[0];
789 } else { /* Cr and Cb blocks */
790 /* don't ask Fabrice why they inverted Cb and Cr ! */
791 data = s->picture.data[6 - j] + c_offset;
792 linesize = s->picture.linesize[6 - j];
793 if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x >= (704 / 8))
794 do_edge_wrap = 1;
795 }
796
797 /* Everything is set up -- now just copy data -> DCT block */
798 if (do_edge_wrap) { /* Edge wrap copy: 4x16 -> 8x8 */
799 uint8_t* d;
800 DCTELEM *b = block;
801 for (i=0;i<8;i++) {
802 d = data + 8 * linesize;
803 b[0] = data[0]; b[1] = data[1]; b[2] = data[2]; b[3] = data[3];
804 b[4] = d[0]; b[5] = d[1]; b[6] = d[2]; b[7] = d[3];
805 data += linesize;
806 b += 8;
807 }
808 } else { /* Simple copy: 8x8 -> 8x8 */
809 s->get_pixels(block, data, linesize);
810 }
811
812 enc_blk->dct_mode = dv_guess_dct_mode(block);
813 enc_blk->mb = &sblock[mb_index*6+j][0];
814 enc_blk->area_q[0] = enc_blk->area_q[1] = enc_blk->area_q[2] = enc_blk->area_q[3] = 0;
815 enc_blk->partial_bit_count = 0;
816 enc_blk->partial_bit_buffer = 0;
817 enc_blk->cur_ac = 1;
818
819 s->fdct[enc_blk->dct_mode](block);
820
821 dv_set_class_number(block, enc_blk,
822 enc_blk->dct_mode ? ff_zigzag248_direct : ff_zigzag_direct,
823 j/4*(j%2));
824
825 init_put_bits(pb, ptr, block_sizes[j]/8);
826 put_bits(pb, 9, (uint16_t)(((enc_blk->mb[0] >> 3) - 1024) >> 2));
827 put_bits(pb, 1, enc_blk->dct_mode);
828 put_bits(pb, 2, enc_blk->cno);
829
830 vs_bit_size += enc_blk->bit_size[0] + enc_blk->bit_size[1] +
831 enc_blk->bit_size[2] + enc_blk->bit_size[3];
832 ++enc_blk;
833 ++pb;
834 ptr += block_sizes[j]/8;
835 }
836 }
837
838 if (vs_total_ac_bits < vs_bit_size)
839 dv_guess_qnos(&enc_blks[0], &qnos[0]);
840
841 for (i=0; i<5; i++) {
842 dif[i*80 + 3] = qnos[i];
843 }
844
845 /* First pass over individual cells only */
846 for (j=0; j<5*6; j++)
847 dv_encode_ac(&enc_blks[j], &pbs[j], 1);
848
849 /* Second pass over each MB space */
850 for (j=0; j<5*6; j++) {
851 if (enc_blks[j].cur_ac < 65 || enc_blks[j].partial_bit_count)
852 dv_encode_ac(&enc_blks[j], &pbs[(j/6)*6], 6);
853 }
854
855 /* Third and final pass over the whole vides segment space */
856 for (j=0; j<5*6; j++) {
857 if (enc_blks[j].cur_ac < 65 || enc_blks[j].partial_bit_count)
858 dv_encode_ac(&enc_blks[j], &pbs[0], 6*5);
859 }
860
861 for (j=0; j<5*6; j++)
862 flush_put_bits(&pbs[j]);
863 }
864
865 static int dv_decode_mt(AVCodecContext *avctx, void* sl)
866 {
867 DVVideoContext *s = avctx->priv_data;
868 int slice = (size_t)sl;
869 dv_decode_video_segment(s, &s->buf[((slice/27)*6+(slice/3)+slice*5+7)*80],
870 &s->sys->video_place[slice*5]);
871 return 0;
872 }
873
874 static int dv_encode_mt(AVCodecContext *avctx, void* sl)
875 {
876 DVVideoContext *s = avctx->priv_data;
877 int slice = (size_t)sl;
878 dv_encode_video_segment(s, &s->buf[((slice/27)*6+(slice/3)+slice*5+7)*80],
879 &s->sys->video_place[slice*5]);
880 return 0;
881 }
882
883 /* NOTE: exactly one frame must be given (120000 bytes for NTSC,
884 144000 bytes for PAL) */
885 static int dvvideo_decode_frame(AVCodecContext *avctx,
886 void *data, int *data_size,
887 uint8_t *buf, int buf_size)
888 {
889 DVVideoContext *s = avctx->priv_data;
890
891 /* special case for last picture */
892 if(buf_size==0)
893 return 0;
894
895 s->sys = dv_frame_profile(buf);
896 if (!s->sys || buf_size < s->sys->frame_size)
897 return -1; /* NOTE: we only accept several full frames */
898
899 if(s->picture.data[0])
900 avctx->release_buffer(avctx, &s->picture);
901
902 s->picture.reference = 0;
903 avctx->pix_fmt = s->sys->pix_fmt;
904 avctx->width = s->sys->width;
905 avctx->height = s->sys->height;
906 if(avctx->get_buffer(avctx, &s->picture) < 0) {
907 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
908 return -1;
909 }
910 s->picture.interlaced_frame = 1;
911 s->picture.top_field_first = 0;
912
913 s->buf = buf;
914 avctx->execute(avctx, dv_decode_mt, (void**)&dv_anchor[0], NULL,
915 s->sys->difseg_size * 27);
916
917 emms_c();
918
919 /* return image */
920 *data_size = sizeof(AVFrame);
921 *(AVFrame*)data= s->picture;
922
923 return s->sys->frame_size;
924 }
925
926 static int dvvideo_encode_frame(AVCodecContext *c, uint8_t *buf, int buf_size,
927 void *data)
928 {
929 DVVideoContext *s = c->priv_data;
930
931 s->sys = dv_codec_profile(c);
932 if (!s->sys)
933 return -1;
934
935 c->pix_fmt = s->sys->pix_fmt;
936 s->picture = *((AVFrame *)data);
937
938 s->buf = buf;
939 c->execute(c, dv_encode_mt, (void**)&dv_anchor[0], NULL,
940 s->sys->difseg_size * 27);
941
942 emms_c();
943 return s->sys->frame_size;
944 }
945
946 AVCodec dvvideo_encoder = {
947 "dvvideo",
948 CODEC_TYPE_VIDEO,
949 CODEC_ID_DVVIDEO,
950 sizeof(DVVideoContext),
951 dvvideo_init,
952 dvvideo_encode_frame,
953 NULL,
954 NULL,
955 CODEC_CAP_DR1,
956 NULL
957 };
958
959 AVCodec dvvideo_decoder = {
960 "dvvideo",
961 CODEC_TYPE_VIDEO,
962 CODEC_ID_DVVIDEO,
963 sizeof(DVVideoContext),
964 dvvideo_init,
965 NULL,
966 NULL,
967 dvvideo_decode_frame,
968 CODEC_CAP_DR1,
969 NULL
970 };