3 * Copyright (c) 2002 Fabrice Bellard.
4 * Copyright (c) 2004 Roman Shaposhnik.
7 * Copyright (c) 2003 Roman Shaposhnik.
9 * Many thanks to Dan Dennedy <dan@dennedy.org> for providing wealth
10 * of DV technical info.
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.
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.
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., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
33 #include "mpegvideo.h"
34 #include "simple_idct.h"
40 typedef struct DVVideoContext {
43 AVCodecContext *avctx;
46 uint8_t dv_zigzag[2][64];
47 uint8_t dv_idct_shift[2][2][22][64];
49 void (*get_pixels)(DCTELEM *block, const uint8_t *pixels, int line_size);
50 void (*fdct[2])(DCTELEM *block);
51 void (*idct_put[2])(uint8_t *dest, int line_size, DCTELEM *block);
54 /* MultiThreading - applies to entire DV codec, not just the avcontext */
57 #define TEX_VLC_BITS 9
59 #ifdef DV_CODEC_TINY_TARGET
60 #define DV_VLC_MAP_RUN_SIZE 15
61 #define DV_VLC_MAP_LEV_SIZE 23
63 #define DV_VLC_MAP_RUN_SIZE 64
64 #define DV_VLC_MAP_LEV_SIZE 512 //FIXME sign was removed so this should be /2 but needs check
67 /* XXX: also include quantization */
68 static RL_VLC_ELEM *dv_rl_vlc;
69 /* VLC encoding lookup table */
70 static struct dv_vlc_pair {
73 } (*dv_vlc_map)[DV_VLC_MAP_LEV_SIZE] = NULL;
75 static void dv_build_unquantize_tables(DVVideoContext *s, uint8_t* perm)
79 /* NOTE: max left shift is 6 */
80 for(q = 0; q < 22; q++) {
82 for(i = 1; i < 64; i++) {
85 s->dv_idct_shift[0][0][q][j] =
86 dv_quant_shifts[q][dv_88_areas[i]] + 1;
87 s->dv_idct_shift[1][0][q][j] = s->dv_idct_shift[0][0][q][j] + 1;
91 for(i = 1; i < 64; i++) {
93 s->dv_idct_shift[0][1][q][i] =
94 dv_quant_shifts[q][dv_248_areas[i]] + 1;
95 s->dv_idct_shift[1][1][q][i] = s->dv_idct_shift[0][1][q][i] + 1;
100 static int dvvideo_init(AVCodecContext *avctx)
102 DVVideoContext *s = avctx->priv_data;
109 uint16_t new_dv_vlc_bits[NB_DV_VLC*2];
110 uint8_t new_dv_vlc_len[NB_DV_VLC*2];
111 uint8_t new_dv_vlc_run[NB_DV_VLC*2];
112 int16_t new_dv_vlc_level[NB_DV_VLC*2];
116 dv_vlc_map = av_mallocz_static(DV_VLC_MAP_LEV_SIZE*DV_VLC_MAP_RUN_SIZE*sizeof(struct dv_vlc_pair));
120 /* dv_anchor lets each thread know its Id */
121 dv_anchor = av_malloc(12*27*sizeof(void*));
125 for (i=0; i<12*27; i++)
126 dv_anchor[i] = (void*)(size_t)i;
128 /* it's faster to include sign bit in a generic VLC parsing scheme */
129 for (i=0, j=0; i<NB_DV_VLC; i++, j++) {
130 new_dv_vlc_bits[j] = dv_vlc_bits[i];
131 new_dv_vlc_len[j] = dv_vlc_len[i];
132 new_dv_vlc_run[j] = dv_vlc_run[i];
133 new_dv_vlc_level[j] = dv_vlc_level[i];
135 if (dv_vlc_level[i]) {
136 new_dv_vlc_bits[j] <<= 1;
140 new_dv_vlc_bits[j] = (dv_vlc_bits[i] << 1) | 1;
141 new_dv_vlc_len[j] = dv_vlc_len[i] + 1;
142 new_dv_vlc_run[j] = dv_vlc_run[i];
143 new_dv_vlc_level[j] = -dv_vlc_level[i];
147 /* NOTE: as a trick, we use the fact the no codes are unused
148 to accelerate the parsing of partial codes */
149 init_vlc(&dv_vlc, TEX_VLC_BITS, j,
150 new_dv_vlc_len, 1, 1, new_dv_vlc_bits, 2, 2, 0);
152 dv_rl_vlc = av_mallocz_static(dv_vlc.table_size * sizeof(RL_VLC_ELEM));
156 for(i = 0; i < dv_vlc.table_size; i++){
157 int code= dv_vlc.table[i][0];
158 int len = dv_vlc.table[i][1];
161 if(len<0){ //more bits needed
165 run= new_dv_vlc_run[code] + 1;
166 level= new_dv_vlc_level[code];
168 dv_rl_vlc[i].len = len;
169 dv_rl_vlc[i].level = level;
170 dv_rl_vlc[i].run = run;
174 for (i = 0; i < NB_DV_VLC - 1; i++) {
175 if (dv_vlc_run[i] >= DV_VLC_MAP_RUN_SIZE)
177 #ifdef DV_CODEC_TINY_TARGET
178 if (dv_vlc_level[i] >= DV_VLC_MAP_LEV_SIZE)
182 if (dv_vlc_map[dv_vlc_run[i]][dv_vlc_level[i]].size != 0)
185 dv_vlc_map[dv_vlc_run[i]][dv_vlc_level[i]].vlc = dv_vlc_bits[i] <<
187 dv_vlc_map[dv_vlc_run[i]][dv_vlc_level[i]].size = dv_vlc_len[i] +
190 for (i = 0; i < DV_VLC_MAP_RUN_SIZE; i++) {
191 #ifdef DV_CODEC_TINY_TARGET
192 for (j = 1; j < DV_VLC_MAP_LEV_SIZE; j++) {
193 if (dv_vlc_map[i][j].size == 0) {
194 dv_vlc_map[i][j].vlc = dv_vlc_map[0][j].vlc |
195 (dv_vlc_map[i-1][0].vlc << (dv_vlc_map[0][j].size));
196 dv_vlc_map[i][j].size = dv_vlc_map[i-1][0].size +
197 dv_vlc_map[0][j].size;
201 for (j = 1; j < DV_VLC_MAP_LEV_SIZE/2; j++) {
202 if (dv_vlc_map[i][j].size == 0) {
203 dv_vlc_map[i][j].vlc = dv_vlc_map[0][j].vlc |
204 (dv_vlc_map[i-1][0].vlc << (dv_vlc_map[0][j].size));
205 dv_vlc_map[i][j].size = dv_vlc_map[i-1][0].size +
206 dv_vlc_map[0][j].size;
208 dv_vlc_map[i][((uint16_t)(-j))&0x1ff].vlc =
209 dv_vlc_map[i][j].vlc | 1;
210 dv_vlc_map[i][((uint16_t)(-j))&0x1ff].size =
211 dv_vlc_map[i][j].size;
217 /* Generic DSP setup */
218 dsputil_init(&dsp, avctx);
219 s->get_pixels = dsp.get_pixels;
222 s->fdct[0] = dsp.fdct;
223 s->idct_put[0] = dsp.idct_put;
225 s->dv_zigzag[0][i] = dsp.idct_permutation[ff_zigzag_direct[i]];
228 s->fdct[1] = dsp.fdct248;
229 s->idct_put[1] = simple_idct248_put; // FIXME: need to add it to DSP
231 for (i=0; i<64; i++){
232 int j= ff_zigzag248_direct[i];
233 s->dv_zigzag[1][i] = dsp.idct_permutation[(j&7) + (j&8)*4 + (j&48)/2];
236 memcpy(s->dv_zigzag[1], ff_zigzag248_direct, 64);
238 /* XXX: do it only for constant case */
239 dv_build_unquantize_tables(s, dsp.idct_permutation);
241 /* FIXME: I really don't think this should be here */
242 if (dv_codec_profile(avctx))
243 avctx->pix_fmt = dv_codec_profile(avctx)->pix_fmt;
244 avctx->coded_frame = &s->picture;
251 // #define printf(...) av_log(NULL, AV_LOG_ERROR, __VA_ARGS__)
253 typedef struct BlockInfo {
254 const uint8_t *shift_table;
255 const uint8_t *scan_table;
256 uint8_t pos; /* position in block */
258 uint8_t partial_bit_count;
259 uint16_t partial_bit_buffer;
263 /* block size in bits */
264 static const uint16_t block_sizes[6] = {
265 112, 112, 112, 112, 80, 80
267 /* bit budget for AC only in 5 MBs */
268 static const int vs_total_ac_bits = (100 * 4 + 68*2) * 5;
269 /* see dv_88_areas and dv_248_areas for details */
270 static const int mb_area_start[5] = { 1, 6, 21, 43, 64 };
272 #ifndef ALT_BITSTREAM_READER
273 #warning only works with ALT_BITSTREAM_READER
274 static int re_index; //Hack to make it compile
277 static inline int get_bits_left(GetBitContext *s)
279 return s->size_in_bits - get_bits_count(s);
282 static inline int get_bits_size(GetBitContext *s)
284 return s->size_in_bits;
287 static inline int put_bits_left(PutBitContext* s)
289 return (s->buf_end - s->buf) * 8 - put_bits_count(s);
292 /* decode ac coefs */
293 static void dv_decode_ac(GetBitContext *gb, BlockInfo *mb, DCTELEM *block)
295 int last_index = get_bits_size(gb);
296 const uint8_t *scan_table = mb->scan_table;
297 const uint8_t *shift_table = mb->shift_table;
299 int partial_bit_count = mb->partial_bit_count;
300 int level, pos1, run, vlc_len, index;
303 UPDATE_CACHE(re, gb);
305 /* if we must parse a partial vlc, we do it here */
306 if (partial_bit_count > 0) {
307 re_cache = ((unsigned)re_cache >> partial_bit_count) |
308 (mb->partial_bit_buffer << (sizeof(re_cache)*8 - partial_bit_count));
309 re_index -= partial_bit_count;
310 mb->partial_bit_count = 0;
313 /* get the AC coefficients until last_index is reached */
316 printf("%2d: bits=%04x index=%d\n", pos, SHOW_UBITS(re, gb, 16), re_index);
318 /* our own optimized GET_RL_VLC */
319 index = NEG_USR32(re_cache, TEX_VLC_BITS);
320 vlc_len = dv_rl_vlc[index].len;
322 index = NEG_USR32((unsigned)re_cache << TEX_VLC_BITS, -vlc_len) + dv_rl_vlc[index].level;
323 vlc_len = TEX_VLC_BITS - vlc_len;
325 level = dv_rl_vlc[index].level;
326 run = dv_rl_vlc[index].run;
328 /* gotta check if we're still within gb boundaries */
329 if (re_index + vlc_len > last_index) {
330 /* should be < 16 bits otherwise a codeword could have been parsed */
331 mb->partial_bit_count = last_index - re_index;
332 mb->partial_bit_buffer = NEG_USR32(re_cache, mb->partial_bit_count);
333 re_index = last_index;
339 printf("run=%d level=%d\n", run, level);
345 pos1 = scan_table[pos];
346 block[pos1] = level << shift_table[pos1];
348 UPDATE_CACHE(re, gb);
350 CLOSE_READER(re, gb);
354 static inline void bit_copy(PutBitContext *pb, GetBitContext *gb)
356 int bits_left = get_bits_left(gb);
357 while (bits_left >= MIN_CACHE_BITS) {
358 put_bits(pb, MIN_CACHE_BITS, get_bits(gb, MIN_CACHE_BITS));
359 bits_left -= MIN_CACHE_BITS;
362 put_bits(pb, bits_left, get_bits(gb, bits_left));
366 /* mb_x and mb_y are in units of 8 pixels */
367 static inline void dv_decode_video_segment(DVVideoContext *s,
369 const uint16_t *mb_pos_ptr)
371 int quant, dc, dct_mode, class1, j;
372 int mb_index, mb_x, mb_y, v, last_index;
373 DCTELEM *block, *block1;
376 void (*idct_put)(uint8_t *dest, int line_size, DCTELEM *block);
378 PutBitContext pb, vs_pb;
380 BlockInfo mb_data[5 * 6], *mb, *mb1;
381 DECLARE_ALIGNED_8(DCTELEM, sblock[5*6][64]);
382 DECLARE_ALIGNED_8(uint8_t, mb_bit_buffer[80 + 4]); /* allow some slack */
383 DECLARE_ALIGNED_8(uint8_t, vs_bit_buffer[5 * 80 + 4]); /* allow some slack */
384 const int log2_blocksize= 3-s->avctx->lowres;
386 assert((((int)mb_bit_buffer)&7)==0);
387 assert((((int)vs_bit_buffer)&7)==0);
389 memset(sblock, 0, sizeof(sblock));
391 /* pass 1 : read DC and AC coefficients in blocks */
393 block1 = &sblock[0][0];
395 init_put_bits(&vs_pb, vs_bit_buffer, 5 * 80);
396 for(mb_index = 0; mb_index < 5; mb_index++, mb1 += 6, block1 += 6 * 64) {
398 quant = buf_ptr[3] & 0x0f;
400 init_put_bits(&pb, mb_bit_buffer, 80);
403 for(j = 0;j < 6; j++) {
404 last_index = block_sizes[j];
405 init_get_bits(&gb, buf_ptr, last_index);
408 dc = get_sbits(&gb, 9);
409 dct_mode = get_bits1(&gb);
410 mb->dct_mode = dct_mode;
411 mb->scan_table = s->dv_zigzag[dct_mode];
412 class1 = get_bits(&gb, 2);
413 mb->shift_table = s->dv_idct_shift[class1 == 3][dct_mode]
414 [quant + dv_quant_offset[class1]];
416 /* convert to unsigned because 128 is not added in the
420 buf_ptr += last_index >> 3;
422 mb->partial_bit_count = 0;
425 printf("MB block: %d, %d ", mb_index, j);
427 dv_decode_ac(&gb, mb, block);
429 /* write the remaining bits in a new buffer only if the
438 /* pass 2 : we can do it just after */
440 printf("***pass 2 size=%d MB#=%d\n", put_bits_count(&pb), mb_index);
444 init_get_bits(&gb, mb_bit_buffer, put_bits_count(&pb));
446 for(j = 0;j < 6; j++, block += 64, mb++) {
447 if (mb->pos < 64 && get_bits_left(&gb) > 0) {
448 dv_decode_ac(&gb, mb, block);
449 /* if still not finished, no need to parse other blocks */
454 /* all blocks are finished, so the extra bytes can be used at
455 the video segment level */
457 bit_copy(&vs_pb, &gb);
460 /* we need a pass other the whole video segment */
462 printf("***pass 3 size=%d\n", put_bits_count(&vs_pb));
464 block = &sblock[0][0];
466 init_get_bits(&gb, vs_bit_buffer, put_bits_count(&vs_pb));
467 flush_put_bits(&vs_pb);
468 for(mb_index = 0; mb_index < 5; mb_index++) {
469 for(j = 0;j < 6; j++) {
472 printf("start %d:%d\n", mb_index, j);
474 dv_decode_ac(&gb, mb, block);
476 if (mb->pos >= 64 && mb->pos < 127)
477 av_log(NULL, AV_LOG_ERROR, "AC EOB marker is absent pos=%d\n", mb->pos);
483 /* compute idct and place blocks */
484 block = &sblock[0][0];
486 for(mb_index = 0; mb_index < 5; mb_index++) {
490 y_ptr = s->picture.data[0] + ((mb_y * s->picture.linesize[0] + mb_x)<<log2_blocksize);
491 if (s->sys->pix_fmt == PIX_FMT_YUV411P)
492 c_offset = ((mb_y * s->picture.linesize[1] + (mb_x >> 2))<<log2_blocksize);
494 c_offset = (((mb_y >> 1) * s->picture.linesize[1] + (mb_x >> 1))<<log2_blocksize);
495 for(j = 0;j < 6; j++) {
496 idct_put = s->idct_put[mb->dct_mode && log2_blocksize==3];
498 if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x < (704 / 8)) {
499 /* NOTE: at end of line, the macroblock is handled as 420 */
500 idct_put(y_ptr + (j<<log2_blocksize), s->picture.linesize[0], block);
502 idct_put(y_ptr + (((j & 1) + (j >> 1) * s->picture.linesize[0])<<log2_blocksize),
503 s->picture.linesize[0], block);
506 if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x >= (704 / 8)) {
507 uint64_t aligned_pixels[64/8];
508 uint8_t *pixels= (uint8_t*)aligned_pixels;
509 uint8_t *c_ptr, *c_ptr1, *ptr, *ptr1;
511 /* NOTE: at end of line, the macroblock is handled as 420 */
512 idct_put(pixels, 8, block);
513 linesize = s->picture.linesize[6 - j];
514 c_ptr = s->picture.data[6 - j] + c_offset;
516 for(y = 0;y < (1<<log2_blocksize); y++) {
517 ptr1= ptr + (1<<(log2_blocksize-1));
518 c_ptr1 = c_ptr + (linesize<<log2_blocksize);
519 for(x=0; x < (1<<(log2_blocksize-1)); x++){
520 c_ptr[x]= ptr[x]; c_ptr1[x]= ptr1[x];
526 /* don't ask me why they inverted Cb and Cr ! */
527 idct_put(s->picture.data[6 - j] + c_offset,
528 s->picture.linesize[6 - j], block);
537 #ifdef DV_CODEC_TINY_TARGET
538 /* Converts run and level (where level != 0) pair into vlc, returning bit size */
539 static always_inline int dv_rl2vlc(int run, int level, int sign, uint32_t* vlc)
542 if (run < DV_VLC_MAP_RUN_SIZE && level < DV_VLC_MAP_LEV_SIZE) {
543 *vlc = dv_vlc_map[run][level].vlc | sign;
544 size = dv_vlc_map[run][level].size;
547 if (level < DV_VLC_MAP_LEV_SIZE) {
548 *vlc = dv_vlc_map[0][level].vlc | sign;
549 size = dv_vlc_map[0][level].size;
551 *vlc = 0xfe00 | (level << 1) | sign;
555 *vlc |= ((run < 16) ? dv_vlc_map[run-1][0].vlc :
556 (0x1f80 | (run - 1))) << size;
557 size += (run < 16) ? dv_vlc_map[run-1][0].size : 13;
564 static always_inline int dv_rl2vlc_size(int run, int level)
568 if (run < DV_VLC_MAP_RUN_SIZE && level < DV_VLC_MAP_LEV_SIZE) {
569 size = dv_vlc_map[run][level].size;
572 size = (level < DV_VLC_MAP_LEV_SIZE) ? dv_vlc_map[0][level].size : 16;
574 size += (run < 16) ? dv_vlc_map[run-1][0].size : 13;
580 static always_inline int dv_rl2vlc(int run, int l, int sign, uint32_t* vlc)
582 *vlc = dv_vlc_map[run][l].vlc | sign;
583 return dv_vlc_map[run][l].size;
586 static always_inline int dv_rl2vlc_size(int run, int l)
588 return dv_vlc_map[run][l].size;
592 typedef struct EncBlockInfo {
602 uint8_t partial_bit_count;
603 uint32_t partial_bit_buffer; /* we can't use uint16_t here */
606 static always_inline PutBitContext* dv_encode_ac(EncBlockInfo* bi, PutBitContext* pb_pool,
607 PutBitContext* pb_end)
611 PutBitContext* pb = pb_pool;
612 int size = bi->partial_bit_count;
613 uint32_t vlc = bi->partial_bit_buffer;
615 bi->partial_bit_count = bi->partial_bit_buffer = 0;
617 /* Find suitable storage space */
618 for (; size > (bits_left = put_bits_left(pb)); pb++) {
621 put_bits(pb, bits_left, vlc >> size);
622 vlc = vlc & ((1<<size)-1);
624 if (pb + 1 >= pb_end) {
625 bi->partial_bit_count = size;
626 bi->partial_bit_buffer = vlc;
632 put_bits(pb, size, vlc);
637 /* Construct the next VLC */
639 bi->cur_ac = bi->next[prev];
641 size = dv_rl2vlc(bi->cur_ac - prev - 1, bi->mb[bi->cur_ac], bi->sign[bi->cur_ac], &vlc);
643 size = 4; vlc = 6; /* End Of Block stamp */
649 static always_inline void dv_set_class_number(DCTELEM* blk, EncBlockInfo* bi,
650 const uint8_t* zigzag_scan, int bias)
653 static const int classes[] = {12, 24, 36, 0xffff};
659 for (area = 0; area < 4; area++) {
660 bi->prev[area] = prev;
661 bi->bit_size[area] = 1; // 4 areas 4 bits for EOB :)
662 for (i=mb_area_start[area]; i<mb_area_start[area+1]; i++) {
663 int level = blk[zigzag_scan[i]];
665 if (level+15 > 30U) {
666 bi->sign[i] = (level>>31)&1;
667 bi->mb[i] = level= ABS(level)>>4;
668 if(level>max) max= level;
669 bi->bit_size[area] += dv_rl2vlc_size(i - prev - 1, level);
676 for(bi->cno = 0; max > classes[bi->cno]; bi->cno++);
684 for (area = 0; area < 4; area++) {
685 bi->prev[area] = prev;
686 bi->bit_size[area] = 1; // 4 areas 4 bits for EOB :)
687 for (; i<mb_area_start[area+1]; i= bi->next[i]) {
691 bi->bit_size[area] += dv_rl2vlc_size(i - prev - 1, bi->mb[i]);
701 //FIXME replace this by dsputil
702 #define SC(x, y) ((s[x] - s[y]) ^ ((s[x] - s[y]) >> 7))
703 static always_inline int dv_guess_dct_mode(DCTELEM *blk) {
709 /* Compute 8-8 score (small values give a better chance for 8-8 DCT) */
712 score88 += SC(0, 8) + SC(1, 9) + SC(2, 10) + SC(3, 11) +
713 SC(4, 12) + SC(5,13) + SC(6, 14) + SC(7, 15);
716 /* Compute 2-4-8 score (small values give a better chance for 2-4-8 DCT) */
719 score248 += SC(0, 16) + SC(1,17) + SC(2, 18) + SC(3, 19) +
720 SC(4, 20) + SC(5,21) + SC(6, 22) + SC(7, 23);
724 return (score88 - score248 > -10);
727 static inline void dv_guess_qnos(EncBlockInfo* blks, int* qnos)
730 int i, j, k, a, prev, a2;
733 size[0] = size[1] = size[2] = size[3] = size[4] = 1<<24;
736 for (i=0; i<5; i++) {
742 for (j=0; j<6; j++, b++) {
743 for (a=0; a<4; a++) {
744 if (b->area_q[a] != dv_quant_shifts[qnos[i] + dv_quant_offset[b->cno]][a]) {
745 b->bit_size[a] = 1; // 4 areas 4 bits for EOB :)
748 assert(b->next[prev] >= mb_area_start[a+1] || b->mb[prev]);
749 for (k= b->next[prev] ; k<mb_area_start[a+1]; k= b->next[k]) {
752 b->bit_size[a] += dv_rl2vlc_size(k - prev - 1, b->mb[k]);
755 if(b->next[k] >= mb_area_start[a+1] && b->next[k]<64){
756 for(a2=a+1; b->next[k] >= mb_area_start[a2+1]; a2++)
759 assert(b->mb[b->next[k]]);
760 b->bit_size[a2] += dv_rl2vlc_size(b->next[k] - prev - 1, b->mb[b->next[k]])
761 -dv_rl2vlc_size(b->next[k] - k - 1, b->mb[b->next[k]]);
762 assert(b->prev[a2]==k && (a2+1 >= 4 || b->prev[a2+1]!=k));
765 b->next[prev] = b->next[k];
770 size[i] += b->bit_size[a];
773 if(vs_total_ac_bits >= size[0] + size[1] + size[2] + size[3] + size[4])
776 } while (qnos[0]|qnos[1]|qnos[2]|qnos[3]|qnos[4]);
779 for(a=2; a==2 || vs_total_ac_bits < size[0]; a+=a){
781 size[0] = 5*6*4; //EOB
782 for (j=0; j<6*5; j++, b++) {
784 for (k= b->next[prev]; k<64; k= b->next[k]) {
785 if(b->mb[k] < a && b->mb[k] > -a){
786 b->next[prev] = b->next[k];
788 size[0] += dv_rl2vlc_size(k - prev - 1, b->mb[k]);
797 * This is a very rough initial implementaion. The performance is
798 * horrible and the weighting is missing. But it's missing from the
799 * decoding step also -- so at least we're on the same page with decoder ;-)
801 static inline void dv_encode_video_segment(DVVideoContext *s,
803 const uint16_t *mb_pos_ptr)
805 int mb_index, i, j, v;
806 int mb_x, mb_y, c_offset, linesize;
811 DECLARE_ALIGNED_8(DCTELEM, block[64]);
812 EncBlockInfo enc_blks[5*6];
813 PutBitContext pbs[5*6];
815 EncBlockInfo* enc_blk;
819 assert((((int)block) & 7) == 0);
821 enc_blk = &enc_blks[0];
823 for(mb_index = 0; mb_index < 5; mb_index++) {
827 y_ptr = s->picture.data[0] + (mb_y * s->picture.linesize[0] * 8) + (mb_x * 8);
828 c_offset = (s->sys->pix_fmt == PIX_FMT_YUV411P) ?
829 ((mb_y * s->picture.linesize[1] * 8) + ((mb_x >> 2) * 8)) :
830 (((mb_y >> 1) * s->picture.linesize[1] * 8) + ((mb_x >> 1) * 8));
832 qnos[mb_index] = 15; /* No quantization */
833 ptr = dif + mb_index*80 + 4;
834 for(j = 0;j < 6; j++) {
835 if (j < 4) { /* Four Y blocks */
836 /* NOTE: at end of line, the macroblock is handled as 420 */
837 if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x < (704 / 8)) {
838 data = y_ptr + (j * 8);
840 data = y_ptr + ((j & 1) * 8) + ((j >> 1) * 8 * s->picture.linesize[0]);
842 linesize = s->picture.linesize[0];
843 } else { /* Cr and Cb blocks */
844 /* don't ask Fabrice why they inverted Cb and Cr ! */
845 data = s->picture.data[6 - j] + c_offset;
846 linesize = s->picture.linesize[6 - j];
847 if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x >= (704 / 8))
851 /* Everything is set up -- now just copy data -> DCT block */
852 if (do_edge_wrap) { /* Edge wrap copy: 4x16 -> 8x8 */
856 d = data + 8 * linesize;
857 b[0] = data[0]; b[1] = data[1]; b[2] = data[2]; b[3] = data[3];
858 b[4] = d[0]; b[5] = d[1]; b[6] = d[2]; b[7] = d[3];
862 } else { /* Simple copy: 8x8 -> 8x8 */
863 s->get_pixels(block, data, linesize);
866 if(s->avctx->flags & CODEC_FLAG_INTERLACED_DCT)
867 enc_blk->dct_mode = dv_guess_dct_mode(block);
869 enc_blk->dct_mode = 0;
870 enc_blk->area_q[0] = enc_blk->area_q[1] = enc_blk->area_q[2] = enc_blk->area_q[3] = 0;
871 enc_blk->partial_bit_count = 0;
872 enc_blk->partial_bit_buffer = 0;
875 s->fdct[enc_blk->dct_mode](block);
877 dv_set_class_number(block, enc_blk,
878 enc_blk->dct_mode ? ff_zigzag248_direct : ff_zigzag_direct, j/4);
880 init_put_bits(pb, ptr, block_sizes[j]/8);
881 put_bits(pb, 9, (uint16_t)(((enc_blk->mb[0] >> 3) - 1024 + 2) >> 2));
882 put_bits(pb, 1, enc_blk->dct_mode);
883 put_bits(pb, 2, enc_blk->cno);
885 vs_bit_size += enc_blk->bit_size[0] + enc_blk->bit_size[1] +
886 enc_blk->bit_size[2] + enc_blk->bit_size[3];
889 ptr += block_sizes[j]/8;
893 if (vs_total_ac_bits < vs_bit_size)
894 dv_guess_qnos(&enc_blks[0], &qnos[0]);
896 for (i=0; i<5; i++) {
897 dif[i*80 + 3] = qnos[i];
900 /* First pass over individual cells only */
901 for (j=0; j<5*6; j++)
902 dv_encode_ac(&enc_blks[j], &pbs[j], &pbs[j+1]);
904 /* Second pass over each MB space */
905 for (j=0; j<5*6; j+=6) {
907 for (i=0; i<6; i++) {
908 if (enc_blks[i+j].partial_bit_count)
909 pb=dv_encode_ac(&enc_blks[i+j], pb, &pbs[j+6]);
913 /* Third and final pass over the whole vides segment space */
915 for (j=0; j<5*6; j++) {
916 if (enc_blks[j].partial_bit_count)
917 pb=dv_encode_ac(&enc_blks[j], pb, &pbs[6*5]);
918 if (enc_blks[j].partial_bit_count)
919 av_log(NULL, AV_LOG_ERROR, "ac bitstream overflow\n");
922 for (j=0; j<5*6; j++)
923 flush_put_bits(&pbs[j]);
926 static int dv_decode_mt(AVCodecContext *avctx, void* sl)
928 DVVideoContext *s = avctx->priv_data;
929 int slice = (size_t)sl;
930 dv_decode_video_segment(s, &s->buf[((slice/27)*6+(slice/3)+slice*5+7)*80],
931 &s->sys->video_place[slice*5]);
935 static int dv_encode_mt(AVCodecContext *avctx, void* sl)
937 DVVideoContext *s = avctx->priv_data;
938 int slice = (size_t)sl;
939 dv_encode_video_segment(s, &s->buf[((slice/27)*6+(slice/3)+slice*5+7)*80],
940 &s->sys->video_place[slice*5]);
944 /* NOTE: exactly one frame must be given (120000 bytes for NTSC,
945 144000 bytes for PAL) */
946 static int dvvideo_decode_frame(AVCodecContext *avctx,
947 void *data, int *data_size,
948 uint8_t *buf, int buf_size)
950 DVVideoContext *s = avctx->priv_data;
952 s->sys = dv_frame_profile(buf);
953 if (!s->sys || buf_size < s->sys->frame_size)
954 return -1; /* NOTE: we only accept several full frames */
956 if(s->picture.data[0])
957 avctx->release_buffer(avctx, &s->picture);
959 s->picture.reference = 0;
960 s->picture.key_frame = 1;
961 s->picture.pict_type = FF_I_TYPE;
962 avctx->pix_fmt = s->sys->pix_fmt;
963 avcodec_set_dimensions(avctx, s->sys->width, s->sys->height);
964 if(avctx->get_buffer(avctx, &s->picture) < 0) {
965 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
968 s->picture.interlaced_frame = 1;
969 s->picture.top_field_first = 0;
972 avctx->execute(avctx, dv_decode_mt, (void**)&dv_anchor[0], NULL,
973 s->sys->difseg_size * 27);
978 *data_size = sizeof(AVFrame);
979 *(AVFrame*)data= s->picture;
981 return s->sys->frame_size;
984 static int dvvideo_encode_frame(AVCodecContext *c, uint8_t *buf, int buf_size,
987 DVVideoContext *s = c->priv_data;
989 s->sys = dv_codec_profile(c);
992 if(buf_size < s->sys->frame_size)
995 c->pix_fmt = s->sys->pix_fmt;
996 s->picture = *((AVFrame *)data);
997 s->picture.key_frame = 1;
998 s->picture.pict_type = FF_I_TYPE;
1001 c->execute(c, dv_encode_mt, (void**)&dv_anchor[0], NULL,
1002 s->sys->difseg_size * 27);
1005 return s->sys->frame_size;
1008 static int dvvideo_close(AVCodecContext *c)
1015 #ifdef CONFIG_DVVIDEO_ENCODER
1016 AVCodec dvvideo_encoder = {
1020 sizeof(DVVideoContext),
1022 dvvideo_encode_frame,
1028 #endif // CONFIG_DVVIDEO_ENCODER
1030 AVCodec dvvideo_decoder = {
1034 sizeof(DVVideoContext),
1038 dvvideo_decode_frame,