2 * H.26L/H.264/AVC/JVT/14496-10/... cavlc bitstream decoding
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * H.264 / AVC / MPEG4 part10 cavlc bitstream decoding.
25 * @author Michael Niedermayer <michaelni@gmx.at>
32 #include "mpegvideo.h"
34 #include "h264data.h" // FIXME FIXME FIXME
35 #include "h264_mvpred.h"
41 static const uint8_t golomb_to_inter_cbp_gray[16]={
42 0, 1, 2, 4, 8, 3, 5,10,12,15, 7,11,13,14, 6, 9,
45 static const uint8_t golomb_to_intra4x4_cbp_gray[16]={
46 15, 0, 7,11,13,14, 3, 5,10,12, 1, 2, 4, 8, 6, 9,
49 static const uint8_t chroma_dc_coeff_token_len[4*5]={
57 static const uint8_t chroma_dc_coeff_token_bits[4*5]={
65 static const uint8_t coeff_token_len[4][4*17]={
68 6, 2, 0, 0, 8, 6, 3, 0, 9, 8, 7, 5, 10, 9, 8, 6,
69 11,10, 9, 7, 13,11,10, 8, 13,13,11, 9, 13,13,13,10,
70 14,14,13,11, 14,14,14,13, 15,15,14,14, 15,15,15,14,
71 16,15,15,15, 16,16,16,15, 16,16,16,16, 16,16,16,16,
75 6, 2, 0, 0, 6, 5, 3, 0, 7, 6, 6, 4, 8, 6, 6, 4,
76 8, 7, 7, 5, 9, 8, 8, 6, 11, 9, 9, 6, 11,11,11, 7,
77 12,11,11, 9, 12,12,12,11, 12,12,12,11, 13,13,13,12,
78 13,13,13,13, 13,14,13,13, 14,14,14,13, 14,14,14,14,
82 6, 4, 0, 0, 6, 5, 4, 0, 6, 5, 5, 4, 7, 5, 5, 4,
83 7, 5, 5, 4, 7, 6, 6, 4, 7, 6, 6, 4, 8, 7, 7, 5,
84 8, 8, 7, 6, 9, 8, 8, 7, 9, 9, 8, 8, 9, 9, 9, 8,
85 10, 9, 9, 9, 10,10,10,10, 10,10,10,10, 10,10,10,10,
89 6, 6, 0, 0, 6, 6, 6, 0, 6, 6, 6, 6, 6, 6, 6, 6,
90 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
91 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
92 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
96 static const uint8_t coeff_token_bits[4][4*17]={
99 5, 1, 0, 0, 7, 4, 1, 0, 7, 6, 5, 3, 7, 6, 5, 3,
100 7, 6, 5, 4, 15, 6, 5, 4, 11,14, 5, 4, 8,10,13, 4,
101 15,14, 9, 4, 11,10,13,12, 15,14, 9,12, 11,10,13, 8,
102 15, 1, 9,12, 11,14,13, 8, 7,10, 9,12, 4, 6, 5, 8,
106 11, 2, 0, 0, 7, 7, 3, 0, 7,10, 9, 5, 7, 6, 5, 4,
107 4, 6, 5, 6, 7, 6, 5, 8, 15, 6, 5, 4, 11,14,13, 4,
108 15,10, 9, 4, 11,14,13,12, 8,10, 9, 8, 15,14,13,12,
109 11,10, 9,12, 7,11, 6, 8, 9, 8,10, 1, 7, 6, 5, 4,
113 15,14, 0, 0, 11,15,13, 0, 8,12,14,12, 15,10,11,11,
114 11, 8, 9,10, 9,14,13, 9, 8,10, 9, 8, 15,14,13,13,
115 11,14,10,12, 15,10,13,12, 11,14, 9,12, 8,10,13, 8,
116 13, 7, 9,12, 9,12,11,10, 5, 8, 7, 6, 1, 4, 3, 2,
120 0, 1, 0, 0, 4, 5, 6, 0, 8, 9,10,11, 12,13,14,15,
121 16,17,18,19, 20,21,22,23, 24,25,26,27, 28,29,30,31,
122 32,33,34,35, 36,37,38,39, 40,41,42,43, 44,45,46,47,
123 48,49,50,51, 52,53,54,55, 56,57,58,59, 60,61,62,63,
127 static const uint8_t total_zeros_len[16][16]= {
128 {1,3,3,4,4,5,5,6,6,7,7,8,8,9,9,9},
129 {3,3,3,3,3,4,4,4,4,5,5,6,6,6,6},
130 {4,3,3,3,4,4,3,3,4,5,5,6,5,6},
131 {5,3,4,4,3,3,3,4,3,4,5,5,5},
132 {4,4,4,3,3,3,3,3,4,5,4,5},
133 {6,5,3,3,3,3,3,3,4,3,6},
134 {6,5,3,3,3,2,3,4,3,6},
145 static const uint8_t total_zeros_bits[16][16]= {
146 {1,3,2,3,2,3,2,3,2,3,2,3,2,3,2,1},
147 {7,6,5,4,3,5,4,3,2,3,2,3,2,1,0},
148 {5,7,6,5,4,3,4,3,2,3,2,1,1,0},
149 {3,7,5,4,6,5,4,3,3,2,2,1,0},
150 {5,4,3,7,6,5,4,3,2,1,1,0},
151 {1,1,7,6,5,4,3,2,1,1,0},
152 {1,1,5,4,3,3,2,1,1,0},
163 static const uint8_t chroma_dc_total_zeros_len[3][4]= {
169 static const uint8_t chroma_dc_total_zeros_bits[3][4]= {
175 static const uint8_t run_len[7][16]={
182 {3,3,3,3,3,3,3,4,5,6,7,8,9,10,11},
185 static const uint8_t run_bits[7][16]={
192 {7,6,5,4,3,2,1,1,1,1,1,1,1,1,1},
195 static VLC coeff_token_vlc[4];
196 static VLC_TYPE coeff_token_vlc_tables[520+332+280+256][2];
197 static const int coeff_token_vlc_tables_size[4]={520,332,280,256};
199 static VLC chroma_dc_coeff_token_vlc;
200 static VLC_TYPE chroma_dc_coeff_token_vlc_table[256][2];
201 static const int chroma_dc_coeff_token_vlc_table_size = 256;
203 static VLC total_zeros_vlc[15];
204 static VLC_TYPE total_zeros_vlc_tables[15][512][2];
205 static const int total_zeros_vlc_tables_size = 512;
207 static VLC chroma_dc_total_zeros_vlc[3];
208 static VLC_TYPE chroma_dc_total_zeros_vlc_tables[3][8][2];
209 static const int chroma_dc_total_zeros_vlc_tables_size = 8;
211 static VLC run_vlc[6];
212 static VLC_TYPE run_vlc_tables[6][8][2];
213 static const int run_vlc_tables_size = 8;
216 static VLC_TYPE run7_vlc_table[96][2];
217 static const int run7_vlc_table_size = 96;
219 #define LEVEL_TAB_BITS 8
220 static int8_t cavlc_level_tab[7][1<<LEVEL_TAB_BITS][2];
224 * gets the predicted number of non-zero coefficients.
225 * @param n block index
227 static inline int pred_non_zero_count(H264Context *h, int n){
228 const int index8= scan8[n];
229 const int left= h->non_zero_count_cache[index8 - 1];
230 const int top = h->non_zero_count_cache[index8 - 8];
233 if(i<64) i= (i+1)>>1;
235 tprintf(h->s.avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
240 static av_cold void init_cavlc_level_tab(void){
241 int suffix_length, mask;
244 for(suffix_length=0; suffix_length<7; suffix_length++){
245 for(i=0; i<(1<<LEVEL_TAB_BITS); i++){
246 int prefix= LEVEL_TAB_BITS - av_log2(2*i);
247 int level_code= (prefix<<suffix_length) + (i>>(LEVEL_TAB_BITS-prefix-1-suffix_length)) - (1<<suffix_length);
249 mask= -(level_code&1);
250 level_code= (((2+level_code)>>1) ^ mask) - mask;
251 if(prefix + 1 + suffix_length <= LEVEL_TAB_BITS){
252 cavlc_level_tab[suffix_length][i][0]= level_code;
253 cavlc_level_tab[suffix_length][i][1]= prefix + 1 + suffix_length;
254 }else if(prefix + 1 <= LEVEL_TAB_BITS){
255 cavlc_level_tab[suffix_length][i][0]= prefix+100;
256 cavlc_level_tab[suffix_length][i][1]= prefix + 1;
258 cavlc_level_tab[suffix_length][i][0]= LEVEL_TAB_BITS+100;
259 cavlc_level_tab[suffix_length][i][1]= LEVEL_TAB_BITS;
265 av_cold void ff_h264_decode_init_vlc(void){
273 chroma_dc_coeff_token_vlc.table = chroma_dc_coeff_token_vlc_table;
274 chroma_dc_coeff_token_vlc.table_allocated = chroma_dc_coeff_token_vlc_table_size;
275 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
276 &chroma_dc_coeff_token_len [0], 1, 1,
277 &chroma_dc_coeff_token_bits[0], 1, 1,
278 INIT_VLC_USE_NEW_STATIC);
282 coeff_token_vlc[i].table = coeff_token_vlc_tables+offset;
283 coeff_token_vlc[i].table_allocated = coeff_token_vlc_tables_size[i];
284 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
285 &coeff_token_len [i][0], 1, 1,
286 &coeff_token_bits[i][0], 1, 1,
287 INIT_VLC_USE_NEW_STATIC);
288 offset += coeff_token_vlc_tables_size[i];
291 * This is a one time safety check to make sure that
292 * the packed static coeff_token_vlc table sizes
293 * were initialized correctly.
295 assert(offset == FF_ARRAY_ELEMS(coeff_token_vlc_tables));
298 chroma_dc_total_zeros_vlc[i].table = chroma_dc_total_zeros_vlc_tables[i];
299 chroma_dc_total_zeros_vlc[i].table_allocated = chroma_dc_total_zeros_vlc_tables_size;
300 init_vlc(&chroma_dc_total_zeros_vlc[i],
301 CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
302 &chroma_dc_total_zeros_len [i][0], 1, 1,
303 &chroma_dc_total_zeros_bits[i][0], 1, 1,
304 INIT_VLC_USE_NEW_STATIC);
307 total_zeros_vlc[i].table = total_zeros_vlc_tables[i];
308 total_zeros_vlc[i].table_allocated = total_zeros_vlc_tables_size;
309 init_vlc(&total_zeros_vlc[i],
310 TOTAL_ZEROS_VLC_BITS, 16,
311 &total_zeros_len [i][0], 1, 1,
312 &total_zeros_bits[i][0], 1, 1,
313 INIT_VLC_USE_NEW_STATIC);
317 run_vlc[i].table = run_vlc_tables[i];
318 run_vlc[i].table_allocated = run_vlc_tables_size;
319 init_vlc(&run_vlc[i],
321 &run_len [i][0], 1, 1,
322 &run_bits[i][0], 1, 1,
323 INIT_VLC_USE_NEW_STATIC);
325 run7_vlc.table = run7_vlc_table,
326 run7_vlc.table_allocated = run7_vlc_table_size;
327 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
328 &run_len [6][0], 1, 1,
329 &run_bits[6][0], 1, 1,
330 INIT_VLC_USE_NEW_STATIC);
332 init_cavlc_level_tab();
339 static inline int get_level_prefix(GetBitContext *gb){
344 UPDATE_CACHE(re, gb);
345 buf=GET_CACHE(re, gb);
347 log= 32 - av_log2(buf);
349 print_bin(buf>>(32-log), log);
350 av_log(NULL, AV_LOG_DEBUG, "%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
353 LAST_SKIP_BITS(re, gb, log);
354 CLOSE_READER(re, gb);
360 * decodes a residual block.
361 * @param n block index
362 * @param scantable scantable
363 * @param max_coeff number of coefficients in the block
364 * @return <0 if an error occurred
366 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff){
367 MpegEncContext * const s = &h->s;
368 static const int coeff_token_table_index[17]= {0, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3};
370 int zeros_left, coeff_token, total_coeff, i, trailing_ones, run_before;
372 //FIXME put trailing_onex into the context
374 if(n >= CHROMA_DC_BLOCK_INDEX){
375 coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
376 total_coeff= coeff_token>>2;
378 if(n == LUMA_DC_BLOCK_INDEX){
379 total_coeff= pred_non_zero_count(h, 0);
380 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
381 total_coeff= coeff_token>>2;
383 total_coeff= pred_non_zero_count(h, n);
384 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
385 total_coeff= coeff_token>>2;
388 h->non_zero_count_cache[ scan8[n] ]= total_coeff;
390 //FIXME set last_non_zero?
394 if(total_coeff > (unsigned)max_coeff) {
395 av_log(h->s.avctx, AV_LOG_ERROR, "corrupted macroblock %d %d (total_coeff=%d)\n", s->mb_x, s->mb_y, total_coeff);
399 trailing_ones= coeff_token&3;
400 tprintf(h->s.avctx, "trailing:%d, total:%d\n", trailing_ones, total_coeff);
401 assert(total_coeff<=16);
403 i = show_bits(gb, 3);
404 skip_bits(gb, trailing_ones);
405 level[0] = 1-((i&4)>>1);
406 level[1] = 1-((i&2) );
407 level[2] = 1-((i&1)<<1);
409 if(trailing_ones<total_coeff) {
411 int suffix_length = total_coeff > 10 & trailing_ones < 3;
412 int bitsi= show_bits(gb, LEVEL_TAB_BITS);
413 int level_code= cavlc_level_tab[suffix_length][bitsi][0];
415 skip_bits(gb, cavlc_level_tab[suffix_length][bitsi][1]);
416 if(level_code >= 100){
417 prefix= level_code - 100;
418 if(prefix == LEVEL_TAB_BITS)
419 prefix += get_level_prefix(gb);
421 //first coefficient has suffix_length equal to 0 or 1
422 if(prefix<14){ //FIXME try to build a large unified VLC table for all this
424 level_code= (prefix<<1) + get_bits1(gb); //part
426 level_code= prefix; //part
427 }else if(prefix==14){
429 level_code= (prefix<<1) + get_bits1(gb); //part
431 level_code= prefix + get_bits(gb, 4); //part
433 level_code= 30 + get_bits(gb, prefix-3); //part
436 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid level prefix\n");
439 level_code += (1<<(prefix-3))-4096;
443 if(trailing_ones < 3) level_code += 2;
446 mask= -(level_code&1);
447 level[trailing_ones]= (((2+level_code)>>1) ^ mask) - mask;
449 level_code += ((level_code>>31)|1) & -(trailing_ones < 3);
451 suffix_length = 1 + (level_code + 3U > 6U);
452 level[trailing_ones]= level_code;
455 //remaining coefficients have suffix_length > 0
456 for(i=trailing_ones+1;i<total_coeff;i++) {
457 static const unsigned int suffix_limit[7] = {0,3,6,12,24,48,INT_MAX };
458 int bitsi= show_bits(gb, LEVEL_TAB_BITS);
459 level_code= cavlc_level_tab[suffix_length][bitsi][0];
461 skip_bits(gb, cavlc_level_tab[suffix_length][bitsi][1]);
462 if(level_code >= 100){
463 prefix= level_code - 100;
464 if(prefix == LEVEL_TAB_BITS){
465 prefix += get_level_prefix(gb);
468 level_code = (prefix<<suffix_length) + get_bits(gb, suffix_length);
470 level_code = (15<<suffix_length) + get_bits(gb, prefix-3);
472 level_code += (1<<(prefix-3))-4096;
474 mask= -(level_code&1);
475 level_code= (((2+level_code)>>1) ^ mask) - mask;
477 level[i]= level_code;
478 suffix_length+= suffix_limit[suffix_length] + level_code > 2U*suffix_limit[suffix_length];
482 if(total_coeff == max_coeff)
485 if(n >= CHROMA_DC_BLOCK_INDEX)
486 zeros_left= get_vlc2(gb, (chroma_dc_total_zeros_vlc-1)[ total_coeff ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
488 zeros_left= get_vlc2(gb, (total_zeros_vlc-1)[ total_coeff ].table, TOTAL_ZEROS_VLC_BITS, 1);
491 #define STORE_BLOCK(type) \
492 scantable += zeros_left + total_coeff - 1; \
493 if(n >= LUMA_DC_BLOCK_INDEX){ \
494 ((type*)block)[*scantable] = level[0]; \
495 for(i=1;i<total_coeff && zeros_left > 0;i++) { \
497 run_before= get_vlc2(gb, (run_vlc-1)[zeros_left].table, RUN_VLC_BITS, 1); \
499 run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2); \
500 zeros_left -= run_before; \
501 scantable -= 1 + run_before; \
502 ((type*)block)[*scantable]= level[i]; \
504 for(;i<total_coeff;i++) { \
506 ((type*)block)[*scantable]= level[i]; \
509 ((type*)block)[*scantable] = ((int)(level[0] * qmul[*scantable] + 32))>>6; \
510 for(i=1;i<total_coeff && zeros_left > 0;i++) { \
512 run_before= get_vlc2(gb, (run_vlc-1)[zeros_left].table, RUN_VLC_BITS, 1); \
514 run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2); \
515 zeros_left -= run_before; \
516 scantable -= 1 + run_before; \
517 ((type*)block)[*scantable]= ((int)(level[i] * qmul[*scantable] + 32))>>6; \
519 for(;i<total_coeff;i++) { \
521 ((type*)block)[*scantable]= ((int)(level[i] * qmul[*scantable] + 32))>>6; \
525 if (h->pixel_shift) {
532 av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
539 int ff_h264_decode_mb_cavlc(H264Context *h){
540 MpegEncContext * const s = &h->s;
543 unsigned int mb_type, cbp;
544 int dct8x8_allowed= h->pps.transform_8x8_mode;
545 const int pixel_shift = h->pixel_shift;
547 mb_xy = h->mb_xy = s->mb_x + s->mb_y*s->mb_stride;
549 tprintf(s->avctx, "pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
550 cbp = 0; /* avoid warning. FIXME: find a solution without slowing
552 if(h->slice_type_nos != AV_PICTURE_TYPE_I){
553 if(s->mb_skip_run==-1)
554 s->mb_skip_run= get_ue_golomb(&s->gb);
556 if (s->mb_skip_run--) {
557 if(FRAME_MBAFF && (s->mb_y&1) == 0){
558 if(s->mb_skip_run==0)
559 h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);
566 if( (s->mb_y&1) == 0 )
567 h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);
570 h->prev_mb_skipped= 0;
572 mb_type= get_ue_golomb(&s->gb);
573 if(h->slice_type_nos == AV_PICTURE_TYPE_B){
575 partition_count= b_mb_type_info[mb_type].partition_count;
576 mb_type= b_mb_type_info[mb_type].type;
579 goto decode_intra_mb;
581 }else if(h->slice_type_nos == AV_PICTURE_TYPE_P){
583 partition_count= p_mb_type_info[mb_type].partition_count;
584 mb_type= p_mb_type_info[mb_type].type;
587 goto decode_intra_mb;
590 assert(h->slice_type_nos == AV_PICTURE_TYPE_I);
591 if(h->slice_type == AV_PICTURE_TYPE_SI && mb_type)
595 av_log(h->s.avctx, AV_LOG_ERROR, "mb_type %d in %c slice too large at %d %d\n", mb_type, av_get_picture_type_char(h->slice_type), s->mb_x, s->mb_y);
599 cbp= i_mb_type_info[mb_type].cbp;
600 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
601 mb_type= i_mb_type_info[mb_type].type;
605 mb_type |= MB_TYPE_INTERLACED;
607 h->slice_table[ mb_xy ]= h->slice_num;
609 if(IS_INTRA_PCM(mb_type)){
612 // We assume these blocks are very rare so we do not optimize it.
613 align_get_bits(&s->gb);
615 // The pixels are stored in the same order as levels in h->mb array.
616 for(x=0; x < (CHROMA ? 384 : 256)*h->sps.bit_depth_luma/8; x++){
617 ((uint8_t*)h->mb)[x]= get_bits(&s->gb, 8);
620 // In deblocking, the quantizer is 0
621 s->current_picture.qscale_table[mb_xy]= 0;
622 // All coeffs are present
623 memset(h->non_zero_count[mb_xy], 16, 32);
625 s->current_picture.mb_type[mb_xy]= mb_type;
630 h->ref_count[0] <<= 1;
631 h->ref_count[1] <<= 1;
634 fill_decode_neighbors(h, mb_type);
635 fill_decode_caches(h, mb_type);
638 if(IS_INTRA(mb_type)){
640 // init_top_left_availability(h);
641 if(IS_INTRA4x4(mb_type)){
644 if(dct8x8_allowed && get_bits1(&s->gb)){
645 mb_type |= MB_TYPE_8x8DCT;
649 // fill_intra4x4_pred_table(h);
650 for(i=0; i<16; i+=di){
651 int mode= pred_intra_mode(h, i);
653 if(!get_bits1(&s->gb)){
654 const int rem_mode= get_bits(&s->gb, 3);
655 mode = rem_mode + (rem_mode >= mode);
659 fill_rectangle( &h->intra4x4_pred_mode_cache[ scan8[i] ], 2, 2, 8, mode, 1 );
661 h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
663 ff_h264_write_back_intra_pred_mode(h);
664 if( ff_h264_check_intra4x4_pred_mode(h) < 0)
667 h->intra16x16_pred_mode= ff_h264_check_intra_pred_mode(h, h->intra16x16_pred_mode);
668 if(h->intra16x16_pred_mode < 0)
672 pred_mode= ff_h264_check_intra_pred_mode(h, get_ue_golomb_31(&s->gb));
675 h->chroma_pred_mode= pred_mode;
677 h->chroma_pred_mode = DC_128_PRED8x8;
679 }else if(partition_count==4){
680 int i, j, sub_partition_count[4], list, ref[2][4];
682 if(h->slice_type_nos == AV_PICTURE_TYPE_B){
684 h->sub_mb_type[i]= get_ue_golomb_31(&s->gb);
685 if(h->sub_mb_type[i] >=13){
686 av_log(h->s.avctx, AV_LOG_ERROR, "B sub_mb_type %u out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
689 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
690 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
692 if( IS_DIRECT(h->sub_mb_type[0]|h->sub_mb_type[1]|h->sub_mb_type[2]|h->sub_mb_type[3])) {
693 ff_h264_pred_direct_motion(h, &mb_type);
694 h->ref_cache[0][scan8[4]] =
695 h->ref_cache[1][scan8[4]] =
696 h->ref_cache[0][scan8[12]] =
697 h->ref_cache[1][scan8[12]] = PART_NOT_AVAILABLE;
700 assert(h->slice_type_nos == AV_PICTURE_TYPE_P); //FIXME SP correct ?
702 h->sub_mb_type[i]= get_ue_golomb_31(&s->gb);
703 if(h->sub_mb_type[i] >=4){
704 av_log(h->s.avctx, AV_LOG_ERROR, "P sub_mb_type %u out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
707 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
708 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
712 for(list=0; list<h->list_count; list++){
713 int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
715 if(IS_DIRECT(h->sub_mb_type[i])) continue;
716 if(IS_DIR(h->sub_mb_type[i], 0, list)){
720 }else if(ref_count == 2){
721 tmp= get_bits1(&s->gb)^1;
723 tmp= get_ue_golomb_31(&s->gb);
725 av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", tmp);
738 dct8x8_allowed = get_dct8x8_allowed(h);
740 for(list=0; list<h->list_count; list++){
742 if(IS_DIRECT(h->sub_mb_type[i])) {
743 h->ref_cache[list][ scan8[4*i] ] = h->ref_cache[list][ scan8[4*i]+1 ];
746 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]=
747 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
749 if(IS_DIR(h->sub_mb_type[i], 0, list)){
750 const int sub_mb_type= h->sub_mb_type[i];
751 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
752 for(j=0; j<sub_partition_count[i]; j++){
754 const int index= 4*i + block_width*j;
755 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
756 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
757 mx += get_se_golomb(&s->gb);
758 my += get_se_golomb(&s->gb);
759 tprintf(s->avctx, "final mv:%d %d\n", mx, my);
761 if(IS_SUB_8X8(sub_mb_type)){
763 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
765 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
766 }else if(IS_SUB_8X4(sub_mb_type)){
767 mv_cache[ 1 ][0]= mx;
768 mv_cache[ 1 ][1]= my;
769 }else if(IS_SUB_4X8(sub_mb_type)){
770 mv_cache[ 8 ][0]= mx;
771 mv_cache[ 8 ][1]= my;
773 mv_cache[ 0 ][0]= mx;
774 mv_cache[ 0 ][1]= my;
777 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
783 }else if(IS_DIRECT(mb_type)){
784 ff_h264_pred_direct_motion(h, &mb_type);
785 dct8x8_allowed &= h->sps.direct_8x8_inference_flag;
788 //FIXME we should set ref_idx_l? to 0 if we use that later ...
789 if(IS_16X16(mb_type)){
790 for(list=0; list<h->list_count; list++){
792 if(IS_DIR(mb_type, 0, list)){
793 if(h->ref_count[list]==1){
795 }else if(h->ref_count[list]==2){
796 val= get_bits1(&s->gb)^1;
798 val= get_ue_golomb_31(&s->gb);
799 if(val >= h->ref_count[list]){
800 av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", val);
804 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
807 for(list=0; list<h->list_count; list++){
808 if(IS_DIR(mb_type, 0, list)){
809 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
810 mx += get_se_golomb(&s->gb);
811 my += get_se_golomb(&s->gb);
812 tprintf(s->avctx, "final mv:%d %d\n", mx, my);
814 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
818 else if(IS_16X8(mb_type)){
819 for(list=0; list<h->list_count; list++){
822 if(IS_DIR(mb_type, i, list)){
823 if(h->ref_count[list] == 1){
825 }else if(h->ref_count[list] == 2){
826 val= get_bits1(&s->gb)^1;
828 val= get_ue_golomb_31(&s->gb);
829 if(val >= h->ref_count[list]){
830 av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", val);
835 val= LIST_NOT_USED&0xFF;
836 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
839 for(list=0; list<h->list_count; list++){
842 if(IS_DIR(mb_type, i, list)){
843 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
844 mx += get_se_golomb(&s->gb);
845 my += get_se_golomb(&s->gb);
846 tprintf(s->avctx, "final mv:%d %d\n", mx, my);
848 val= pack16to32(mx,my);
851 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 4);
855 assert(IS_8X16(mb_type));
856 for(list=0; list<h->list_count; list++){
859 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
860 if(h->ref_count[list]==1){
862 }else if(h->ref_count[list]==2){
863 val= get_bits1(&s->gb)^1;
865 val= get_ue_golomb_31(&s->gb);
866 if(val >= h->ref_count[list]){
867 av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", val);
872 val= LIST_NOT_USED&0xFF;
873 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
876 for(list=0; list<h->list_count; list++){
879 if(IS_DIR(mb_type, i, list)){
880 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
881 mx += get_se_golomb(&s->gb);
882 my += get_se_golomb(&s->gb);
883 tprintf(s->avctx, "final mv:%d %d\n", mx, my);
885 val= pack16to32(mx,my);
888 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 4);
894 if(IS_INTER(mb_type))
895 write_back_motion(h, mb_type);
897 if(!IS_INTRA16x16(mb_type)){
898 cbp= get_ue_golomb(&s->gb);
900 av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%u) at %d %d\n", cbp, s->mb_x, s->mb_y);
905 if(IS_INTRA4x4(mb_type)) cbp= golomb_to_intra4x4_cbp[cbp];
906 else cbp= golomb_to_inter_cbp [cbp];
908 if(IS_INTRA4x4(mb_type)) cbp= golomb_to_intra4x4_cbp_gray[cbp];
909 else cbp= golomb_to_inter_cbp_gray[cbp];
913 if(dct8x8_allowed && (cbp&15) && !IS_INTRA(mb_type)){
914 mb_type |= MB_TYPE_8x8DCT*get_bits1(&s->gb);
917 h->cbp_table[mb_xy]= cbp;
918 s->current_picture.mb_type[mb_xy]= mb_type;
920 if(cbp || IS_INTRA16x16(mb_type)){
921 int i8x8, i4x4, chroma_idx;
923 GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
924 const uint8_t *scan, *scan8x8;
925 const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
927 if(IS_INTERLACED(mb_type)){
928 scan8x8= s->qscale ? h->field_scan8x8_cavlc : h->field_scan8x8_cavlc_q0;
929 scan= s->qscale ? h->field_scan : h->field_scan_q0;
931 scan8x8= s->qscale ? h->zigzag_scan8x8_cavlc : h->zigzag_scan8x8_cavlc_q0;
932 scan= s->qscale ? h->zigzag_scan : h->zigzag_scan_q0;
935 dquant= get_se_golomb(&s->gb);
939 if(((unsigned)s->qscale) > max_qp){
940 if(s->qscale<0) s->qscale+= max_qp+1;
941 else s->qscale-= max_qp+1;
942 if(((unsigned)s->qscale) > max_qp){
943 av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
948 h->chroma_qp[0]= get_chroma_qp(h, 0, s->qscale);
949 h->chroma_qp[1]= get_chroma_qp(h, 1, s->qscale);
950 if(IS_INTRA16x16(mb_type)){
951 AV_ZERO128(h->mb_luma_dc+0);
952 AV_ZERO128(h->mb_luma_dc+8);
953 AV_ZERO128(h->mb_luma_dc+16);
954 AV_ZERO128(h->mb_luma_dc+24);
955 if( decode_residual(h, h->intra_gb_ptr, h->mb_luma_dc, LUMA_DC_BLOCK_INDEX, scan, h->dequant4_coeff[0][s->qscale], 16) < 0){
956 return -1; //FIXME continue if partitioned and other return -1 too
959 assert((cbp&15) == 0 || (cbp&15) == 15);
962 for(i8x8=0; i8x8<4; i8x8++){
963 for(i4x4=0; i4x4<4; i4x4++){
964 const int index= i4x4 + 4*i8x8;
965 if( decode_residual(h, h->intra_gb_ptr, h->mb + (16*index << pixel_shift), index, scan + 1, h->dequant4_coeff[0][s->qscale], 15) < 0 ){
971 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
974 for(i8x8=0; i8x8<4; i8x8++){
976 if(IS_8x8DCT(mb_type)){
977 DCTELEM *buf = &h->mb[64*i8x8 << pixel_shift];
979 for(i4x4=0; i4x4<4; i4x4++){
980 if( decode_residual(h, gb, buf, i4x4+4*i8x8, scan8x8+16*i4x4,
981 h->dequant8_coeff[IS_INTRA( mb_type ) ? 0:1][s->qscale], 16) <0 )
984 nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
985 nnz[0] += nnz[1] + nnz[8] + nnz[9];
987 for(i4x4=0; i4x4<4; i4x4++){
988 const int index= i4x4 + 4*i8x8;
990 if( decode_residual(h, gb, h->mb + (16*index << pixel_shift), index, scan, h->dequant4_coeff[IS_INTRA( mb_type ) ? 0:3][s->qscale], 16) <0 ){
996 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
997 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
1003 for(chroma_idx=0; chroma_idx<2; chroma_idx++)
1004 if( decode_residual(h, gb, h->mb + ((256 + 16*4*chroma_idx) << pixel_shift), CHROMA_DC_BLOCK_INDEX+chroma_idx, chroma_dc_scan, NULL, 4) < 0){
1010 for(chroma_idx=0; chroma_idx<2; chroma_idx++){
1011 const uint32_t *qmul = h->dequant4_coeff[chroma_idx+1+(IS_INTRA( mb_type ) ? 0:3)][h->chroma_qp[chroma_idx]];
1012 for(i4x4=0; i4x4<4; i4x4++){
1013 const int index= 16 + 4*chroma_idx + i4x4;
1014 if( decode_residual(h, gb, h->mb + (16*index << pixel_shift), index, scan + 1, qmul, 15) < 0){
1020 uint8_t * const nnz= &h->non_zero_count_cache[0];
1021 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
1022 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
1025 uint8_t * const nnz= &h->non_zero_count_cache[0];
1026 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
1027 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
1028 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
1030 s->current_picture.qscale_table[mb_xy]= s->qscale;
1031 write_back_non_zero_count(h);
1034 h->ref_count[0] >>= 1;
1035 h->ref_count[1] >>= 1;