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>
29 #define UNCHECKED_BITSTREAM_READER 1
33 #include "mpegvideo.h"
35 #include "h264data.h" // FIXME FIXME FIXME
36 #include "h264_mvpred.h"
38 #include "libavutil/avassert.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 chroma422_dc_coeff_token_len[4*9]={
77 static const uint8_t chroma422_dc_coeff_token_bits[4*9]={
89 static const uint8_t coeff_token_len[4][4*17]={
92 6, 2, 0, 0, 8, 6, 3, 0, 9, 8, 7, 5, 10, 9, 8, 6,
93 11,10, 9, 7, 13,11,10, 8, 13,13,11, 9, 13,13,13,10,
94 14,14,13,11, 14,14,14,13, 15,15,14,14, 15,15,15,14,
95 16,15,15,15, 16,16,16,15, 16,16,16,16, 16,16,16,16,
99 6, 2, 0, 0, 6, 5, 3, 0, 7, 6, 6, 4, 8, 6, 6, 4,
100 8, 7, 7, 5, 9, 8, 8, 6, 11, 9, 9, 6, 11,11,11, 7,
101 12,11,11, 9, 12,12,12,11, 12,12,12,11, 13,13,13,12,
102 13,13,13,13, 13,14,13,13, 14,14,14,13, 14,14,14,14,
106 6, 4, 0, 0, 6, 5, 4, 0, 6, 5, 5, 4, 7, 5, 5, 4,
107 7, 5, 5, 4, 7, 6, 6, 4, 7, 6, 6, 4, 8, 7, 7, 5,
108 8, 8, 7, 6, 9, 8, 8, 7, 9, 9, 8, 8, 9, 9, 9, 8,
109 10, 9, 9, 9, 10,10,10,10, 10,10,10,10, 10,10,10,10,
113 6, 6, 0, 0, 6, 6, 6, 0, 6, 6, 6, 6, 6, 6, 6, 6,
114 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
115 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
116 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
120 static const uint8_t coeff_token_bits[4][4*17]={
123 5, 1, 0, 0, 7, 4, 1, 0, 7, 6, 5, 3, 7, 6, 5, 3,
124 7, 6, 5, 4, 15, 6, 5, 4, 11,14, 5, 4, 8,10,13, 4,
125 15,14, 9, 4, 11,10,13,12, 15,14, 9,12, 11,10,13, 8,
126 15, 1, 9,12, 11,14,13, 8, 7,10, 9,12, 4, 6, 5, 8,
130 11, 2, 0, 0, 7, 7, 3, 0, 7,10, 9, 5, 7, 6, 5, 4,
131 4, 6, 5, 6, 7, 6, 5, 8, 15, 6, 5, 4, 11,14,13, 4,
132 15,10, 9, 4, 11,14,13,12, 8,10, 9, 8, 15,14,13,12,
133 11,10, 9,12, 7,11, 6, 8, 9, 8,10, 1, 7, 6, 5, 4,
137 15,14, 0, 0, 11,15,13, 0, 8,12,14,12, 15,10,11,11,
138 11, 8, 9,10, 9,14,13, 9, 8,10, 9, 8, 15,14,13,13,
139 11,14,10,12, 15,10,13,12, 11,14, 9,12, 8,10,13, 8,
140 13, 7, 9,12, 9,12,11,10, 5, 8, 7, 6, 1, 4, 3, 2,
144 0, 1, 0, 0, 4, 5, 6, 0, 8, 9,10,11, 12,13,14,15,
145 16,17,18,19, 20,21,22,23, 24,25,26,27, 28,29,30,31,
146 32,33,34,35, 36,37,38,39, 40,41,42,43, 44,45,46,47,
147 48,49,50,51, 52,53,54,55, 56,57,58,59, 60,61,62,63,
151 static const uint8_t total_zeros_len[16][16]= {
152 {1,3,3,4,4,5,5,6,6,7,7,8,8,9,9,9},
153 {3,3,3,3,3,4,4,4,4,5,5,6,6,6,6},
154 {4,3,3,3,4,4,3,3,4,5,5,6,5,6},
155 {5,3,4,4,3,3,3,4,3,4,5,5,5},
156 {4,4,4,3,3,3,3,3,4,5,4,5},
157 {6,5,3,3,3,3,3,3,4,3,6},
158 {6,5,3,3,3,2,3,4,3,6},
169 static const uint8_t total_zeros_bits[16][16]= {
170 {1,3,2,3,2,3,2,3,2,3,2,3,2,3,2,1},
171 {7,6,5,4,3,5,4,3,2,3,2,3,2,1,0},
172 {5,7,6,5,4,3,4,3,2,3,2,1,1,0},
173 {3,7,5,4,6,5,4,3,3,2,2,1,0},
174 {5,4,3,7,6,5,4,3,2,1,1,0},
175 {1,1,7,6,5,4,3,2,1,1,0},
176 {1,1,5,4,3,3,2,1,1,0},
187 static const uint8_t chroma_dc_total_zeros_len[3][4]= {
193 static const uint8_t chroma_dc_total_zeros_bits[3][4]= {
199 static const uint8_t chroma422_dc_total_zeros_len[7][8]= {
200 { 1, 3, 3, 4, 4, 4, 5, 5 },
201 { 3, 2, 3, 3, 3, 3, 3 },
202 { 3, 3, 2, 2, 3, 3 },
209 static const uint8_t chroma422_dc_total_zeros_bits[7][8]= {
210 { 1, 2, 3, 2, 3, 1, 1, 0 },
211 { 0, 1, 1, 4, 5, 6, 7 },
212 { 0, 1, 1, 2, 6, 7 },
219 static const uint8_t run_len[7][16]={
226 {3,3,3,3,3,3,3,4,5,6,7,8,9,10,11},
229 static const uint8_t run_bits[7][16]={
236 {7,6,5,4,3,2,1,1,1,1,1,1,1,1,1},
239 static VLC coeff_token_vlc[4];
240 static VLC_TYPE coeff_token_vlc_tables[520+332+280+256][2];
241 static const int coeff_token_vlc_tables_size[4]={520,332,280,256};
243 static VLC chroma_dc_coeff_token_vlc;
244 static VLC_TYPE chroma_dc_coeff_token_vlc_table[256][2];
245 static const int chroma_dc_coeff_token_vlc_table_size = 256;
247 static VLC chroma422_dc_coeff_token_vlc;
248 static VLC_TYPE chroma422_dc_coeff_token_vlc_table[8192][2];
249 static const int chroma422_dc_coeff_token_vlc_table_size = 8192;
251 static VLC total_zeros_vlc[15];
252 static VLC_TYPE total_zeros_vlc_tables[15][512][2];
253 static const int total_zeros_vlc_tables_size = 512;
255 static VLC chroma_dc_total_zeros_vlc[3];
256 static VLC_TYPE chroma_dc_total_zeros_vlc_tables[3][8][2];
257 static const int chroma_dc_total_zeros_vlc_tables_size = 8;
259 static VLC chroma422_dc_total_zeros_vlc[7];
260 static VLC_TYPE chroma422_dc_total_zeros_vlc_tables[7][32][2];
261 static const int chroma422_dc_total_zeros_vlc_tables_size = 32;
263 static VLC run_vlc[6];
264 static VLC_TYPE run_vlc_tables[6][8][2];
265 static const int run_vlc_tables_size = 8;
268 static VLC_TYPE run7_vlc_table[96][2];
269 static const int run7_vlc_table_size = 96;
271 #define LEVEL_TAB_BITS 8
272 static int8_t cavlc_level_tab[7][1<<LEVEL_TAB_BITS][2];
274 #define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
275 #define CHROMA422_DC_COEFF_TOKEN_VLC_BITS 13
276 #define COEFF_TOKEN_VLC_BITS 8
277 #define TOTAL_ZEROS_VLC_BITS 9
278 #define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
279 #define CHROMA422_DC_TOTAL_ZEROS_VLC_BITS 5
280 #define RUN_VLC_BITS 3
281 #define RUN7_VLC_BITS 6
284 * Get the predicted number of non-zero coefficients.
285 * @param n block index
287 static inline int pred_non_zero_count(H264Context *h, int n){
288 const int index8= scan8[n];
289 const int left= h->non_zero_count_cache[index8 - 1];
290 const int top = h->non_zero_count_cache[index8 - 8];
293 if(i<64) i= (i+1)>>1;
295 tprintf(h->avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
300 static av_cold void init_cavlc_level_tab(void){
304 for(suffix_length=0; suffix_length<7; suffix_length++){
305 for(i=0; i<(1<<LEVEL_TAB_BITS); i++){
306 int prefix= LEVEL_TAB_BITS - av_log2(2*i);
308 if(prefix + 1 + suffix_length <= LEVEL_TAB_BITS){
309 int level_code = (prefix << suffix_length) +
310 (i >> (av_log2(i) - suffix_length)) - (1 << suffix_length);
311 int mask = -(level_code&1);
312 level_code = (((2 + level_code) >> 1) ^ mask) - mask;
313 cavlc_level_tab[suffix_length][i][0]= level_code;
314 cavlc_level_tab[suffix_length][i][1]= prefix + 1 + suffix_length;
315 }else if(prefix + 1 <= LEVEL_TAB_BITS){
316 cavlc_level_tab[suffix_length][i][0]= prefix+100;
317 cavlc_level_tab[suffix_length][i][1]= prefix + 1;
319 cavlc_level_tab[suffix_length][i][0]= LEVEL_TAB_BITS+100;
320 cavlc_level_tab[suffix_length][i][1]= LEVEL_TAB_BITS;
326 av_cold void ff_h264_decode_init_vlc(void){
334 chroma_dc_coeff_token_vlc.table = chroma_dc_coeff_token_vlc_table;
335 chroma_dc_coeff_token_vlc.table_allocated = chroma_dc_coeff_token_vlc_table_size;
336 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
337 &chroma_dc_coeff_token_len [0], 1, 1,
338 &chroma_dc_coeff_token_bits[0], 1, 1,
339 INIT_VLC_USE_NEW_STATIC);
341 chroma422_dc_coeff_token_vlc.table = chroma422_dc_coeff_token_vlc_table;
342 chroma422_dc_coeff_token_vlc.table_allocated = chroma422_dc_coeff_token_vlc_table_size;
343 init_vlc(&chroma422_dc_coeff_token_vlc, CHROMA422_DC_COEFF_TOKEN_VLC_BITS, 4*9,
344 &chroma422_dc_coeff_token_len [0], 1, 1,
345 &chroma422_dc_coeff_token_bits[0], 1, 1,
346 INIT_VLC_USE_NEW_STATIC);
350 coeff_token_vlc[i].table = coeff_token_vlc_tables+offset;
351 coeff_token_vlc[i].table_allocated = coeff_token_vlc_tables_size[i];
352 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
353 &coeff_token_len [i][0], 1, 1,
354 &coeff_token_bits[i][0], 1, 1,
355 INIT_VLC_USE_NEW_STATIC);
356 offset += coeff_token_vlc_tables_size[i];
359 * This is a one time safety check to make sure that
360 * the packed static coeff_token_vlc table sizes
361 * were initialized correctly.
363 av_assert0(offset == FF_ARRAY_ELEMS(coeff_token_vlc_tables));
366 chroma_dc_total_zeros_vlc[i].table = chroma_dc_total_zeros_vlc_tables[i];
367 chroma_dc_total_zeros_vlc[i].table_allocated = chroma_dc_total_zeros_vlc_tables_size;
368 init_vlc(&chroma_dc_total_zeros_vlc[i],
369 CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
370 &chroma_dc_total_zeros_len [i][0], 1, 1,
371 &chroma_dc_total_zeros_bits[i][0], 1, 1,
372 INIT_VLC_USE_NEW_STATIC);
376 chroma422_dc_total_zeros_vlc[i].table = chroma422_dc_total_zeros_vlc_tables[i];
377 chroma422_dc_total_zeros_vlc[i].table_allocated = chroma422_dc_total_zeros_vlc_tables_size;
378 init_vlc(&chroma422_dc_total_zeros_vlc[i],
379 CHROMA422_DC_TOTAL_ZEROS_VLC_BITS, 8,
380 &chroma422_dc_total_zeros_len [i][0], 1, 1,
381 &chroma422_dc_total_zeros_bits[i][0], 1, 1,
382 INIT_VLC_USE_NEW_STATIC);
386 total_zeros_vlc[i].table = total_zeros_vlc_tables[i];
387 total_zeros_vlc[i].table_allocated = total_zeros_vlc_tables_size;
388 init_vlc(&total_zeros_vlc[i],
389 TOTAL_ZEROS_VLC_BITS, 16,
390 &total_zeros_len [i][0], 1, 1,
391 &total_zeros_bits[i][0], 1, 1,
392 INIT_VLC_USE_NEW_STATIC);
396 run_vlc[i].table = run_vlc_tables[i];
397 run_vlc[i].table_allocated = run_vlc_tables_size;
398 init_vlc(&run_vlc[i],
400 &run_len [i][0], 1, 1,
401 &run_bits[i][0], 1, 1,
402 INIT_VLC_USE_NEW_STATIC);
404 run7_vlc.table = run7_vlc_table,
405 run7_vlc.table_allocated = run7_vlc_table_size;
406 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
407 &run_len [6][0], 1, 1,
408 &run_bits[6][0], 1, 1,
409 INIT_VLC_USE_NEW_STATIC);
411 init_cavlc_level_tab();
418 static inline int get_level_prefix(GetBitContext *gb){
423 UPDATE_CACHE(re, gb);
424 buf=GET_CACHE(re, gb);
426 log= 32 - av_log2(buf);
428 print_bin(buf>>(32-log), log);
429 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__);
432 LAST_SKIP_BITS(re, gb, log);
433 CLOSE_READER(re, gb);
439 * Decode a residual block.
440 * @param n block index
441 * @param scantable scantable
442 * @param max_coeff number of coefficients in the block
443 * @return <0 if an error occurred
445 static int decode_residual(H264Context *h, GetBitContext *gb, int16_t *block, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff){
446 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};
448 int zeros_left, coeff_token, total_coeff, i, trailing_ones, run_before;
450 //FIXME put trailing_onex into the context
454 coeff_token = get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
456 coeff_token = get_vlc2(gb, chroma422_dc_coeff_token_vlc.table, CHROMA422_DC_COEFF_TOKEN_VLC_BITS, 1);
457 total_coeff= coeff_token>>2;
459 if(n >= LUMA_DC_BLOCK_INDEX){
460 total_coeff= pred_non_zero_count(h, (n - LUMA_DC_BLOCK_INDEX)*16);
461 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
462 total_coeff= coeff_token>>2;
464 total_coeff= pred_non_zero_count(h, n);
465 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
466 total_coeff= coeff_token>>2;
469 h->non_zero_count_cache[ scan8[n] ]= total_coeff;
471 //FIXME set last_non_zero?
475 if(total_coeff > (unsigned)max_coeff) {
476 av_log(h->avctx, AV_LOG_ERROR, "corrupted macroblock %d %d (total_coeff=%d)\n", h->mb_x, h->mb_y, total_coeff);
480 trailing_ones= coeff_token&3;
481 tprintf(h->avctx, "trailing:%d, total:%d\n", trailing_ones, total_coeff);
482 av_assert2(total_coeff<=16);
484 i = show_bits(gb, 3);
485 skip_bits(gb, trailing_ones);
486 level[0] = 1-((i&4)>>1);
487 level[1] = 1-((i&2) );
488 level[2] = 1-((i&1)<<1);
490 if(trailing_ones<total_coeff) {
492 int suffix_length = total_coeff > 10 & trailing_ones < 3;
493 int bitsi= show_bits(gb, LEVEL_TAB_BITS);
494 int level_code= cavlc_level_tab[suffix_length][bitsi][0];
496 skip_bits(gb, cavlc_level_tab[suffix_length][bitsi][1]);
497 if(level_code >= 100){
498 prefix= level_code - 100;
499 if(prefix == LEVEL_TAB_BITS)
500 prefix += get_level_prefix(gb);
502 //first coefficient has suffix_length equal to 0 or 1
503 if(prefix<14){ //FIXME try to build a large unified VLC table for all this
505 level_code= (prefix<<1) + get_bits1(gb); //part
507 level_code= prefix; //part
508 }else if(prefix==14){
510 level_code= (prefix<<1) + get_bits1(gb); //part
512 level_code= prefix + get_bits(gb, 4); //part
517 av_log(h->avctx, AV_LOG_ERROR, "Invalid level prefix\n");
520 level_code += (1<<(prefix-3))-4096;
522 level_code += get_bits(gb, prefix-3); //part
525 if(trailing_ones < 3) level_code += 2;
528 mask= -(level_code&1);
529 level[trailing_ones]= (((2+level_code)>>1) ^ mask) - mask;
531 level_code += ((level_code>>31)|1) & -(trailing_ones < 3);
533 suffix_length = 1 + (level_code + 3U > 6U);
534 level[trailing_ones]= level_code;
537 //remaining coefficients have suffix_length > 0
538 for(i=trailing_ones+1;i<total_coeff;i++) {
539 static const unsigned int suffix_limit[7] = {0,3,6,12,24,48,INT_MAX };
540 int bitsi= show_bits(gb, LEVEL_TAB_BITS);
541 level_code= cavlc_level_tab[suffix_length][bitsi][0];
543 skip_bits(gb, cavlc_level_tab[suffix_length][bitsi][1]);
544 if(level_code >= 100){
545 prefix= level_code - 100;
546 if(prefix == LEVEL_TAB_BITS){
547 prefix += get_level_prefix(gb);
550 level_code = (prefix<<suffix_length) + get_bits(gb, suffix_length);
552 level_code = (15<<suffix_length) + get_bits(gb, prefix-3);
554 level_code += (1<<(prefix-3))-4096;
556 mask= -(level_code&1);
557 level_code= (((2+level_code)>>1) ^ mask) - mask;
559 level[i]= level_code;
560 suffix_length+= suffix_limit[suffix_length] + level_code > 2U*suffix_limit[suffix_length];
564 if(total_coeff == max_coeff)
567 if (max_coeff <= 8) {
569 zeros_left = get_vlc2(gb, (chroma_dc_total_zeros_vlc-1)[total_coeff].table,
570 CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
572 zeros_left = get_vlc2(gb, (chroma422_dc_total_zeros_vlc-1)[total_coeff].table,
573 CHROMA422_DC_TOTAL_ZEROS_VLC_BITS, 1);
575 zeros_left= get_vlc2(gb, (total_zeros_vlc-1)[ total_coeff ].table, TOTAL_ZEROS_VLC_BITS, 1);
579 #define STORE_BLOCK(type) \
580 scantable += zeros_left + total_coeff - 1; \
581 if(n >= LUMA_DC_BLOCK_INDEX){ \
582 ((type*)block)[*scantable] = level[0]; \
583 for(i=1;i<total_coeff && zeros_left > 0;i++) { \
585 run_before= get_vlc2(gb, (run_vlc-1)[zeros_left].table, RUN_VLC_BITS, 1); \
587 run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2); \
588 zeros_left -= run_before; \
589 scantable -= 1 + run_before; \
590 ((type*)block)[*scantable]= level[i]; \
592 for(;i<total_coeff;i++) { \
594 ((type*)block)[*scantable]= level[i]; \
597 ((type*)block)[*scantable] = ((int)(level[0] * qmul[*scantable] + 32))>>6; \
598 for(i=1;i<total_coeff && zeros_left > 0;i++) { \
600 run_before= get_vlc2(gb, (run_vlc-1)[zeros_left].table, RUN_VLC_BITS, 1); \
602 run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2); \
603 zeros_left -= run_before; \
604 scantable -= 1 + run_before; \
605 ((type*)block)[*scantable]= ((int)(level[i] * qmul[*scantable] + 32))>>6; \
607 for(;i<total_coeff;i++) { \
609 ((type*)block)[*scantable]= ((int)(level[i] * qmul[*scantable] + 32))>>6; \
613 if (h->pixel_shift) {
620 av_log(h->avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", h->mb_x, h->mb_y);
627 static av_always_inline int decode_luma_residual(H264Context *h, GetBitContext *gb, const uint8_t *scan, const uint8_t *scan8x8, int pixel_shift, int mb_type, int cbp, int p){
629 int qscale = p == 0 ? h->qscale : h->chroma_qp[p-1];
630 if(IS_INTRA16x16(mb_type)){
631 AV_ZERO128(h->mb_luma_dc[p]+0);
632 AV_ZERO128(h->mb_luma_dc[p]+8);
633 AV_ZERO128(h->mb_luma_dc[p]+16);
634 AV_ZERO128(h->mb_luma_dc[p]+24);
635 if( decode_residual(h, h->intra_gb_ptr, h->mb_luma_dc[p], LUMA_DC_BLOCK_INDEX+p, scan, NULL, 16) < 0){
636 return -1; //FIXME continue if partitioned and other return -1 too
639 av_assert2((cbp&15) == 0 || (cbp&15) == 15);
642 for(i8x8=0; i8x8<4; i8x8++){
643 for(i4x4=0; i4x4<4; i4x4++){
644 const int index= i4x4 + 4*i8x8 + p*16;
645 if( decode_residual(h, h->intra_gb_ptr, h->mb + (16*index << pixel_shift),
646 index, scan + 1, h->dequant4_coeff[p][qscale], 15) < 0 ){
653 fill_rectangle(&h->non_zero_count_cache[scan8[p*16]], 4, 4, 8, 0, 1);
657 int cqm = (IS_INTRA( mb_type ) ? 0:3)+p;
658 /* For CAVLC 4:4:4, we need to keep track of the luma 8x8 CBP for deblocking nnz purposes. */
660 for(i8x8=0; i8x8<4; i8x8++){
662 if(IS_8x8DCT(mb_type)){
663 int16_t *buf = &h->mb[64*i8x8+256*p << pixel_shift];
665 for(i4x4=0; i4x4<4; i4x4++){
666 const int index= i4x4 + 4*i8x8 + p*16;
667 if( decode_residual(h, gb, buf, index, scan8x8+16*i4x4,
668 h->dequant8_coeff[cqm][qscale], 16) < 0 )
671 nnz= &h->non_zero_count_cache[ scan8[4*i8x8+p*16] ];
672 nnz[0] += nnz[1] + nnz[8] + nnz[9];
673 new_cbp |= !!nnz[0] << i8x8;
675 for(i4x4=0; i4x4<4; i4x4++){
676 const int index= i4x4 + 4*i8x8 + p*16;
677 if( decode_residual(h, gb, h->mb + (16*index << pixel_shift), index,
678 scan, h->dequant4_coeff[cqm][qscale], 16) < 0 ){
681 new_cbp |= h->non_zero_count_cache[ scan8[index] ] << i8x8;
685 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8+p*16] ];
686 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
693 int ff_h264_decode_mb_cavlc(H264Context *h){
696 unsigned int mb_type, cbp;
697 int dct8x8_allowed= h->pps.transform_8x8_mode;
698 int decode_chroma = h->sps.chroma_format_idc == 1 || h->sps.chroma_format_idc == 2;
699 const int pixel_shift = h->pixel_shift;
700 unsigned local_ref_count[2];
702 mb_xy = h->mb_xy = h->mb_x + h->mb_y*h->mb_stride;
704 tprintf(h->avctx, "pic:%d mb:%d/%d\n", h->frame_num, h->mb_x, h->mb_y);
705 cbp = 0; /* avoid warning. FIXME: find a solution without slowing
707 if(h->slice_type_nos != AV_PICTURE_TYPE_I){
708 if(h->mb_skip_run==-1)
709 h->mb_skip_run= get_ue_golomb(&h->gb);
711 if (h->mb_skip_run--) {
712 if(FRAME_MBAFF && (h->mb_y&1) == 0){
713 if(h->mb_skip_run==0)
714 h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&h->gb);
721 if( (h->mb_y&1) == 0 )
722 h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&h->gb);
725 h->prev_mb_skipped= 0;
727 mb_type= get_ue_golomb(&h->gb);
728 if(h->slice_type_nos == AV_PICTURE_TYPE_B){
730 partition_count= b_mb_type_info[mb_type].partition_count;
731 mb_type= b_mb_type_info[mb_type].type;
734 goto decode_intra_mb;
736 }else if(h->slice_type_nos == AV_PICTURE_TYPE_P){
738 partition_count= p_mb_type_info[mb_type].partition_count;
739 mb_type= p_mb_type_info[mb_type].type;
742 goto decode_intra_mb;
745 av_assert2(h->slice_type_nos == AV_PICTURE_TYPE_I);
746 if(h->slice_type == AV_PICTURE_TYPE_SI && mb_type)
750 av_log(h->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), h->mb_x, h->mb_y);
754 cbp= i_mb_type_info[mb_type].cbp;
755 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
756 mb_type= i_mb_type_info[mb_type].type;
760 mb_type |= MB_TYPE_INTERLACED;
762 h->slice_table[ mb_xy ]= h->slice_num;
764 if(IS_INTRA_PCM(mb_type)){
765 const int mb_size = ff_h264_mb_sizes[h->sps.chroma_format_idc] *
766 h->sps.bit_depth_luma;
768 // We assume these blocks are very rare so we do not optimize it.
769 h->intra_pcm_ptr = align_get_bits(&h->gb);
770 skip_bits_long(&h->gb, mb_size);
772 // In deblocking, the quantizer is 0
773 h->cur_pic.f.qscale_table[mb_xy] = 0;
774 // All coeffs are present
775 memset(h->non_zero_count[mb_xy], 16, 48);
777 h->cur_pic.f.mb_type[mb_xy] = mb_type;
781 local_ref_count[0] = h->ref_count[0] << MB_MBAFF;
782 local_ref_count[1] = h->ref_count[1] << MB_MBAFF;
784 fill_decode_neighbors(h, mb_type);
785 fill_decode_caches(h, mb_type);
788 if(IS_INTRA(mb_type)){
790 // init_top_left_availability(h);
791 if(IS_INTRA4x4(mb_type)){
794 if(dct8x8_allowed && get_bits1(&h->gb)){
795 mb_type |= MB_TYPE_8x8DCT;
799 // fill_intra4x4_pred_table(h);
800 for(i=0; i<16; i+=di){
801 int mode= pred_intra_mode(h, i);
803 if(!get_bits1(&h->gb)){
804 const int rem_mode= get_bits(&h->gb, 3);
805 mode = rem_mode + (rem_mode >= mode);
809 fill_rectangle( &h->intra4x4_pred_mode_cache[ scan8[i] ], 2, 2, 8, mode, 1 );
811 h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
813 write_back_intra_pred_mode(h);
814 if( ff_h264_check_intra4x4_pred_mode(h) < 0)
817 h->intra16x16_pred_mode= ff_h264_check_intra_pred_mode(h, h->intra16x16_pred_mode, 0);
818 if(h->intra16x16_pred_mode < 0)
822 pred_mode= ff_h264_check_intra_pred_mode(h, get_ue_golomb_31(&h->gb), 1);
825 h->chroma_pred_mode= pred_mode;
827 h->chroma_pred_mode = DC_128_PRED8x8;
829 }else if(partition_count==4){
830 int i, j, sub_partition_count[4], list, ref[2][4];
832 if(h->slice_type_nos == AV_PICTURE_TYPE_B){
834 h->sub_mb_type[i]= get_ue_golomb_31(&h->gb);
835 if(h->sub_mb_type[i] >=13){
836 av_log(h->avctx, AV_LOG_ERROR, "B sub_mb_type %u out of range at %d %d\n", h->sub_mb_type[i], h->mb_x, h->mb_y);
839 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
840 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
842 if( IS_DIRECT(h->sub_mb_type[0]|h->sub_mb_type[1]|h->sub_mb_type[2]|h->sub_mb_type[3])) {
843 ff_h264_pred_direct_motion(h, &mb_type);
844 h->ref_cache[0][scan8[4]] =
845 h->ref_cache[1][scan8[4]] =
846 h->ref_cache[0][scan8[12]] =
847 h->ref_cache[1][scan8[12]] = PART_NOT_AVAILABLE;
850 av_assert2(h->slice_type_nos == AV_PICTURE_TYPE_P); //FIXME SP correct ?
852 h->sub_mb_type[i]= get_ue_golomb_31(&h->gb);
853 if(h->sub_mb_type[i] >=4){
854 av_log(h->avctx, AV_LOG_ERROR, "P sub_mb_type %u out of range at %d %d\n", h->sub_mb_type[i], h->mb_x, h->mb_y);
857 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
858 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
862 for(list=0; list<h->list_count; list++){
863 int ref_count= IS_REF0(mb_type) ? 1 : local_ref_count[list];
865 if(IS_DIRECT(h->sub_mb_type[i])) continue;
866 if(IS_DIR(h->sub_mb_type[i], 0, list)){
870 }else if(ref_count == 2){
871 tmp= get_bits1(&h->gb)^1;
873 tmp= get_ue_golomb_31(&h->gb);
875 av_log(h->avctx, AV_LOG_ERROR, "ref %u overflow\n", tmp);
888 dct8x8_allowed = get_dct8x8_allowed(h);
890 for(list=0; list<h->list_count; list++){
892 if(IS_DIRECT(h->sub_mb_type[i])) {
893 h->ref_cache[list][ scan8[4*i] ] = h->ref_cache[list][ scan8[4*i]+1 ];
896 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]=
897 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
899 if(IS_DIR(h->sub_mb_type[i], 0, list)){
900 const int sub_mb_type= h->sub_mb_type[i];
901 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
902 for(j=0; j<sub_partition_count[i]; j++){
904 const int index= 4*i + block_width*j;
905 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
906 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
907 mx += get_se_golomb(&h->gb);
908 my += get_se_golomb(&h->gb);
909 tprintf(h->avctx, "final mv:%d %d\n", mx, my);
911 if(IS_SUB_8X8(sub_mb_type)){
913 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
915 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
916 }else if(IS_SUB_8X4(sub_mb_type)){
917 mv_cache[ 1 ][0]= mx;
918 mv_cache[ 1 ][1]= my;
919 }else if(IS_SUB_4X8(sub_mb_type)){
920 mv_cache[ 8 ][0]= mx;
921 mv_cache[ 8 ][1]= my;
923 mv_cache[ 0 ][0]= mx;
924 mv_cache[ 0 ][1]= my;
927 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
933 }else if(IS_DIRECT(mb_type)){
934 ff_h264_pred_direct_motion(h, &mb_type);
935 dct8x8_allowed &= h->sps.direct_8x8_inference_flag;
938 //FIXME we should set ref_idx_l? to 0 if we use that later ...
939 if(IS_16X16(mb_type)){
940 for(list=0; list<h->list_count; list++){
942 if(IS_DIR(mb_type, 0, list)){
943 if(local_ref_count[list]==1){
945 }else if(local_ref_count[list]==2){
946 val= get_bits1(&h->gb)^1;
948 val= get_ue_golomb_31(&h->gb);
949 if(val >= local_ref_count[list]){
950 av_log(h->avctx, AV_LOG_ERROR, "ref %u overflow\n", val);
954 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
957 for(list=0; list<h->list_count; list++){
958 if(IS_DIR(mb_type, 0, list)){
959 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
960 mx += get_se_golomb(&h->gb);
961 my += get_se_golomb(&h->gb);
962 tprintf(h->avctx, "final mv:%d %d\n", mx, my);
964 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
968 else if(IS_16X8(mb_type)){
969 for(list=0; list<h->list_count; list++){
972 if(IS_DIR(mb_type, i, list)){
973 if(local_ref_count[list] == 1){
975 }else if(local_ref_count[list] == 2){
976 val= get_bits1(&h->gb)^1;
978 val= get_ue_golomb_31(&h->gb);
979 if(val >= local_ref_count[list]){
980 av_log(h->avctx, AV_LOG_ERROR, "ref %u overflow\n", val);
985 val= LIST_NOT_USED&0xFF;
986 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
989 for(list=0; list<h->list_count; list++){
992 if(IS_DIR(mb_type, i, list)){
993 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
994 mx += get_se_golomb(&h->gb);
995 my += get_se_golomb(&h->gb);
996 tprintf(h->avctx, "final mv:%d %d\n", mx, my);
998 val= pack16to32(mx,my);
1001 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 4);
1005 av_assert2(IS_8X16(mb_type));
1006 for(list=0; list<h->list_count; list++){
1009 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
1010 if(local_ref_count[list]==1){
1012 }else if(local_ref_count[list]==2){
1013 val= get_bits1(&h->gb)^1;
1015 val= get_ue_golomb_31(&h->gb);
1016 if(val >= local_ref_count[list]){
1017 av_log(h->avctx, AV_LOG_ERROR, "ref %u overflow\n", val);
1022 val= LIST_NOT_USED&0xFF;
1023 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
1026 for(list=0; list<h->list_count; list++){
1029 if(IS_DIR(mb_type, i, list)){
1030 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
1031 mx += get_se_golomb(&h->gb);
1032 my += get_se_golomb(&h->gb);
1033 tprintf(h->avctx, "final mv:%d %d\n", mx, my);
1035 val= pack16to32(mx,my);
1038 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 4);
1044 if(IS_INTER(mb_type))
1045 write_back_motion(h, mb_type);
1047 if(!IS_INTRA16x16(mb_type)){
1048 cbp= get_ue_golomb(&h->gb);
1052 av_log(h->avctx, AV_LOG_ERROR, "cbp too large (%u) at %d %d\n", cbp, h->mb_x, h->mb_y);
1055 if(IS_INTRA4x4(mb_type)) cbp= golomb_to_intra4x4_cbp[cbp];
1056 else cbp= golomb_to_inter_cbp [cbp];
1059 av_log(h->avctx, AV_LOG_ERROR, "cbp too large (%u) at %d %d\n", cbp, h->mb_x, h->mb_y);
1062 if(IS_INTRA4x4(mb_type)) cbp= golomb_to_intra4x4_cbp_gray[cbp];
1063 else cbp= golomb_to_inter_cbp_gray[cbp];
1066 if (!decode_chroma && cbp>15) {
1067 av_log(h->avctx, AV_LOG_ERROR, "gray chroma\n");
1068 return AVERROR_INVALIDDATA;
1072 if(dct8x8_allowed && (cbp&15) && !IS_INTRA(mb_type)){
1073 mb_type |= MB_TYPE_8x8DCT*get_bits1(&h->gb);
1076 h->cbp_table[mb_xy]= cbp;
1077 h->cur_pic.f.mb_type[mb_xy] = mb_type;
1079 if(cbp || IS_INTRA16x16(mb_type)){
1080 int i4x4, i8x8, chroma_idx;
1083 GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
1084 const uint8_t *scan, *scan8x8;
1085 const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
1087 if(IS_INTERLACED(mb_type)){
1088 scan8x8= h->qscale ? h->field_scan8x8_cavlc : h->field_scan8x8_cavlc_q0;
1089 scan= h->qscale ? h->field_scan : h->field_scan_q0;
1091 scan8x8= h->qscale ? h->zigzag_scan8x8_cavlc : h->zigzag_scan8x8_cavlc_q0;
1092 scan= h->qscale ? h->zigzag_scan : h->zigzag_scan_q0;
1095 dquant= get_se_golomb(&h->gb);
1097 h->qscale += dquant;
1099 if(((unsigned)h->qscale) > max_qp){
1100 if(h->qscale<0) h->qscale+= max_qp+1;
1101 else h->qscale-= max_qp+1;
1102 if(((unsigned)h->qscale) > max_qp){
1103 av_log(h->avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, h->mb_x, h->mb_y);
1108 h->chroma_qp[0]= get_chroma_qp(h, 0, h->qscale);
1109 h->chroma_qp[1]= get_chroma_qp(h, 1, h->qscale);
1111 if( (ret = decode_luma_residual(h, gb, scan, scan8x8, pixel_shift, mb_type, cbp, 0)) < 0 ){
1114 h->cbp_table[mb_xy] |= ret << 12;
1116 if( decode_luma_residual(h, gb, scan, scan8x8, pixel_shift, mb_type, cbp, 1) < 0 ){
1119 if( decode_luma_residual(h, gb, scan, scan8x8, pixel_shift, mb_type, cbp, 2) < 0 ){
1123 const int num_c8x8 = h->sps.chroma_format_idc;
1126 for(chroma_idx=0; chroma_idx<2; chroma_idx++)
1127 if (decode_residual(h, gb, h->mb + ((256 + 16*16*chroma_idx) << pixel_shift),
1128 CHROMA_DC_BLOCK_INDEX+chroma_idx,
1129 CHROMA422 ? chroma422_dc_scan : chroma_dc_scan,
1130 NULL, 4*num_c8x8) < 0) {
1136 for(chroma_idx=0; chroma_idx<2; chroma_idx++){
1137 const uint32_t *qmul = h->dequant4_coeff[chroma_idx+1+(IS_INTRA( mb_type ) ? 0:3)][h->chroma_qp[chroma_idx]];
1138 int16_t *mb = h->mb + (16*(16 + 16*chroma_idx) << pixel_shift);
1139 for (i8x8=0; i8x8<num_c8x8; i8x8++) {
1140 for (i4x4=0; i4x4<4; i4x4++) {
1141 const int index= 16 + 16*chroma_idx + 8*i8x8 + i4x4;
1142 if (decode_residual(h, gb, mb, index, scan + 1, qmul, 15) < 0)
1144 mb += 16<<pixel_shift;
1149 fill_rectangle(&h->non_zero_count_cache[scan8[16]], 4, 4, 8, 0, 1);
1150 fill_rectangle(&h->non_zero_count_cache[scan8[32]], 4, 4, 8, 0, 1);
1154 fill_rectangle(&h->non_zero_count_cache[scan8[ 0]], 4, 4, 8, 0, 1);
1155 fill_rectangle(&h->non_zero_count_cache[scan8[16]], 4, 4, 8, 0, 1);
1156 fill_rectangle(&h->non_zero_count_cache[scan8[32]], 4, 4, 8, 0, 1);
1158 h->cur_pic.f.qscale_table[mb_xy] = h->qscale;
1159 write_back_non_zero_count(h);