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 Libav.
7 * Libav 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 * Libav 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 Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * H.264 / AVC / MPEG-4 part10 cavlc bitstream decoding.
25 * @author Michael Niedermayer <michaelni@gmx.at>
33 #include "h264_mvpred.h"
35 #include "golomb_legacy.h"
36 #include "mpegutils.h"
40 static const uint8_t golomb_to_inter_cbp_gray[16]={
41 0, 1, 2, 4, 8, 3, 5,10,12,15, 7,11,13,14, 6, 9,
44 static const uint8_t golomb_to_intra4x4_cbp_gray[16]={
45 15, 0, 7,11,13,14, 3, 5,10,12, 1, 2, 4, 8, 6, 9,
48 static const uint8_t chroma_dc_coeff_token_len[4*5]={
56 static const uint8_t chroma_dc_coeff_token_bits[4*5]={
64 static const uint8_t chroma422_dc_coeff_token_len[4*9]={
76 static const uint8_t chroma422_dc_coeff_token_bits[4*9]={
88 static const uint8_t coeff_token_len[4][4*17]={
91 6, 2, 0, 0, 8, 6, 3, 0, 9, 8, 7, 5, 10, 9, 8, 6,
92 11,10, 9, 7, 13,11,10, 8, 13,13,11, 9, 13,13,13,10,
93 14,14,13,11, 14,14,14,13, 15,15,14,14, 15,15,15,14,
94 16,15,15,15, 16,16,16,15, 16,16,16,16, 16,16,16,16,
98 6, 2, 0, 0, 6, 5, 3, 0, 7, 6, 6, 4, 8, 6, 6, 4,
99 8, 7, 7, 5, 9, 8, 8, 6, 11, 9, 9, 6, 11,11,11, 7,
100 12,11,11, 9, 12,12,12,11, 12,12,12,11, 13,13,13,12,
101 13,13,13,13, 13,14,13,13, 14,14,14,13, 14,14,14,14,
105 6, 4, 0, 0, 6, 5, 4, 0, 6, 5, 5, 4, 7, 5, 5, 4,
106 7, 5, 5, 4, 7, 6, 6, 4, 7, 6, 6, 4, 8, 7, 7, 5,
107 8, 8, 7, 6, 9, 8, 8, 7, 9, 9, 8, 8, 9, 9, 9, 8,
108 10, 9, 9, 9, 10,10,10,10, 10,10,10,10, 10,10,10,10,
112 6, 6, 0, 0, 6, 6, 6, 0, 6, 6, 6, 6, 6, 6, 6, 6,
113 6, 6, 6, 6, 6, 6, 6, 6, 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,
119 static const uint8_t coeff_token_bits[4][4*17]={
122 5, 1, 0, 0, 7, 4, 1, 0, 7, 6, 5, 3, 7, 6, 5, 3,
123 7, 6, 5, 4, 15, 6, 5, 4, 11,14, 5, 4, 8,10,13, 4,
124 15,14, 9, 4, 11,10,13,12, 15,14, 9,12, 11,10,13, 8,
125 15, 1, 9,12, 11,14,13, 8, 7,10, 9,12, 4, 6, 5, 8,
129 11, 2, 0, 0, 7, 7, 3, 0, 7,10, 9, 5, 7, 6, 5, 4,
130 4, 6, 5, 6, 7, 6, 5, 8, 15, 6, 5, 4, 11,14,13, 4,
131 15,10, 9, 4, 11,14,13,12, 8,10, 9, 8, 15,14,13,12,
132 11,10, 9,12, 7,11, 6, 8, 9, 8,10, 1, 7, 6, 5, 4,
136 15,14, 0, 0, 11,15,13, 0, 8,12,14,12, 15,10,11,11,
137 11, 8, 9,10, 9,14,13, 9, 8,10, 9, 8, 15,14,13,13,
138 11,14,10,12, 15,10,13,12, 11,14, 9,12, 8,10,13, 8,
139 13, 7, 9,12, 9,12,11,10, 5, 8, 7, 6, 1, 4, 3, 2,
143 0, 1, 0, 0, 4, 5, 6, 0, 8, 9,10,11, 12,13,14,15,
144 16,17,18,19, 20,21,22,23, 24,25,26,27, 28,29,30,31,
145 32,33,34,35, 36,37,38,39, 40,41,42,43, 44,45,46,47,
146 48,49,50,51, 52,53,54,55, 56,57,58,59, 60,61,62,63,
150 static const uint8_t total_zeros_len[16][16]= {
151 {1,3,3,4,4,5,5,6,6,7,7,8,8,9,9,9},
152 {3,3,3,3,3,4,4,4,4,5,5,6,6,6,6},
153 {4,3,3,3,4,4,3,3,4,5,5,6,5,6},
154 {5,3,4,4,3,3,3,4,3,4,5,5,5},
155 {4,4,4,3,3,3,3,3,4,5,4,5},
156 {6,5,3,3,3,3,3,3,4,3,6},
157 {6,5,3,3,3,2,3,4,3,6},
168 static const uint8_t total_zeros_bits[16][16]= {
169 {1,3,2,3,2,3,2,3,2,3,2,3,2,3,2,1},
170 {7,6,5,4,3,5,4,3,2,3,2,3,2,1,0},
171 {5,7,6,5,4,3,4,3,2,3,2,1,1,0},
172 {3,7,5,4,6,5,4,3,3,2,2,1,0},
173 {5,4,3,7,6,5,4,3,2,1,1,0},
174 {1,1,7,6,5,4,3,2,1,1,0},
175 {1,1,5,4,3,3,2,1,1,0},
186 static const uint8_t chroma_dc_total_zeros_len[3][4]= {
192 static const uint8_t chroma_dc_total_zeros_bits[3][4]= {
198 static const uint8_t chroma422_dc_total_zeros_len[7][8]= {
199 { 1, 3, 3, 4, 4, 4, 5, 5 },
200 { 3, 2, 3, 3, 3, 3, 3 },
201 { 3, 3, 2, 2, 3, 3 },
208 static const uint8_t chroma422_dc_total_zeros_bits[7][8]= {
209 { 1, 2, 3, 2, 3, 1, 1, 0 },
210 { 0, 1, 1, 4, 5, 6, 7 },
211 { 0, 1, 1, 2, 6, 7 },
218 static const uint8_t run_len[7][16]={
225 {3,3,3,3,3,3,3,4,5,6,7,8,9,10,11},
228 static const uint8_t run_bits[7][16]={
235 {7,6,5,4,3,2,1,1,1,1,1,1,1,1,1},
238 static VLC coeff_token_vlc[4];
239 static VLC_TYPE coeff_token_vlc_tables[520+332+280+256][2];
240 static const int coeff_token_vlc_tables_size[4]={520,332,280,256};
242 static VLC chroma_dc_coeff_token_vlc;
243 static VLC_TYPE chroma_dc_coeff_token_vlc_table[256][2];
244 static const int chroma_dc_coeff_token_vlc_table_size = 256;
246 static VLC chroma422_dc_coeff_token_vlc;
247 static VLC_TYPE chroma422_dc_coeff_token_vlc_table[8192][2];
248 static const int chroma422_dc_coeff_token_vlc_table_size = 8192;
250 static VLC total_zeros_vlc[15];
251 static VLC_TYPE total_zeros_vlc_tables[15][512][2];
252 static const int total_zeros_vlc_tables_size = 512;
254 static VLC chroma_dc_total_zeros_vlc[3];
255 static VLC_TYPE chroma_dc_total_zeros_vlc_tables[3][8][2];
256 static const int chroma_dc_total_zeros_vlc_tables_size = 8;
258 static VLC chroma422_dc_total_zeros_vlc[7];
259 static VLC_TYPE chroma422_dc_total_zeros_vlc_tables[7][32][2];
260 static const int chroma422_dc_total_zeros_vlc_tables_size = 32;
262 static VLC run_vlc[6];
263 static VLC_TYPE run_vlc_tables[6][8][2];
264 static const int run_vlc_tables_size = 8;
267 static VLC_TYPE run7_vlc_table[96][2];
268 static const int run7_vlc_table_size = 96;
270 #define LEVEL_TAB_BITS 8
271 static int8_t cavlc_level_tab[7][1<<LEVEL_TAB_BITS][2];
273 #define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
274 #define CHROMA422_DC_COEFF_TOKEN_VLC_BITS 13
275 #define COEFF_TOKEN_VLC_BITS 8
276 #define TOTAL_ZEROS_VLC_BITS 9
277 #define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
278 #define CHROMA422_DC_TOTAL_ZEROS_VLC_BITS 5
279 #define RUN_VLC_BITS 3
280 #define RUN7_VLC_BITS 6
283 * Get the predicted number of non-zero coefficients.
284 * @param n block index
286 static inline int pred_non_zero_count(const H264Context *h, H264SliceContext *sl, int n)
288 const int index8= scan8[n];
289 const int left = sl->non_zero_count_cache[index8 - 1];
290 const int top = sl->non_zero_count_cache[index8 - 8];
293 if(i<64) i= (i+1)>>1;
295 ff_tlog(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 assert(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();
415 static inline int get_level_prefix(GetBitContext *gb){
420 UPDATE_CACHE(re, gb);
421 buf=GET_CACHE(re, gb);
423 log= 32 - av_log2(buf);
425 LAST_SKIP_BITS(re, gb, log);
426 CLOSE_READER(re, gb);
432 * Decode a residual block.
433 * @param n block index
434 * @param scantable scantable
435 * @param max_coeff number of coefficients in the block
436 * @return <0 if an error occurred
438 static int decode_residual(const H264Context *h, H264SliceContext *sl,
439 GetBitContext *gb, int16_t *block, int n,
440 const uint8_t *scantable, const uint32_t *qmul,
443 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};
445 int zeros_left, coeff_token, total_coeff, i, trailing_ones, run_before;
447 //FIXME put trailing_onex into the context
451 coeff_token = get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
453 coeff_token = get_vlc2(gb, chroma422_dc_coeff_token_vlc.table, CHROMA422_DC_COEFF_TOKEN_VLC_BITS, 1);
454 total_coeff= coeff_token>>2;
456 if(n >= LUMA_DC_BLOCK_INDEX){
457 total_coeff= pred_non_zero_count(h, sl, (n - LUMA_DC_BLOCK_INDEX)*16);
458 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
459 total_coeff= coeff_token>>2;
461 total_coeff= pred_non_zero_count(h, sl, n);
462 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
463 total_coeff= coeff_token>>2;
466 sl->non_zero_count_cache[scan8[n]] = total_coeff;
468 //FIXME set last_non_zero?
472 if(total_coeff > (unsigned)max_coeff) {
473 av_log(h->avctx, AV_LOG_ERROR, "corrupted macroblock %d %d (total_coeff=%d)\n", sl->mb_x, sl->mb_y, total_coeff);
477 trailing_ones= coeff_token&3;
478 ff_tlog(h->avctx, "trailing:%d, total:%d\n", trailing_ones, total_coeff);
479 assert(total_coeff<=16);
481 i = show_bits(gb, 3);
482 skip_bits(gb, trailing_ones);
483 level[0] = 1-((i&4)>>1);
484 level[1] = 1-((i&2) );
485 level[2] = 1-((i&1)<<1);
487 if(trailing_ones<total_coeff) {
489 int suffix_length = total_coeff > 10 & trailing_ones < 3;
490 int bitsi= show_bits(gb, LEVEL_TAB_BITS);
491 int level_code= cavlc_level_tab[suffix_length][bitsi][0];
493 skip_bits(gb, cavlc_level_tab[suffix_length][bitsi][1]);
494 if(level_code >= 100){
495 prefix= level_code - 100;
496 if(prefix == LEVEL_TAB_BITS)
497 prefix += get_level_prefix(gb);
499 //first coefficient has suffix_length equal to 0 or 1
500 if(prefix<14){ //FIXME try to build a large unified VLC table for all this
502 level_code= (prefix<<1) + get_bits1(gb); //part
504 level_code= prefix; //part
505 }else if(prefix==14){
507 level_code= (prefix<<1) + get_bits1(gb); //part
509 level_code= prefix + get_bits(gb, 4); //part
511 level_code= 30 + get_bits(gb, prefix-3); //part
514 av_log(h->avctx, AV_LOG_ERROR, "Invalid level prefix\n");
517 level_code += (1<<(prefix-3))-4096;
521 if(trailing_ones < 3) level_code += 2;
524 mask= -(level_code&1);
525 level[trailing_ones]= (((2+level_code)>>1) ^ mask) - mask;
527 level_code += ((level_code>>31)|1) & -(trailing_ones < 3);
529 suffix_length = 1 + (level_code + 3U > 6U);
530 level[trailing_ones]= level_code;
533 //remaining coefficients have suffix_length > 0
534 for(i=trailing_ones+1;i<total_coeff;i++) {
535 static const unsigned int suffix_limit[7] = {0,3,6,12,24,48,INT_MAX };
536 int bitsi= show_bits(gb, LEVEL_TAB_BITS);
537 level_code= cavlc_level_tab[suffix_length][bitsi][0];
539 skip_bits(gb, cavlc_level_tab[suffix_length][bitsi][1]);
540 if(level_code >= 100){
541 prefix= level_code - 100;
542 if(prefix == LEVEL_TAB_BITS){
543 prefix += get_level_prefix(gb);
546 level_code = (prefix<<suffix_length) + get_bits(gb, suffix_length);
548 level_code = (15<<suffix_length) + get_bits(gb, prefix-3);
550 level_code += (1<<(prefix-3))-4096;
552 mask= -(level_code&1);
553 level_code= (((2+level_code)>>1) ^ mask) - mask;
555 level[i]= level_code;
556 suffix_length+= suffix_limit[suffix_length] + level_code > 2U*suffix_limit[suffix_length];
560 if(total_coeff == max_coeff)
563 if (max_coeff <= 8) {
565 zeros_left = get_vlc2(gb, chroma_dc_total_zeros_vlc[total_coeff - 1].table,
566 CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
568 zeros_left = get_vlc2(gb, chroma422_dc_total_zeros_vlc[total_coeff - 1].table,
569 CHROMA422_DC_TOTAL_ZEROS_VLC_BITS, 1);
571 zeros_left= get_vlc2(gb, total_zeros_vlc[total_coeff - 1].table, TOTAL_ZEROS_VLC_BITS, 1);
575 #define STORE_BLOCK(type) \
576 scantable += zeros_left + total_coeff - 1; \
577 if(n >= LUMA_DC_BLOCK_INDEX){ \
578 ((type*)block)[*scantable] = level[0]; \
579 for(i=1;i<total_coeff && zeros_left > 0;i++) { \
581 run_before= get_vlc2(gb, run_vlc[zeros_left - 1].table, RUN_VLC_BITS, 1); \
583 run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2); \
584 run_before = FFMIN(zeros_left, run_before);\
586 zeros_left -= run_before; \
587 scantable -= 1 + run_before; \
588 ((type*)block)[*scantable]= level[i]; \
590 for(;i<total_coeff;i++) { \
592 ((type*)block)[*scantable]= level[i]; \
595 ((type*)block)[*scantable] = ((int)(level[0] * qmul[*scantable] + 32))>>6; \
596 for(i=1;i<total_coeff && zeros_left > 0;i++) { \
598 run_before= get_vlc2(gb, run_vlc[zeros_left - 1].table, RUN_VLC_BITS, 1); \
600 run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2); \
601 run_before = FFMIN(zeros_left, run_before);\
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 (zeros_left < 0) {
614 av_log(h->avctx, AV_LOG_ERROR,
615 "negative number of zero coeffs at %d %d\n", sl->mb_x, sl->mb_y);
616 return AVERROR_INVALIDDATA;
619 if (h->pixel_shift) {
628 static av_always_inline
629 int decode_luma_residual(const H264Context *h, H264SliceContext *sl,
630 GetBitContext *gb, const uint8_t *scan,
631 const uint8_t *scan8x8, int pixel_shift,
632 int mb_type, int cbp, int p)
635 int qscale = p == 0 ? sl->qscale : sl->chroma_qp[p - 1];
636 if(IS_INTRA16x16(mb_type)){
637 AV_ZERO128(sl->mb_luma_dc[p]+0);
638 AV_ZERO128(sl->mb_luma_dc[p]+8);
639 AV_ZERO128(sl->mb_luma_dc[p]+16);
640 AV_ZERO128(sl->mb_luma_dc[p]+24);
641 if (decode_residual(h, sl, gb, sl->mb_luma_dc[p], LUMA_DC_BLOCK_INDEX + p, scan, NULL, 16) < 0) {
642 return -1; //FIXME continue if partitioned and other return -1 too
645 assert((cbp&15) == 0 || (cbp&15) == 15);
648 for(i8x8=0; i8x8<4; i8x8++){
649 for(i4x4=0; i4x4<4; i4x4++){
650 const int index= i4x4 + 4*i8x8 + p*16;
651 if( decode_residual(h, sl, gb, sl->mb + (16*index << pixel_shift),
652 index, scan + 1, h->ps.pps->dequant4_coeff[p][qscale], 15) < 0 ){
659 fill_rectangle(&sl->non_zero_count_cache[scan8[p*16]], 4, 4, 8, 0, 1);
663 int cqm = (IS_INTRA( mb_type ) ? 0:3)+p;
664 /* For CAVLC 4:4:4, we need to keep track of the luma 8x8 CBP for deblocking nnz purposes. */
666 for(i8x8=0; i8x8<4; i8x8++){
668 if(IS_8x8DCT(mb_type)){
669 int16_t *buf = &sl->mb[64*i8x8+256*p << pixel_shift];
671 for(i4x4=0; i4x4<4; i4x4++){
672 const int index= i4x4 + 4*i8x8 + p*16;
673 if( decode_residual(h, sl, gb, buf, index, scan8x8+16*i4x4,
674 h->ps.pps->dequant8_coeff[cqm][qscale], 16) < 0 )
677 nnz = &sl->non_zero_count_cache[scan8[4 * i8x8 + p * 16]];
678 nnz[0] += nnz[1] + nnz[8] + nnz[9];
679 new_cbp |= !!nnz[0] << i8x8;
681 for(i4x4=0; i4x4<4; i4x4++){
682 const int index= i4x4 + 4*i8x8 + p*16;
683 if( decode_residual(h, sl, gb, sl->mb + (16*index << pixel_shift), index,
684 scan, h->ps.pps->dequant4_coeff[cqm][qscale], 16) < 0 ){
687 new_cbp |= sl->non_zero_count_cache[scan8[index]] << i8x8;
691 uint8_t * const nnz = &sl->non_zero_count_cache[scan8[4 * i8x8 + p * 16]];
692 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
699 int ff_h264_decode_mb_cavlc(const H264Context *h, H264SliceContext *sl)
703 unsigned int mb_type, cbp;
704 int dct8x8_allowed= h->ps.pps->transform_8x8_mode;
705 int decode_chroma = h->ps.sps->chroma_format_idc == 1 || h->ps.sps->chroma_format_idc == 2;
706 const int pixel_shift = h->pixel_shift;
708 mb_xy = sl->mb_xy = sl->mb_x + sl->mb_y*h->mb_stride;
710 cbp = 0; /* avoid warning. FIXME: find a solution without slowing
712 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
713 if (sl->mb_skip_run == -1)
714 sl->mb_skip_run = get_ue_golomb(&sl->gb);
716 if (sl->mb_skip_run--) {
717 if (FRAME_MBAFF(h) && (sl->mb_y & 1) == 0) {
718 if (sl->mb_skip_run == 0)
719 sl->mb_mbaff = sl->mb_field_decoding_flag = get_bits1(&sl->gb);
721 decode_mb_skip(h, sl);
725 if (FRAME_MBAFF(h)) {
726 if ((sl->mb_y & 1) == 0)
727 sl->mb_mbaff = sl->mb_field_decoding_flag = get_bits1(&sl->gb);
730 sl->prev_mb_skipped = 0;
732 mb_type= get_ue_golomb(&sl->gb);
733 if (sl->slice_type_nos == AV_PICTURE_TYPE_B) {
735 partition_count = ff_h264_b_mb_type_info[mb_type].partition_count;
736 mb_type = ff_h264_b_mb_type_info[mb_type].type;
739 goto decode_intra_mb;
741 } else if (sl->slice_type_nos == AV_PICTURE_TYPE_P) {
743 partition_count = ff_h264_p_mb_type_info[mb_type].partition_count;
744 mb_type = ff_h264_p_mb_type_info[mb_type].type;
747 goto decode_intra_mb;
750 assert(sl->slice_type_nos == AV_PICTURE_TYPE_I);
751 if (sl->slice_type == AV_PICTURE_TYPE_SI && mb_type)
755 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(sl->slice_type), sl->mb_x, sl->mb_y);
759 cbp = ff_h264_i_mb_type_info[mb_type].cbp;
760 sl->intra16x16_pred_mode = ff_h264_i_mb_type_info[mb_type].pred_mode;
761 mb_type = ff_h264_i_mb_type_info[mb_type].type;
765 mb_type |= MB_TYPE_INTERLACED;
767 h->slice_table[mb_xy] = sl->slice_num;
769 if(IS_INTRA_PCM(mb_type)){
770 const int mb_size = ff_h264_mb_sizes[h->ps.sps->chroma_format_idc] *
771 h->ps.sps->bit_depth_luma;
773 // We assume these blocks are very rare so we do not optimize it.
774 sl->intra_pcm_ptr = align_get_bits(&sl->gb);
775 if (get_bits_left(&sl->gb) < mb_size) {
776 av_log(h->avctx, AV_LOG_ERROR, "Not enough data for an intra PCM block.\n");
777 return AVERROR_INVALIDDATA;
779 skip_bits_long(&sl->gb, mb_size);
781 // In deblocking, the quantizer is 0
782 h->cur_pic.qscale_table[mb_xy] = 0;
783 // All coeffs are present
784 memset(h->non_zero_count[mb_xy], 16, 48);
786 h->cur_pic.mb_type[mb_xy] = mb_type;
790 fill_decode_neighbors(h, sl, mb_type);
791 fill_decode_caches(h, sl, mb_type);
794 if(IS_INTRA(mb_type)){
796 // init_top_left_availability(h);
797 if(IS_INTRA4x4(mb_type)){
800 if(dct8x8_allowed && get_bits1(&sl->gb)){
801 mb_type |= MB_TYPE_8x8DCT;
805 // fill_intra4x4_pred_table(h);
806 for(i=0; i<16; i+=di){
807 int mode = pred_intra_mode(h, sl, i);
809 if(!get_bits1(&sl->gb)){
810 const int rem_mode= get_bits(&sl->gb, 3);
811 mode = rem_mode + (rem_mode >= mode);
815 fill_rectangle(&sl->intra4x4_pred_mode_cache[ scan8[i] ], 2, 2, 8, mode, 1);
817 sl->intra4x4_pred_mode_cache[scan8[i]] = mode;
819 write_back_intra_pred_mode(h, sl);
820 if (ff_h264_check_intra4x4_pred_mode(sl->intra4x4_pred_mode_cache, h->avctx,
821 sl->top_samples_available, sl->left_samples_available) < 0)
824 sl->intra16x16_pred_mode = ff_h264_check_intra_pred_mode(h->avctx, sl->top_samples_available,
825 sl->left_samples_available, sl->intra16x16_pred_mode, 0);
826 if (sl->intra16x16_pred_mode < 0)
830 pred_mode= ff_h264_check_intra_pred_mode(h->avctx, sl->top_samples_available,
831 sl->left_samples_available, get_ue_golomb_31(&sl->gb), 1);
834 sl->chroma_pred_mode = pred_mode;
836 sl->chroma_pred_mode = DC_128_PRED8x8;
838 }else if(partition_count==4){
839 int i, j, sub_partition_count[4], list, ref[2][4];
841 if (sl->slice_type_nos == AV_PICTURE_TYPE_B) {
843 sl->sub_mb_type[i]= get_ue_golomb_31(&sl->gb);
844 if(sl->sub_mb_type[i] >=13){
845 av_log(h->avctx, AV_LOG_ERROR, "B sub_mb_type %u out of range at %d %d\n", sl->sub_mb_type[i], sl->mb_x, sl->mb_y);
848 sub_partition_count[i] = ff_h264_b_sub_mb_type_info[sl->sub_mb_type[i]].partition_count;
849 sl->sub_mb_type[i] = ff_h264_b_sub_mb_type_info[sl->sub_mb_type[i]].type;
851 if( IS_DIRECT(sl->sub_mb_type[0]|sl->sub_mb_type[1]|sl->sub_mb_type[2]|sl->sub_mb_type[3])) {
852 ff_h264_pred_direct_motion(h, sl, &mb_type);
853 sl->ref_cache[0][scan8[4]] =
854 sl->ref_cache[1][scan8[4]] =
855 sl->ref_cache[0][scan8[12]] =
856 sl->ref_cache[1][scan8[12]] = PART_NOT_AVAILABLE;
859 assert(sl->slice_type_nos == AV_PICTURE_TYPE_P); //FIXME SP correct ?
861 sl->sub_mb_type[i]= get_ue_golomb_31(&sl->gb);
862 if(sl->sub_mb_type[i] >=4){
863 av_log(h->avctx, AV_LOG_ERROR, "P sub_mb_type %u out of range at %d %d\n", sl->sub_mb_type[i], sl->mb_x, sl->mb_y);
866 sub_partition_count[i] = ff_h264_p_sub_mb_type_info[sl->sub_mb_type[i]].partition_count;
867 sl->sub_mb_type[i] = ff_h264_p_sub_mb_type_info[sl->sub_mb_type[i]].type;
871 for (list = 0; list < sl->list_count; list++) {
872 int ref_count = IS_REF0(mb_type) ? 1 : sl->ref_count[list] << MB_MBAFF(sl);
874 if(IS_DIRECT(sl->sub_mb_type[i])) continue;
875 if(IS_DIR(sl->sub_mb_type[i], 0, list)){
879 }else if(ref_count == 2){
880 tmp= get_bits1(&sl->gb)^1;
882 tmp= get_ue_golomb_31(&sl->gb);
884 av_log(h->avctx, AV_LOG_ERROR, "ref %u overflow\n", tmp);
897 dct8x8_allowed = get_dct8x8_allowed(h, sl);
899 for (list = 0; list < sl->list_count; list++) {
901 if(IS_DIRECT(sl->sub_mb_type[i])) {
902 sl->ref_cache[list][ scan8[4*i] ] = sl->ref_cache[list][ scan8[4*i]+1 ];
905 sl->ref_cache[list][ scan8[4*i] ]=sl->ref_cache[list][ scan8[4*i]+1 ]=
906 sl->ref_cache[list][ scan8[4*i]+8 ]=sl->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
908 if(IS_DIR(sl->sub_mb_type[i], 0, list)){
909 const int sub_mb_type= sl->sub_mb_type[i];
910 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
911 for(j=0; j<sub_partition_count[i]; j++){
913 const int index= 4*i + block_width*j;
914 int16_t (* mv_cache)[2]= &sl->mv_cache[list][ scan8[index] ];
915 pred_motion(h, sl, index, block_width, list, sl->ref_cache[list][ scan8[index] ], &mx, &my);
916 mx += get_se_golomb(&sl->gb);
917 my += get_se_golomb(&sl->gb);
918 ff_tlog(h->avctx, "final mv:%d %d\n", mx, my);
920 if(IS_SUB_8X8(sub_mb_type)){
922 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
924 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
925 }else if(IS_SUB_8X4(sub_mb_type)){
926 mv_cache[ 1 ][0]= mx;
927 mv_cache[ 1 ][1]= my;
928 }else if(IS_SUB_4X8(sub_mb_type)){
929 mv_cache[ 8 ][0]= mx;
930 mv_cache[ 8 ][1]= my;
932 mv_cache[ 0 ][0]= mx;
933 mv_cache[ 0 ][1]= my;
936 uint32_t *p= (uint32_t *)&sl->mv_cache[list][ scan8[4*i] ][0];
942 }else if(IS_DIRECT(mb_type)){
943 ff_h264_pred_direct_motion(h, sl, &mb_type);
944 dct8x8_allowed &= h->ps.sps->direct_8x8_inference_flag;
947 //FIXME we should set ref_idx_l? to 0 if we use that later ...
948 if(IS_16X16(mb_type)){
949 for (list = 0; list < sl->list_count; list++) {
951 if(IS_DIR(mb_type, 0, list)){
952 int rc = sl->ref_count[list] << MB_MBAFF(sl);
955 } else if (rc == 2) {
956 val= get_bits1(&sl->gb)^1;
958 val= get_ue_golomb_31(&sl->gb);
960 av_log(h->avctx, AV_LOG_ERROR, "ref %u overflow\n", val);
964 fill_rectangle(&sl->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
967 for (list = 0; list < sl->list_count; list++) {
968 if(IS_DIR(mb_type, 0, list)){
969 pred_motion(h, sl, 0, 4, list, sl->ref_cache[list][ scan8[0] ], &mx, &my);
970 mx += get_se_golomb(&sl->gb);
971 my += get_se_golomb(&sl->gb);
972 ff_tlog(h->avctx, "final mv:%d %d\n", mx, my);
974 fill_rectangle(sl->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
978 else if(IS_16X8(mb_type)){
979 for (list = 0; list < sl->list_count; list++) {
982 if(IS_DIR(mb_type, i, list)){
983 int rc = sl->ref_count[list] << MB_MBAFF(sl);
986 } else if (rc == 2) {
987 val= get_bits1(&sl->gb)^1;
989 val= get_ue_golomb_31(&sl->gb);
991 av_log(h->avctx, AV_LOG_ERROR, "ref %u overflow\n", val);
996 val= LIST_NOT_USED&0xFF;
997 fill_rectangle(&sl->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
1000 for (list = 0; list < sl->list_count; list++) {
1003 if(IS_DIR(mb_type, i, list)){
1004 pred_16x8_motion(h, sl, 8*i, list, sl->ref_cache[list][scan8[0] + 16*i], &mx, &my);
1005 mx += get_se_golomb(&sl->gb);
1006 my += get_se_golomb(&sl->gb);
1007 ff_tlog(h->avctx, "final mv:%d %d\n", mx, my);
1009 val= pack16to32(mx,my);
1012 fill_rectangle(sl->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 4);
1016 assert(IS_8X16(mb_type));
1017 for (list = 0; list < sl->list_count; list++) {
1020 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
1021 int rc = sl->ref_count[list] << MB_MBAFF(sl);
1024 } else if (rc == 2) {
1025 val= get_bits1(&sl->gb)^1;
1027 val= get_ue_golomb_31(&sl->gb);
1029 av_log(h->avctx, AV_LOG_ERROR, "ref %u overflow\n", val);
1034 val= LIST_NOT_USED&0xFF;
1035 fill_rectangle(&sl->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
1038 for (list = 0; list < sl->list_count; list++) {
1041 if(IS_DIR(mb_type, i, list)){
1042 pred_8x16_motion(h, sl, i*4, list, sl->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
1043 mx += get_se_golomb(&sl->gb);
1044 my += get_se_golomb(&sl->gb);
1045 ff_tlog(h->avctx, "final mv:%d %d\n", mx, my);
1047 val= pack16to32(mx,my);
1050 fill_rectangle(sl->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 4);
1056 if(IS_INTER(mb_type))
1057 write_back_motion(h, sl, mb_type);
1059 if(!IS_INTRA16x16(mb_type)){
1060 cbp= get_ue_golomb(&sl->gb);
1064 av_log(h->avctx, AV_LOG_ERROR, "cbp too large (%u) at %d %d\n", cbp, sl->mb_x, sl->mb_y);
1067 if (IS_INTRA4x4(mb_type))
1068 cbp = ff_h264_golomb_to_intra4x4_cbp[cbp];
1070 cbp = ff_h264_golomb_to_inter_cbp[cbp];
1073 av_log(h->avctx, AV_LOG_ERROR, "cbp too large (%u) at %d %d\n", cbp, sl->mb_x, sl->mb_y);
1076 if(IS_INTRA4x4(mb_type)) cbp= golomb_to_intra4x4_cbp_gray[cbp];
1077 else cbp= golomb_to_inter_cbp_gray[cbp];
1081 if(dct8x8_allowed && (cbp&15) && !IS_INTRA(mb_type)){
1082 mb_type |= MB_TYPE_8x8DCT*get_bits1(&sl->gb);
1085 h->cbp_table[mb_xy]= cbp;
1086 h->cur_pic.mb_type[mb_xy] = mb_type;
1088 if(cbp || IS_INTRA16x16(mb_type)){
1089 int i4x4, i8x8, chroma_idx;
1092 GetBitContext *gb = &sl->gb;
1093 const uint8_t *scan, *scan8x8;
1094 const int max_qp = 51 + 6 * (h->ps.sps->bit_depth_luma - 8);
1096 dquant= get_se_golomb(&sl->gb);
1098 sl->qscale += dquant;
1100 if (((unsigned)sl->qscale) > max_qp){
1101 if (sl->qscale < 0) sl->qscale += max_qp + 1;
1102 else sl->qscale -= max_qp+1;
1103 if (((unsigned)sl->qscale) > max_qp){
1104 av_log(h->avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, sl->mb_x, sl->mb_y);
1109 sl->chroma_qp[0] = get_chroma_qp(h->ps.pps, 0, sl->qscale);
1110 sl->chroma_qp[1] = get_chroma_qp(h->ps.pps, 1, sl->qscale);
1112 if(IS_INTERLACED(mb_type)){
1113 scan8x8 = sl->qscale ? h->field_scan8x8_cavlc : h->field_scan8x8_cavlc_q0;
1114 scan = sl->qscale ? h->field_scan : h->field_scan_q0;
1116 scan8x8 = sl->qscale ? h->zigzag_scan8x8_cavlc : h->zigzag_scan8x8_cavlc_q0;
1117 scan = sl->qscale ? h->zigzag_scan : h->zigzag_scan_q0;
1120 if ((ret = decode_luma_residual(h, sl, gb, scan, scan8x8, pixel_shift, mb_type, cbp, 0)) < 0 ) {
1123 h->cbp_table[mb_xy] |= ret << 12;
1125 if (decode_luma_residual(h, sl, gb, scan, scan8x8, pixel_shift, mb_type, cbp, 1) < 0 ) {
1128 if (decode_luma_residual(h, sl, gb, scan, scan8x8, pixel_shift, mb_type, cbp, 2) < 0 ) {
1131 } else if (CHROMA422(h)) {
1133 for(chroma_idx=0; chroma_idx<2; chroma_idx++)
1134 if (decode_residual(h, sl, gb, sl->mb + ((256 + 16*16*chroma_idx) << pixel_shift),
1135 CHROMA_DC_BLOCK_INDEX + chroma_idx, ff_h264_chroma422_dc_scan,
1142 for(chroma_idx=0; chroma_idx<2; chroma_idx++){
1143 const uint32_t *qmul = h->ps.pps->dequant4_coeff[chroma_idx+1+(IS_INTRA( mb_type ) ? 0:3)][sl->chroma_qp[chroma_idx]];
1144 int16_t *mb = sl->mb + (16*(16 + 16*chroma_idx) << pixel_shift);
1145 for (i8x8 = 0; i8x8 < 2; i8x8++) {
1146 for (i4x4 = 0; i4x4 < 4; i4x4++) {
1147 const int index = 16 + 16*chroma_idx + 8*i8x8 + i4x4;
1148 if (decode_residual(h, sl, gb, mb, index, scan + 1, qmul, 15) < 0)
1150 mb += 16 << pixel_shift;
1155 fill_rectangle(&sl->non_zero_count_cache[scan8[16]], 4, 4, 8, 0, 1);
1156 fill_rectangle(&sl->non_zero_count_cache[scan8[32]], 4, 4, 8, 0, 1);
1158 } else /* yuv420 */ {
1160 for(chroma_idx=0; chroma_idx<2; chroma_idx++)
1161 if (decode_residual(h, sl, gb, sl->mb + ((256 + 16 * 16 * chroma_idx) << pixel_shift),
1162 CHROMA_DC_BLOCK_INDEX + chroma_idx, ff_h264_chroma_dc_scan, NULL, 4) < 0) {
1168 for(chroma_idx=0; chroma_idx<2; chroma_idx++){
1169 const uint32_t *qmul = h->ps.pps->dequant4_coeff[chroma_idx+1+(IS_INTRA( mb_type ) ? 0:3)][sl->chroma_qp[chroma_idx]];
1170 for(i4x4=0; i4x4<4; i4x4++){
1171 const int index= 16 + 16*chroma_idx + i4x4;
1172 if( decode_residual(h, sl, gb, sl->mb + (16*index << pixel_shift), index, scan + 1, qmul, 15) < 0){
1178 fill_rectangle(&sl->non_zero_count_cache[scan8[16]], 4, 4, 8, 0, 1);
1179 fill_rectangle(&sl->non_zero_count_cache[scan8[32]], 4, 4, 8, 0, 1);
1183 fill_rectangle(&sl->non_zero_count_cache[scan8[ 0]], 4, 4, 8, 0, 1);
1184 fill_rectangle(&sl->non_zero_count_cache[scan8[16]], 4, 4, 8, 0, 1);
1185 fill_rectangle(&sl->non_zero_count_cache[scan8[32]], 4, 4, 8, 0, 1);
1187 h->cur_pic.qscale_table[mb_xy] = sl->qscale;
1188 write_back_non_zero_count(h, sl);