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
34 #include "h264data.h" // FIXME FIXME FIXME
35 #include "h264_mvpred.h"
37 #include "mpegutils.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, H264SliceContext *sl, int n)
289 const int index8= scan8[n];
290 const int left = sl->non_zero_count_cache[index8 - 1];
291 const int top = sl->non_zero_count_cache[index8 - 8];
294 if(i<64) i= (i+1)>>1;
296 tprintf(h->avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
301 static av_cold void init_cavlc_level_tab(void){
305 for(suffix_length=0; suffix_length<7; suffix_length++){
306 for(i=0; i<(1<<LEVEL_TAB_BITS); i++){
307 int prefix= LEVEL_TAB_BITS - av_log2(2*i);
309 if(prefix + 1 + suffix_length <= LEVEL_TAB_BITS){
310 int level_code = (prefix << suffix_length) +
311 (i >> (av_log2(i) - suffix_length)) - (1 << suffix_length);
312 int mask = -(level_code&1);
313 level_code = (((2 + level_code) >> 1) ^ mask) - mask;
314 cavlc_level_tab[suffix_length][i][0]= level_code;
315 cavlc_level_tab[suffix_length][i][1]= prefix + 1 + suffix_length;
316 }else if(prefix + 1 <= LEVEL_TAB_BITS){
317 cavlc_level_tab[suffix_length][i][0]= prefix+100;
318 cavlc_level_tab[suffix_length][i][1]= prefix + 1;
320 cavlc_level_tab[suffix_length][i][0]= LEVEL_TAB_BITS+100;
321 cavlc_level_tab[suffix_length][i][1]= LEVEL_TAB_BITS;
327 av_cold void ff_h264_decode_init_vlc(void){
335 chroma_dc_coeff_token_vlc.table = chroma_dc_coeff_token_vlc_table;
336 chroma_dc_coeff_token_vlc.table_allocated = chroma_dc_coeff_token_vlc_table_size;
337 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
338 &chroma_dc_coeff_token_len [0], 1, 1,
339 &chroma_dc_coeff_token_bits[0], 1, 1,
340 INIT_VLC_USE_NEW_STATIC);
342 chroma422_dc_coeff_token_vlc.table = chroma422_dc_coeff_token_vlc_table;
343 chroma422_dc_coeff_token_vlc.table_allocated = chroma422_dc_coeff_token_vlc_table_size;
344 init_vlc(&chroma422_dc_coeff_token_vlc, CHROMA422_DC_COEFF_TOKEN_VLC_BITS, 4*9,
345 &chroma422_dc_coeff_token_len [0], 1, 1,
346 &chroma422_dc_coeff_token_bits[0], 1, 1,
347 INIT_VLC_USE_NEW_STATIC);
351 coeff_token_vlc[i].table = coeff_token_vlc_tables+offset;
352 coeff_token_vlc[i].table_allocated = coeff_token_vlc_tables_size[i];
353 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
354 &coeff_token_len [i][0], 1, 1,
355 &coeff_token_bits[i][0], 1, 1,
356 INIT_VLC_USE_NEW_STATIC);
357 offset += coeff_token_vlc_tables_size[i];
360 * This is a one time safety check to make sure that
361 * the packed static coeff_token_vlc table sizes
362 * were initialized correctly.
364 av_assert0(offset == FF_ARRAY_ELEMS(coeff_token_vlc_tables));
367 chroma_dc_total_zeros_vlc[i].table = chroma_dc_total_zeros_vlc_tables[i];
368 chroma_dc_total_zeros_vlc[i].table_allocated = chroma_dc_total_zeros_vlc_tables_size;
369 init_vlc(&chroma_dc_total_zeros_vlc[i],
370 CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
371 &chroma_dc_total_zeros_len [i][0], 1, 1,
372 &chroma_dc_total_zeros_bits[i][0], 1, 1,
373 INIT_VLC_USE_NEW_STATIC);
377 chroma422_dc_total_zeros_vlc[i].table = chroma422_dc_total_zeros_vlc_tables[i];
378 chroma422_dc_total_zeros_vlc[i].table_allocated = chroma422_dc_total_zeros_vlc_tables_size;
379 init_vlc(&chroma422_dc_total_zeros_vlc[i],
380 CHROMA422_DC_TOTAL_ZEROS_VLC_BITS, 8,
381 &chroma422_dc_total_zeros_len [i][0], 1, 1,
382 &chroma422_dc_total_zeros_bits[i][0], 1, 1,
383 INIT_VLC_USE_NEW_STATIC);
387 total_zeros_vlc[i].table = total_zeros_vlc_tables[i];
388 total_zeros_vlc[i].table_allocated = total_zeros_vlc_tables_size;
389 init_vlc(&total_zeros_vlc[i],
390 TOTAL_ZEROS_VLC_BITS, 16,
391 &total_zeros_len [i][0], 1, 1,
392 &total_zeros_bits[i][0], 1, 1,
393 INIT_VLC_USE_NEW_STATIC);
397 run_vlc[i].table = run_vlc_tables[i];
398 run_vlc[i].table_allocated = run_vlc_tables_size;
399 init_vlc(&run_vlc[i],
401 &run_len [i][0], 1, 1,
402 &run_bits[i][0], 1, 1,
403 INIT_VLC_USE_NEW_STATIC);
405 run7_vlc.table = run7_vlc_table,
406 run7_vlc.table_allocated = run7_vlc_table_size;
407 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
408 &run_len [6][0], 1, 1,
409 &run_bits[6][0], 1, 1,
410 INIT_VLC_USE_NEW_STATIC);
412 init_cavlc_level_tab();
419 static inline int get_level_prefix(GetBitContext *gb){
424 UPDATE_CACHE(re, gb);
425 buf=GET_CACHE(re, gb);
427 log= 32 - av_log2(buf);
429 print_bin(buf>>(32-log), log);
430 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__);
433 LAST_SKIP_BITS(re, gb, log);
434 CLOSE_READER(re, gb);
440 * Decode a residual block.
441 * @param n block index
442 * @param scantable scantable
443 * @param max_coeff number of coefficients in the block
444 * @return <0 if an error occurred
446 static int decode_residual(H264Context *h, H264SliceContext *sl,
447 GetBitContext *gb, int16_t *block, int n,
448 const uint8_t *scantable, const uint32_t *qmul,
451 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};
453 int zeros_left, coeff_token, total_coeff, i, trailing_ones, run_before;
455 //FIXME put trailing_onex into the context
459 coeff_token = get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
461 coeff_token = get_vlc2(gb, chroma422_dc_coeff_token_vlc.table, CHROMA422_DC_COEFF_TOKEN_VLC_BITS, 1);
462 total_coeff= coeff_token>>2;
464 if(n >= LUMA_DC_BLOCK_INDEX){
465 total_coeff= pred_non_zero_count(h, sl, (n - LUMA_DC_BLOCK_INDEX)*16);
466 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
467 total_coeff= coeff_token>>2;
469 total_coeff= pred_non_zero_count(h, sl, n);
470 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
471 total_coeff= coeff_token>>2;
474 sl->non_zero_count_cache[scan8[n]] = total_coeff;
476 //FIXME set last_non_zero?
480 if(total_coeff > (unsigned)max_coeff) {
481 av_log(h->avctx, AV_LOG_ERROR, "corrupted macroblock %d %d (total_coeff=%d)\n", h->mb_x, h->mb_y, total_coeff);
485 trailing_ones= coeff_token&3;
486 tprintf(h->avctx, "trailing:%d, total:%d\n", trailing_ones, total_coeff);
487 av_assert2(total_coeff<=16);
489 i = show_bits(gb, 3);
490 skip_bits(gb, trailing_ones);
491 level[0] = 1-((i&4)>>1);
492 level[1] = 1-((i&2) );
493 level[2] = 1-((i&1)<<1);
495 if(trailing_ones<total_coeff) {
497 int suffix_length = total_coeff > 10 & trailing_ones < 3;
498 int bitsi= show_bits(gb, LEVEL_TAB_BITS);
499 int level_code= cavlc_level_tab[suffix_length][bitsi][0];
501 skip_bits(gb, cavlc_level_tab[suffix_length][bitsi][1]);
502 if(level_code >= 100){
503 prefix= level_code - 100;
504 if(prefix == LEVEL_TAB_BITS)
505 prefix += get_level_prefix(gb);
507 //first coefficient has suffix_length equal to 0 or 1
508 if(prefix<14){ //FIXME try to build a large unified VLC table for all this
510 level_code= (prefix<<1) + get_bits1(gb); //part
512 level_code= prefix; //part
513 }else if(prefix==14){
515 level_code= (prefix<<1) + get_bits1(gb); //part
517 level_code= prefix + get_bits(gb, 4); //part
522 av_log(h->avctx, AV_LOG_ERROR, "Invalid level prefix\n");
525 level_code += (1<<(prefix-3))-4096;
527 level_code += get_bits(gb, prefix-3); //part
530 if(trailing_ones < 3) level_code += 2;
533 mask= -(level_code&1);
534 level[trailing_ones]= (((2+level_code)>>1) ^ mask) - mask;
536 level_code += ((level_code>>31)|1) & -(trailing_ones < 3);
538 suffix_length = 1 + (level_code + 3U > 6U);
539 level[trailing_ones]= level_code;
542 //remaining coefficients have suffix_length > 0
543 for(i=trailing_ones+1;i<total_coeff;i++) {
544 static const unsigned int suffix_limit[7] = {0,3,6,12,24,48,INT_MAX };
545 int bitsi= show_bits(gb, LEVEL_TAB_BITS);
546 level_code= cavlc_level_tab[suffix_length][bitsi][0];
548 skip_bits(gb, cavlc_level_tab[suffix_length][bitsi][1]);
549 if(level_code >= 100){
550 prefix= level_code - 100;
551 if(prefix == LEVEL_TAB_BITS){
552 prefix += get_level_prefix(gb);
555 level_code = (prefix<<suffix_length) + get_bits(gb, suffix_length);
557 level_code = 15<<suffix_length;
560 av_log(h->avctx, AV_LOG_ERROR, "Invalid level prefix\n");
561 return AVERROR_INVALIDDATA;
563 level_code += (1<<(prefix-3))-4096;
565 level_code += get_bits(gb, prefix-3);
567 mask= -(level_code&1);
568 level_code= (((2+level_code)>>1) ^ mask) - mask;
570 level[i]= level_code;
571 suffix_length+= suffix_limit[suffix_length] + level_code > 2U*suffix_limit[suffix_length];
575 if(total_coeff == max_coeff)
578 if (max_coeff <= 8) {
580 zeros_left = get_vlc2(gb, (chroma_dc_total_zeros_vlc-1)[total_coeff].table,
581 CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
583 zeros_left = get_vlc2(gb, (chroma422_dc_total_zeros_vlc-1)[total_coeff].table,
584 CHROMA422_DC_TOTAL_ZEROS_VLC_BITS, 1);
586 zeros_left= get_vlc2(gb, (total_zeros_vlc-1)[ total_coeff ].table, TOTAL_ZEROS_VLC_BITS, 1);
590 #define STORE_BLOCK(type) \
591 scantable += zeros_left + total_coeff - 1; \
592 if(n >= LUMA_DC_BLOCK_INDEX){ \
593 ((type*)block)[*scantable] = level[0]; \
594 for(i=1;i<total_coeff && zeros_left > 0;i++) { \
596 run_before= get_vlc2(gb, (run_vlc-1)[zeros_left].table, RUN_VLC_BITS, 1); \
598 run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2); \
599 zeros_left -= run_before; \
600 scantable -= 1 + run_before; \
601 ((type*)block)[*scantable]= level[i]; \
603 for(;i<total_coeff;i++) { \
605 ((type*)block)[*scantable]= level[i]; \
608 ((type*)block)[*scantable] = ((int)(level[0] * qmul[*scantable] + 32))>>6; \
609 for(i=1;i<total_coeff && zeros_left > 0;i++) { \
611 run_before= get_vlc2(gb, (run_vlc-1)[zeros_left].table, RUN_VLC_BITS, 1); \
613 run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2); \
614 zeros_left -= run_before; \
615 scantable -= 1 + run_before; \
616 ((type*)block)[*scantable]= ((int)(level[i] * qmul[*scantable] + 32))>>6; \
618 for(;i<total_coeff;i++) { \
620 ((type*)block)[*scantable]= ((int)(level[i] * qmul[*scantable] + 32))>>6; \
624 if (h->pixel_shift) {
631 av_log(h->avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", h->mb_x, h->mb_y);
638 static av_always_inline int decode_luma_residual(H264Context *h, H264SliceContext *sl, GetBitContext *gb, const uint8_t *scan, const uint8_t *scan8x8, int pixel_shift, int mb_type, int cbp, int p){
640 int qscale = p == 0 ? sl->qscale : sl->chroma_qp[p - 1];
641 if(IS_INTRA16x16(mb_type)){
642 AV_ZERO128(h->mb_luma_dc[p]+0);
643 AV_ZERO128(h->mb_luma_dc[p]+8);
644 AV_ZERO128(h->mb_luma_dc[p]+16);
645 AV_ZERO128(h->mb_luma_dc[p]+24);
646 if( decode_residual(h, sl, h->intra_gb_ptr, h->mb_luma_dc[p], LUMA_DC_BLOCK_INDEX+p, scan, NULL, 16) < 0){
647 return -1; //FIXME continue if partitioned and other return -1 too
650 av_assert2((cbp&15) == 0 || (cbp&15) == 15);
653 for(i8x8=0; i8x8<4; i8x8++){
654 for(i4x4=0; i4x4<4; i4x4++){
655 const int index= i4x4 + 4*i8x8 + p*16;
656 if( decode_residual(h, sl, h->intra_gb_ptr, h->mb + (16*index << pixel_shift),
657 index, scan + 1, h->dequant4_coeff[p][qscale], 15) < 0 ){
664 fill_rectangle(&sl->non_zero_count_cache[scan8[p*16]], 4, 4, 8, 0, 1);
668 int cqm = (IS_INTRA( mb_type ) ? 0:3)+p;
669 /* For CAVLC 4:4:4, we need to keep track of the luma 8x8 CBP for deblocking nnz purposes. */
671 for(i8x8=0; i8x8<4; i8x8++){
673 if(IS_8x8DCT(mb_type)){
674 int16_t *buf = &h->mb[64*i8x8+256*p << pixel_shift];
676 for(i4x4=0; i4x4<4; i4x4++){
677 const int index= i4x4 + 4*i8x8 + p*16;
678 if( decode_residual(h, sl, gb, buf, index, scan8x8+16*i4x4,
679 h->dequant8_coeff[cqm][qscale], 16) < 0 )
682 nnz = &sl->non_zero_count_cache[scan8[4 * i8x8 + p * 16]];
683 nnz[0] += nnz[1] + nnz[8] + nnz[9];
684 new_cbp |= !!nnz[0] << i8x8;
686 for(i4x4=0; i4x4<4; i4x4++){
687 const int index= i4x4 + 4*i8x8 + p*16;
688 if( decode_residual(h, sl, gb, h->mb + (16*index << pixel_shift), index,
689 scan, h->dequant4_coeff[cqm][qscale], 16) < 0 ){
692 new_cbp |= sl->non_zero_count_cache[scan8[index]] << i8x8;
696 uint8_t * const nnz = &sl->non_zero_count_cache[scan8[4 * i8x8 + p * 16]];
697 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
704 int ff_h264_decode_mb_cavlc(H264Context *h, H264SliceContext *sl)
708 unsigned int mb_type, cbp;
709 int dct8x8_allowed= h->pps.transform_8x8_mode;
710 int decode_chroma = h->sps.chroma_format_idc == 1 || h->sps.chroma_format_idc == 2;
711 const int pixel_shift = h->pixel_shift;
712 unsigned local_ref_count[2];
714 mb_xy = h->mb_xy = h->mb_x + h->mb_y*h->mb_stride;
716 tprintf(h->avctx, "pic:%d mb:%d/%d\n", h->frame_num, h->mb_x, h->mb_y);
717 cbp = 0; /* avoid warning. FIXME: find a solution without slowing
719 if(h->slice_type_nos != AV_PICTURE_TYPE_I){
720 if(h->mb_skip_run==-1)
721 h->mb_skip_run= get_ue_golomb_long(&h->gb);
723 if (h->mb_skip_run--) {
724 if(FRAME_MBAFF(h) && (h->mb_y&1) == 0){
725 if(h->mb_skip_run==0)
726 h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&h->gb);
728 decode_mb_skip(h, sl);
732 if (FRAME_MBAFF(h)) {
733 if( (h->mb_y&1) == 0 )
734 h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&h->gb);
737 sl->prev_mb_skipped = 0;
739 mb_type= get_ue_golomb(&h->gb);
740 if(h->slice_type_nos == AV_PICTURE_TYPE_B){
742 partition_count= b_mb_type_info[mb_type].partition_count;
743 mb_type= b_mb_type_info[mb_type].type;
746 goto decode_intra_mb;
748 }else if(h->slice_type_nos == AV_PICTURE_TYPE_P){
750 partition_count= p_mb_type_info[mb_type].partition_count;
751 mb_type= p_mb_type_info[mb_type].type;
754 goto decode_intra_mb;
757 av_assert2(h->slice_type_nos == AV_PICTURE_TYPE_I);
758 if(h->slice_type == AV_PICTURE_TYPE_SI && mb_type)
762 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);
766 cbp= i_mb_type_info[mb_type].cbp;
767 sl->intra16x16_pred_mode = i_mb_type_info[mb_type].pred_mode;
768 mb_type= i_mb_type_info[mb_type].type;
772 mb_type |= MB_TYPE_INTERLACED;
774 h->slice_table[ mb_xy ]= h->slice_num;
776 if(IS_INTRA_PCM(mb_type)){
777 const int mb_size = ff_h264_mb_sizes[h->sps.chroma_format_idc] *
778 h->sps.bit_depth_luma;
780 // We assume these blocks are very rare so we do not optimize it.
781 h->intra_pcm_ptr = align_get_bits(&h->gb);
782 if (get_bits_left(&h->gb) < mb_size) {
783 av_log(h->avctx, AV_LOG_ERROR, "Not enough data for an intra PCM block.\n");
784 return AVERROR_INVALIDDATA;
786 skip_bits_long(&h->gb, mb_size);
788 // In deblocking, the quantizer is 0
789 h->cur_pic.qscale_table[mb_xy] = 0;
790 // All coeffs are present
791 memset(h->non_zero_count[mb_xy], 16, 48);
793 h->cur_pic.mb_type[mb_xy] = mb_type;
797 local_ref_count[0] = h->ref_count[0] << MB_MBAFF(h);
798 local_ref_count[1] = h->ref_count[1] << MB_MBAFF(h);
800 fill_decode_neighbors(h, sl, mb_type);
801 fill_decode_caches(h, sl, mb_type);
804 if(IS_INTRA(mb_type)){
806 // init_top_left_availability(h);
807 if(IS_INTRA4x4(mb_type)){
810 if(dct8x8_allowed && get_bits1(&h->gb)){
811 mb_type |= MB_TYPE_8x8DCT;
815 // fill_intra4x4_pred_table(h);
816 for(i=0; i<16; i+=di){
817 int mode = pred_intra_mode(h, sl, i);
819 if(!get_bits1(&h->gb)){
820 const int rem_mode= get_bits(&h->gb, 3);
821 mode = rem_mode + (rem_mode >= mode);
825 fill_rectangle(&sl->intra4x4_pred_mode_cache[ scan8[i] ], 2, 2, 8, mode, 1);
827 sl->intra4x4_pred_mode_cache[scan8[i]] = mode;
829 write_back_intra_pred_mode(h, sl);
830 if (ff_h264_check_intra4x4_pred_mode(h, sl) < 0)
833 sl->intra16x16_pred_mode = ff_h264_check_intra_pred_mode(h, sl, sl->intra16x16_pred_mode, 0);
834 if (sl->intra16x16_pred_mode < 0)
838 pred_mode= ff_h264_check_intra_pred_mode(h, sl, get_ue_golomb_31(&h->gb), 1);
841 sl->chroma_pred_mode = pred_mode;
843 sl->chroma_pred_mode = DC_128_PRED8x8;
845 }else if(partition_count==4){
846 int i, j, sub_partition_count[4], list, ref[2][4];
848 if(h->slice_type_nos == AV_PICTURE_TYPE_B){
850 h->sub_mb_type[i]= get_ue_golomb_31(&h->gb);
851 if(h->sub_mb_type[i] >=13){
852 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);
855 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
856 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
858 if( IS_DIRECT(h->sub_mb_type[0]|h->sub_mb_type[1]|h->sub_mb_type[2]|h->sub_mb_type[3])) {
859 ff_h264_pred_direct_motion(h, &mb_type);
860 h->ref_cache[0][scan8[4]] =
861 h->ref_cache[1][scan8[4]] =
862 h->ref_cache[0][scan8[12]] =
863 h->ref_cache[1][scan8[12]] = PART_NOT_AVAILABLE;
866 av_assert2(h->slice_type_nos == AV_PICTURE_TYPE_P); //FIXME SP correct ?
868 h->sub_mb_type[i]= get_ue_golomb_31(&h->gb);
869 if(h->sub_mb_type[i] >=4){
870 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);
873 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
874 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
878 for(list=0; list<h->list_count; list++){
879 int ref_count = IS_REF0(mb_type) ? 1 : local_ref_count[list];
881 if(IS_DIRECT(h->sub_mb_type[i])) continue;
882 if(IS_DIR(h->sub_mb_type[i], 0, list)){
886 }else if(ref_count == 2){
887 tmp= get_bits1(&h->gb)^1;
889 tmp= get_ue_golomb_31(&h->gb);
891 av_log(h->avctx, AV_LOG_ERROR, "ref %u overflow\n", tmp);
904 dct8x8_allowed = get_dct8x8_allowed(h);
906 for(list=0; list<h->list_count; list++){
908 if(IS_DIRECT(h->sub_mb_type[i])) {
909 h->ref_cache[list][ scan8[4*i] ] = h->ref_cache[list][ scan8[4*i]+1 ];
912 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]=
913 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
915 if(IS_DIR(h->sub_mb_type[i], 0, list)){
916 const int sub_mb_type= h->sub_mb_type[i];
917 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
918 for(j=0; j<sub_partition_count[i]; j++){
920 const int index= 4*i + block_width*j;
921 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
922 pred_motion(h, sl, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
923 mx += get_se_golomb(&h->gb);
924 my += get_se_golomb(&h->gb);
925 tprintf(h->avctx, "final mv:%d %d\n", mx, my);
927 if(IS_SUB_8X8(sub_mb_type)){
929 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
931 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
932 }else if(IS_SUB_8X4(sub_mb_type)){
933 mv_cache[ 1 ][0]= mx;
934 mv_cache[ 1 ][1]= my;
935 }else if(IS_SUB_4X8(sub_mb_type)){
936 mv_cache[ 8 ][0]= mx;
937 mv_cache[ 8 ][1]= my;
939 mv_cache[ 0 ][0]= mx;
940 mv_cache[ 0 ][1]= my;
943 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
949 }else if(IS_DIRECT(mb_type)){
950 ff_h264_pred_direct_motion(h, &mb_type);
951 dct8x8_allowed &= h->sps.direct_8x8_inference_flag;
954 //FIXME we should set ref_idx_l? to 0 if we use that later ...
955 if(IS_16X16(mb_type)){
956 for(list=0; list<h->list_count; list++){
958 if(IS_DIR(mb_type, 0, list)){
959 if(local_ref_count[list]==1){
961 } else if(local_ref_count[list]==2){
962 val= get_bits1(&h->gb)^1;
964 val= get_ue_golomb_31(&h->gb);
965 if (val >= local_ref_count[list]){
966 av_log(h->avctx, AV_LOG_ERROR, "ref %u overflow\n", val);
970 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
973 for(list=0; list<h->list_count; list++){
974 if(IS_DIR(mb_type, 0, list)){
975 pred_motion(h, sl, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
976 mx += get_se_golomb(&h->gb);
977 my += get_se_golomb(&h->gb);
978 tprintf(h->avctx, "final mv:%d %d\n", mx, my);
980 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
984 else if(IS_16X8(mb_type)){
985 for(list=0; list<h->list_count; list++){
988 if(IS_DIR(mb_type, i, list)){
989 if(local_ref_count[list] == 1) {
991 } else if(local_ref_count[list] == 2) {
992 val= get_bits1(&h->gb)^1;
994 val= get_ue_golomb_31(&h->gb);
995 if (val >= local_ref_count[list]){
996 av_log(h->avctx, AV_LOG_ERROR, "ref %u overflow\n", val);
1001 val= LIST_NOT_USED&0xFF;
1002 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
1005 for(list=0; list<h->list_count; list++){
1008 if(IS_DIR(mb_type, i, list)){
1009 pred_16x8_motion(h, sl, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
1010 mx += get_se_golomb(&h->gb);
1011 my += get_se_golomb(&h->gb);
1012 tprintf(h->avctx, "final mv:%d %d\n", mx, my);
1014 val= pack16to32(mx,my);
1017 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 4);
1021 av_assert2(IS_8X16(mb_type));
1022 for(list=0; list<h->list_count; list++){
1025 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
1026 if(local_ref_count[list]==1){
1028 } else if(local_ref_count[list]==2){
1029 val= get_bits1(&h->gb)^1;
1031 val= get_ue_golomb_31(&h->gb);
1032 if (val >= local_ref_count[list]){
1033 av_log(h->avctx, AV_LOG_ERROR, "ref %u overflow\n", val);
1038 val= LIST_NOT_USED&0xFF;
1039 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
1042 for(list=0; list<h->list_count; list++){
1045 if(IS_DIR(mb_type, i, list)){
1046 pred_8x16_motion(h, sl, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
1047 mx += get_se_golomb(&h->gb);
1048 my += get_se_golomb(&h->gb);
1049 tprintf(h->avctx, "final mv:%d %d\n", mx, my);
1051 val= pack16to32(mx,my);
1054 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 4);
1060 if(IS_INTER(mb_type))
1061 write_back_motion(h, mb_type);
1063 if(!IS_INTRA16x16(mb_type)){
1064 cbp= get_ue_golomb(&h->gb);
1068 av_log(h->avctx, AV_LOG_ERROR, "cbp too large (%u) at %d %d\n", cbp, h->mb_x, h->mb_y);
1071 if(IS_INTRA4x4(mb_type)) cbp= golomb_to_intra4x4_cbp[cbp];
1072 else cbp= golomb_to_inter_cbp [cbp];
1075 av_log(h->avctx, AV_LOG_ERROR, "cbp too large (%u) at %d %d\n", cbp, h->mb_x, h->mb_y);
1078 if(IS_INTRA4x4(mb_type)) cbp= golomb_to_intra4x4_cbp_gray[cbp];
1079 else cbp= golomb_to_inter_cbp_gray[cbp];
1082 if (!decode_chroma && cbp>15) {
1083 av_log(h->avctx, AV_LOG_ERROR, "gray chroma\n");
1084 return AVERROR_INVALIDDATA;
1088 if(dct8x8_allowed && (cbp&15) && !IS_INTRA(mb_type)){
1089 mb_type |= MB_TYPE_8x8DCT*get_bits1(&h->gb);
1092 h->cbp_table[mb_xy]= cbp;
1093 h->cur_pic.mb_type[mb_xy] = mb_type;
1095 if(cbp || IS_INTRA16x16(mb_type)){
1096 int i4x4, i8x8, chroma_idx;
1099 GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
1100 const uint8_t *scan, *scan8x8;
1101 const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
1103 if(IS_INTERLACED(mb_type)){
1104 scan8x8 = sl->qscale ? h->field_scan8x8_cavlc : h->field_scan8x8_cavlc_q0;
1105 scan = sl->qscale ? h->field_scan : h->field_scan_q0;
1107 scan8x8 = sl->qscale ? h->zigzag_scan8x8_cavlc : h->zigzag_scan8x8_cavlc_q0;
1108 scan = sl->qscale ? h->zigzag_scan : h->zigzag_scan_q0;
1111 dquant= get_se_golomb(&h->gb);
1113 sl->qscale += dquant;
1115 if (((unsigned)sl->qscale) > max_qp){
1116 if (sl->qscale < 0) sl->qscale += max_qp + 1;
1117 else sl->qscale -= max_qp+1;
1118 if (((unsigned)sl->qscale) > max_qp){
1119 av_log(h->avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, h->mb_x, h->mb_y);
1124 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
1125 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
1127 if ((ret = decode_luma_residual(h, sl, gb, scan, scan8x8, pixel_shift, mb_type, cbp, 0)) < 0 ) {
1130 h->cbp_table[mb_xy] |= ret << 12;
1132 if (decode_luma_residual(h, sl, gb, scan, scan8x8, pixel_shift, mb_type, cbp, 1) < 0 ) {
1135 if (decode_luma_residual(h, sl, gb, scan, scan8x8, pixel_shift, mb_type, cbp, 2) < 0 ) {
1139 const int num_c8x8 = h->sps.chroma_format_idc;
1142 for(chroma_idx=0; chroma_idx<2; chroma_idx++)
1143 if (decode_residual(h, sl, gb, h->mb + ((256 + 16*16*chroma_idx) << pixel_shift),
1144 CHROMA_DC_BLOCK_INDEX+chroma_idx,
1145 CHROMA422(h) ? chroma422_dc_scan : chroma_dc_scan,
1146 NULL, 4*num_c8x8) < 0) {
1152 for(chroma_idx=0; chroma_idx<2; chroma_idx++){
1153 const uint32_t *qmul = h->dequant4_coeff[chroma_idx+1+(IS_INTRA( mb_type ) ? 0:3)][sl->chroma_qp[chroma_idx]];
1154 int16_t *mb = h->mb + (16*(16 + 16*chroma_idx) << pixel_shift);
1155 for (i8x8 = 0; i8x8<num_c8x8; i8x8++) {
1156 for (i4x4 = 0; i4x4 < 4; i4x4++) {
1157 const int index = 16 + 16*chroma_idx + 8*i8x8 + i4x4;
1158 if (decode_residual(h, sl, gb, mb, index, scan + 1, qmul, 15) < 0)
1160 mb += 16 << pixel_shift;
1165 fill_rectangle(&sl->non_zero_count_cache[scan8[16]], 4, 4, 8, 0, 1);
1166 fill_rectangle(&sl->non_zero_count_cache[scan8[32]], 4, 4, 8, 0, 1);
1170 fill_rectangle(&sl->non_zero_count_cache[scan8[ 0]], 4, 4, 8, 0, 1);
1171 fill_rectangle(&sl->non_zero_count_cache[scan8[16]], 4, 4, 8, 0, 1);
1172 fill_rectangle(&sl->non_zero_count_cache[scan8[32]], 4, 4, 8, 0, 1);
1174 h->cur_pic.qscale_table[mb_xy] = sl->qscale;
1175 write_back_non_zero_count(h, sl);