2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
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
23 * @file libavcodec/h264.c
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
31 #include "mpegvideo.h"
34 #include "h264_parser.h"
37 #include "rectangle.h"
38 #include "vdpau_internal.h"
42 #include "x86/h264_i386.h"
49 * Value of Picture.reference when Picture is not a reference picture, but
50 * is held for delayed output.
52 #define DELAYED_PIC_REF 4
54 static VLC coeff_token_vlc[4];
55 static VLC_TYPE coeff_token_vlc_tables[520+332+280+256][2];
56 static const int coeff_token_vlc_tables_size[4]={520,332,280,256};
58 static VLC chroma_dc_coeff_token_vlc;
59 static VLC_TYPE chroma_dc_coeff_token_vlc_table[256][2];
60 static const int chroma_dc_coeff_token_vlc_table_size = 256;
62 static VLC total_zeros_vlc[15];
63 static VLC_TYPE total_zeros_vlc_tables[15][512][2];
64 static const int total_zeros_vlc_tables_size = 512;
66 static VLC chroma_dc_total_zeros_vlc[3];
67 static VLC_TYPE chroma_dc_total_zeros_vlc_tables[3][8][2];
68 static const int chroma_dc_total_zeros_vlc_tables_size = 8;
70 static VLC run_vlc[6];
71 static VLC_TYPE run_vlc_tables[6][8][2];
72 static const int run_vlc_tables_size = 8;
75 static VLC_TYPE run7_vlc_table[96][2];
76 static const int run7_vlc_table_size = 96;
78 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
79 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
80 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
81 static void filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
82 static Picture * remove_long(H264Context *h, int i, int ref_mask);
84 static av_always_inline uint32_t pack16to32(int a, int b){
85 #ifdef WORDS_BIGENDIAN
86 return (b&0xFFFF) + (a<<16);
88 return (a&0xFFFF) + (b<<16);
92 static const uint8_t rem6[52]={
93 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
96 static const uint8_t div6[52]={
97 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8,
100 static const uint8_t left_block_options[4][8]={
107 #define LEVEL_TAB_BITS 8
108 static int8_t cavlc_level_tab[7][1<<LEVEL_TAB_BITS][2];
110 static void fill_caches(H264Context *h, int mb_type, int for_deblock){
111 MpegEncContext * const s = &h->s;
112 const int mb_xy= h->mb_xy;
113 int topleft_xy, top_xy, topright_xy, left_xy[2];
114 int topleft_type, top_type, topright_type, left_type[2];
115 const uint8_t * left_block;
116 int topleft_partition= -1;
119 top_xy = mb_xy - (s->mb_stride << FIELD_PICTURE);
121 //FIXME deblocking could skip the intra and nnz parts.
122 if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[top_xy]) && !FRAME_MBAFF)
125 /* Wow, what a mess, why didn't they simplify the interlacing & intra
126 * stuff, I can't imagine that these complex rules are worth it. */
128 topleft_xy = top_xy - 1;
129 topright_xy= top_xy + 1;
130 left_xy[1] = left_xy[0] = mb_xy-1;
131 left_block = left_block_options[0];
133 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride;
134 const int top_pair_xy = pair_xy - s->mb_stride;
135 const int topleft_pair_xy = top_pair_xy - 1;
136 const int topright_pair_xy = top_pair_xy + 1;
137 const int topleft_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
138 const int top_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
139 const int topright_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
140 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
141 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
142 const int bottom = (s->mb_y & 1);
143 tprintf(s->avctx, "fill_caches: curr_mb_field_flag:%d, left_mb_field_flag:%d, topleft_mb_field_flag:%d, top_mb_field_flag:%d, topright_mb_field_flag:%d\n", curr_mb_field_flag, left_mb_field_flag, topleft_mb_field_flag, top_mb_field_flag, topright_mb_field_flag);
145 if (curr_mb_field_flag && (bottom || top_mb_field_flag)){
146 top_xy -= s->mb_stride;
148 if (curr_mb_field_flag && (bottom || topleft_mb_field_flag)){
149 topleft_xy -= s->mb_stride;
150 } else if(bottom && !curr_mb_field_flag && left_mb_field_flag) {
151 topleft_xy += s->mb_stride;
152 // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
153 topleft_partition = 0;
155 if (curr_mb_field_flag && (bottom || topright_mb_field_flag)){
156 topright_xy -= s->mb_stride;
158 if (left_mb_field_flag != curr_mb_field_flag) {
159 left_xy[1] = left_xy[0] = pair_xy - 1;
160 if (curr_mb_field_flag) {
161 left_xy[1] += s->mb_stride;
162 left_block = left_block_options[3];
164 left_block= left_block_options[2 - bottom];
169 h->top_mb_xy = top_xy;
170 h->left_mb_xy[0] = left_xy[0];
171 h->left_mb_xy[1] = left_xy[1];
175 top_type = h->slice_table[top_xy ] < 0xFFFF ? s->current_picture.mb_type[top_xy] : 0;
176 left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0;
177 left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0;
179 if(MB_MBAFF && !IS_INTRA(mb_type)){
181 for(list=0; list<h->list_count; list++){
182 //These values where changed for ease of performing MC, we need to change them back
183 //FIXME maybe we can make MC and loop filter use the same values or prevent
184 //the MC code from changing ref_cache and rather use a temporary array.
185 if(USES_LIST(mb_type,list)){
186 int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
187 *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
188 *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
190 *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
191 *(uint32_t*)&h->ref_cache[list][scan8[10]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
196 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
197 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
198 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
199 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
200 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
202 if(IS_INTRA(mb_type)){
203 int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
204 h->topleft_samples_available=
205 h->top_samples_available=
206 h->left_samples_available= 0xFFFF;
207 h->topright_samples_available= 0xEEEA;
209 if(!(top_type & type_mask)){
210 h->topleft_samples_available= 0xB3FF;
211 h->top_samples_available= 0x33FF;
212 h->topright_samples_available= 0x26EA;
214 if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
215 if(IS_INTERLACED(mb_type)){
216 if(!(left_type[0] & type_mask)){
217 h->topleft_samples_available&= 0xDFFF;
218 h->left_samples_available&= 0x5FFF;
220 if(!(left_type[1] & type_mask)){
221 h->topleft_samples_available&= 0xFF5F;
222 h->left_samples_available&= 0xFF5F;
225 int left_typei = h->slice_table[left_xy[0] + s->mb_stride ] == h->slice_num
226 ? s->current_picture.mb_type[left_xy[0] + s->mb_stride] : 0;
227 assert(left_xy[0] == left_xy[1]);
228 if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
229 h->topleft_samples_available&= 0xDF5F;
230 h->left_samples_available&= 0x5F5F;
234 if(!(left_type[0] & type_mask)){
235 h->topleft_samples_available&= 0xDF5F;
236 h->left_samples_available&= 0x5F5F;
240 if(!(topleft_type & type_mask))
241 h->topleft_samples_available&= 0x7FFF;
243 if(!(topright_type & type_mask))
244 h->topright_samples_available&= 0xFBFF;
246 if(IS_INTRA4x4(mb_type)){
247 if(IS_INTRA4x4(top_type)){
248 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
249 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
250 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
251 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
254 if(!(top_type & type_mask))
259 h->intra4x4_pred_mode_cache[4+8*0]=
260 h->intra4x4_pred_mode_cache[5+8*0]=
261 h->intra4x4_pred_mode_cache[6+8*0]=
262 h->intra4x4_pred_mode_cache[7+8*0]= pred;
265 if(IS_INTRA4x4(left_type[i])){
266 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
267 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
270 if(!(left_type[i] & type_mask))
275 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
276 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
292 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
294 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
295 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
296 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
297 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
299 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
300 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
302 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
303 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
306 h->non_zero_count_cache[4+8*0]=
307 h->non_zero_count_cache[5+8*0]=
308 h->non_zero_count_cache[6+8*0]=
309 h->non_zero_count_cache[7+8*0]=
311 h->non_zero_count_cache[1+8*0]=
312 h->non_zero_count_cache[2+8*0]=
314 h->non_zero_count_cache[1+8*3]=
315 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
319 for (i=0; i<2; i++) {
321 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
322 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
323 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
324 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
326 h->non_zero_count_cache[3+8*1 + 2*8*i]=
327 h->non_zero_count_cache[3+8*2 + 2*8*i]=
328 h->non_zero_count_cache[0+8*1 + 8*i]=
329 h->non_zero_count_cache[0+8*4 + 8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
336 h->top_cbp = h->cbp_table[top_xy];
337 } else if(IS_INTRA(mb_type)) {
344 h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
345 } else if(IS_INTRA(mb_type)) {
351 h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
354 h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
359 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
361 for(list=0; list<h->list_count; list++){
362 if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
363 /*if(!h->mv_cache_clean[list]){
364 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
365 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
366 h->mv_cache_clean[list]= 1;
370 h->mv_cache_clean[list]= 0;
372 if(USES_LIST(top_type, list)){
373 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
374 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
375 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
376 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
377 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
378 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
379 h->ref_cache[list][scan8[0] + 0 - 1*8]=
380 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
381 h->ref_cache[list][scan8[0] + 2 - 1*8]=
382 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
384 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
385 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
386 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
387 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
388 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
392 int cache_idx = scan8[0] - 1 + i*2*8;
393 if(USES_LIST(left_type[i], list)){
394 const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
395 const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
396 *(uint32_t*)h->mv_cache[list][cache_idx ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]];
397 *(uint32_t*)h->mv_cache[list][cache_idx+8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1+i*2]];
398 h->ref_cache[list][cache_idx ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
399 h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
401 *(uint32_t*)h->mv_cache [list][cache_idx ]=
402 *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
403 h->ref_cache[list][cache_idx ]=
404 h->ref_cache[list][cache_idx+8]= left_type[i] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
408 if(for_deblock || ((IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred) && !FRAME_MBAFF))
411 if(USES_LIST(topleft_type, list)){
412 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + h->b_stride + (topleft_partition & 2*h->b_stride);
413 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + (topleft_partition & h->b8_stride);
414 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
415 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
417 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
418 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
421 if(USES_LIST(topright_type, list)){
422 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
423 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
424 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
425 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
427 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
428 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
431 if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
434 h->ref_cache[list][scan8[5 ]+1] =
435 h->ref_cache[list][scan8[7 ]+1] =
436 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewhere else)
437 h->ref_cache[list][scan8[4 ]] =
438 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
439 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
440 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
441 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
442 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
443 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
446 /* XXX beurk, Load mvd */
447 if(USES_LIST(top_type, list)){
448 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
449 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
450 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
451 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
452 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
454 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
455 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
456 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
457 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
459 if(USES_LIST(left_type[0], list)){
460 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
461 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[0]];
462 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[1]];
464 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
465 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
467 if(USES_LIST(left_type[1], list)){
468 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
469 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[2]];
470 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[3]];
472 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
473 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
475 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
476 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
477 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
478 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
479 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
481 if(h->slice_type_nos == FF_B_TYPE){
482 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
484 if(IS_DIRECT(top_type)){
485 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
486 }else if(IS_8X8(top_type)){
487 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
488 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
489 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
491 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
494 if(IS_DIRECT(left_type[0]))
495 h->direct_cache[scan8[0] - 1 + 0*8]= 1;
496 else if(IS_8X8(left_type[0]))
497 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[h->mb2b8_xy[left_xy[0]] + 1 + h->b8_stride*(left_block[0]>>1)];
499 h->direct_cache[scan8[0] - 1 + 0*8]= 0;
501 if(IS_DIRECT(left_type[1]))
502 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
503 else if(IS_8X8(left_type[1]))
504 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[h->mb2b8_xy[left_xy[1]] + 1 + h->b8_stride*(left_block[2]>>1)];
506 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
512 MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
513 MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
514 MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
515 MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
516 MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
517 MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
518 MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
519 MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
520 MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
521 MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
523 #define MAP_F2F(idx, mb_type)\
524 if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
525 h->ref_cache[list][idx] <<= 1;\
526 h->mv_cache[list][idx][1] /= 2;\
527 h->mvd_cache[list][idx][1] /= 2;\
532 #define MAP_F2F(idx, mb_type)\
533 if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
534 h->ref_cache[list][idx] >>= 1;\
535 h->mv_cache[list][idx][1] <<= 1;\
536 h->mvd_cache[list][idx][1] <<= 1;\
546 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
549 static inline void write_back_intra_pred_mode(H264Context *h){
550 const int mb_xy= h->mb_xy;
552 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
553 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
554 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
555 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
556 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
557 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
558 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
562 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
564 static inline int check_intra4x4_pred_mode(H264Context *h){
565 MpegEncContext * const s = &h->s;
566 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
567 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
570 if(!(h->top_samples_available&0x8000)){
572 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
574 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
577 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
582 if((h->left_samples_available&0x8888)!=0x8888){
583 static const int mask[4]={0x8000,0x2000,0x80,0x20};
585 if(!(h->left_samples_available&mask[i])){
586 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
588 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
591 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
598 } //FIXME cleanup like next
601 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
603 static inline int check_intra_pred_mode(H264Context *h, int mode){
604 MpegEncContext * const s = &h->s;
605 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
606 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
609 av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
613 if(!(h->top_samples_available&0x8000)){
616 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
621 if((h->left_samples_available&0x8080) != 0x8080){
623 if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
624 mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
627 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
636 * gets the predicted intra4x4 prediction mode.
638 static inline int pred_intra_mode(H264Context *h, int n){
639 const int index8= scan8[n];
640 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
641 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
642 const int min= FFMIN(left, top);
644 tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
646 if(min<0) return DC_PRED;
650 static inline void write_back_non_zero_count(H264Context *h){
651 const int mb_xy= h->mb_xy;
653 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
654 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
655 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
656 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
657 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
658 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
659 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
661 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
662 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
663 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
665 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
666 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
667 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
671 * gets the predicted number of non-zero coefficients.
672 * @param n block index
674 static inline int pred_non_zero_count(H264Context *h, int n){
675 const int index8= scan8[n];
676 const int left= h->non_zero_count_cache[index8 - 1];
677 const int top = h->non_zero_count_cache[index8 - 8];
680 if(i<64) i= (i+1)>>1;
682 tprintf(h->s.avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
687 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
688 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
689 MpegEncContext *s = &h->s;
691 /* there is no consistent mapping of mvs to neighboring locations that will
692 * make mbaff happy, so we can't move all this logic to fill_caches */
694 const uint32_t *mb_types = s->current_picture_ptr->mb_type;
696 *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
697 *C = h->mv_cache[list][scan8[0]-2];
700 && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
701 int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
702 if(IS_INTERLACED(mb_types[topright_xy])){
703 #define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
704 const int x4 = X4, y4 = Y4;\
705 const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
706 if(!USES_LIST(mb_type,list))\
707 return LIST_NOT_USED;\
708 mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
709 h->mv_cache[list][scan8[0]-2][0] = mv[0];\
710 h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
711 return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;
713 SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
716 if(topright_ref == PART_NOT_AVAILABLE
717 && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
718 && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
720 && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
721 SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
724 && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
726 // left shift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's OK.
727 SET_DIAG_MV(/2, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
733 if(topright_ref != PART_NOT_AVAILABLE){
734 *C= h->mv_cache[list][ i - 8 + part_width ];
737 tprintf(s->avctx, "topright MV not available\n");
739 *C= h->mv_cache[list][ i - 8 - 1 ];
740 return h->ref_cache[list][ i - 8 - 1 ];
745 * gets the predicted MV.
746 * @param n the block index
747 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
748 * @param mx the x component of the predicted motion vector
749 * @param my the y component of the predicted motion vector
751 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
752 const int index8= scan8[n];
753 const int top_ref= h->ref_cache[list][ index8 - 8 ];
754 const int left_ref= h->ref_cache[list][ index8 - 1 ];
755 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
756 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
758 int diagonal_ref, match_count;
760 assert(part_width==1 || part_width==2 || part_width==4);
770 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
771 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
772 tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count);
773 if(match_count > 1){ //most common
774 *mx= mid_pred(A[0], B[0], C[0]);
775 *my= mid_pred(A[1], B[1], C[1]);
776 }else if(match_count==1){
780 }else if(top_ref==ref){
788 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
792 *mx= mid_pred(A[0], B[0], C[0]);
793 *my= mid_pred(A[1], B[1], C[1]);
797 tprintf(h->s.avctx, "pred_motion (%2d %2d %2d) (%2d %2d %2d) (%2d %2d %2d) -> (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], diagonal_ref, C[0], C[1], left_ref, A[0], A[1], ref, *mx, *my, h->s.mb_x, h->s.mb_y, n, list);
801 * gets the directionally predicted 16x8 MV.
802 * @param n the block index
803 * @param mx the x component of the predicted motion vector
804 * @param my the y component of the predicted motion vector
806 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
808 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
809 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
811 tprintf(h->s.avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list);
819 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
820 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
822 tprintf(h->s.avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
832 pred_motion(h, n, 4, list, ref, mx, my);
836 * gets the directionally predicted 8x16 MV.
837 * @param n the block index
838 * @param mx the x component of the predicted motion vector
839 * @param my the y component of the predicted motion vector
841 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
843 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
844 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
846 tprintf(h->s.avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
857 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
859 tprintf(h->s.avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list);
861 if(diagonal_ref == ref){
869 pred_motion(h, n, 2, list, ref, mx, my);
872 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
873 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
874 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
876 tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
878 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
879 || !( top_ref | *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ])
880 || !(left_ref | *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ])){
886 pred_motion(h, 0, 4, 0, 0, mx, my);
891 static int get_scale_factor(H264Context * const h, int poc, int poc1, int i){
892 int poc0 = h->ref_list[0][i].poc;
893 int td = av_clip(poc1 - poc0, -128, 127);
894 if(td == 0 || h->ref_list[0][i].long_ref){
897 int tb = av_clip(poc - poc0, -128, 127);
898 int tx = (16384 + (FFABS(td) >> 1)) / td;
899 return av_clip((tb*tx + 32) >> 6, -1024, 1023);
903 static inline void direct_dist_scale_factor(H264Context * const h){
904 MpegEncContext * const s = &h->s;
905 const int poc = h->s.current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
906 const int poc1 = h->ref_list[1][0].poc;
908 for(field=0; field<2; field++){
909 const int poc = h->s.current_picture_ptr->field_poc[field];
910 const int poc1 = h->ref_list[1][0].field_poc[field];
911 for(i=0; i < 2*h->ref_count[0]; i++)
912 h->dist_scale_factor_field[field][i^field] = get_scale_factor(h, poc, poc1, i+16);
915 for(i=0; i<h->ref_count[0]; i++){
916 h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i);
920 static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field, int colfield, int mbafi){
921 MpegEncContext * const s = &h->s;
922 Picture * const ref1 = &h->ref_list[1][0];
923 int j, old_ref, rfield;
924 int start= mbafi ? 16 : 0;
925 int end = mbafi ? 16+2*h->ref_count[list] : h->ref_count[list];
926 int interl= mbafi || s->picture_structure != PICT_FRAME;
928 /* bogus; fills in for missing frames */
929 memset(map[list], 0, sizeof(map[list]));
931 for(rfield=0; rfield<2; rfield++){
932 for(old_ref=0; old_ref<ref1->ref_count[colfield][list]; old_ref++){
933 int poc = ref1->ref_poc[colfield][list][old_ref];
937 else if( interl && (poc&3) == 3) //FIXME store all MBAFF references so this isnt needed
938 poc= (poc&~3) + rfield + 1;
940 for(j=start; j<end; j++){
941 if(4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3) == poc){
942 int cur_ref= mbafi ? (j-16)^field : j;
943 map[list][2*old_ref + (rfield^field) + 16] = cur_ref;
945 map[list][old_ref] = cur_ref;
953 static inline void direct_ref_list_init(H264Context * const h){
954 MpegEncContext * const s = &h->s;
955 Picture * const ref1 = &h->ref_list[1][0];
956 Picture * const cur = s->current_picture_ptr;
958 int sidx= (s->picture_structure&1)^1;
959 int ref1sidx= (ref1->reference&1)^1;
961 for(list=0; list<2; list++){
962 cur->ref_count[sidx][list] = h->ref_count[list];
963 for(j=0; j<h->ref_count[list]; j++)
964 cur->ref_poc[sidx][list][j] = 4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3);
967 if(s->picture_structure == PICT_FRAME){
968 memcpy(cur->ref_count[1], cur->ref_count[0], sizeof(cur->ref_count[0]));
969 memcpy(cur->ref_poc [1], cur->ref_poc [0], sizeof(cur->ref_poc [0]));
972 cur->mbaff= FRAME_MBAFF;
974 if(cur->pict_type != FF_B_TYPE || h->direct_spatial_mv_pred)
977 for(list=0; list<2; list++){
978 fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0);
979 for(field=0; field<2; field++)
980 fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1);
984 static inline void pred_direct_motion(H264Context * const h, int *mb_type){
985 MpegEncContext * const s = &h->s;
986 int b8_stride = h->b8_stride;
987 int b4_stride = h->b_stride;
988 int mb_xy = h->mb_xy;
990 const int16_t (*l1mv0)[2], (*l1mv1)[2];
991 const int8_t *l1ref0, *l1ref1;
992 const int is_b8x8 = IS_8X8(*mb_type);
993 unsigned int sub_mb_type;
996 #define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
998 if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL
999 if(!IS_INTERLACED(*mb_type)){ // AFR/FR -> AFL/FL
1000 int cur_poc = s->current_picture_ptr->poc;
1001 int *col_poc = h->ref_list[1]->field_poc;
1002 int col_parity = FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc);
1003 mb_xy= s->mb_x + ((s->mb_y&~1) + col_parity)*s->mb_stride;
1005 }else if(!(s->picture_structure & h->ref_list[1][0].reference) && !h->ref_list[1][0].mbaff){// FL -> FL & differ parity
1006 int fieldoff= 2*(h->ref_list[1][0].reference)-3;
1007 mb_xy += s->mb_stride*fieldoff;
1010 }else{ // AFL/AFR/FR/FL -> AFR/FR
1011 if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR
1012 mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
1013 mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
1014 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
1017 //FIXME IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag
1018 if( (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
1019 && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
1021 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1022 *mb_type |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */
1024 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1025 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
1027 }else{ // AFR/FR -> AFR/FR
1030 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
1031 if(IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag){
1032 /* FIXME save sub mb types from previous frames (or derive from MVs)
1033 * so we know exactly what block size to use */
1034 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1035 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
1036 }else if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
1037 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1038 *mb_type |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1040 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1041 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
1046 l1mv0 = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
1047 l1mv1 = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
1048 l1ref0 = &h->ref_list[1][0].ref_index [0][h->mb2b8_xy[mb_xy]];
1049 l1ref1 = &h->ref_list[1][0].ref_index [1][h->mb2b8_xy[mb_xy]];
1052 l1ref0 += h->b8_stride;
1053 l1ref1 += h->b8_stride;
1054 l1mv0 += 2*b4_stride;
1055 l1mv1 += 2*b4_stride;
1059 if(h->direct_spatial_mv_pred){
1064 /* FIXME interlacing + spatial direct uses wrong colocated block positions */
1066 /* ref = min(neighbors) */
1067 for(list=0; list<2; list++){
1068 int refa = h->ref_cache[list][scan8[0] - 1];
1069 int refb = h->ref_cache[list][scan8[0] - 8];
1070 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1071 if(refc == PART_NOT_AVAILABLE)
1072 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1073 ref[list] = FFMIN3((unsigned)refa, (unsigned)refb, (unsigned)refc);
1078 if(ref[0] < 0 && ref[1] < 0){
1079 ref[0] = ref[1] = 0;
1080 mv[0][0] = mv[0][1] =
1081 mv[1][0] = mv[1][1] = 0;
1083 for(list=0; list<2; list++){
1085 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1087 mv[list][0] = mv[list][1] = 0;
1093 *mb_type &= ~MB_TYPE_L1;
1094 sub_mb_type &= ~MB_TYPE_L1;
1095 }else if(ref[0] < 0){
1097 *mb_type &= ~MB_TYPE_L0;
1098 sub_mb_type &= ~MB_TYPE_L0;
1101 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
1102 for(i8=0; i8<4; i8++){
1105 int xy8 = x8+y8*b8_stride;
1106 int xy4 = 3*x8+y8*b4_stride;
1109 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1111 h->sub_mb_type[i8] = sub_mb_type;
1113 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1114 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1115 if(!IS_INTRA(mb_type_col[y8])
1116 && ( (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
1117 || (l1ref0[xy8] < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
1119 a= pack16to32(mv[0][0],mv[0][1]);
1121 b= pack16to32(mv[1][0],mv[1][1]);
1123 a= pack16to32(mv[0][0],mv[0][1]);
1124 b= pack16to32(mv[1][0],mv[1][1]);
1126 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
1127 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
1129 }else if(IS_16X16(*mb_type)){
1132 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1133 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
1134 if(!IS_INTRA(mb_type_col[0])
1135 && ( (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
1136 || (l1ref0[0] < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
1137 && (h->x264_build>33 || !h->x264_build)))){
1139 a= pack16to32(mv[0][0],mv[0][1]);
1141 b= pack16to32(mv[1][0],mv[1][1]);
1143 a= pack16to32(mv[0][0],mv[0][1]);
1144 b= pack16to32(mv[1][0],mv[1][1]);
1146 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
1147 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
1149 for(i8=0; i8<4; i8++){
1150 const int x8 = i8&1;
1151 const int y8 = i8>>1;
1153 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1155 h->sub_mb_type[i8] = sub_mb_type;
1157 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1158 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1159 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1160 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1163 if(!IS_INTRA(mb_type_col[0]) && ( l1ref0[x8 + y8*b8_stride] == 0
1164 || (l1ref0[x8 + y8*b8_stride] < 0 && l1ref1[x8 + y8*b8_stride] == 0
1165 && (h->x264_build>33 || !h->x264_build)))){
1166 const int16_t (*l1mv)[2]= l1ref0[x8 + y8*b8_stride] == 0 ? l1mv0 : l1mv1;
1167 if(IS_SUB_8X8(sub_mb_type)){
1168 const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
1169 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1171 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1173 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1176 for(i4=0; i4<4; i4++){
1177 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
1178 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1180 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1182 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1188 }else{ /* direct temporal mv pred */
1189 const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
1190 const int *dist_scale_factor = h->dist_scale_factor;
1193 if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){
1194 map_col_to_list0[0] = h->map_col_to_list0_field[s->mb_y&1][0];
1195 map_col_to_list0[1] = h->map_col_to_list0_field[s->mb_y&1][1];
1196 dist_scale_factor =h->dist_scale_factor_field[s->mb_y&1];
1198 if(h->ref_list[1][0].mbaff && IS_INTERLACED(mb_type_col[0]))
1201 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
1202 /* FIXME assumes direct_8x8_inference == 1 */
1203 int y_shift = 2*!IS_INTERLACED(*mb_type);
1205 for(i8=0; i8<4; i8++){
1206 const int x8 = i8&1;
1207 const int y8 = i8>>1;
1209 const int16_t (*l1mv)[2]= l1mv0;
1211 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1213 h->sub_mb_type[i8] = sub_mb_type;
1215 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1216 if(IS_INTRA(mb_type_col[y8])){
1217 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1218 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1219 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1223 ref0 = l1ref0[x8 + y8*b8_stride];
1225 ref0 = map_col_to_list0[0][ref0 + ref_offset];
1227 ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
1230 scale = dist_scale_factor[ref0];
1231 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1234 const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
1235 int my_col = (mv_col[1]<<y_shift)/2;
1236 int mx = (scale * mv_col[0] + 128) >> 8;
1237 int my = (scale * my_col + 128) >> 8;
1238 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1239 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
1245 /* one-to-one mv scaling */
1247 if(IS_16X16(*mb_type)){
1250 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1251 if(IS_INTRA(mb_type_col[0])){
1254 const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset]
1255 : map_col_to_list0[1][l1ref1[0] + ref_offset];
1256 const int scale = dist_scale_factor[ref0];
1257 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1259 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1260 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1262 mv0= pack16to32(mv_l0[0],mv_l0[1]);
1263 mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1265 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
1266 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
1267 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
1269 for(i8=0; i8<4; i8++){
1270 const int x8 = i8&1;
1271 const int y8 = i8>>1;
1273 const int16_t (*l1mv)[2]= l1mv0;
1275 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1277 h->sub_mb_type[i8] = sub_mb_type;
1278 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1279 if(IS_INTRA(mb_type_col[0])){
1280 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1281 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1282 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1286 ref0 = l1ref0[x8 + y8*b8_stride] + ref_offset;
1288 ref0 = map_col_to_list0[0][ref0];
1290 ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
1293 scale = dist_scale_factor[ref0];
1295 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1296 if(IS_SUB_8X8(sub_mb_type)){
1297 const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
1298 int mx = (scale * mv_col[0] + 128) >> 8;
1299 int my = (scale * mv_col[1] + 128) >> 8;
1300 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1301 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1303 for(i4=0; i4<4; i4++){
1304 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
1305 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1306 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1307 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1308 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1309 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1316 static inline void write_back_motion(H264Context *h, int mb_type){
1317 MpegEncContext * const s = &h->s;
1318 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1319 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1322 if(!USES_LIST(mb_type, 0))
1323 fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1325 for(list=0; list<h->list_count; list++){
1327 if(!USES_LIST(mb_type, list))
1331 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+0 + 8*y];
1332 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+2 + 8*y];
1334 if( h->pps.cabac ) {
1335 if(IS_SKIP(mb_type))
1336 fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
1339 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1340 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1345 int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1346 ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1347 ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1348 ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1349 ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1353 if(h->slice_type_nos == FF_B_TYPE && h->pps.cabac){
1354 if(IS_8X8(mb_type)){
1355 uint8_t *direct_table = &h->direct_table[b8_xy];
1356 direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1357 direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1358 direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1363 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
1368 // src[0]&0x80; //forbidden bit
1369 h->nal_ref_idc= src[0]>>5;
1370 h->nal_unit_type= src[0]&0x1F;
1374 for(i=0; i<length; i++)
1375 printf("%2X ", src[i]);
1378 #if HAVE_FAST_UNALIGNED
1379 # if HAVE_FAST_64BIT
1381 for(i=0; i+1<length; i+=9){
1382 if(!((~*(const uint64_t*)(src+i) & (*(const uint64_t*)(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
1385 for(i=0; i+1<length; i+=5){
1386 if(!((~*(const uint32_t*)(src+i) & (*(const uint32_t*)(src+i) - 0x01000101U)) & 0x80008080U))
1389 if(i>0 && !src[i]) i--;
1393 for(i=0; i+1<length; i+=2){
1394 if(src[i]) continue;
1395 if(i>0 && src[i-1]==0) i--;
1397 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1399 /* startcode, so we must be past the end */
1407 if(i>=length-1){ //no escaped 0
1408 *dst_length= length;
1409 *consumed= length+1; //+1 for the header
1413 bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
1414 h->rbsp_buffer[bufidx]= av_fast_realloc(h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
1415 dst= h->rbsp_buffer[bufidx];
1421 //printf("decoding esc\n");
1422 memcpy(dst, src, i);
1425 //remove escapes (very rare 1:2^22)
1427 dst[di++]= src[si++];
1428 dst[di++]= src[si++];
1429 }else if(src[si]==0 && src[si+1]==0){
1430 if(src[si+2]==3){ //escape
1435 }else //next start code
1439 dst[di++]= src[si++];
1442 dst[di++]= src[si++];
1445 memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
1448 *consumed= si + 1;//+1 for the header
1449 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
1453 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
1457 tprintf(h->s.avctx, "rbsp trailing %X\n", v);
1467 * IDCT transforms the 16 dc values and dequantizes them.
1468 * @param qp quantization parameter
1470 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1473 int temp[16]; //FIXME check if this is a good idea
1474 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1475 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1477 //memset(block, 64, 2*256);
1480 const int offset= y_offset[i];
1481 const int z0= block[offset+stride*0] + block[offset+stride*4];
1482 const int z1= block[offset+stride*0] - block[offset+stride*4];
1483 const int z2= block[offset+stride*1] - block[offset+stride*5];
1484 const int z3= block[offset+stride*1] + block[offset+stride*5];
1493 const int offset= x_offset[i];
1494 const int z0= temp[4*0+i] + temp[4*2+i];
1495 const int z1= temp[4*0+i] - temp[4*2+i];
1496 const int z2= temp[4*1+i] - temp[4*3+i];
1497 const int z3= temp[4*1+i] + temp[4*3+i];
1499 block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
1500 block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1501 block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1502 block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1508 * DCT transforms the 16 dc values.
1509 * @param qp quantization parameter ??? FIXME
1511 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1512 // const int qmul= dequant_coeff[qp][0];
1514 int temp[16]; //FIXME check if this is a good idea
1515 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1516 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1519 const int offset= y_offset[i];
1520 const int z0= block[offset+stride*0] + block[offset+stride*4];
1521 const int z1= block[offset+stride*0] - block[offset+stride*4];
1522 const int z2= block[offset+stride*1] - block[offset+stride*5];
1523 const int z3= block[offset+stride*1] + block[offset+stride*5];
1532 const int offset= x_offset[i];
1533 const int z0= temp[4*0+i] + temp[4*2+i];
1534 const int z1= temp[4*0+i] - temp[4*2+i];
1535 const int z2= temp[4*1+i] - temp[4*3+i];
1536 const int z3= temp[4*1+i] + temp[4*3+i];
1538 block[stride*0 +offset]= (z0 + z3)>>1;
1539 block[stride*2 +offset]= (z1 + z2)>>1;
1540 block[stride*8 +offset]= (z1 - z2)>>1;
1541 block[stride*10+offset]= (z0 - z3)>>1;
1549 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1550 const int stride= 16*2;
1551 const int xStride= 16;
1554 a= block[stride*0 + xStride*0];
1555 b= block[stride*0 + xStride*1];
1556 c= block[stride*1 + xStride*0];
1557 d= block[stride*1 + xStride*1];
1564 block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1565 block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1566 block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1567 block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
1571 static void chroma_dc_dct_c(DCTELEM *block){
1572 const int stride= 16*2;
1573 const int xStride= 16;
1576 a= block[stride*0 + xStride*0];
1577 b= block[stride*0 + xStride*1];
1578 c= block[stride*1 + xStride*0];
1579 d= block[stride*1 + xStride*1];
1586 block[stride*0 + xStride*0]= (a+c);
1587 block[stride*0 + xStride*1]= (e+b);
1588 block[stride*1 + xStride*0]= (a-c);
1589 block[stride*1 + xStride*1]= (e-b);
1594 * gets the chroma qp.
1596 static inline int get_chroma_qp(H264Context *h, int t, int qscale){
1597 return h->pps.chroma_qp_table[t][qscale];
1600 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1601 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1602 int src_x_offset, int src_y_offset,
1603 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1604 MpegEncContext * const s = &h->s;
1605 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1606 int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1607 const int luma_xy= (mx&3) + ((my&3)<<2);
1608 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
1609 uint8_t * src_cb, * src_cr;
1610 int extra_width= h->emu_edge_width;
1611 int extra_height= h->emu_edge_height;
1613 const int full_mx= mx>>2;
1614 const int full_my= my>>2;
1615 const int pic_width = 16*s->mb_width;
1616 const int pic_height = 16*s->mb_height >> MB_FIELD;
1618 if(mx&7) extra_width -= 3;
1619 if(my&7) extra_height -= 3;
1621 if( full_mx < 0-extra_width
1622 || full_my < 0-extra_height
1623 || full_mx + 16/*FIXME*/ > pic_width + extra_width
1624 || full_my + 16/*FIXME*/ > pic_height + extra_height){
1625 ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
1626 src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
1630 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
1632 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
1635 if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
1638 // chroma offset when predicting from a field of opposite parity
1639 my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
1640 emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
1642 src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1643 src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1646 ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
1647 src_cb= s->edge_emu_buffer;
1649 chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1652 ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
1653 src_cr= s->edge_emu_buffer;
1655 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1658 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
1659 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1660 int x_offset, int y_offset,
1661 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1662 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1663 int list0, int list1){
1664 MpegEncContext * const s = &h->s;
1665 qpel_mc_func *qpix_op= qpix_put;
1666 h264_chroma_mc_func chroma_op= chroma_put;
1668 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize;
1669 dest_cb += x_offset + y_offset*h->mb_uvlinesize;
1670 dest_cr += x_offset + y_offset*h->mb_uvlinesize;
1671 x_offset += 8*s->mb_x;
1672 y_offset += 8*(s->mb_y >> MB_FIELD);
1675 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1676 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1677 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1678 qpix_op, chroma_op);
1681 chroma_op= chroma_avg;
1685 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
1686 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1687 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1688 qpix_op, chroma_op);
1692 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
1693 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1694 int x_offset, int y_offset,
1695 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1696 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
1697 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
1698 int list0, int list1){
1699 MpegEncContext * const s = &h->s;
1701 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize;
1702 dest_cb += x_offset + y_offset*h->mb_uvlinesize;
1703 dest_cr += x_offset + y_offset*h->mb_uvlinesize;
1704 x_offset += 8*s->mb_x;
1705 y_offset += 8*(s->mb_y >> MB_FIELD);
1708 /* don't optimize for luma-only case, since B-frames usually
1709 * use implicit weights => chroma too. */
1710 uint8_t *tmp_cb = s->obmc_scratchpad;
1711 uint8_t *tmp_cr = s->obmc_scratchpad + 8;
1712 uint8_t *tmp_y = s->obmc_scratchpad + 8*h->mb_uvlinesize;
1713 int refn0 = h->ref_cache[0][ scan8[n] ];
1714 int refn1 = h->ref_cache[1][ scan8[n] ];
1716 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
1717 dest_y, dest_cb, dest_cr,
1718 x_offset, y_offset, qpix_put, chroma_put);
1719 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
1720 tmp_y, tmp_cb, tmp_cr,
1721 x_offset, y_offset, qpix_put, chroma_put);
1723 if(h->use_weight == 2){
1724 int weight0 = h->implicit_weight[refn0][refn1];
1725 int weight1 = 64 - weight0;
1726 luma_weight_avg( dest_y, tmp_y, h-> mb_linesize, 5, weight0, weight1, 0);
1727 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
1728 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
1730 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
1731 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
1732 h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
1733 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1734 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
1735 h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
1736 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1737 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
1738 h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
1741 int list = list1 ? 1 : 0;
1742 int refn = h->ref_cache[list][ scan8[n] ];
1743 Picture *ref= &h->ref_list[list][refn];
1744 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
1745 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1746 qpix_put, chroma_put);
1748 luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
1749 h->luma_weight[list][refn], h->luma_offset[list][refn]);
1750 if(h->use_weight_chroma){
1751 chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1752 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
1753 chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1754 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
1759 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1760 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1761 int x_offset, int y_offset,
1762 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1763 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1764 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
1765 int list0, int list1){
1766 if((h->use_weight==2 && list0 && list1
1767 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
1768 || h->use_weight==1)
1769 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1770 x_offset, y_offset, qpix_put, chroma_put,
1771 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
1773 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1774 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
1777 static inline void prefetch_motion(H264Context *h, int list){
1778 /* fetch pixels for estimated mv 4 macroblocks ahead
1779 * optimized for 64byte cache lines */
1780 MpegEncContext * const s = &h->s;
1781 const int refn = h->ref_cache[list][scan8[0]];
1783 const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
1784 const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
1785 uint8_t **src= h->ref_list[list][refn].data;
1786 int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
1787 s->dsp.prefetch(src[0]+off, s->linesize, 4);
1788 off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
1789 s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
1793 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1794 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1795 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
1796 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
1797 MpegEncContext * const s = &h->s;
1798 const int mb_xy= h->mb_xy;
1799 const int mb_type= s->current_picture.mb_type[mb_xy];
1801 assert(IS_INTER(mb_type));
1803 prefetch_motion(h, 0);
1805 if(IS_16X16(mb_type)){
1806 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1807 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1808 &weight_op[0], &weight_avg[0],
1809 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1810 }else if(IS_16X8(mb_type)){
1811 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1812 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1813 &weight_op[1], &weight_avg[1],
1814 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1815 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1816 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1817 &weight_op[1], &weight_avg[1],
1818 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1819 }else if(IS_8X16(mb_type)){
1820 mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
1821 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1822 &weight_op[2], &weight_avg[2],
1823 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1824 mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
1825 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1826 &weight_op[2], &weight_avg[2],
1827 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1831 assert(IS_8X8(mb_type));
1834 const int sub_mb_type= h->sub_mb_type[i];
1836 int x_offset= (i&1)<<2;
1837 int y_offset= (i&2)<<1;
1839 if(IS_SUB_8X8(sub_mb_type)){
1840 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1841 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1842 &weight_op[3], &weight_avg[3],
1843 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1844 }else if(IS_SUB_8X4(sub_mb_type)){
1845 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1846 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1847 &weight_op[4], &weight_avg[4],
1848 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1849 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
1850 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1851 &weight_op[4], &weight_avg[4],
1852 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1853 }else if(IS_SUB_4X8(sub_mb_type)){
1854 mc_part(h, n , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1855 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1856 &weight_op[5], &weight_avg[5],
1857 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1858 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
1859 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1860 &weight_op[5], &weight_avg[5],
1861 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1864 assert(IS_SUB_4X4(sub_mb_type));
1866 int sub_x_offset= x_offset + 2*(j&1);
1867 int sub_y_offset= y_offset + (j&2);
1868 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
1869 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1870 &weight_op[6], &weight_avg[6],
1871 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1877 prefetch_motion(h, 1);
1880 static av_cold void init_cavlc_level_tab(void){
1881 int suffix_length, mask;
1884 for(suffix_length=0; suffix_length<7; suffix_length++){
1885 for(i=0; i<(1<<LEVEL_TAB_BITS); i++){
1886 int prefix= LEVEL_TAB_BITS - av_log2(2*i);
1887 int level_code= (prefix<<suffix_length) + (i>>(LEVEL_TAB_BITS-prefix-1-suffix_length)) - (1<<suffix_length);
1889 mask= -(level_code&1);
1890 level_code= (((2+level_code)>>1) ^ mask) - mask;
1891 if(prefix + 1 + suffix_length <= LEVEL_TAB_BITS){
1892 cavlc_level_tab[suffix_length][i][0]= level_code;
1893 cavlc_level_tab[suffix_length][i][1]= prefix + 1 + suffix_length;
1894 }else if(prefix + 1 <= LEVEL_TAB_BITS){
1895 cavlc_level_tab[suffix_length][i][0]= prefix+100;
1896 cavlc_level_tab[suffix_length][i][1]= prefix + 1;
1898 cavlc_level_tab[suffix_length][i][0]= LEVEL_TAB_BITS+100;
1899 cavlc_level_tab[suffix_length][i][1]= LEVEL_TAB_BITS;
1905 static av_cold void decode_init_vlc(void){
1906 static int done = 0;
1913 chroma_dc_coeff_token_vlc.table = chroma_dc_coeff_token_vlc_table;
1914 chroma_dc_coeff_token_vlc.table_allocated = chroma_dc_coeff_token_vlc_table_size;
1915 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
1916 &chroma_dc_coeff_token_len [0], 1, 1,
1917 &chroma_dc_coeff_token_bits[0], 1, 1,
1918 INIT_VLC_USE_NEW_STATIC);
1922 coeff_token_vlc[i].table = coeff_token_vlc_tables+offset;
1923 coeff_token_vlc[i].table_allocated = coeff_token_vlc_tables_size[i];
1924 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
1925 &coeff_token_len [i][0], 1, 1,
1926 &coeff_token_bits[i][0], 1, 1,
1927 INIT_VLC_USE_NEW_STATIC);
1928 offset += coeff_token_vlc_tables_size[i];
1931 * This is a one time safety check to make sure that
1932 * the packed static coeff_token_vlc table sizes
1933 * were initialized correctly.
1935 assert(offset == FF_ARRAY_ELEMS(coeff_token_vlc_tables));
1938 chroma_dc_total_zeros_vlc[i].table = chroma_dc_total_zeros_vlc_tables[i];
1939 chroma_dc_total_zeros_vlc[i].table_allocated = chroma_dc_total_zeros_vlc_tables_size;
1940 init_vlc(&chroma_dc_total_zeros_vlc[i],
1941 CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
1942 &chroma_dc_total_zeros_len [i][0], 1, 1,
1943 &chroma_dc_total_zeros_bits[i][0], 1, 1,
1944 INIT_VLC_USE_NEW_STATIC);
1946 for(i=0; i<15; i++){
1947 total_zeros_vlc[i].table = total_zeros_vlc_tables[i];
1948 total_zeros_vlc[i].table_allocated = total_zeros_vlc_tables_size;
1949 init_vlc(&total_zeros_vlc[i],
1950 TOTAL_ZEROS_VLC_BITS, 16,
1951 &total_zeros_len [i][0], 1, 1,
1952 &total_zeros_bits[i][0], 1, 1,
1953 INIT_VLC_USE_NEW_STATIC);
1957 run_vlc[i].table = run_vlc_tables[i];
1958 run_vlc[i].table_allocated = run_vlc_tables_size;
1959 init_vlc(&run_vlc[i],
1961 &run_len [i][0], 1, 1,
1962 &run_bits[i][0], 1, 1,
1963 INIT_VLC_USE_NEW_STATIC);
1965 run7_vlc.table = run7_vlc_table,
1966 run7_vlc.table_allocated = run7_vlc_table_size;
1967 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
1968 &run_len [6][0], 1, 1,
1969 &run_bits[6][0], 1, 1,
1970 INIT_VLC_USE_NEW_STATIC);
1972 init_cavlc_level_tab();
1976 static void free_tables(H264Context *h){
1979 av_freep(&h->intra4x4_pred_mode);
1980 av_freep(&h->chroma_pred_mode_table);
1981 av_freep(&h->cbp_table);
1982 av_freep(&h->mvd_table[0]);
1983 av_freep(&h->mvd_table[1]);
1984 av_freep(&h->direct_table);
1985 av_freep(&h->non_zero_count);
1986 av_freep(&h->slice_table_base);
1987 h->slice_table= NULL;
1989 av_freep(&h->mb2b_xy);
1990 av_freep(&h->mb2b8_xy);
1992 for(i = 0; i < h->s.avctx->thread_count; i++) {
1993 hx = h->thread_context[i];
1995 av_freep(&hx->top_borders[1]);
1996 av_freep(&hx->top_borders[0]);
1997 av_freep(&hx->s.obmc_scratchpad);
2001 static void init_dequant8_coeff_table(H264Context *h){
2003 const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
2004 h->dequant8_coeff[0] = h->dequant8_buffer[0];
2005 h->dequant8_coeff[1] = h->dequant8_buffer[1];
2007 for(i=0; i<2; i++ ){
2008 if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
2009 h->dequant8_coeff[1] = h->dequant8_buffer[0];
2013 for(q=0; q<52; q++){
2014 int shift = div6[q];
2017 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
2018 ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
2019 h->pps.scaling_matrix8[i][x]) << shift;
2024 static void init_dequant4_coeff_table(H264Context *h){
2026 const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
2027 for(i=0; i<6; i++ ){
2028 h->dequant4_coeff[i] = h->dequant4_buffer[i];
2030 if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
2031 h->dequant4_coeff[i] = h->dequant4_buffer[j];
2038 for(q=0; q<52; q++){
2039 int shift = div6[q] + 2;
2042 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
2043 ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
2044 h->pps.scaling_matrix4[i][x]) << shift;
2049 static void init_dequant_tables(H264Context *h){
2051 init_dequant4_coeff_table(h);
2052 if(h->pps.transform_8x8_mode)
2053 init_dequant8_coeff_table(h);
2054 if(h->sps.transform_bypass){
2057 h->dequant4_coeff[i][0][x] = 1<<6;
2058 if(h->pps.transform_8x8_mode)
2061 h->dequant8_coeff[i][0][x] = 1<<6;
2068 * needs width/height
2070 static int alloc_tables(H264Context *h){
2071 MpegEncContext * const s = &h->s;
2072 const int big_mb_num= s->mb_stride * (s->mb_height+1);
2075 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
2077 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
2078 CHECKED_ALLOCZ(h->slice_table_base , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base))
2079 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2081 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2082 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2083 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2084 CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2086 memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base));
2087 h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
2089 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint32_t));
2090 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
2091 for(y=0; y<s->mb_height; y++){
2092 for(x=0; x<s->mb_width; x++){
2093 const int mb_xy= x + y*s->mb_stride;
2094 const int b_xy = 4*x + 4*y*h->b_stride;
2095 const int b8_xy= 2*x + 2*y*h->b8_stride;
2097 h->mb2b_xy [mb_xy]= b_xy;
2098 h->mb2b8_xy[mb_xy]= b8_xy;
2102 s->obmc_scratchpad = NULL;
2104 if(!h->dequant4_coeff[0])
2105 init_dequant_tables(h);
2114 * Mimic alloc_tables(), but for every context thread.
2116 static void clone_tables(H264Context *dst, H264Context *src){
2117 dst->intra4x4_pred_mode = src->intra4x4_pred_mode;
2118 dst->non_zero_count = src->non_zero_count;
2119 dst->slice_table = src->slice_table;
2120 dst->cbp_table = src->cbp_table;
2121 dst->mb2b_xy = src->mb2b_xy;
2122 dst->mb2b8_xy = src->mb2b8_xy;
2123 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
2124 dst->mvd_table[0] = src->mvd_table[0];
2125 dst->mvd_table[1] = src->mvd_table[1];
2126 dst->direct_table = src->direct_table;
2128 dst->s.obmc_scratchpad = NULL;
2129 ff_h264_pred_init(&dst->hpc, src->s.codec_id);
2134 * Allocate buffers which are not shared amongst multiple threads.
2136 static int context_init(H264Context *h){
2137 CHECKED_ALLOCZ(h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2138 CHECKED_ALLOCZ(h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2142 return -1; // free_tables will clean up for us
2145 static av_cold void common_init(H264Context *h){
2146 MpegEncContext * const s = &h->s;
2148 s->width = s->avctx->width;
2149 s->height = s->avctx->height;
2150 s->codec_id= s->avctx->codec->id;
2152 ff_h264_pred_init(&h->hpc, s->codec_id);
2154 h->dequant_coeff_pps= -1;
2155 s->unrestricted_mv=1;
2156 s->decode=1; //FIXME
2158 dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
2160 memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
2161 memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
2165 * Reset SEI values at the beginning of the frame.
2167 * @param h H.264 context.
2169 static void reset_sei(H264Context *h) {
2170 h->sei_recovery_frame_cnt = -1;
2171 h->sei_dpb_output_delay = 0;
2172 h->sei_cpb_removal_delay = -1;
2173 h->sei_buffering_period_present = 0;
2176 static av_cold int decode_init(AVCodecContext *avctx){
2177 H264Context *h= avctx->priv_data;
2178 MpegEncContext * const s = &h->s;
2180 MPV_decode_defaults(s);
2185 s->out_format = FMT_H264;
2186 s->workaround_bugs= avctx->workaround_bugs;
2189 // s->decode_mb= ff_h263_decode_mb;
2190 s->quarter_sample = 1;
2191 if(!avctx->has_b_frames)
2194 if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2195 avctx->pix_fmt= PIX_FMT_VDPAU_H264;
2197 avctx->pix_fmt= avctx->get_format(avctx, avctx->codec->pix_fmts);
2198 avctx->hwaccel = ff_find_hwaccel(avctx->codec->id, avctx->pix_fmt);
2202 if(avctx->extradata_size > 0 && avctx->extradata &&
2203 *(char *)avctx->extradata == 1){
2210 h->thread_context[0] = h;
2211 h->outputed_poc = INT_MIN;
2212 h->prev_poc_msb= 1<<16;
2214 if(avctx->codec_id == CODEC_ID_H264){
2215 if(avctx->ticks_per_frame == 1){
2216 s->avctx->time_base.den *=2;
2218 avctx->ticks_per_frame = 2;
2223 static int frame_start(H264Context *h){
2224 MpegEncContext * const s = &h->s;
2227 if(MPV_frame_start(s, s->avctx) < 0)
2229 ff_er_frame_start(s);
2231 * MPV_frame_start uses pict_type to derive key_frame.
2232 * This is incorrect for H.264; IDR markings must be used.
2233 * Zero here; IDR markings per slice in frame or fields are ORed in later.
2234 * See decode_nal_units().
2236 s->current_picture_ptr->key_frame= 0;
2238 assert(s->linesize && s->uvlinesize);
2240 for(i=0; i<16; i++){
2241 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2242 h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
2245 h->block_offset[16+i]=
2246 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2247 h->block_offset[24+16+i]=
2248 h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2251 /* can't be in alloc_tables because linesize isn't known there.
2252 * FIXME: redo bipred weight to not require extra buffer? */
2253 for(i = 0; i < s->avctx->thread_count; i++)
2254 if(!h->thread_context[i]->s.obmc_scratchpad)
2255 h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
2257 /* some macroblocks will be accessed before they're available */
2258 if(FRAME_MBAFF || s->avctx->thread_count > 1)
2259 memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
2261 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2263 // We mark the current picture as non-reference after allocating it, so
2264 // that if we break out due to an error it can be released automatically
2265 // in the next MPV_frame_start().
2266 // SVQ3 as well as most other codecs have only last/next/current and thus
2267 // get released even with set reference, besides SVQ3 and others do not
2268 // mark frames as reference later "naturally".
2269 if(s->codec_id != CODEC_ID_SVQ3)
2270 s->current_picture_ptr->reference= 0;
2272 s->current_picture_ptr->field_poc[0]=
2273 s->current_picture_ptr->field_poc[1]= INT_MAX;
2274 assert(s->current_picture_ptr->long_ref==0);
2279 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int simple){
2280 MpegEncContext * const s = &h->s;
2289 src_cb -= uvlinesize;
2290 src_cr -= uvlinesize;
2292 if(!simple && FRAME_MBAFF){
2294 offset = MB_MBAFF ? 1 : 17;
2295 uvoffset= MB_MBAFF ? 1 : 9;
2297 *(uint64_t*)(h->top_borders[0][s->mb_x]+ 0)= *(uint64_t*)(src_y + 15*linesize);
2298 *(uint64_t*)(h->top_borders[0][s->mb_x]+ 8)= *(uint64_t*)(src_y +8+15*linesize);
2299 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2300 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+7*uvlinesize);
2301 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+7*uvlinesize);
2306 h->left_border[0]= h->top_borders[0][s->mb_x][15];
2307 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2308 h->left_border[34 ]= h->top_borders[0][s->mb_x][16+7 ];
2309 h->left_border[34+18]= h->top_borders[0][s->mb_x][16+8+7];
2315 top_idx = MB_MBAFF ? 0 : 1;
2317 step= MB_MBAFF ? 2 : 1;
2320 // There are two lines saved, the line above the the top macroblock of a pair,
2321 // and the line above the bottom macroblock
2322 h->left_border[offset]= h->top_borders[top_idx][s->mb_x][15];
2323 for(i=1; i<17 - skiplast; i++){
2324 h->left_border[offset+i*step]= src_y[15+i* linesize];
2327 *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
2328 *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2330 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2331 h->left_border[uvoffset+34 ]= h->top_borders[top_idx][s->mb_x][16+7];
2332 h->left_border[uvoffset+34+18]= h->top_borders[top_idx][s->mb_x][24+7];
2333 for(i=1; i<9 - skiplast; i++){
2334 h->left_border[uvoffset+34 +i*step]= src_cb[7+i*uvlinesize];
2335 h->left_border[uvoffset+34+18+i*step]= src_cr[7+i*uvlinesize];
2337 *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2338 *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2342 static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg, int simple){
2343 MpegEncContext * const s = &h->s;
2354 if(!simple && FRAME_MBAFF){
2356 offset = MB_MBAFF ? 1 : 17;
2357 uvoffset= MB_MBAFF ? 1 : 9;
2361 top_idx = MB_MBAFF ? 0 : 1;
2363 step= MB_MBAFF ? 2 : 1;
2366 if(h->deblocking_filter == 2) {
2368 deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
2369 deblock_top = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
2371 deblock_left = (s->mb_x > 0);
2372 deblock_top = (s->mb_y > !!MB_FIELD);
2375 src_y -= linesize + 1;
2376 src_cb -= uvlinesize + 1;
2377 src_cr -= uvlinesize + 1;
2379 #define XCHG(a,b,t,xchg)\
2386 for(i = !deblock_top; i<16; i++){
2387 XCHG(h->left_border[offset+i*step], src_y [i* linesize], temp8, xchg);
2389 XCHG(h->left_border[offset+i*step], src_y [i* linesize], temp8, 1);
2393 XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2394 XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2395 if(s->mb_x+1 < s->mb_width){
2396 XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
2400 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2402 for(i = !deblock_top; i<8; i++){
2403 XCHG(h->left_border[uvoffset+34 +i*step], src_cb[i*uvlinesize], temp8, xchg);
2404 XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, xchg);
2406 XCHG(h->left_border[uvoffset+34 +i*step], src_cb[i*uvlinesize], temp8, 1);
2407 XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, 1);
2410 XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2411 XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2416 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
2417 MpegEncContext * const s = &h->s;
2418 const int mb_x= s->mb_x;
2419 const int mb_y= s->mb_y;
2420 const int mb_xy= h->mb_xy;
2421 const int mb_type= s->current_picture.mb_type[mb_xy];
2422 uint8_t *dest_y, *dest_cb, *dest_cr;
2423 int linesize, uvlinesize /*dct_offset*/;
2425 int *block_offset = &h->block_offset[0];
2426 const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
2427 /* is_h264 should always be true if SVQ3 is disabled. */
2428 const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
2429 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
2430 void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
2432 dest_y = s->current_picture.data[0] + (mb_x + mb_y * s->linesize ) * 16;
2433 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2434 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2436 s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
2437 s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
2439 if (!simple && MB_FIELD) {
2440 linesize = h->mb_linesize = s->linesize * 2;
2441 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2442 block_offset = &h->block_offset[24];
2443 if(mb_y&1){ //FIXME move out of this function?
2444 dest_y -= s->linesize*15;
2445 dest_cb-= s->uvlinesize*7;
2446 dest_cr-= s->uvlinesize*7;
2450 for(list=0; list<h->list_count; list++){
2451 if(!USES_LIST(mb_type, list))
2453 if(IS_16X16(mb_type)){
2454 int8_t *ref = &h->ref_cache[list][scan8[0]];
2455 fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
2457 for(i=0; i<16; i+=4){
2458 int ref = h->ref_cache[list][scan8[i]];
2460 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
2466 linesize = h->mb_linesize = s->linesize;
2467 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2468 // dct_offset = s->linesize * 16;
2471 if (!simple && IS_INTRA_PCM(mb_type)) {
2472 for (i=0; i<16; i++) {
2473 memcpy(dest_y + i* linesize, h->mb + i*8, 16);
2475 for (i=0; i<8; i++) {
2476 memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4, 8);
2477 memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4, 8);
2480 if(IS_INTRA(mb_type)){
2481 if(h->deblocking_filter)
2482 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
2484 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2485 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2486 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2489 if(IS_INTRA4x4(mb_type)){
2490 if(simple || !s->encoding){
2491 if(IS_8x8DCT(mb_type)){
2492 if(transform_bypass){
2494 idct_add = s->dsp.add_pixels8;
2496 idct_dc_add = s->dsp.h264_idct8_dc_add;
2497 idct_add = s->dsp.h264_idct8_add;
2499 for(i=0; i<16; i+=4){
2500 uint8_t * const ptr= dest_y + block_offset[i];
2501 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2502 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
2503 h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
2505 const int nnz = h->non_zero_count_cache[ scan8[i] ];
2506 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
2507 (h->topright_samples_available<<i)&0x4000, linesize);
2509 if(nnz == 1 && h->mb[i*16])
2510 idct_dc_add(ptr, h->mb + i*16, linesize);
2512 idct_add (ptr, h->mb + i*16, linesize);
2517 if(transform_bypass){
2519 idct_add = s->dsp.add_pixels4;
2521 idct_dc_add = s->dsp.h264_idct_dc_add;
2522 idct_add = s->dsp.h264_idct_add;
2524 for(i=0; i<16; i++){
2525 uint8_t * const ptr= dest_y + block_offset[i];
2526 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2528 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
2529 h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
2533 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2534 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2535 assert(mb_y || linesize <= block_offset[i]);
2536 if(!topright_avail){
2537 tr= ptr[3 - linesize]*0x01010101;
2538 topright= (uint8_t*) &tr;
2540 topright= ptr + 4 - linesize;
2544 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
2545 nnz = h->non_zero_count_cache[ scan8[i] ];
2548 if(nnz == 1 && h->mb[i*16])
2549 idct_dc_add(ptr, h->mb + i*16, linesize);
2551 idct_add (ptr, h->mb + i*16, linesize);
2553 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2560 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2562 if(!transform_bypass)
2563 h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
2565 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2567 if(h->deblocking_filter)
2568 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
2570 hl_motion(h, dest_y, dest_cb, dest_cr,
2571 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2572 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2573 s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
2577 if(!IS_INTRA4x4(mb_type)){
2579 if(IS_INTRA16x16(mb_type)){
2580 if(transform_bypass){
2581 if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
2582 h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
2584 for(i=0; i<16; i++){
2585 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
2586 s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
2590 s->dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2592 }else if(h->cbp&15){
2593 if(transform_bypass){
2594 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
2595 idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
2596 for(i=0; i<16; i+=di){
2597 if(h->non_zero_count_cache[ scan8[i] ]){
2598 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2602 if(IS_8x8DCT(mb_type)){
2603 s->dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2605 s->dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2610 for(i=0; i<16; i++){
2611 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2612 uint8_t * const ptr= dest_y + block_offset[i];
2613 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2619 if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
2620 uint8_t *dest[2] = {dest_cb, dest_cr};
2621 if(transform_bypass){
2622 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
2623 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
2624 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
2626 idct_add = s->dsp.add_pixels4;
2627 for(i=16; i<16+8; i++){
2628 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
2629 idct_add (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2633 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp[0], h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
2634 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp[1], h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
2636 idct_add = s->dsp.h264_idct_add;
2637 idct_dc_add = s->dsp.h264_idct_dc_add;
2638 for(i=16; i<16+8; i++){
2639 if(h->non_zero_count_cache[ scan8[i] ])
2640 idct_add (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2641 else if(h->mb[i*16])
2642 idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2645 for(i=16; i<16+8; i++){
2646 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2647 uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
2648 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2655 if(h->cbp || IS_INTRA(mb_type))
2656 s->dsp.clear_blocks(h->mb);
2658 if(h->deblocking_filter) {
2659 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, simple);
2660 fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
2661 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2662 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2663 if (!simple && FRAME_MBAFF) {
2664 filter_mb (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2666 filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2672 * Process a macroblock; this case avoids checks for expensive uncommon cases.
2674 static void hl_decode_mb_simple(H264Context *h){
2675 hl_decode_mb_internal(h, 1);
2679 * Process a macroblock; this handles edge cases, such as interlacing.
2681 static void av_noinline hl_decode_mb_complex(H264Context *h){
2682 hl_decode_mb_internal(h, 0);
2685 static void hl_decode_mb(H264Context *h){
2686 MpegEncContext * const s = &h->s;
2687 const int mb_xy= h->mb_xy;
2688 const int mb_type= s->current_picture.mb_type[mb_xy];
2689 int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
2692 hl_decode_mb_complex(h);
2693 else hl_decode_mb_simple(h);
2696 static void pic_as_field(Picture *pic, const int parity){
2698 for (i = 0; i < 4; ++i) {
2699 if (parity == PICT_BOTTOM_FIELD)
2700 pic->data[i] += pic->linesize[i];
2701 pic->reference = parity;
2702 pic->linesize[i] *= 2;
2704 pic->poc= pic->field_poc[parity == PICT_BOTTOM_FIELD];
2707 static int split_field_copy(Picture *dest, Picture *src,
2708 int parity, int id_add){
2709 int match = !!(src->reference & parity);
2713 if(parity != PICT_FRAME){
2714 pic_as_field(dest, parity);
2716 dest->pic_id += id_add;
2723 static int build_def_list(Picture *def, Picture **in, int len, int is_long, int sel){
2727 while(i[0]<len || i[1]<len){
2728 while(i[0]<len && !(in[ i[0] ] && (in[ i[0] ]->reference & sel)))
2730 while(i[1]<len && !(in[ i[1] ] && (in[ i[1] ]->reference & (sel^3))))
2733 in[ i[0] ]->pic_id= is_long ? i[0] : in[ i[0] ]->frame_num;
2734 split_field_copy(&def[index++], in[ i[0]++ ], sel , 1);
2737 in[ i[1] ]->pic_id= is_long ? i[1] : in[ i[1] ]->frame_num;
2738 split_field_copy(&def[index++], in[ i[1]++ ], sel^3, 0);
2745 static int add_sorted(Picture **sorted, Picture **src, int len, int limit, int dir){
2750 best_poc= dir ? INT_MIN : INT_MAX;
2752 for(i=0; i<len; i++){
2753 const int poc= src[i]->poc;
2754 if(((poc > limit) ^ dir) && ((poc < best_poc) ^ dir)){
2756 sorted[out_i]= src[i];
2759 if(best_poc == (dir ? INT_MIN : INT_MAX))
2761 limit= sorted[out_i++]->poc - dir;
2767 * fills the default_ref_list.
2769 static int fill_default_ref_list(H264Context *h){
2770 MpegEncContext * const s = &h->s;
2773 if(h->slice_type_nos==FF_B_TYPE){
2774 Picture *sorted[32];
2779 cur_poc= s->current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
2781 cur_poc= s->current_picture_ptr->poc;
2783 for(list= 0; list<2; list++){
2784 len= add_sorted(sorted , h->short_ref, h->short_ref_count, cur_poc, 1^list);
2785 len+=add_sorted(sorted+len, h->short_ref, h->short_ref_count, cur_poc, 0^list);
2787 len= build_def_list(h->default_ref_list[list] , sorted , len, 0, s->picture_structure);
2788 len+=build_def_list(h->default_ref_list[list]+len, h->long_ref, 16 , 1, s->picture_structure);
2791 if(len < h->ref_count[list])
2792 memset(&h->default_ref_list[list][len], 0, sizeof(Picture)*(h->ref_count[list] - len));
2796 if(lens[0] == lens[1] && lens[1] > 1){
2797 for(i=0; h->default_ref_list[0][i].data[0] == h->default_ref_list[1][i].data[0] && i<lens[0]; i++);
2799 FFSWAP(Picture, h->default_ref_list[1][0], h->default_ref_list[1][1]);
2802 len = build_def_list(h->default_ref_list[0] , h->short_ref, h->short_ref_count, 0, s->picture_structure);
2803 len+= build_def_list(h->default_ref_list[0]+len, h-> long_ref, 16 , 1, s->picture_structure);
2805 if(len < h->ref_count[0])
2806 memset(&h->default_ref_list[0][len], 0, sizeof(Picture)*(h->ref_count[0] - len));
2809 for (i=0; i<h->ref_count[0]; i++) {
2810 tprintf(h->s.avctx, "List0: %s fn:%d 0x%p\n", (h->default_ref_list[0][i].long_ref ? "LT" : "ST"), h->default_ref_list[0][i].pic_id, h->default_ref_list[0][i].data[0]);
2812 if(h->slice_type_nos==FF_B_TYPE){
2813 for (i=0; i<h->ref_count[1]; i++) {
2814 tprintf(h->s.avctx, "List1: %s fn:%d 0x%p\n", (h->default_ref_list[1][i].long_ref ? "LT" : "ST"), h->default_ref_list[1][i].pic_id, h->default_ref_list[1][i].data[0]);
2821 static void print_short_term(H264Context *h);
2822 static void print_long_term(H264Context *h);
2825 * Extract structure information about the picture described by pic_num in
2826 * the current decoding context (frame or field). Note that pic_num is
2827 * picture number without wrapping (so, 0<=pic_num<max_pic_num).
2828 * @param pic_num picture number for which to extract structure information
2829 * @param structure one of PICT_XXX describing structure of picture
2831 * @return frame number (short term) or long term index of picture
2832 * described by pic_num
2834 static int pic_num_extract(H264Context *h, int pic_num, int *structure){
2835 MpegEncContext * const s = &h->s;
2837 *structure = s->picture_structure;
2840 /* opposite field */
2841 *structure ^= PICT_FRAME;
2848 static int decode_ref_pic_list_reordering(H264Context *h){
2849 MpegEncContext * const s = &h->s;
2850 int list, index, pic_structure;
2852 print_short_term(h);
2855 for(list=0; list<h->list_count; list++){
2856 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2858 if(get_bits1(&s->gb)){
2859 int pred= h->curr_pic_num;
2861 for(index=0; ; index++){
2862 unsigned int reordering_of_pic_nums_idc= get_ue_golomb_31(&s->gb);
2863 unsigned int pic_id;
2865 Picture *ref = NULL;
2867 if(reordering_of_pic_nums_idc==3)
2870 if(index >= h->ref_count[list]){
2871 av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
2875 if(reordering_of_pic_nums_idc<3){
2876 if(reordering_of_pic_nums_idc<2){
2877 const unsigned int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2880 if(abs_diff_pic_num > h->max_pic_num){
2881 av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
2885 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2886 else pred+= abs_diff_pic_num;
2887 pred &= h->max_pic_num - 1;
2889 frame_num = pic_num_extract(h, pred, &pic_structure);
2891 for(i= h->short_ref_count-1; i>=0; i--){
2892 ref = h->short_ref[i];
2893 assert(ref->reference);
2894 assert(!ref->long_ref);
2896 ref->frame_num == frame_num &&
2897 (ref->reference & pic_structure)
2905 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2907 long_idx= pic_num_extract(h, pic_id, &pic_structure);
2910 av_log(h->s.avctx, AV_LOG_ERROR, "long_term_pic_idx overflow\n");
2913 ref = h->long_ref[long_idx];
2914 assert(!(ref && !ref->reference));
2915 if(ref && (ref->reference & pic_structure)){
2916 ref->pic_id= pic_id;
2917 assert(ref->long_ref);
2925 av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
2926 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2928 for(i=index; i+1<h->ref_count[list]; i++){
2929 if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)
2932 for(; i > index; i--){
2933 h->ref_list[list][i]= h->ref_list[list][i-1];
2935 h->ref_list[list][index]= *ref;
2937 pic_as_field(&h->ref_list[list][index], pic_structure);
2941 av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
2947 for(list=0; list<h->list_count; list++){
2948 for(index= 0; index < h->ref_count[list]; index++){
2949 if(!h->ref_list[list][index].data[0]){
2950 av_log(h->s.avctx, AV_LOG_ERROR, "Missing reference picture\n");
2951 h->ref_list[list][index]= s->current_picture; //FIXME this is not a sensible solution
2959 static void fill_mbaff_ref_list(H264Context *h){
2961 for(list=0; list<2; list++){ //FIXME try list_count
2962 for(i=0; i<h->ref_count[list]; i++){
2963 Picture *frame = &h->ref_list[list][i];
2964 Picture *field = &h->ref_list[list][16+2*i];
2967 field[0].linesize[j] <<= 1;
2968 field[0].reference = PICT_TOP_FIELD;
2969 field[0].poc= field[0].field_poc[0];
2970 field[1] = field[0];
2972 field[1].data[j] += frame->linesize[j];
2973 field[1].reference = PICT_BOTTOM_FIELD;
2974 field[1].poc= field[1].field_poc[1];
2976 h->luma_weight[list][16+2*i] = h->luma_weight[list][16+2*i+1] = h->luma_weight[list][i];
2977 h->luma_offset[list][16+2*i] = h->luma_offset[list][16+2*i+1] = h->luma_offset[list][i];
2979 h->chroma_weight[list][16+2*i][j] = h->chroma_weight[list][16+2*i+1][j] = h->chroma_weight[list][i][j];
2980 h->chroma_offset[list][16+2*i][j] = h->chroma_offset[list][16+2*i+1][j] = h->chroma_offset[list][i][j];
2984 for(j=0; j<h->ref_count[1]; j++){
2985 for(i=0; i<h->ref_count[0]; i++)
2986 h->implicit_weight[j][16+2*i] = h->implicit_weight[j][16+2*i+1] = h->implicit_weight[j][i];
2987 memcpy(h->implicit_weight[16+2*j], h->implicit_weight[j], sizeof(*h->implicit_weight));
2988 memcpy(h->implicit_weight[16+2*j+1], h->implicit_weight[j], sizeof(*h->implicit_weight));
2992 static int pred_weight_table(H264Context *h){
2993 MpegEncContext * const s = &h->s;
2995 int luma_def, chroma_def;
2998 h->use_weight_chroma= 0;
2999 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
3000 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
3001 luma_def = 1<<h->luma_log2_weight_denom;
3002 chroma_def = 1<<h->chroma_log2_weight_denom;
3004 for(list=0; list<2; list++){
3005 h->luma_weight_flag[list] = 0;
3006 h->chroma_weight_flag[list] = 0;
3007 for(i=0; i<h->ref_count[list]; i++){
3008 int luma_weight_flag, chroma_weight_flag;
3010 luma_weight_flag= get_bits1(&s->gb);
3011 if(luma_weight_flag){
3012 h->luma_weight[list][i]= get_se_golomb(&s->gb);
3013 h->luma_offset[list][i]= get_se_golomb(&s->gb);
3014 if( h->luma_weight[list][i] != luma_def
3015 || h->luma_offset[list][i] != 0) {
3017 h->luma_weight_flag[list]= 1;
3020 h->luma_weight[list][i]= luma_def;
3021 h->luma_offset[list][i]= 0;
3025 chroma_weight_flag= get_bits1(&s->gb);
3026 if(chroma_weight_flag){
3029 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3030 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3031 if( h->chroma_weight[list][i][j] != chroma_def
3032 || h->chroma_offset[list][i][j] != 0) {
3033 h->use_weight_chroma= 1;
3034 h->chroma_weight_flag[list]= 1;
3040 h->chroma_weight[list][i][j]= chroma_def;
3041 h->chroma_offset[list][i][j]= 0;
3046 if(h->slice_type_nos != FF_B_TYPE) break;
3048 h->use_weight= h->use_weight || h->use_weight_chroma;
3052 static void implicit_weight_table(H264Context *h){
3053 MpegEncContext * const s = &h->s;
3055 int cur_poc = s->current_picture_ptr->poc;
3057 for (i = 0; i < 2; i++) {
3058 h->luma_weight_flag[i] = 0;
3059 h->chroma_weight_flag[i] = 0;
3062 if( h->ref_count[0] == 1 && h->ref_count[1] == 1
3063 && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3065 h->use_weight_chroma= 0;
3070 h->use_weight_chroma= 2;
3071 h->luma_log2_weight_denom= 5;
3072 h->chroma_log2_weight_denom= 5;
3074 for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3075 int poc0 = h->ref_list[0][ref0].poc;
3076 for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3077 int poc1 = h->ref_list[1][ref1].poc;
3078 int td = av_clip(poc1 - poc0, -128, 127);
3080 int tb = av_clip(cur_poc - poc0, -128, 127);
3081 int tx = (16384 + (FFABS(td) >> 1)) / td;
3082 int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3083 if(dist_scale_factor < -64 || dist_scale_factor > 128)
3084 h->implicit_weight[ref0][ref1] = 32;
3086 h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3088 h->implicit_weight[ref0][ref1] = 32;
3094 * Mark a picture as no longer needed for reference. The refmask
3095 * argument allows unreferencing of individual fields or the whole frame.
3096 * If the picture becomes entirely unreferenced, but is being held for
3097 * display purposes, it is marked as such.
3098 * @param refmask mask of fields to unreference; the mask is bitwise
3099 * anded with the reference marking of pic
3100 * @return non-zero if pic becomes entirely unreferenced (except possibly
3101 * for display purposes) zero if one of the fields remains in
3104 static inline int unreference_pic(H264Context *h, Picture *pic, int refmask){
3106 if (pic->reference &= refmask) {
3109 for(i = 0; h->delayed_pic[i]; i++)
3110 if(pic == h->delayed_pic[i]){
3111 pic->reference=DELAYED_PIC_REF;
3119 * instantaneous decoder refresh.
3121 static void idr(H264Context *h){
3124 for(i=0; i<16; i++){
3125 remove_long(h, i, 0);
3127 assert(h->long_ref_count==0);
3129 for(i=0; i<h->short_ref_count; i++){
3130 unreference_pic(h, h->short_ref[i], 0);
3131 h->short_ref[i]= NULL;
3133 h->short_ref_count=0;
3134 h->prev_frame_num= 0;
3135 h->prev_frame_num_offset= 0;
3140 /* forget old pics after a seek */
3141 static void flush_dpb(AVCodecContext *avctx){
3142 H264Context *h= avctx->priv_data;
3144 for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
3145 if(h->delayed_pic[i])
3146 h->delayed_pic[i]->reference= 0;
3147 h->delayed_pic[i]= NULL;
3149 h->outputed_poc= INT_MIN;
3151 if(h->s.current_picture_ptr)
3152 h->s.current_picture_ptr->reference= 0;
3153 h->s.first_field= 0;
3155 ff_mpeg_flush(avctx);
3159 * Find a Picture in the short term reference list by frame number.
3160 * @param frame_num frame number to search for
3161 * @param idx the index into h->short_ref where returned picture is found
3162 * undefined if no picture found.
3163 * @return pointer to the found picture, or NULL if no pic with the provided
3164 * frame number is found
3166 static Picture * find_short(H264Context *h, int frame_num, int *idx){
3167 MpegEncContext * const s = &h->s;
3170 for(i=0; i<h->short_ref_count; i++){
3171 Picture *pic= h->short_ref[i];
3172 if(s->avctx->debug&FF_DEBUG_MMCO)
3173 av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3174 if(pic->frame_num == frame_num) {
3183 * Remove a picture from the short term reference list by its index in
3184 * that list. This does no checking on the provided index; it is assumed
3185 * to be valid. Other list entries are shifted down.
3186 * @param i index into h->short_ref of picture to remove.
3188 static void remove_short_at_index(H264Context *h, int i){
3189 assert(i >= 0 && i < h->short_ref_count);
3190 h->short_ref[i]= NULL;
3191 if (--h->short_ref_count)
3192 memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i)*sizeof(Picture*));
3197 * @return the removed picture or NULL if an error occurs
3199 static Picture * remove_short(H264Context *h, int frame_num, int ref_mask){
3200 MpegEncContext * const s = &h->s;
3204 if(s->avctx->debug&FF_DEBUG_MMCO)
3205 av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3207 pic = find_short(h, frame_num, &i);
3209 if(unreference_pic(h, pic, ref_mask))
3210 remove_short_at_index(h, i);
3217 * Remove a picture from the long term reference list by its index in
3219 * @return the removed picture or NULL if an error occurs
3221 static Picture * remove_long(H264Context *h, int i, int ref_mask){
3224 pic= h->long_ref[i];
3226 if(unreference_pic(h, pic, ref_mask)){
3227 assert(h->long_ref[i]->long_ref == 1);
3228 h->long_ref[i]->long_ref= 0;
3229 h->long_ref[i]= NULL;
3230 h->long_ref_count--;
3238 * print short term list
3240 static void print_short_term(H264Context *h) {
3242 if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3243 av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3244 for(i=0; i<h->short_ref_count; i++){
3245 Picture *pic= h->short_ref[i];
3246 av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3252 * print long term list
3254 static void print_long_term(H264Context *h) {
3256 if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3257 av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3258 for(i = 0; i < 16; i++){
3259 Picture *pic= h->long_ref[i];
3261 av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3268 * Executes the reference picture marking (memory management control operations).
3270 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3271 MpegEncContext * const s = &h->s;
3273 int current_ref_assigned=0;
3274 Picture *av_uninit(pic);
3276 if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3277 av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3279 for(i=0; i<mmco_count; i++){
3280 int structure, av_uninit(frame_num);
3281 if(s->avctx->debug&FF_DEBUG_MMCO)
3282 av_log(h->s.avctx, AV_LOG_DEBUG, "mmco:%d %d %d\n", h->mmco[i].opcode, h->mmco[i].short_pic_num, h->mmco[i].long_arg);
3284 if( mmco[i].opcode == MMCO_SHORT2UNUSED
3285 || mmco[i].opcode == MMCO_SHORT2LONG){
3286 frame_num = pic_num_extract(h, mmco[i].short_pic_num, &structure);
3287 pic = find_short(h, frame_num, &j);
3289 if(mmco[i].opcode != MMCO_SHORT2LONG || !h->long_ref[mmco[i].long_arg]
3290 || h->long_ref[mmco[i].long_arg]->frame_num != frame_num)
3291 av_log(h->s.avctx, AV_LOG_ERROR, "mmco: unref short failure\n");
3296 switch(mmco[i].opcode){
3297 case MMCO_SHORT2UNUSED:
3298 if(s->avctx->debug&FF_DEBUG_MMCO)
3299 av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: unref short %d count %d\n", h->mmco[i].short_pic_num, h->short_ref_count);
3300 remove_short(h, frame_num, structure ^ PICT_FRAME);
3302 case MMCO_SHORT2LONG:
3303 if (h->long_ref[mmco[i].long_arg] != pic)
3304 remove_long(h, mmco[i].long_arg, 0);
3306 remove_short_at_index(h, j);
3307 h->long_ref[ mmco[i].long_arg ]= pic;
3308 if (h->long_ref[ mmco[i].long_arg ]){
3309 h->long_ref[ mmco[i].long_arg ]->long_ref=1;
3310 h->long_ref_count++;
3313 case MMCO_LONG2UNUSED:
3314 j = pic_num_extract(h, mmco[i].long_arg, &structure);
3315 pic = h->long_ref[j];
3317 remove_long(h, j, structure ^ PICT_FRAME);
3318 } else if(s->avctx->debug&FF_DEBUG_MMCO)
3319 av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: unref long failure\n");
3322 // Comment below left from previous code as it is an interresting note.
3323 /* First field in pair is in short term list or
3324 * at a different long term index.
3325 * This is not allowed; see 7.4.3.3, notes 2 and 3.
3326 * Report the problem and keep the pair where it is,
3327 * and mark this field valid.
3330 if (h->long_ref[mmco[i].long_arg] != s->current_picture_ptr) {
3331 remove_long(h, mmco[i].long_arg, 0);
3333 h->long_ref[ mmco[i].long_arg ]= s->current_picture_ptr;
3334 h->long_ref[ mmco[i].long_arg ]->long_ref=1;
3335 h->long_ref_count++;
3338 s->current_picture_ptr->reference |= s->picture_structure;
3339 current_ref_assigned=1;
3341 case MMCO_SET_MAX_LONG:
3342 assert(mmco[i].long_arg <= 16);
3343 // just remove the long term which index is greater than new max
3344 for(j = mmco[i].long_arg; j<16; j++){
3345 remove_long(h, j, 0);
3349 while(h->short_ref_count){
3350 remove_short(h, h->short_ref[0]->frame_num, 0);
3352 for(j = 0; j < 16; j++) {
3353 remove_long(h, j, 0);
3355 s->current_picture_ptr->poc=
3356 s->current_picture_ptr->field_poc[0]=
3357 s->current_picture_ptr->field_poc[1]=
3361 s->current_picture_ptr->frame_num= 0;
3367 if (!current_ref_assigned) {
3368 /* Second field of complementary field pair; the first field of
3369 * which is already referenced. If short referenced, it
3370 * should be first entry in short_ref. If not, it must exist
3371 * in long_ref; trying to put it on the short list here is an
3372 * error in the encoded bit stream (ref: 7.4.3.3, NOTE 2 and 3).
3374 if (h->short_ref_count && h->short_ref[0] == s->current_picture_ptr) {
3375 /* Just mark the second field valid */
3376 s->current_picture_ptr->reference = PICT_FRAME;
3377 } else if (s->current_picture_ptr->long_ref) {
3378 av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term reference "
3379 "assignment for second field "
3380 "in complementary field pair "
3381 "(first field is long term)\n");
3383 pic= remove_short(h, s->current_picture_ptr->frame_num, 0);
3385 av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3388 if(h->short_ref_count)
3389 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3391 h->short_ref[0]= s->current_picture_ptr;
3392 h->short_ref_count++;
3393 s->current_picture_ptr->reference |= s->picture_structure;
3397 if (h->long_ref_count + h->short_ref_count > h->sps.ref_frame_count){
3399 /* We have too many reference frames, probably due to corrupted
3400 * stream. Need to discard one frame. Prevents overrun of the
3401 * short_ref and long_ref buffers.
3403 av_log(h->s.avctx, AV_LOG_ERROR,
3404 "number of reference frames exceeds max (probably "
3405 "corrupt input), discarding one\n");
3407 if (h->long_ref_count && !h->short_ref_count) {
3408 for (i = 0; i < 16; ++i)
3413 remove_long(h, i, 0);
3415 pic = h->short_ref[h->short_ref_count - 1];
3416 remove_short(h, pic->frame_num, 0);
3420 print_short_term(h);
3425 static int decode_ref_pic_marking(H264Context *h, GetBitContext *gb){
3426 MpegEncContext * const s = &h->s;
3430 if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3431 s->broken_link= get_bits1(gb) -1;
3433 h->mmco[0].opcode= MMCO_LONG;
3434 h->mmco[0].long_arg= 0;
3438 if(get_bits1(gb)){ // adaptive_ref_pic_marking_mode_flag
3439 for(i= 0; i<MAX_MMCO_COUNT; i++) {
3440 MMCOOpcode opcode= get_ue_golomb_31(gb);
3442 h->mmco[i].opcode= opcode;
3443 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3444 h->mmco[i].short_pic_num= (h->curr_pic_num - get_ue_golomb(gb) - 1) & (h->max_pic_num - 1);
3445 /* if(h->mmco[i].short_pic_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_pic_num ] == NULL){
3446 av_log(s->avctx, AV_LOG_ERROR, "illegal short ref in memory management control operation %d\n", mmco);
3450 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3451 unsigned int long_arg= get_ue_golomb_31(gb);
3452 if(long_arg >= 32 || (long_arg >= 16 && !(opcode == MMCO_LONG2UNUSED && FIELD_PICTURE))){
3453 av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
3456 h->mmco[i].long_arg= long_arg;
3459 if(opcode > (unsigned)MMCO_LONG){
3460 av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
3463 if(opcode == MMCO_END)
3468 assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
3470 if(h->short_ref_count && h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count &&
3471 !(FIELD_PICTURE && !s->first_field && s->current_picture_ptr->reference)) {
3472 h->mmco[0].opcode= MMCO_SHORT2UNUSED;
3473 h->mmco[0].short_pic_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
3475 if (FIELD_PICTURE) {
3476 h->mmco[0].short_pic_num *= 2;
3477 h->mmco[1].opcode= MMCO_SHORT2UNUSED;
3478 h->mmco[1].short_pic_num= h->mmco[0].short_pic_num + 1;
3488 static int init_poc(H264Context *h){
3489 MpegEncContext * const s = &h->s;
3490 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
3492 Picture *cur = s->current_picture_ptr;
3494 h->frame_num_offset= h->prev_frame_num_offset;
3495 if(h->frame_num < h->prev_frame_num)
3496 h->frame_num_offset += max_frame_num;
3498 if(h->sps.poc_type==0){
3499 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
3501 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
3502 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
3503 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
3504 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
3506 h->poc_msb = h->prev_poc_msb;
3507 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
3509 field_poc[1] = h->poc_msb + h->poc_lsb;
3510 if(s->picture_structure == PICT_FRAME)
3511 field_poc[1] += h->delta_poc_bottom;
3512 }else if(h->sps.poc_type==1){
3513 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
3516 if(h->sps.poc_cycle_length != 0)
3517 abs_frame_num = h->frame_num_offset + h->frame_num;
3521 if(h->nal_ref_idc==0 && abs_frame_num > 0)
3524 expected_delta_per_poc_cycle = 0;
3525 for(i=0; i < h->sps.poc_cycle_length; i++)
3526 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
3528 if(abs_frame_num > 0){
3529 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
3530 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
3532 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
3533 for(i = 0; i <= frame_num_in_poc_cycle; i++)
3534 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
3538 if(h->nal_ref_idc == 0)
3539 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
3541 field_poc[0] = expectedpoc + h->delta_poc[0];
3542 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
3544 if(s->picture_structure == PICT_FRAME)
3545 field_poc[1] += h->delta_poc[1];
3547 int poc= 2*(h->frame_num_offset + h->frame_num);
3556 if(s->picture_structure != PICT_BOTTOM_FIELD)
3557 s->current_picture_ptr->field_poc[0]= field_poc[0];
3558 if(s->picture_structure != PICT_TOP_FIELD)
3559 s->current_picture_ptr->field_poc[1]= field_poc[1];
3560 cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
3567 * initialize scan tables
3569 static void init_scan_tables(H264Context *h){
3570 MpegEncContext * const s = &h->s;
3572 if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
3573 memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
3574 memcpy(h-> field_scan, field_scan, 16*sizeof(uint8_t));
3576 for(i=0; i<16; i++){
3577 #define T(x) (x>>2) | ((x<<2) & 0xF)
3578 h->zigzag_scan[i] = T(zigzag_scan[i]);
3579 h-> field_scan[i] = T( field_scan[i]);
3583 if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
3584 memcpy(h->zigzag_scan8x8, ff_zigzag_direct, 64*sizeof(uint8_t));
3585 memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
3586 memcpy(h->field_scan8x8, field_scan8x8, 64*sizeof(uint8_t));
3587 memcpy(h->field_scan8x8_cavlc, field_scan8x8_cavlc, 64*sizeof(uint8_t));
3589 for(i=0; i<64; i++){
3590 #define T(x) (x>>3) | ((x&7)<<3)
3591 h->zigzag_scan8x8[i] = T(ff_zigzag_direct[i]);
3592 h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
3593 h->field_scan8x8[i] = T(field_scan8x8[i]);
3594 h->field_scan8x8_cavlc[i] = T(field_scan8x8_cavlc[i]);
3598 if(h->sps.transform_bypass){ //FIXME same ugly
3599 h->zigzag_scan_q0 = zigzag_scan;
3600 h->zigzag_scan8x8_q0 = ff_zigzag_direct;
3601 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
3602 h->field_scan_q0 = field_scan;
3603 h->field_scan8x8_q0 = field_scan8x8;
3604 h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc;
3606 h->zigzag_scan_q0 = h->zigzag_scan;
3607 h->zigzag_scan8x8_q0 = h->zigzag_scan8x8;
3608 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
3609 h->field_scan_q0 = h->field_scan;
3610 h->field_scan8x8_q0 = h->field_scan8x8;
3611 h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc;
3616 * Replicates H264 "master" context to thread contexts.
3618 static void clone_slice(H264Context *dst, H264Context *src)
3620 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
3621 dst->s.current_picture_ptr = src->s.current_picture_ptr;
3622 dst->s.current_picture = src->s.current_picture;
3623 dst->s.linesize = src->s.linesize;
3624 dst->s.uvlinesize = src->s.uvlinesize;
3625 dst->s.first_field = src->s.first_field;
3627 dst->prev_poc_msb = src->prev_poc_msb;
3628 dst->prev_poc_lsb = src->prev_poc_lsb;
3629 dst->prev_frame_num_offset = src->prev_frame_num_offset;
3630 dst->prev_frame_num = src->prev_frame_num;
3631 dst->short_ref_count = src->short_ref_count;
3633 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
3634 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
3635 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
3636 memcpy(dst->ref_list, src->ref_list, sizeof(dst->ref_list));
3638 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
3639 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
3643 * decodes a slice header.
3644 * This will also call MPV_common_init() and frame_start() as needed.
3646 * @param h h264context
3647 * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
3649 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
3651 static int decode_slice_header(H264Context *h, H264Context *h0){
3652 MpegEncContext * const s = &h->s;
3653 MpegEncContext * const s0 = &h0->s;
3654 unsigned int first_mb_in_slice;
3655 unsigned int pps_id;
3656 int num_ref_idx_active_override_flag;
3657 unsigned int slice_type, tmp, i, j;
3658 int default_ref_list_done = 0;
3659 int last_pic_structure;
3661 s->dropable= h->nal_ref_idc == 0;
3663 if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
3664 s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
3665 s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
3667 s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
3668 s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
3671 first_mb_in_slice= get_ue_golomb(&s->gb);
3673 if((s->flags2 & CODEC_FLAG2_CHUNKS) && first_mb_in_slice == 0){
3674 h0->current_slice = 0;
3675 if (!s0->first_field)
3676 s->current_picture_ptr= NULL;
3679 slice_type= get_ue_golomb_31(&s->gb);
3681 av_log(h->s.avctx, AV_LOG_ERROR, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);
3686 h->slice_type_fixed=1;
3688 h->slice_type_fixed=0;
3690 slice_type= golomb_to_pict_type[ slice_type ];
3691 if (slice_type == FF_I_TYPE
3692 || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
3693 default_ref_list_done = 1;
3695 h->slice_type= slice_type;
3696 h->slice_type_nos= slice_type & 3;
3698 s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
3699 if (s->pict_type == FF_B_TYPE && s0->last_picture_ptr == NULL) {
3700 av_log(h->s.avctx, AV_LOG_ERROR,
3701 "B picture before any references, skipping\n");
3705 pps_id= get_ue_golomb(&s->gb);
3706 if(pps_id>=MAX_PPS_COUNT){
3707 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3710 if(!h0->pps_buffers[pps_id]) {
3711 av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS referenced\n");
3714 h->pps= *h0->pps_buffers[pps_id];
3716 if(!h0->sps_buffers[h->pps.sps_id]) {
3717 av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS referenced\n");
3720 h->sps = *h0->sps_buffers[h->pps.sps_id];
3722 if(h == h0 && h->dequant_coeff_pps != pps_id){
3723 h->dequant_coeff_pps = pps_id;
3724 init_dequant_tables(h);
3727 s->mb_width= h->sps.mb_width;
3728 s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
3730 h->b_stride= s->mb_width*4;
3731 h->b8_stride= s->mb_width*2;
3733 s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
3734 if(h->sps.frame_mbs_only_flag)
3735 s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
3737 s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
3739 if (s->context_initialized
3740 && ( s->width != s->avctx->width || s->height != s->avctx->height)) {
3742 return -1; // width / height changed during parallelized decoding
3744 flush_dpb(s->avctx);
3747 if (!s->context_initialized) {
3749 return -1; // we cant (re-)initialize context during parallel decoding
3750 if (MPV_common_init(s) < 0)
3754 init_scan_tables(h);
3757 for(i = 1; i < s->avctx->thread_count; i++) {
3759 c = h->thread_context[i] = av_malloc(sizeof(H264Context));
3760 memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
3761 memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
3764 init_scan_tables(c);
3768 for(i = 0; i < s->avctx->thread_count; i++)
3769 if(context_init(h->thread_context[i]) < 0)
3772 s->avctx->width = s->width;
3773 s->avctx->height = s->height;
3774 s->avctx->sample_aspect_ratio= h->sps.sar;
3775 if(!s->avctx->sample_aspect_ratio.den)
3776 s->avctx->sample_aspect_ratio.den = 1;
3778 if(h->sps.timing_info_present_flag){
3779 s->avctx->time_base= (AVRational){h->sps.num_units_in_tick, h->sps.time_scale};
3780 if(h->x264_build > 0 && h->x264_build < 44)
3781 s->avctx->time_base.den *= 2;
3782 av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
3783 s->avctx->time_base.num, s->avctx->time_base.den, 1<<30);
3787 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3790 h->mb_aff_frame = 0;
3791 last_pic_structure = s0->picture_structure;
3792 if(h->sps.frame_mbs_only_flag){
3793 s->picture_structure= PICT_FRAME;
3795 if(get_bits1(&s->gb)) { //field_pic_flag
3796 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3798 s->picture_structure= PICT_FRAME;
3799 h->mb_aff_frame = h->sps.mb_aff;
3802 h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
3804 if(h0->current_slice == 0){
3805 while(h->frame_num != h->prev_frame_num &&
3806 h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
3807 av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
3808 if (frame_start(h) < 0)
3810 h->prev_frame_num++;
3811 h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
3812 s->current_picture_ptr->frame_num= h->prev_frame_num;
3813 execute_ref_pic_marking(h, NULL, 0);
3816 /* See if we have a decoded first field looking for a pair... */
3817 if (s0->first_field) {
3818 assert(s0->current_picture_ptr);
3819 assert(s0->current_picture_ptr->data[0]);
3820 assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
3822 /* figure out if we have a complementary field pair */
3823 if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
3825 * Previous field is unmatched. Don't display it, but let it
3826 * remain for reference if marked as such.
3828 s0->current_picture_ptr = NULL;
3829 s0->first_field = FIELD_PICTURE;
3832 if (h->nal_ref_idc &&
3833 s0->current_picture_ptr->reference &&
3834 s0->current_picture_ptr->frame_num != h->frame_num) {
3836 * This and previous field were reference, but had
3837 * different frame_nums. Consider this field first in
3838 * pair. Throw away previous field except for reference
3841 s0->first_field = 1;
3842 s0->current_picture_ptr = NULL;
3845 /* Second field in complementary pair */
3846 s0->first_field = 0;
3851 /* Frame or first field in a potentially complementary pair */
3852 assert(!s0->current_picture_ptr);
3853 s0->first_field = FIELD_PICTURE;
3856 if((!FIELD_PICTURE || s0->first_field) && frame_start(h) < 0) {
3857 s0->first_field = 0;
3864 s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
3866 assert(s->mb_num == s->mb_width * s->mb_height);
3867 if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
3868 first_mb_in_slice >= s->mb_num){
3869 av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
3872 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
3873 s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
3874 if (s->picture_structure == PICT_BOTTOM_FIELD)
3875 s->resync_mb_y = s->mb_y = s->mb_y + 1;
3876 assert(s->mb_y < s->mb_height);
3878 if(s->picture_structure==PICT_FRAME){
3879 h->curr_pic_num= h->frame_num;
3880 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3882 h->curr_pic_num= 2*h->frame_num + 1;
3883 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3886 if(h->nal_unit_type == NAL_IDR_SLICE){
3887 get_ue_golomb(&s->gb); /* idr_pic_id */
3890 if(h->sps.poc_type==0){
3891 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3893 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3894 h->delta_poc_bottom= get_se_golomb(&s->gb);
3898 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3899 h->delta_poc[0]= get_se_golomb(&s->gb);
3901 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3902 h->delta_poc[1]= get_se_golomb(&s->gb);
3907 if(h->pps.redundant_pic_cnt_present){
3908 h->redundant_pic_count= get_ue_golomb(&s->gb);
3911 //set defaults, might be overridden a few lines later
3912 h->ref_count[0]= h->pps.ref_count[0];
3913 h->ref_count[1]= h->pps.ref_count[1];
3915 if(h->slice_type_nos != FF_I_TYPE){
3916 if(h->slice_type_nos == FF_B_TYPE){
3917 h->direct_spatial_mv_pred= get_bits1(&s->gb);
3919 num_ref_idx_active_override_flag= get_bits1(&s->gb);
3921 if(num_ref_idx_active_override_flag){
3922 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3923 if(h->slice_type_nos==FF_B_TYPE)
3924 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3926 if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
3927 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3928 h->ref_count[0]= h->ref_count[1]= 1;
3932 if(h->slice_type_nos == FF_B_TYPE)
3939 if(!default_ref_list_done){
3940 fill_default_ref_list(h);
3943 if(h->slice_type_nos!=FF_I_TYPE && decode_ref_pic_list_reordering(h) < 0)
3946 if(h->slice_type_nos!=FF_I_TYPE){
3947 s->last_picture_ptr= &h->ref_list[0][0];
3948 ff_copy_picture(&s->last_picture, s->last_picture_ptr);
3950 if(h->slice_type_nos==FF_B_TYPE){
3951 s->next_picture_ptr= &h->ref_list[1][0];
3952 ff_copy_picture(&s->next_picture, s->next_picture_ptr);
3955 if( (h->pps.weighted_pred && h->slice_type_nos == FF_P_TYPE )
3956 || (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
3957 pred_weight_table(h);
3958 else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE)
3959 implicit_weight_table(h);
3962 for (i = 0; i < 2; i++) {
3963 h->luma_weight_flag[i] = 0;
3964 h->chroma_weight_flag[i] = 0;
3969 decode_ref_pic_marking(h0, &s->gb);
3972 fill_mbaff_ref_list(h);
3974 if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
3975 direct_dist_scale_factor(h);
3976 direct_ref_list_init(h);
3978 if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
3979 tmp = get_ue_golomb_31(&s->gb);
3981 av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3984 h->cabac_init_idc= tmp;
3987 h->last_qscale_diff = 0;
3988 tmp = h->pps.init_qp + get_se_golomb(&s->gb);
3990 av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3994 h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3995 h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3996 //FIXME qscale / qp ... stuff
3997 if(h->slice_type == FF_SP_TYPE){
3998 get_bits1(&s->gb); /* sp_for_switch_flag */
4000 if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
4001 get_se_golomb(&s->gb); /* slice_qs_delta */
4004 h->deblocking_filter = 1;
4005 h->slice_alpha_c0_offset = 0;
4006 h->slice_beta_offset = 0;
4007 if( h->pps.deblocking_filter_parameters_present ) {
4008 tmp= get_ue_golomb_31(&s->gb);
4010 av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
4013 h->deblocking_filter= tmp;
4014 if(h->deblocking_filter < 2)
4015 h->deblocking_filter^= 1; // 1<->0
4017 if( h->deblocking_filter ) {
4018 h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
4019 h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
4023 if( s->avctx->skip_loop_filter >= AVDISCARD_ALL
4024 ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
4025 ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR && h->slice_type_nos == FF_B_TYPE)
4026 ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
4027 h->deblocking_filter= 0;
4029 if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
4030 if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
4031 /* Cheat slightly for speed:
4032 Do not bother to deblock across slices. */
4033 h->deblocking_filter = 2;
4035 h0->max_contexts = 1;
4036 if(!h0->single_decode_warning) {
4037 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
4038 h0->single_decode_warning = 1;
4041 return 1; // deblocking switched inside frame
4046 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
4047 slice_group_change_cycle= get_bits(&s->gb, ?);
4050 h0->last_slice_type = slice_type;
4051 h->slice_num = ++h0->current_slice;
4052 if(h->slice_num >= MAX_SLICES){
4053 av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
4057 int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
4061 ref2frm[i+2]= 4*h->ref_list[j][i].frame_num
4062 +(h->ref_list[j][i].reference&3);
4065 for(i=16; i<48; i++)
4066 ref2frm[i+4]= 4*h->ref_list[j][i].frame_num
4067 +(h->ref_list[j][i].reference&3);
4070 h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
4071 h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
4073 s->avctx->refs= h->sps.ref_frame_count;
4075 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
4076 av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d %s mb:%d %c%s%s pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
4078 (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
4080 av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
4081 pps_id, h->frame_num,
4082 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
4083 h->ref_count[0], h->ref_count[1],
4085 h->deblocking_filter, h->slice_alpha_c0_offset/2, h->slice_beta_offset/2,
4087 h->use_weight==1 && h->use_weight_chroma ? "c" : "",
4088 h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
4098 static inline int get_level_prefix(GetBitContext *gb){
4102 OPEN_READER(re, gb);
4103 UPDATE_CACHE(re, gb);
4104 buf=GET_CACHE(re, gb);
4106 log= 32 - av_log2(buf);
4108 print_bin(buf>>(32-log), log);
4109 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__);
4112 LAST_SKIP_BITS(re, gb, log);
4113 CLOSE_READER(re, gb);
4118 static inline int get_dct8x8_allowed(H264Context *h){
4119 if(h->sps.direct_8x8_inference_flag)
4120 return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8 )*0x0001000100010001ULL));
4122 return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
4126 * decodes a residual block.
4127 * @param n block index
4128 * @param scantable scantable
4129 * @param max_coeff number of coefficients in the block
4130 * @return <0 if an error occurred
4132 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff){
4133 MpegEncContext * const s = &h->s;
4134 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};
4136 int zeros_left, coeff_num, coeff_token, total_coeff, i, j, trailing_ones, run_before;
4138 //FIXME put trailing_onex into the context
4140 if(n == CHROMA_DC_BLOCK_INDEX){
4141 coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
4142 total_coeff= coeff_token>>2;
4144 if(n == LUMA_DC_BLOCK_INDEX){
4145 total_coeff= pred_non_zero_count(h, 0);
4146 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4147 total_coeff= coeff_token>>2;
4149 total_coeff= pred_non_zero_count(h, n);
4150 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4151 total_coeff= coeff_token>>2;
4152 h->non_zero_count_cache[ scan8[n] ]= total_coeff;
4156 //FIXME set last_non_zero?
4160 if(total_coeff > (unsigned)max_coeff) {
4161 av_log(h->s.avctx, AV_LOG_ERROR, "corrupted macroblock %d %d (total_coeff=%d)\n", s->mb_x, s->mb_y, total_coeff);
4165 trailing_ones= coeff_token&3;
4166 tprintf(h->s.avctx, "trailing:%d, total:%d\n", trailing_ones, total_coeff);
4167 assert(total_coeff<=16);
4169 i = show_bits(gb, 3);
4170 skip_bits(gb, trailing_ones);
4171 level[0] = 1-((i&4)>>1);
4172 level[1] = 1-((i&2) );
4173 level[2] = 1-((i&1)<<1);
4175 if(trailing_ones<total_coeff) {
4177 int suffix_length = total_coeff > 10 && trailing_ones < 3;
4178 int bitsi= show_bits(gb, LEVEL_TAB_BITS);
4179 int level_code= cavlc_level_tab[suffix_length][bitsi][0];
4181 skip_bits(gb, cavlc_level_tab[suffix_length][bitsi][1]);
4182 if(level_code >= 100){
4183 prefix= level_code - 100;
4184 if(prefix == LEVEL_TAB_BITS)
4185 prefix += get_level_prefix(gb);
4187 //first coefficient has suffix_length equal to 0 or 1
4188 if(prefix<14){ //FIXME try to build a large unified VLC table for all this
4190 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4192 level_code= (prefix<<suffix_length); //part
4193 }else if(prefix==14){
4195 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4197 level_code= prefix + get_bits(gb, 4); //part
4199 level_code= (15<<suffix_length) + get_bits(gb, prefix-3); //part
4200 if(suffix_length==0) level_code+=15; //FIXME doesn't make (much)sense
4202 level_code += (1<<(prefix-3))-4096;
4205 if(trailing_ones < 3) level_code += 2;
4208 mask= -(level_code&1);
4209 level[trailing_ones]= (((2+level_code)>>1) ^ mask) - mask;
4211 if(trailing_ones < 3) level_code += (level_code>>31)|1;
4214 if(level_code + 3U > 6U)
4216 level[trailing_ones]= level_code;
4219 //remaining coefficients have suffix_length > 0
4220 for(i=trailing_ones+1;i<total_coeff;i++) {
4221 static const unsigned int suffix_limit[7] = {0,3,6,12,24,48,INT_MAX };
4222 int bitsi= show_bits(gb, LEVEL_TAB_BITS);
4223 level_code= cavlc_level_tab[suffix_length][bitsi][0];
4225 skip_bits(gb, cavlc_level_tab[suffix_length][bitsi][1]);
4226 if(level_code >= 100){
4227 prefix= level_code - 100;
4228 if(prefix == LEVEL_TAB_BITS){
4229 prefix += get_level_prefix(gb);
4232 level_code = (prefix<<suffix_length) + get_bits(gb, suffix_length);
4234 level_code = (15<<suffix_length) + get_bits(gb, prefix-3);
4236 level_code += (1<<(prefix-3))-4096;
4238 mask= -(level_code&1);
4239 level_code= (((2+level_code)>>1) ^ mask) - mask;
4241 level[i]= level_code;
4243 if(suffix_limit[suffix_length] + level_code > 2U*suffix_limit[suffix_length])
4248 if(total_coeff == max_coeff)
4251 if(n == CHROMA_DC_BLOCK_INDEX)
4252 zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
4254 zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
4257 coeff_num = zeros_left + total_coeff - 1;
4258 j = scantable[coeff_num];
4260 block[j] = level[0];
4261 for(i=1;i<total_coeff;i++) {
4264 else if(zeros_left < 7){
4265 run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
4267 run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
4269 zeros_left -= run_before;
4270 coeff_num -= 1 + run_before;
4271 j= scantable[ coeff_num ];
4276 block[j] = (level[0] * qmul[j] + 32)>>6;
4277 for(i=1;i<total_coeff;i++) {
4280 else if(zeros_left < 7){
4281 run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
4283 run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
4285 zeros_left -= run_before;
4286 coeff_num -= 1 + run_before;
4287 j= scantable[ coeff_num ];
4289 block[j]= (level[i] * qmul[j] + 32)>>6;
4294 av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
4301 static void predict_field_decoding_flag(H264Context *h){
4302 MpegEncContext * const s = &h->s;
4303 const int mb_xy= h->mb_xy;
4304 int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
4305 ? s->current_picture.mb_type[mb_xy-1]
4306 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
4307 ? s->current_picture.mb_type[mb_xy-s->mb_stride]
4309 h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
4313 * decodes a P_SKIP or B_SKIP macroblock
4315 static void decode_mb_skip(H264Context *h){
4316 MpegEncContext * const s = &h->s;
4317 const int mb_xy= h->mb_xy;
4320 memset(h->non_zero_count[mb_xy], 0, 16);
4321 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
4324 mb_type|= MB_TYPE_INTERLACED;
4326 if( h->slice_type_nos == FF_B_TYPE )
4328 // just for fill_caches. pred_direct_motion will set the real mb_type
4329 mb_type|= MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
4331 fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4332 pred_direct_motion(h, &mb_type);
4333 mb_type|= MB_TYPE_SKIP;
4338 mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
4340 fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4341 pred_pskip_motion(h, &mx, &my);
4342 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
4343 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
4346 write_back_motion(h, mb_type);
4347 s->current_picture.mb_type[mb_xy]= mb_type;
4348 s->current_picture.qscale_table[mb_xy]= s->qscale;
4349 h->slice_table[ mb_xy ]= h->slice_num;
4350 h->prev_mb_skipped= 1;
4354 * decodes a macroblock
4355 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
4357 static int decode_mb_cavlc(H264Context *h){
4358 MpegEncContext * const s = &h->s;
4360 int partition_count;
4361 unsigned int mb_type, cbp;
4362 int dct8x8_allowed= h->pps.transform_8x8_mode;
4364 mb_xy = h->mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4366 tprintf(s->avctx, "pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
4367 cbp = 0; /* avoid warning. FIXME: find a solution without slowing
4369 if(h->slice_type_nos != FF_I_TYPE){
4370 if(s->mb_skip_run==-1)
4371 s->mb_skip_run= get_ue_golomb(&s->gb);
4373 if (s->mb_skip_run--) {
4374 if(FRAME_MBAFF && (s->mb_y&1) == 0){
4375 if(s->mb_skip_run==0)
4376 h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);
4378 predict_field_decoding_flag(h);
4385 if( (s->mb_y&1) == 0 )
4386 h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);
4389 h->prev_mb_skipped= 0;
4391 mb_type= get_ue_golomb(&s->gb);
4392 if(h->slice_type_nos == FF_B_TYPE){
4394 partition_count= b_mb_type_info[mb_type].partition_count;
4395 mb_type= b_mb_type_info[mb_type].type;
4398 goto decode_intra_mb;
4400 }else if(h->slice_type_nos == FF_P_TYPE){
4402 partition_count= p_mb_type_info[mb_type].partition_count;
4403 mb_type= p_mb_type_info[mb_type].type;
4406 goto decode_intra_mb;
4409 assert(h->slice_type_nos == FF_I_TYPE);
4410 if(h->slice_type == FF_SI_TYPE && mb_type)
4414 av_log(h->s.avctx, AV_LOG_ERROR, "mb_type %d in %c slice too large at %d %d\n", mb_type, av_get_pict_type_char(h->slice_type), s->mb_x, s->mb_y);
4418 cbp= i_mb_type_info[mb_type].cbp;
4419 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
4420 mb_type= i_mb_type_info[mb_type].type;
4424 mb_type |= MB_TYPE_INTERLACED;
4426 h->slice_table[ mb_xy ]= h->slice_num;
4428 if(IS_INTRA_PCM(mb_type)){
4431 // We assume these blocks are very rare so we do not optimize it.
4432 align_get_bits(&s->gb);
4434 // The pixels are stored in the same order as levels in h->mb array.
4435 for(x=0; x < (CHROMA ? 384 : 256); x++){
4436 ((uint8_t*)h->mb)[x]= get_bits(&s->gb, 8);
4439 // In deblocking, the quantizer is 0
4440 s->current_picture.qscale_table[mb_xy]= 0;
4441 // All coeffs are present
4442 memset(h->non_zero_count[mb_xy], 16, 16);
4444 s->current_picture.mb_type[mb_xy]= mb_type;
4449 h->ref_count[0] <<= 1;
4450 h->ref_count[1] <<= 1;
4453 fill_caches(h, mb_type, 0);
4456 if(IS_INTRA(mb_type)){
4458 // init_top_left_availability(h);
4459 if(IS_INTRA4x4(mb_type)){
4462 if(dct8x8_allowed && get_bits1(&s->gb)){
4463 mb_type |= MB_TYPE_8x8DCT;
4467 // fill_intra4x4_pred_table(h);
4468 for(i=0; i<16; i+=di){
4469 int mode= pred_intra_mode(h, i);
4471 if(!get_bits1(&s->gb)){
4472 const int rem_mode= get_bits(&s->gb, 3);
4473 mode = rem_mode + (rem_mode >= mode);
4477 fill_rectangle( &h->intra4x4_pred_mode_cache[ scan8[i] ], 2, 2, 8, mode, 1 );
4479 h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
4481 write_back_intra_pred_mode(h);
4482 if( check_intra4x4_pred_mode(h) < 0)
4485 h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
4486 if(h->intra16x16_pred_mode < 0)
4490 pred_mode= check_intra_pred_mode(h, get_ue_golomb_31(&s->gb));
4493 h->chroma_pred_mode= pred_mode;
4495 }else if(partition_count==4){
4496 int i, j, sub_partition_count[4], list, ref[2][4];
4498 if(h->slice_type_nos == FF_B_TYPE){
4500 h->sub_mb_type[i]= get_ue_golomb_31(&s->gb);
4501 if(h->sub_mb_type[i] >=13){
4502 av_log(h->s.avctx, AV_LOG_ERROR, "B sub_mb_type %u out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
4505 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4506 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4508 if( IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])
4509 || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3])) {
4510 pred_direct_motion(h, &mb_type);
4511 h->ref_cache[0][scan8[4]] =
4512 h->ref_cache[1][scan8[4]] =
4513 h->ref_cache[0][scan8[12]] =
4514 h->ref_cache[1][scan8[12]] = PART_NOT_AVAILABLE;
4517 assert(h->slice_type_nos == FF_P_TYPE); //FIXME SP correct ?
4519 h->sub_mb_type[i]= get_ue_golomb_31(&s->gb);
4520 if(h->sub_mb_type[i] >=4){
4521 av_log(h->s.avctx, AV_LOG_ERROR, "P sub_mb_type %u out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
4524 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4525 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4529 for(list=0; list<h->list_count; list++){
4530 int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
4532 if(IS_DIRECT(h->sub_mb_type[i])) continue;
4533 if(IS_DIR(h->sub_mb_type[i], 0, list)){
4537 }else if(ref_count == 2){
4538 tmp= get_bits1(&s->gb)^1;
4540 tmp= get_ue_golomb_31(&s->gb);
4542 av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", tmp);
4555 dct8x8_allowed = get_dct8x8_allowed(h);
4557 for(list=0; list<h->list_count; list++){
4559 if(IS_DIRECT(h->sub_mb_type[i])) {
4560 h->ref_cache[list][ scan8[4*i] ] = h->ref_cache[list][ scan8[4*i]+1 ];
4563 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]=
4564 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
4566 if(IS_DIR(h->sub_mb_type[i], 0, list)){
4567 const int sub_mb_type= h->sub_mb_type[i];
4568 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
4569 for(j=0; j<sub_partition_count[i]; j++){
4571 const int index= 4*i + block_width*j;
4572 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
4573 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
4574 mx += get_se_golomb(&s->gb);
4575 my += get_se_golomb(&s->gb);
4576 tprintf(s->avctx, "final mv:%d %d\n", mx, my);
4578 if(IS_SUB_8X8(sub_mb_type)){
4580 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
4582 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
4583 }else if(IS_SUB_8X4(sub_mb_type)){
4584 mv_cache[ 1 ][0]= mx;
4585 mv_cache[ 1 ][1]= my;
4586 }else if(IS_SUB_4X8(sub_mb_type)){
4587 mv_cache[ 8 ][0]= mx;
4588 mv_cache[ 8 ][1]= my;
4590 mv_cache[ 0 ][0]= mx;
4591 mv_cache[ 0 ][1]= my;
4594 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
4600 }else if(IS_DIRECT(mb_type)){
4601 pred_direct_motion(h, &mb_type);
4602 dct8x8_allowed &= h->sps.direct_8x8_inference_flag;
4604 int list, mx, my, i;
4605 //FIXME we should set ref_idx_l? to 0 if we use that later ...
4606 if(IS_16X16(mb_type)){
4607 for(list=0; list<h->list_count; list++){
4609 if(IS_DIR(mb_type, 0, list)){
4610 if(h->ref_count[list]==1){
4612 }else if(h->ref_count[list]==2){
4613 val= get_bits1(&s->gb)^1;
4615 val= get_ue_golomb_31(&s->gb);
4616 if(val >= h->ref_count[list]){
4617 av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", val);
4622 val= LIST_NOT_USED&0xFF;
4623 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
4625 for(list=0; list<h->list_count; list++){
4627 if(IS_DIR(mb_type, 0, list)){
4628 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
4629 mx += get_se_golomb(&s->gb);
4630 my += get_se_golomb(&s->gb);
4631 tprintf(s->avctx, "final mv:%d %d\n", mx, my);
4633 val= pack16to32(mx,my);
4636 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, val, 4);
4639 else if(IS_16X8(mb_type)){
4640 for(list=0; list<h->list_count; list++){
4643 if(IS_DIR(mb_type, i, list)){
4644 if(h->ref_count[list] == 1){
4646 }else if(h->ref_count[list] == 2){
4647 val= get_bits1(&s->gb)^1;
4649 val= get_ue_golomb_31(&s->gb);
4650 if(val >= h->ref_count[list]){
4651 av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", val);
4656 val= LIST_NOT_USED&0xFF;
4657 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
4660 for(list=0; list<h->list_count; list++){
4663 if(IS_DIR(mb_type, i, list)){
4664 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
4665 mx += get_se_golomb(&s->gb);
4666 my += get_se_golomb(&s->gb);
4667 tprintf(s->avctx, "final mv:%d %d\n", mx, my);
4669 val= pack16to32(mx,my);
4672 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 4);
4676 assert(IS_8X16(mb_type));
4677 for(list=0; list<h->list_count; list++){
4680 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
4681 if(h->ref_count[list]==1){
4683 }else if(h->ref_count[list]==2){
4684 val= get_bits1(&s->gb)^1;
4686 val= get_ue_golomb_31(&s->gb);
4687 if(val >= h->ref_count[list]){
4688 av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", val);
4693 val= LIST_NOT_USED&0xFF;
4694 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
4697 for(list=0; list<h->list_count; list++){
4700 if(IS_DIR(mb_type, i, list)){
4701 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
4702 mx += get_se_golomb(&s->gb);
4703 my += get_se_golomb(&s->gb);
4704 tprintf(s->avctx, "final mv:%d %d\n", mx, my);
4706 val= pack16to32(mx,my);
4709 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 4);
4715 if(IS_INTER(mb_type))
4716 write_back_motion(h, mb_type);
4718 if(!IS_INTRA16x16(mb_type)){
4719 cbp= get_ue_golomb(&s->gb);
4721 av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%u) at %d %d\n", cbp, s->mb_x, s->mb_y);
4726 if(IS_INTRA4x4(mb_type)) cbp= golomb_to_intra4x4_cbp[cbp];
4727 else cbp= golomb_to_inter_cbp [cbp];
4729 if(IS_INTRA4x4(mb_type)) cbp= golomb_to_intra4x4_cbp_gray[cbp];
4730 else cbp= golomb_to_inter_cbp_gray[cbp];
4735 if(dct8x8_allowed && (cbp&15) && !IS_INTRA(mb_type)){
4736 if(get_bits1(&s->gb)){
4737 mb_type |= MB_TYPE_8x8DCT;
4738 h->cbp_table[mb_xy]= cbp;
4741 s->current_picture.mb_type[mb_xy]= mb_type;
4743 if(cbp || IS_INTRA16x16(mb_type)){
4744 int i8x8, i4x4, chroma_idx;
4746 GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
4747 const uint8_t *scan, *scan8x8, *dc_scan;
4749 // fill_non_zero_count_cache(h);
4751 if(IS_INTERLACED(mb_type)){
4752 scan8x8= s->qscale ? h->field_scan8x8_cavlc : h->field_scan8x8_cavlc_q0;
4753 scan= s->qscale ? h->field_scan : h->field_scan_q0;
4754 dc_scan= luma_dc_field_scan;
4756 scan8x8= s->qscale ? h->zigzag_scan8x8_cavlc : h->zigzag_scan8x8_cavlc_q0;
4757 scan= s->qscale ? h->zigzag_scan : h->zigzag_scan_q0;
4758 dc_scan= luma_dc_zigzag_scan;
4761 dquant= get_se_golomb(&s->gb);
4763 if( dquant > 25 || dquant < -26 ){
4764 av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
4768 s->qscale += dquant;
4769 if(((unsigned)s->qscale) > 51){
4770 if(s->qscale<0) s->qscale+= 52;
4771 else s->qscale-= 52;
4774 h->chroma_qp[0]= get_chroma_qp(h, 0, s->qscale);
4775 h->chroma_qp[1]= get_chroma_qp(h, 1, s->qscale);
4776 if(IS_INTRA16x16(mb_type)){
4777 if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, h->dequant4_coeff[0][s->qscale], 16) < 0){
4778 return -1; //FIXME continue if partitioned and other return -1 too
4781 assert((cbp&15) == 0 || (cbp&15) == 15);
4784 for(i8x8=0; i8x8<4; i8x8++){
4785 for(i4x4=0; i4x4<4; i4x4++){
4786 const int index= i4x4 + 4*i8x8;
4787 if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, h->dequant4_coeff[0][s->qscale], 15) < 0 ){
4793 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
4796 for(i8x8=0; i8x8<4; i8x8++){
4797 if(cbp & (1<<i8x8)){
4798 if(IS_8x8DCT(mb_type)){
4799 DCTELEM *buf = &h->mb[64*i8x8];
4801 for(i4x4=0; i4x4<4; i4x4++){
4802 if( decode_residual(h, gb, buf, i4x4+4*i8x8, scan8x8+16*i4x4,
4803 h->dequant8_coeff[IS_INTRA( mb_type ) ? 0:1][s->qscale], 16) <0 )
4806 nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
4807 nnz[0] += nnz[1] + nnz[8] + nnz[9];
4809 for(i4x4=0; i4x4<4; i4x4++){
4810 const int index= i4x4 + 4*i8x8;
4812 if( decode_residual(h, gb, h->mb + 16*index, index, scan, h->dequant4_coeff[IS_INTRA( mb_type ) ? 0:3][s->qscale], 16) <0 ){
4818 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
4819 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
4825 for(chroma_idx=0; chroma_idx<2; chroma_idx++)
4826 if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, NULL, 4) < 0){
4832 for(chroma_idx=0; chroma_idx<2; chroma_idx++){
4833 const uint32_t *qmul = h->dequant4_coeff[chroma_idx+1+(IS_INTRA( mb_type ) ? 0:3)][h->chroma_qp[chroma_idx]];
4834 for(i4x4=0; i4x4<4; i4x4++){
4835 const int index= 16 + 4*chroma_idx + i4x4;
4836 if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, qmul, 15) < 0){
4842 uint8_t * const nnz= &h->non_zero_count_cache[0];
4843 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
4844 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
4847 uint8_t * const nnz= &h->non_zero_count_cache[0];
4848 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
4849 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
4850 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
4852 s->current_picture.qscale_table[mb_xy]= s->qscale;
4853 write_back_non_zero_count(h);
4856 h->ref_count[0] >>= 1;
4857 h->ref_count[1] >>= 1;
4863 static int decode_cabac_field_decoding_flag(H264Context *h) {
4864 MpegEncContext * const s = &h->s;
4865 const int mb_x = s->mb_x;
4866 const int mb_y = s->mb_y & ~1;
4867 const int mba_xy = mb_x - 1 + mb_y *s->mb_stride;
4868 const int mbb_xy = mb_x + (mb_y-2)*s->mb_stride;
4870 unsigned int ctx = 0;
4872 if( h->slice_table[mba_xy] == h->slice_num && IS_INTERLACED( s->current_picture.mb_type[mba_xy] ) ) {
4875 if( h->slice_table[mbb_xy] == h->slice_num && IS_INTERLACED( s->current_picture.mb_type[mbb_xy] ) ) {
4879 return get_cabac_noinline( &h->cabac, &h->cabac_state[70 + ctx] );
4882 static int decode_cabac_intra_mb_type(H264Context *h, int ctx_base, int intra_slice) {
4883 uint8_t *state= &h->cabac_state[ctx_base];
4887 MpegEncContext * const s = &h->s;
4888 const int mba_xy = h->left_mb_xy[0];
4889 const int mbb_xy = h->top_mb_xy;
4891 if( h->slice_table[mba_xy] == h->slice_num && !IS_INTRA4x4( s->current_picture.mb_type[mba_xy] ) )
4893 if( h->slice_table[mbb_xy] == h->slice_num && !IS_INTRA4x4( s->current_picture.mb_type[mbb_xy] ) )
4895 if( get_cabac_noinline( &h->cabac, &state[ctx] ) == 0 )
4896 return 0; /* I4x4 */
4899 if( get_cabac_noinline( &h->cabac, &state[0] ) == 0 )
4900 return 0; /* I4x4 */
4903 if( get_cabac_terminate( &h->cabac ) )
4904 return 25; /* PCM */
4906 mb_type = 1; /* I16x16 */
4907 mb_type += 12 * get_cabac_noinline( &h->cabac, &state[1] ); /* cbp_luma != 0 */
4908 if( get_cabac_noinline( &h->cabac, &state[2] ) ) /* cbp_chroma */
4909 mb_type += 4 + 4 * get_cabac_noinline( &h->cabac, &state[2+intra_slice] );
4910 mb_type += 2 * get_cabac_noinline( &h->cabac, &state[3+intra_slice] );
4911 mb_type += 1 * get_cabac_noinline( &h->cabac, &state[3+2*intra_slice] );
4915 static int decode_cabac_mb_type_b( H264Context *h ) {
4916 MpegEncContext * const s = &h->s;
4918 const int mba_xy = h->left_mb_xy[0];
4919 const int mbb_xy = h->top_mb_xy;
4922 assert(h->slice_type_nos == FF_B_TYPE);
4924 if( h->slice_table[mba_xy] == h->slice_num && !IS_DIRECT( s->current_picture.mb_type[mba_xy] ) )
4926 if( h->slice_table[mbb_xy] == h->slice_num && !IS_DIRECT( s->current_picture.mb_type[mbb_xy] ) )
4929 if( !get_cabac_noinline( &h->cabac, &h->cabac_state[27+ctx] ) )
4930 return 0; /* B_Direct_16x16 */
4932 if( !get_cabac_noinline( &h->cabac, &h->cabac_state[27+3] ) ) {
4933 return 1 + get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] ); /* B_L[01]_16x16 */
4936 bits = get_cabac_noinline( &h->cabac, &h->cabac_state[27+4] ) << 3;
4937 bits|= get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] ) << 2;
4938 bits|= get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] ) << 1;
4939 bits|= get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] );
4941 return bits + 3; /* B_Bi_16x16 through B_L1_L0_16x8 */
4942 else if( bits == 13 ) {
4943 return decode_cabac_intra_mb_type(h, 32, 0) + 23;
4944 } else if( bits == 14 )
4945 return 11; /* B_L1_L0_8x16 */
4946 else if( bits == 15 )
4947 return 22; /* B_8x8 */
4949 bits= ( bits<<1 ) | get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] );
4950 return bits - 4; /* B_L0_Bi_* through B_Bi_Bi_* */
4953 static int decode_cabac_mb_skip( H264Context *h, int mb_x, int mb_y ) {
4954 MpegEncContext * const s = &h->s;
4958 if(FRAME_MBAFF){ //FIXME merge with the stuff in fill_caches?
4959 int mb_xy = mb_x + (mb_y&~1)*s->mb_stride;
4962 && h->slice_table[mba_xy] == h->slice_num
4963 && MB_FIELD == !!IS_INTERLACED( s->current_picture.mb_type[mba_xy] ) )
4964 mba_xy += s->mb_stride;
4966 mbb_xy = mb_xy - s->mb_stride;
4968 && h->slice_table[mbb_xy] == h->slice_num
4969 && IS_INTERLACED( s->current_picture.mb_type[mbb_xy] ) )
4970 mbb_xy -= s->mb_stride;
4972 mbb_xy = mb_x + (mb_y-1)*s->mb_stride;
4974 int mb_xy = h->mb_xy;
4976 mbb_xy = mb_xy - (s->mb_stride << FIELD_PICTURE);
4979 if( h->slice_table[mba_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mba_xy] ))
4981 if( h->slice_table[mbb_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mbb_xy] ))
4984 if( h->slice_type_nos == FF_B_TYPE )
4986 return get_cabac_noinline( &h->cabac, &h->cabac_state[11+ctx] );
4989 static int decode_cabac_mb_intra4x4_pred_mode( H264Context *h, int pred_mode ) {
4992 if( get_cabac( &h->cabac, &h->cabac_state[68] ) )
4995 mode += 1 * get_cabac( &h->cabac, &h->cabac_state[69] );
4996 mode += 2 * get_cabac( &h->cabac, &h->cabac_state[69] );
4997 mode += 4 * get_cabac( &h->cabac, &h->cabac_state[69] );
4999 if( mode >= pred_mode )
5005 static int decode_cabac_mb_chroma_pre_mode( H264Context *h) {
5006 const int mba_xy = h->left_mb_xy[0];
5007 const int mbb_xy = h->top_mb_xy;
5011 /* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */
5012 if( h->slice_table[mba_xy] == h->slice_num && h->chroma_pred_mode_table[mba_xy] != 0 )
5015 if( h->slice_table[mbb_xy] == h->slice_num && h->chroma_pred_mode_table[mbb_xy] != 0 )
5018 if( get_cabac_noinline( &h->cabac, &h->cabac_state[64+ctx] ) == 0 )
5021 if( get_cabac_noinline( &h->cabac, &h->cabac_state[64+3] ) == 0 )
5023 if( get_cabac_noinline( &h->cabac, &h->cabac_state[64+3] ) == 0 )
5029 static int decode_cabac_mb_cbp_luma( H264Context *h) {
5030 int cbp_b, cbp_a, ctx, cbp = 0;
5032 cbp_a = h->slice_table[h->left_mb_xy[0]] == h->slice_num ? h->left_cbp : -1;
5033 cbp_b = h->slice_table[h->top_mb_xy] == h->slice_num ? h->top_cbp : -1;
5035 ctx = !(cbp_a & 0x02) + 2 * !(cbp_b & 0x04);
5036 cbp |= get_cabac_noinline(&h->cabac, &h->cabac_state[73 + ctx]);
5037 ctx = !(cbp & 0x01) + 2 * !(cbp_b & 0x08);
5038 cbp |= get_cabac_noinline(&h->cabac, &h->cabac_state[73 + ctx]) << 1;
5039 ctx = !(cbp_a & 0x08) + 2 * !(cbp & 0x01);
5040 cbp |= get_cabac_noinline(&h->cabac, &h->cabac_state[73 + ctx]) << 2;
5041 ctx = !(cbp & 0x04) + 2 * !(cbp & 0x02);
5042 cbp |= get_cabac_noinline(&h->cabac, &h->cabac_state[73 + ctx]) << 3;
5045 static int decode_cabac_mb_cbp_chroma( H264Context *h) {
5049 cbp_a = (h->left_cbp>>4)&0x03;
5050 cbp_b = (h-> top_cbp>>4)&0x03;
5053 if( cbp_a > 0 ) ctx++;
5054 if( cbp_b > 0 ) ctx += 2;
5055 if( get_cabac_noinline( &h->cabac, &h->cabac_state[77 + ctx] ) == 0 )
5059 if( cbp_a == 2 ) ctx++;
5060 if( cbp_b == 2 ) ctx += 2;
5061 return 1 + get_cabac_noinline( &h->cabac, &h->cabac_state[77 + ctx] );
5063 static int decode_cabac_mb_dqp( H264Context *h) {
5064 int ctx= h->last_qscale_diff != 0;
5067 while( get_cabac_noinline( &h->cabac, &h->cabac_state[60 + ctx] ) ) {
5070 if(val > 102) //prevent infinite loop
5075 return (val + 1)>>1 ;
5077 return -((val + 1)>>1);
5079 static int decode_cabac_p_mb_sub_type( H264Context *h ) {
5080 if( get_cabac( &h->cabac, &h->cabac_state[21] ) )
5082 if( !get_cabac( &h->cabac, &h->cabac_state[22] ) )
5084 if( get_cabac( &h->cabac, &h->cabac_state[23] ) )
5088 static int decode_cabac_b_mb_sub_type( H264Context *h ) {
5090 if( !get_cabac( &h->cabac, &h->cabac_state[36] ) )
5091 return 0; /* B_Direct_8x8 */
5092 if( !get_cabac( &h->cabac, &h->cabac_state[37] ) )
5093 return 1 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L0_8x8, B_L1_8x8 */
5095 if( get_cabac( &h->cabac, &h->cabac_state[38] ) ) {
5096 if( get_cabac( &h->cabac, &h->cabac_state[39] ) )
5097 return 11 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L1_4x4, B_Bi_4x4 */
5100 type += 2*get_cabac( &h->cabac, &h->cabac_state[39] );
5101 type += get_cabac( &h->cabac, &h->cabac_state[39] );
5105 static inline int decode_cabac_mb_transform_size( H264Context *h ) {
5106 return get_cabac_noinline( &h->cabac, &h->cabac_state[399 + h->neighbor_transform_size] );
5109 static int decode_cabac_mb_ref( H264Context *h, int list, int n ) {
5110 int refa = h->ref_cache[list][scan8[n] - 1];
5111 int refb = h->ref_cache[list][scan8[n] - 8];
5115 if( h->slice_type_nos == FF_B_TYPE) {
5116 if( refa > 0 && !h->direct_cache[scan8[n] - 1] )
5118 if( refb > 0 && !h->direct_cache[scan8[n] - 8] )
5127 while( get_cabac( &h->cabac, &h->cabac_state[54+ctx] ) ) {
5130 if(ref >= 32 /*h->ref_list[list]*/){
5137 static int decode_cabac_mb_mvd( H264Context *h, int list, int n, int l ) {
5138 int amvd = abs( h->mvd_cache[list][scan8[n] - 1][l] ) +
5139 abs( h->mvd_cache[list][scan8[n] - 8][l] );
5140 int ctxbase = (l == 0) ? 40 : 47;
5142 int ctx = (amvd>2) + (amvd>32);
5144 if(!get_cabac(&h->cabac, &h->cabac_state[ctxbase+ctx]))
5149 while( mvd < 9 && get_cabac( &h->cabac, &h->cabac_state[ctxbase+ctx] ) ) {
5157 while( get_cabac_bypass( &h->cabac ) ) {
5161 av_log(h->s.avctx, AV_LOG_ERROR, "overflow in decode_cabac_mb_mvd\n");
5166 if( get_cabac_bypass( &h->cabac ) )
5170 return get_cabac_bypass_sign( &h->cabac, -mvd );
5173 static av_always_inline int get_cabac_cbf_ctx( H264Context *h, int cat, int idx, int is_dc ) {
5179 nza = h->left_cbp&0x100;
5180 nzb = h-> top_cbp&0x100;
5182 nza = (h->left_cbp>>(6+idx))&0x01;
5183 nzb = (h-> top_cbp>>(6+idx))&0x01;
5186 assert(cat == 1 || cat == 2 || cat == 4);
5187 nza = h->non_zero_count_cache[scan8[idx] - 1];
5188 nzb = h->non_zero_count_cache[scan8[idx] - 8];
5197 return ctx + 4 * cat;
5200 DECLARE_ASM_CONST(1, uint8_t, last_coeff_flag_offset_8x8[63]) = {
5201 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
5202 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
5203 3, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4,
5204 5, 5, 5, 5, 6, 6, 6, 6, 7, 7, 7, 7, 8, 8, 8
5207 static av_always_inline void decode_cabac_residual_internal( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff, int is_dc ) {
5208 static const int significant_coeff_flag_offset[2][6] = {
5209 { 105+0, 105+15, 105+29, 105+44, 105+47, 402 },
5210 { 277+0, 277+15, 277+29, 277+44, 277+47, 436 }
5212 static const int last_coeff_flag_offset[2][6] = {
5213 { 166+0, 166+15, 166+29, 166+44, 166+47, 417 },
5214 { 338+0, 338+15, 338+29, 338+44, 338+47, 451 }
5216 static const int coeff_abs_level_m1_offset[6] = {
5217 227+0, 227+10, 227+20, 227+30, 227+39, 426
5219 static const uint8_t significant_coeff_flag_offset_8x8[2][63] = {
5220 { 0, 1, 2, 3, 4, 5, 5, 4, 4, 3, 3, 4, 4, 4, 5, 5,
5221 4, 4, 4, 4, 3, 3, 6, 7, 7, 7, 8, 9,10, 9, 8, 7,
5222 7, 6,11,12,13,11, 6, 7, 8, 9,14,10, 9, 8, 6,11,
5223 12,13,11, 6, 9,14,10, 9,11,12,13,11,14,10,12 },
5224 { 0, 1, 1, 2, 2, 3, 3, 4, 5, 6, 7, 7, 7, 8, 4, 5,
5225 6, 9,10,10, 8,11,12,11, 9, 9,10,10, 8,11,12,11,
5226 9, 9,10,10, 8,11,12,11, 9, 9,10,10, 8,13,13, 9,
5227 9,10,10, 8,13,13, 9, 9,10,10,14,14,14,14,14 }
5229 /* node ctx: 0..3: abslevel1 (with abslevelgt1 == 0).
5230 * 4..7: abslevelgt1 + 3 (and abslevel1 doesn't matter).
5231 * map node ctx => cabac ctx for level=1 */
5232 static const uint8_t coeff_abs_level1_ctx[8] = { 1, 2, 3, 4, 0, 0, 0, 0 };
5233 /* map node ctx => cabac ctx for level>1 */
5234 static const uint8_t coeff_abs_levelgt1_ctx[8] = { 5, 5, 5, 5, 6, 7, 8, 9 };
5235 static const uint8_t coeff_abs_level_transition[2][8] = {
5236 /* update node ctx after decoding a level=1 */
5237 { 1, 2, 3, 3, 4, 5, 6, 7 },
5238 /* update node ctx after decoding a level>1 */
5239 { 4, 4, 4, 4, 5, 6, 7, 7 }
5245 int coeff_count = 0;
5248 uint8_t *significant_coeff_ctx_base;
5249 uint8_t *last_coeff_ctx_base;
5250 uint8_t *abs_level_m1_ctx_base;
5253 #define CABAC_ON_STACK
5255 #ifdef CABAC_ON_STACK
5258 cc.range = h->cabac.range;
5259 cc.low = h->cabac.low;
5260 cc.bytestream= h->cabac.bytestream;
5262 #define CC &h->cabac
5266 /* cat: 0-> DC 16x16 n = 0
5267 * 1-> AC 16x16 n = luma4x4idx
5268 * 2-> Luma4x4 n = luma4x4idx
5269 * 3-> DC Chroma n = iCbCr
5270 * 4-> AC Chroma n = 16 + 4 * iCbCr + chroma4x4idx
5271 * 5-> Luma8x8 n = 4 * luma8x8idx
5274 /* read coded block flag */
5275 if( is_dc || cat != 5 ) {
5276 if( get_cabac( CC, &h->cabac_state[85 + get_cabac_cbf_ctx( h, cat, n, is_dc ) ] ) == 0 ) {
5278 h->non_zero_count_cache[scan8[n]] = 0;
5280 #ifdef CABAC_ON_STACK
5281 h->cabac.range = cc.range ;
5282 h->cabac.low = cc.low ;
5283 h->cabac.bytestream= cc.bytestream;
5289 significant_coeff_ctx_base = h->cabac_state
5290 + significant_coeff_flag_offset[MB_FIELD][cat];
5291 last_coeff_ctx_base = h->cabac_state
5292 + last_coeff_flag_offset[MB_FIELD][cat];
5293 abs_level_m1_ctx_base = h->cabac_state
5294 + coeff_abs_level_m1_offset[cat];
5296 if( !is_dc && cat == 5 ) {
5297 #define DECODE_SIGNIFICANCE( coefs, sig_off, last_off ) \
5298 for(last= 0; last < coefs; last++) { \
5299 uint8_t *sig_ctx = significant_coeff_ctx_base + sig_off; \
5300 if( get_cabac( CC, sig_ctx )) { \
5301 uint8_t *last_ctx = last_coeff_ctx_base + last_off; \
5302 index[coeff_count++] = last; \
5303 if( get_cabac( CC, last_ctx ) ) { \
5309 if( last == max_coeff -1 ) {\
5310 index[coeff_count++] = last;\
5312 const uint8_t *sig_off = significant_coeff_flag_offset_8x8[MB_FIELD];
5313 #if ARCH_X86 && HAVE_7REGS && HAVE_EBX_AVAILABLE && !defined(BROKEN_RELOCATIONS)
5314 coeff_count= decode_significance_8x8_x86(CC, significant_coeff_ctx_base, index, sig_off);
5316 coeff_count= decode_significance_x86(CC, max_coeff, significant_coeff_ctx_base, index);
5318 DECODE_SIGNIFICANCE( 63, sig_off[last], last_coeff_flag_offset_8x8[last] );
5320 DECODE_SIGNIFICANCE( max_coeff - 1, last, last );
5323 assert(coeff_count > 0);
5327 h->cbp_table[h->mb_xy] |= 0x100;
5329 h->cbp_table[h->mb_xy] |= 0x40 << n;
5332 fill_rectangle(&h->non_zero_count_cache[scan8[n]], 2, 2, 8, coeff_count, 1);
5334 assert( cat == 1 || cat == 2 || cat == 4 );
5335 h->non_zero_count_cache[scan8[n]] = coeff_count;
5340 uint8_t *ctx = coeff_abs_level1_ctx[node_ctx] + abs_level_m1_ctx_base;
5342 int j= scantable[index[--coeff_count]];
5344 if( get_cabac( CC, ctx ) == 0 ) {
5345 node_ctx = coeff_abs_level_transition[0][node_ctx];
5347 block[j] = get_cabac_bypass_sign( CC, -1);
5349 block[j] = (get_cabac_bypass_sign( CC, -qmul[j]) + 32) >> 6;
5353 ctx = coeff_abs_levelgt1_ctx[node_ctx] + abs_level_m1_ctx_base;
5354 node_ctx = coeff_abs_level_transition[1][node_ctx];
5356 while( coeff_abs < 15 && get_cabac( CC, ctx ) ) {
5360 if( coeff_abs >= 15 ) {
5362 while( get_cabac_bypass( CC ) ) {
5368 coeff_abs += coeff_abs + get_cabac_bypass( CC );
5374 block[j] = get_cabac_bypass_sign( CC, -coeff_abs );
5376 block[j] = (get_cabac_bypass_sign( CC, -coeff_abs ) * qmul[j] + 32) >> 6;
5379 } while( coeff_count );
5380 #ifdef CABAC_ON_STACK
5381 h->cabac.range = cc.range ;
5382 h->cabac.low = cc.low ;
5383 h->cabac.bytestream= cc.bytestream;
5389 static void decode_cabac_residual_dc( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff ) {
5390 decode_cabac_residual_internal(h, block, cat, n, scantable, qmul, max_coeff, 1);
5393 static void decode_cabac_residual_nondc( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff ) {
5394 decode_cabac_residual_internal(h, block, cat, n, scantable, qmul, max_coeff, 0);
5398 static void decode_cabac_residual( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff ) {
5400 decode_cabac_residual_internal(h, block, cat, n, scantable, qmul, max_coeff, cat == 0 || cat == 3);
5402 if( cat == 0 || cat == 3 ) decode_cabac_residual_dc(h, block, cat, n, scantable, qmul, max_coeff);
5403 else decode_cabac_residual_nondc(h, block, cat, n, scantable, qmul, max_coeff);
5407 static inline void compute_mb_neighbors(H264Context *h)
5409 MpegEncContext * const s = &h->s;
5410 const int mb_xy = h->mb_xy;
5411 h->top_mb_xy = mb_xy - s->mb_stride;
5412 h->left_mb_xy[0] = mb_xy - 1;
5414 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride;
5415 const int top_pair_xy = pair_xy - s->mb_stride;
5416 const int top_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
5417 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
5418 const int curr_mb_field_flag = MB_FIELD;
5419 const int bottom = (s->mb_y & 1);
5421 if (curr_mb_field_flag && (bottom || top_mb_field_flag)){
5422 h->top_mb_xy -= s->mb_stride;
5424 if (!left_mb_field_flag == curr_mb_field_flag) {
5425 h->left_mb_xy[0] = pair_xy - 1;
5427 } else if (FIELD_PICTURE) {
5428 h->top_mb_xy -= s->mb_stride;
5434 * decodes a macroblock
5435 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
5437 static int decode_mb_cabac(H264Context *h) {
5438 MpegEncContext * const s = &h->s;
5440 int mb_type, partition_count, cbp = 0;
5441 int dct8x8_allowed= h->pps.transform_8x8_mode;
5443 mb_xy = h->mb_xy = s->mb_x + s->mb_y*s->mb_stride;
5445 tprintf(s->avctx, "pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
5446 if( h->slice_type_nos != FF_I_TYPE ) {
5448 /* a skipped mb needs the aff flag from the following mb */
5449 if( FRAME_MBAFF && s->mb_x==0 && (s->mb_y&1)==0 )
5450 predict_field_decoding_flag(h);
5451 if( FRAME_MBAFF && (s->mb_y&1)==1 && h->prev_mb_skipped )
5452 skip = h->next_mb_skipped;
5454 skip = decode_cabac_mb_skip( h, s->mb_x, s->mb_y );
5455 /* read skip flags */
5457 if( FRAME_MBAFF && (s->mb_y&1)==0 ){
5458 s->current_picture.mb_type[mb_xy] = MB_TYPE_SKIP;
5459 h->next_mb_skipped = decode_cabac_mb_skip( h, s->mb_x, s->mb_y+1 );
5460 if(!h->next_mb_skipped)
5461 h->mb_mbaff = h->mb_field_decoding_flag = decode_cabac_field_decoding_flag(h);
5466 h->cbp_table[mb_xy] = 0;
5467 h->chroma_pred_mode_table[mb_xy] = 0;
5468 h->last_qscale_diff = 0;
5475 if( (s->mb_y&1) == 0 )
5477 h->mb_field_decoding_flag = decode_cabac_field_decoding_flag(h);
5480 h->prev_mb_skipped = 0;
5482 compute_mb_neighbors(h);
5484 if( h->slice_type_nos == FF_B_TYPE ) {
5485 mb_type = decode_cabac_mb_type_b( h );
5487 partition_count= b_mb_type_info[mb_type].partition_count;
5488 mb_type= b_mb_type_info[mb_type].type;
5491 goto decode_intra_mb;
5493 } else if( h->slice_type_nos == FF_P_TYPE ) {
5494 if( get_cabac_noinline( &h->cabac, &h->cabac_state[14] ) == 0 ) {
5496 if( get_cabac_noinline( &h->cabac, &h->cabac_state[15] ) == 0 ) {
5497 /* P_L0_D16x16, P_8x8 */
5498 mb_type= 3 * get_cabac_noinline( &h->cabac, &h->cabac_state[16] );
5500 /* P_L0_D8x16, P_L0_D16x8 */
5501 mb_type= 2 - get_cabac_noinline( &h->cabac, &h->cabac_state[17] );
5503 partition_count= p_mb_type_info[mb_type].partition_count;
5504 mb_type= p_mb_type_info[mb_type].type;
5506 mb_type= decode_cabac_intra_mb_type(h, 17, 0);
5507 goto decode_intra_mb;
5510 mb_type= decode_cabac_intra_mb_type(h, 3, 1);
5511 if(h->slice_type == FF_SI_TYPE && mb_type)
5513 assert(h->slice_type_nos == FF_I_TYPE);
5515 partition_count = 0;
5516 cbp= i_mb_type_info[mb_type].cbp;
5517 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
5518 mb_type= i_mb_type_info[mb_type].type;
5521 mb_type |= MB_TYPE_INTERLACED;
5523 h->slice_table[ mb_xy ]= h->slice_num;
5525 if(IS_INTRA_PCM(mb_type)) {
5528 // We assume these blocks are very rare so we do not optimize it.
5529 // FIXME The two following lines get the bitstream position in the cabac
5530 // decode, I think it should be done by a function in cabac.h (or cabac.c).
5531 ptr= h->cabac.bytestream;
5532 if(h->cabac.low&0x1) ptr--;
5534 if(h->cabac.low&0x1FF) ptr--;
5537 // The pixels are stored in the same order as levels in h->mb array.
5538 memcpy(h->mb, ptr, 256); ptr+=256;
5540 memcpy(h->mb+128, ptr, 128); ptr+=128;
5543 ff_init_cabac_decoder(&h->cabac, ptr, h->cabac.bytestream_end - ptr);
5545 // All blocks are present
5546 h->cbp_table[mb_xy] = 0x1ef;
5547 h->chroma_pred_mode_table[mb_xy] = 0;
5548 // In deblocking, the quantizer is 0
5549 s->current_picture.qscale_table[mb_xy]= 0;
5550 // All coeffs are present
5551 memset(h->non_zero_count[mb_xy], 16, 16);
5552 s->current_picture.mb_type[mb_xy]= mb_type;
5553 h->last_qscale_diff = 0;
5558 h->ref_count[0] <<= 1;
5559 h->ref_count[1] <<= 1;
5562 fill_caches(h, mb_type, 0);
5564 if( IS_INTRA( mb_type ) ) {
5566 if( IS_INTRA4x4( mb_type ) ) {
5567 if( dct8x8_allowed && decode_cabac_mb_transform_size( h ) ) {
5568 mb_type |= MB_TYPE_8x8DCT;
5569 for( i = 0; i < 16; i+=4 ) {
5570 int pred = pred_intra_mode( h, i );
5571 int mode = decode_cabac_mb_intra4x4_pred_mode( h, pred );
5572 fill_rectangle( &h->intra4x4_pred_mode_cache[ scan8[i] ], 2, 2, 8, mode, 1 );
5575 for( i = 0; i < 16; i++ ) {
5576 int pred = pred_intra_mode( h, i );
5577 h->intra4x4_pred_mode_cache[ scan8[i] ] = decode_cabac_mb_intra4x4_pred_mode( h, pred );
5579 //av_log( s->avctx, AV_LOG_ERROR, "i4x4 pred=%d mode=%d\n", pred, h->intra4x4_pred_mode_cache[ scan8[i] ] );
5582 write_back_intra_pred_mode(h);
5583 if( check_intra4x4_pred_mode(h) < 0 ) return -1;
5585 h->intra16x16_pred_mode= check_intra_pred_mode( h, h->intra16x16_pred_mode );
5586 if( h->intra16x16_pred_mode < 0 ) return -1;
5589 h->chroma_pred_mode_table[mb_xy] =
5590 pred_mode = decode_cabac_mb_chroma_pre_mode( h );
5592 pred_mode= check_intra_pred_mode( h, pred_mode );
5593 if( pred_mode < 0 ) return -1;
5594 h->chroma_pred_mode= pred_mode;
5596 } else if( partition_count == 4 ) {
5597 int i, j, sub_partition_count[4], list, ref[2][4];
5599 if( h->slice_type_nos == FF_B_TYPE ) {
5600 for( i = 0; i < 4; i++ ) {
5601 h->sub_mb_type[i] = decode_cabac_b_mb_sub_type( h );
5602 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
5603 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
5605 if( IS_DIRECT(h->sub_mb_type[0] | h->sub_mb_type[1] |
5606 h->sub_mb_type[2] | h->sub_mb_type[3]) ) {
5607 pred_direct_motion(h, &mb_type);
5608 h->ref_cache[0][scan8[4]] =
5609 h->ref_cache[1][scan8[4]] =
5610 h->ref_cache[0][scan8[12]] =
5611 h->ref_cache[1][scan8[12]] = PART_NOT_AVAILABLE;
5612 if( h->ref_count[0] > 1 || h->ref_count[1] > 1 ) {
5613 for( i = 0; i < 4; i++ )
5614 if( IS_DIRECT(h->sub_mb_type[i]) )
5615 fill_rectangle( &h->direct_cache[scan8[4*i]], 2, 2, 8, 1, 1 );
5619 for( i = 0; i < 4; i++ ) {
5620 h->sub_mb_type[i] = decode_cabac_p_mb_sub_type( h );
5621 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
5622 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
5626 for( list = 0; list < h->list_count; list++ ) {
5627 for( i = 0; i < 4; i++ ) {
5628 if(IS_DIRECT(h->sub_mb_type[i])) continue;
5629 if(IS_DIR(h->sub_mb_type[i], 0, list)){
5630 if( h->ref_count[list] > 1 ){
5631 ref[list][i] = decode_cabac_mb_ref( h, list, 4*i );
5632 if(ref[list][i] >= (unsigned)h->ref_count[list]){
5633 av_log(s->avctx, AV_LOG_ERROR, "Reference %d >= %d\n", ref[list][i], h->ref_count[list]);
5641 h->ref_cache[list][ scan8[4*i]+1 ]=
5642 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
5647 dct8x8_allowed = get_dct8x8_allowed(h);
5649 for(list=0; list<h->list_count; list++){
5651 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ];
5652 if(IS_DIRECT(h->sub_mb_type[i])){
5653 fill_rectangle(h->mvd_cache[list][scan8[4*i]], 2, 2, 8, 0, 4);
5657 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
5658 const int sub_mb_type= h->sub_mb_type[i];
5659 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
5660 for(j=0; j<sub_partition_count[i]; j++){
5663 const int index= 4*i + block_width*j;
5664 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
5665 int16_t (* mvd_cache)[2]= &h->mvd_cache[list][ scan8[index] ];
5666 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mpx, &mpy);
5668 mx = mpx + decode_cabac_mb_mvd( h, list, index, 0 );
5669 my = mpy + decode_cabac_mb_mvd( h, list, index, 1 );
5670 tprintf(s->avctx, "final mv:%d %d\n", mx, my);
5672 if(IS_SUB_8X8(sub_mb_type)){
5674 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
5676 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
5679 mvd_cache[ 8 ][0]= mvd_cache[ 9 ][0]= mx - mpx;
5681 mvd_cache[ 8 ][1]= mvd_cache[ 9 ][1]= my - mpy;
5682 }else if(IS_SUB_8X4(sub_mb_type)){
5683 mv_cache[ 1 ][0]= mx;
5684 mv_cache[ 1 ][1]= my;
5686 mvd_cache[ 1 ][0]= mx - mpx;
5687 mvd_cache[ 1 ][1]= my - mpy;
5688 }else if(IS_SUB_4X8(sub_mb_type)){
5689 mv_cache[ 8 ][0]= mx;
5690 mv_cache[ 8 ][1]= my;
5692 mvd_cache[ 8 ][0]= mx - mpx;
5693 mvd_cache[ 8 ][1]= my - mpy;
5695 mv_cache[ 0 ][0]= mx;
5696 mv_cache[ 0 ][1]= my;
5698 mvd_cache[ 0 ][0]= mx - mpx;
5699 mvd_cache[ 0 ][1]= my - mpy;
5702 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
5703 uint32_t *pd= (uint32_t *)&h->mvd_cache[list][ scan8[4*i] ][0];
5704 p[0] = p[1] = p[8] = p[9] = 0;
5705 pd[0]= pd[1]= pd[8]= pd[9]= 0;
5709 } else if( IS_DIRECT(mb_type) ) {
5710 pred_direct_motion(h, &mb_type);
5711 fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
5712 fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4);
5713 dct8x8_allowed &= h->sps.direct_8x8_inference_flag;
5715 int list, mx, my, i, mpx, mpy;
5716 if(IS_16X16(mb_type)){
5717 for(list=0; list<h->list_count; list++){
5718 if(IS_DIR(mb_type, 0, list)){
5720 if(h->ref_count[list] > 1){
5721 ref= decode_cabac_mb_ref(h, list, 0);
5722 if(ref >= (unsigned)h->ref_count[list]){
5723 av_log(s->avctx, AV_LOG_ERROR, "Reference %d >= %d\n", ref, h->ref_count[list]);
5728 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, ref, 1);
5730 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1); //FIXME factorize and the other fill_rect below too
5732 for(list=0; list<h->list_count; list++){
5733 if(IS_DIR(mb_type, 0, list)){
5734 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mpx, &mpy);
5736 mx = mpx + decode_cabac_mb_mvd( h, list, 0, 0 );
5737 my = mpy + decode_cabac_mb_mvd( h, list, 0, 1 );
5738 tprintf(s->avctx, "final mv:%d %d\n", mx, my);
5740 fill_rectangle(h->mvd_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx-mpx,my-mpy), 4);
5741 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
5743 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, 0, 4);
5746 else if(IS_16X8(mb_type)){
5747 for(list=0; list<h->list_count; list++){
5749 if(IS_DIR(mb_type, i, list)){
5751 if(h->ref_count[list] > 1){
5752 ref= decode_cabac_mb_ref( h, list, 8*i );
5753 if(ref >= (unsigned)h->ref_count[list]){
5754 av_log(s->avctx, AV_LOG_ERROR, "Reference %d >= %d\n", ref, h->ref_count[list]);
5759 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, ref, 1);
5761 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1);
5764 for(list=0; list<h->list_count; list++){
5766 if(IS_DIR(mb_type, i, list)){
5767 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mpx, &mpy);
5768 mx = mpx + decode_cabac_mb_mvd( h, list, 8*i, 0 );
5769 my = mpy + decode_cabac_mb_mvd( h, list, 8*i, 1 );
5770 tprintf(s->avctx, "final mv:%d %d\n", mx, my);
5772 fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx-mpx,my-mpy), 4);
5773 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
5775 fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
5776 fill_rectangle(h-> mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
5781 assert(IS_8X16(mb_type));
5782 for(list=0; list<h->list_count; list++){
5784 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
5786 if(h->ref_count[list] > 1){
5787 ref= decode_cabac_mb_ref( h, list, 4*i );
5788 if(ref >= (unsigned)h->ref_count[list]){
5789 av_log(s->avctx, AV_LOG_ERROR, "Reference %d >= %d\n", ref, h->ref_count[list]);
5794 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, ref, 1);
5796 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1);
5799 for(list=0; list<h->list_count; list++){
5801 if(IS_DIR(mb_type, i, list)){
5802 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mpx, &mpy);
5803 mx = mpx + decode_cabac_mb_mvd( h, list, 4*i, 0 );
5804 my = mpy + decode_cabac_mb_mvd( h, list, 4*i, 1 );
5806 tprintf(s->avctx, "final mv:%d %d\n", mx, my);
5807 fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx-mpx,my-mpy), 4);
5808 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
5810 fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
5811 fill_rectangle(h-> mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
5818 if( IS_INTER( mb_type ) ) {
5819 h->chroma_pred_mode_table[mb_xy] = 0;
5820 write_back_motion( h, mb_type );
5823 if( !IS_INTRA16x16( mb_type ) ) {
5824 cbp = decode_cabac_mb_cbp_luma( h );
5826 cbp |= decode_cabac_mb_cbp_chroma( h ) << 4;
5829 h->cbp_table[mb_xy] = h->cbp = cbp;
5831 if( dct8x8_allowed && (cbp&15) && !IS_INTRA( mb_type ) ) {
5832 if( decode_cabac_mb_transform_size( h ) )
5833 mb_type |= MB_TYPE_8x8DCT;
5835 s->current_picture.mb_type[mb_xy]= mb_type;
5837 if( cbp || IS_INTRA16x16( mb_type ) ) {
5838 const uint8_t *scan, *scan8x8, *dc_scan;
5839 const uint32_t *qmul;
5842 if(IS_INTERLACED(mb_type)){
5843 scan8x8= s->qscale ? h->field_scan8x8 : h->field_scan8x8_q0;
5844 scan= s->qscale ? h->field_scan : h->field_scan_q0;
5845 dc_scan= luma_dc_field_scan;
5847 scan8x8= s->qscale ? h->zigzag_scan8x8 : h->zigzag_scan8x8_q0;
5848 scan= s->qscale ? h->zigzag_scan : h->zigzag_scan_q0;
5849 dc_scan= luma_dc_zigzag_scan;
5852 h->last_qscale_diff = dqp = decode_cabac_mb_dqp( h );
5853 if( dqp == INT_MIN ){
5854 av_log(h->s.avctx, AV_LOG_ERROR, "cabac decode of qscale diff failed at %d %d\n", s->mb_x, s->mb_y);
5858 if(((unsigned)s->qscale) > 51){
5859 if(s->qscale<0) s->qscale+= 52;
5860 else s->qscale-= 52;
5862 h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
5863 h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
5865 if( IS_INTRA16x16( mb_type ) ) {
5867 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 DC\n" );
5868 decode_cabac_residual( h, h->mb, 0, 0, dc_scan, NULL, 16);
5871 qmul = h->dequant4_coeff[0][s->qscale];
5872 for( i = 0; i < 16; i++ ) {
5873 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 AC:%d\n", i );
5874 decode_cabac_residual(h, h->mb + 16*i, 1, i, scan + 1, qmul, 15);
5877 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
5881 for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
5882 if( cbp & (1<<i8x8) ) {
5883 if( IS_8x8DCT(mb_type) ) {
5884 decode_cabac_residual(h, h->mb + 64*i8x8, 5, 4*i8x8,
5885 scan8x8, h->dequant8_coeff[IS_INTRA( mb_type ) ? 0:1][s->qscale], 64);
5887 qmul = h->dequant4_coeff[IS_INTRA( mb_type ) ? 0:3][s->qscale];
5888 for( i4x4 = 0; i4x4 < 4; i4x4++ ) {
5889 const int index = 4*i8x8 + i4x4;
5890 //av_log( s->avctx, AV_LOG_ERROR, "Luma4x4: %d\n", index );
5892 decode_cabac_residual(h, h->mb + 16*index, 2, index, scan, qmul, 16);
5893 //STOP_TIMER("decode_residual")
5897 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
5898 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
5905 for( c = 0; c < 2; c++ ) {
5906 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-DC\n",c );
5907 decode_cabac_residual(h, h->mb + 256 + 16*4*c, 3, c, chroma_dc_scan, NULL, 4);
5913 for( c = 0; c < 2; c++ ) {
5914 qmul = h->dequant4_coeff[c+1+(IS_INTRA( mb_type ) ? 0:3)][h->chroma_qp[c]];
5915 for( i = 0; i < 4; i++ ) {
5916 const int index = 16 + 4 * c + i;
5917 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-AC %d\n",c, index - 16 );
5918 decode_cabac_residual(h, h->mb + 16*index, 4, index, scan + 1, qmul, 15);
5922 uint8_t * const nnz= &h->non_zero_count_cache[0];
5923 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
5924 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
5927 uint8_t * const nnz= &h->non_zero_count_cache[0];
5928 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
5929 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
5930 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
5931 h->last_qscale_diff = 0;
5934 s->current_picture.qscale_table[mb_xy]= s->qscale;
5935 write_back_non_zero_count(h);
5938 h->ref_count[0] >>= 1;
5939 h->ref_count[1] >>= 1;
5946 static void filter_mb_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) {
5947 const int index_a = qp + h->slice_alpha_c0_offset;
5948 const int alpha = (alpha_table+52)[index_a];
5949 const int beta = (beta_table+52)[qp + h->slice_beta_offset];
5953 tc[0] = (tc0_table+52)[index_a][bS[0]];
5954 tc[1] = (tc0_table+52)[index_a][bS[1]];
5955 tc[2] = (tc0_table+52)[index_a][bS[2]];
5956 tc[3] = (tc0_table+52)[index_a][bS[3]];
5957 h->s.dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc);
5959 h->s.dsp.h264_h_loop_filter_luma_intra(pix, stride, alpha, beta);
5962 static void filter_mb_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) {
5963 const int index_a = qp + h->slice_alpha_c0_offset;
5964 const int alpha = (alpha_table+52)[index_a];
5965 const int beta = (beta_table+52)[qp + h->slice_beta_offset];
5969 tc[0] = (tc0_table+52)[index_a][bS[0]]+1;
5970 tc[1] = (tc0_table+52)[index_a][bS[1]]+1;
5971 tc[2] = (tc0_table+52)[index_a][bS[2]]+1;
5972 tc[3] = (tc0_table+52)[index_a][bS[3]]+1;
5973 h->s.dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc);
5975 h->s.dsp.h264_h_loop_filter_chroma_intra(pix, stride, alpha, beta);
5979 static void filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[8], int qp[2] ) {
5981 for( i = 0; i < 16; i++, pix += stride) {
5987 int bS_index = (i >> 1);
5990 bS_index |= (i & 1);
5993 if( bS[bS_index] == 0 ) {
5997 qp_index = MB_FIELD ? (i >> 3) : (i & 1);
5998 index_a = qp[qp_index] + h->slice_alpha_c0_offset;
5999 alpha = (alpha_table+52)[index_a];
6000 beta = (beta_table+52)[qp[qp_index] + h->slice_beta_offset];
6002 if( bS[bS_index] < 4 ) {
6003 const int tc0 = (tc0_table+52)[index_a][bS[bS_index]];
6004 const int p0 = pix[-1];
6005 const int p1 = pix[-2];
6006 const int p2 = pix[-3];
6007 const int q0 = pix[0];
6008 const int q1 = pix[1];
6009 const int q2 = pix[2];
6011 if( FFABS( p0 - q0 ) < alpha &&
6012 FFABS( p1 - p0 ) < beta &&
6013 FFABS( q1 - q0 ) < beta ) {
6017 if( FFABS( p2 - p0 ) < beta ) {
6018 pix[-2] = p1 + av_clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
6021 if( FFABS( q2 - q0 ) < beta ) {
6022 pix[1] = q1 + av_clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
6026 i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
6027 pix[-1] = av_clip_uint8( p0 + i_delta ); /* p0' */
6028 pix[0] = av_clip_uint8( q0 - i_delta ); /* q0' */
6029 tprintf(h->s.avctx, "filter_mb_mbaff_edgev i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, tc, bS[bS_index], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);
6032 const int p0 = pix[-1];
6033 const int p1 = pix[-2];
6034 const int p2 = pix[-3];
6036 const int q0 = pix[0];
6037 const int q1 = pix[1];
6038 const int q2 = pix[2];
6040 if( FFABS( p0 - q0 ) < alpha &&
6041 FFABS( p1 - p0 ) < beta &&
6042 FFABS( q1 - q0 ) < beta ) {
6044 if(FFABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
6045 if( FFABS( p2 - p0 ) < beta)
6047 const int p3 = pix[-4];
6049 pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
6050 pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
6051 pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
6054 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
6056 if( FFABS( q2 - q0 ) < beta)
6058 const int q3 = pix[3];
6060 pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
6061 pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
6062 pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
6065 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
6069 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
6070 pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
6072 tprintf(h->s.avctx, "filter_mb_mbaff_edgev i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, p2, p1, p0, q0, q1, q2, pix[-3], pix[-2], pix[-1], pix[0], pix[1], pix[2]);
6077 static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[8], int qp[2] ) {
6079 for( i = 0; i < 8; i++, pix += stride) {
6087 if( bS[bS_index] == 0 ) {
6091 qp_index = MB_FIELD ? (i >> 2) : (i & 1);
6092 index_a = qp[qp_index] + h->slice_alpha_c0_offset;
6093 alpha = (alpha_table+52)[index_a];
6094 beta = (beta_table+52)[qp[qp_index] + h->slice_beta_offset];
6096 if( bS[bS_index] < 4 ) {
6097 const int tc = (tc0_table+52)[index_a][bS[bS_index]] + 1;
6098 const int p0 = pix[-1];
6099 const int p1 = pix[-2];
6100 const int q0 = pix[0];
6101 const int q1 = pix[1];
6103 if( FFABS( p0 - q0 ) < alpha &&
6104 FFABS( p1 - p0 ) < beta &&
6105 FFABS( q1 - q0 ) < beta ) {
6106 const int i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
6108 pix[-1] = av_clip_uint8( p0 + i_delta ); /* p0' */
6109 pix[0] = av_clip_uint8( q0 - i_delta ); /* q0' */
6110 tprintf(h->s.avctx, "filter_mb_mbaff_edgecv i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, tc, bS[bS_index], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);
6113 const int p0 = pix[-1];
6114 const int p1 = pix[-2];
6115 const int q0 = pix[0];
6116 const int q1 = pix[1];
6118 if( FFABS( p0 - q0 ) < alpha &&
6119 FFABS( p1 - p0 ) < beta &&
6120 FFABS( q1 - q0 ) < beta ) {
6122 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
6123 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
6124 tprintf(h->s.avctx, "filter_mb_mbaff_edgecv i:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x, %02x, %02x]\n", i, pix[-3], p1, p0, q0, q1, pix[2], pix[-3], pix[-2], pix[-1], pix[0], pix[1], pix[2]);
6130 static void filter_mb_edgeh( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) {
6131 const int index_a = qp + h->slice_alpha_c0_offset;
6132 const int alpha = (alpha_table+52)[index_a];
6133 const int beta = (beta_table+52)[qp + h->slice_beta_offset];
6137 tc[0] = (tc0_table+52)[index_a][bS[0]];
6138 tc[1] = (tc0_table+52)[index_a][bS[1]];
6139 tc[2] = (tc0_table+52)[index_a][bS[2]];
6140 tc[3] = (tc0_table+52)[index_a][bS[3]];
6141 h->s.dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc);
6143 h->s.dsp.h264_v_loop_filter_luma_intra(pix, stride, alpha, beta);
6147 static void filter_mb_edgech( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) {
6148 const int index_a = qp + h->slice_alpha_c0_offset;
6149 const int alpha = (alpha_table+52)[index_a];
6150 const int beta = (beta_table+52)[qp + h->slice_beta_offset];
6154 tc[0] = (tc0_table+52)[index_a][bS[0]]+1;
6155 tc[1] = (tc0_table+52)[index_a][bS[1]]+1;
6156 tc[2] = (tc0_table+52)[index_a][bS[2]]+1;
6157 tc[3] = (tc0_table+52)[index_a][bS[3]]+1;
6158 h->s.dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc);
6160 h->s.dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta);
6164 static void filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize) {
6165 MpegEncContext * const s = &h->s;
6166 int mb_y_firstrow = s->picture_structure == PICT_BOTTOM_FIELD;
6168 int qp, qp0, qp1, qpc, qpc0, qpc1, qp_thresh;
6172 if(mb_x==0 || mb_y==mb_y_firstrow || !s->dsp.h264_loop_filter_strength || h->pps.chroma_qp_diff ||
6173 !(s->flags2 & CODEC_FLAG2_FAST) || //FIXME filter_mb_fast is broken, thus hasto be, but should not under CODEC_FLAG2_FAST
6174 (h->deblocking_filter == 2 && (h->slice_table[mb_xy] != h->slice_table[h->top_mb_xy] ||
6175 h->slice_table[mb_xy] != h->slice_table[mb_xy - 1]))) {
6176 filter_mb(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize);
6179 assert(!FRAME_MBAFF);
6181 mb_type = s->current_picture.mb_type[mb_xy];
6182 qp = s->current_picture.qscale_table[mb_xy];
6183 qp0 = s->current_picture.qscale_table[mb_xy-1];
6184 qp1 = s->current_picture.qscale_table[h->top_mb_xy];
6185 qpc = get_chroma_qp( h, 0, qp );
6186 qpc0 = get_chroma_qp( h, 0, qp0 );
6187 qpc1 = get_chroma_qp( h, 0, qp1 );
6188 qp0 = (qp + qp0 + 1) >> 1;
6189 qp1 = (qp + qp1 + 1) >> 1;
6190 qpc0 = (qpc + qpc0 + 1) >> 1;
6191 qpc1 = (qpc + qpc1 + 1) >> 1;
6192 qp_thresh = 15 - h->slice_alpha_c0_offset;
6193 if(qp <= qp_thresh && qp0 <= qp_thresh && qp1 <= qp_thresh &&
6194 qpc <= qp_thresh && qpc0 <= qp_thresh && qpc1 <= qp_thresh)
6197 if( IS_INTRA(mb_type) ) {
6198 int16_t bS4[4] = {4,4,4,4};
6199 int16_t bS3[4] = {3,3,3,3};
6200 int16_t *bSH = FIELD_PICTURE ? bS3 : bS4;
6201 if( IS_8x8DCT(mb_type) ) {
6202 filter_mb_edgev( h, &img_y[4*0], linesize, bS4, qp0 );
6203 filter_mb_edgev( h, &img_y[4*2], linesize, bS3, qp );
6204 filter_mb_edgeh( h, &img_y[4*0*linesize], linesize, bSH, qp1 );
6205 filter_mb_edgeh( h, &img_y[4*2*linesize], linesize, bS3, qp );
6207 filter_mb_edgev( h, &img_y[4*0], linesize, bS4, qp0 );
6208 filter_mb_edgev( h, &img_y[4*1], linesize, bS3, qp );
6209 filter_mb_edgev( h, &img_y[4*2], linesize, bS3, qp );
6210 filter_mb_edgev( h, &img_y[4*3], linesize, bS3, qp );
6211 filter_mb_edgeh( h, &img_y[4*0*linesize], linesize, bSH, qp1 );
6212 filter_mb_edgeh( h, &img_y[4*1*linesize], linesize, bS3, qp );
6213 filter_mb_edgeh( h, &img_y[4*2*linesize], linesize, bS3, qp );
6214 filter_mb_edgeh( h, &img_y[4*3*linesize], linesize, bS3, qp );
6216 filter_mb_edgecv( h, &img_cb[2*0], uvlinesize, bS4, qpc0 );
6217 filter_mb_edgecv( h, &img_cb[2*2], uvlinesize, bS3, qpc );
6218 filter_mb_edgecv( h, &img_cr[2*0], uvlinesize, bS4, qpc0 );
6219 filter_mb_edgecv( h, &img_cr[2*2], uvlinesize, bS3, qpc );
6220 filter_mb_edgech( h, &img_cb[2*0*uvlinesize], uvlinesize, bSH, qpc1 );
6221 filter_mb_edgech( h, &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc );
6222 filter_mb_edgech( h, &img_cr[2*0*uvlinesize], uvlinesize, bSH, qpc1 );
6223 filter_mb_edgech( h, &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc );
6226 DECLARE_ALIGNED_8(int16_t, bS[2][4][4]);
6227 uint64_t (*bSv)[4] = (uint64_t(*)[4])bS;
6229 if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 ) {
6231 bSv[0][0] = bSv[0][2] = bSv[1][0] = bSv[1][2] = 0x0002000200020002ULL;
6233 int mask_edge1 = (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 :
6234 (mb_type & MB_TYPE_16x8) ? 1 : 0;
6235 int mask_edge0 = (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16))
6236 && (s->current_picture.mb_type[mb_xy-1] & (MB_TYPE_16x16 | MB_TYPE_8x16))
6238 int step = IS_8x8DCT(mb_type) ? 2 : 1;
6239 edges = (mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4;
6240 s->dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache,
6241 (h->slice_type_nos == FF_B_TYPE), edges, step, mask_edge0, mask_edge1, FIELD_PICTURE);
6243 if( IS_INTRA(s->current_picture.mb_type[mb_xy-1]) )
6244 bSv[0][0] = 0x0004000400040004ULL;
6245 if( IS_INTRA(s->current_picture.mb_type[h->top_mb_xy]) )
6246 bSv[1][0] = FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL;
6248 #define FILTER(hv,dir,edge)\
6249 if(bSv[dir][edge]) {\
6250 filter_mb_edge##hv( h, &img_y[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qp : qp##dir );\
6252 filter_mb_edgec##hv( h, &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir );\
6253 filter_mb_edgec##hv( h, &img_cr[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir );\
6259 } else if( IS_8x8DCT(mb_type) ) {
6279 static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize, int mb_xy, int mb_type, int mvy_limit, int first_vertical_edge_done, int dir) {
6280 MpegEncContext * const s = &h->s;
6282 const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy;
6283 const int mbm_type = s->current_picture.mb_type[mbm_xy];
6284 int (*ref2frm) [64] = h->ref2frm[ h->slice_num &(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
6285 int (*ref2frmm)[64] = h->ref2frm[ h->slice_table[mbm_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
6286 int start = h->slice_table[mbm_xy] == 0xFFFF ? 1 : 0;
6288 const int edges = (mb_type & (MB_TYPE_16x16|MB_TYPE_SKIP))
6289 == (MB_TYPE_16x16|MB_TYPE_SKIP) ? 1 : 4;
6290 // how often to recheck mv-based bS when iterating between edges
6291 const int mask_edge = (mb_type & (MB_TYPE_16x16 | (MB_TYPE_16x8 << dir))) ? 3 :
6292 (mb_type & (MB_TYPE_8x16 >> dir)) ? 1 : 0;
6293 // how often to recheck mv-based bS when iterating along each edge
6294 const int mask_par0 = mb_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir));
6296 if (first_vertical_edge_done) {
6300 if (h->deblocking_filter==2 && h->slice_table[mbm_xy] != h->slice_table[mb_xy])
6303 if (FRAME_MBAFF && (dir == 1) && ((mb_y&1) == 0) && start == 0
6304 && !IS_INTERLACED(mb_type)
6305 && IS_INTERLACED(mbm_type)
6307 // This is a special case in the norm where the filtering must
6308 // be done twice (one each of the field) even if we are in a
6309 // frame macroblock.
6311 static const int nnz_idx[4] = {4,5,6,3};
6312 unsigned int tmp_linesize = 2 * linesize;
6313 unsigned int tmp_uvlinesize = 2 * uvlinesize;
6314 int mbn_xy = mb_xy - 2 * s->mb_stride;
6319 for(j=0; j<2; j++, mbn_xy += s->mb_stride){
6320 if( IS_INTRA(mb_type) ||
6321 IS_INTRA(s->current_picture.mb_type[mbn_xy]) ) {
6322 bS[0] = bS[1] = bS[2] = bS[3] = 3;
6324 const uint8_t *mbn_nnz = h->non_zero_count[mbn_xy];
6325 for( i = 0; i < 4; i++ ) {
6326 if( h->non_zero_count_cache[scan8[0]+i] != 0 ||
6327 mbn_nnz[nnz_idx[i]] != 0 )
6333 // Do not use s->qscale as luma quantizer because it has not the same
6334 // value in IPCM macroblocks.
6335 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
6336 tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, tmp_linesize, tmp_uvlinesize);
6337 { int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
6338 filter_mb_edgeh( h, &img_y[j*linesize], tmp_linesize, bS, qp );
6339 filter_mb_edgech( h, &img_cb[j*uvlinesize], tmp_uvlinesize, bS,
6340 ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1);
6341 filter_mb_edgech( h, &img_cr[j*uvlinesize], tmp_uvlinesize, bS,
6342 ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1);
6349 for( edge = start; edge < edges; edge++ ) {
6350 /* mbn_xy: neighbor macroblock */
6351 const int mbn_xy = edge > 0 ? mb_xy : mbm_xy;
6352 const int mbn_type = s->current_picture.mb_type[mbn_xy];
6353 int (*ref2frmn)[64] = edge > 0 ? ref2frm : ref2frmm;
6357 if( (edge&1) && IS_8x8DCT(mb_type) )
6360 if( IS_INTRA(mb_type) ||
6361 IS_INTRA(mbn_type) ) {
6364 if ( (!IS_INTERLACED(mb_type) && !IS_INTERLACED(mbm_type))
6365 || ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0))
6374 bS[0] = bS[1] = bS[2] = bS[3] = value;
6379 if( edge & mask_edge ) {
6380 bS[0] = bS[1] = bS[2] = bS[3] = 0;
6383 else if( FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbn_type)) {
6384 bS[0] = bS[1] = bS[2] = bS[3] = 1;
6387 else if( mask_par0 && (edge || (mbn_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) {
6388 int b_idx= 8 + 4 + edge * (dir ? 8:1);
6389 int bn_idx= b_idx - (dir ? 8:1);
6392 for( l = 0; !v && l < 1 + (h->slice_type_nos == FF_B_TYPE); l++ ) {
6393 v |= ref2frm[l][h->ref_cache[l][b_idx]] != ref2frmn[l][h->ref_cache[l][bn_idx]] ||
6394 FFABS( h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] ) >= 4 ||
6395 FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= mvy_limit;
6398 if(h->slice_type_nos == FF_B_TYPE && v){
6400 for( l = 0; !v && l < 2; l++ ) {
6402 v |= ref2frm[l][h->ref_cache[l][b_idx]] != ref2frmn[ln][h->ref_cache[ln][bn_idx]] ||
6403 FFABS( h->mv_cache[l][b_idx][0] - h->mv_cache[ln][bn_idx][0] ) >= 4 ||
6404 FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[ln][bn_idx][1] ) >= mvy_limit;
6408 bS[0] = bS[1] = bS[2] = bS[3] = v;
6414 for( i = 0; i < 4; i++ ) {
6415 int x = dir == 0 ? edge : i;
6416 int y = dir == 0 ? i : edge;
6417 int b_idx= 8 + 4 + x + 8*y;
6418 int bn_idx= b_idx - (dir ? 8:1);
6420 if( h->non_zero_count_cache[b_idx] |
6421 h->non_zero_count_cache[bn_idx] ) {
6427 for( l = 0; l < 1 + (h->slice_type_nos == FF_B_TYPE); l++ ) {
6428 if( ref2frm[l][h->ref_cache[l][b_idx]] != ref2frmn[l][h->ref_cache[l][bn_idx]] ||
6429 FFABS( h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] ) >= 4 ||
6430 FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= mvy_limit ) {
6436 if(h->slice_type_nos == FF_B_TYPE && bS[i]){
6438 for( l = 0; l < 2; l++ ) {
6440 if( ref2frm[l][h->ref_cache[l][b_idx]] != ref2frmn[ln][h->ref_cache[ln][bn_idx]] ||
6441 FFABS( h->mv_cache[l][b_idx][0] - h->mv_cache[ln][bn_idx][0] ) >= 4 ||
6442 FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[ln][bn_idx][1] ) >= mvy_limit ) {
6451 if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
6456 // Do not use s->qscale as luma quantizer because it has not the same
6457 // value in IPCM macroblocks.
6458 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
6459 //tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d, QPc:%d, QPcn:%d\n", mb_x, mb_y, dir, edge, qp, h->chroma_qp, s->current_picture.qscale_table[mbn_xy]);
6460 tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, linesize, uvlinesize);
6461 { int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
6463 filter_mb_edgev( h, &img_y[4*edge], linesize, bS, qp );
6464 if( (edge&1) == 0 ) {
6465 filter_mb_edgecv( h, &img_cb[2*edge], uvlinesize, bS,
6466 ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1);
6467 filter_mb_edgecv( h, &img_cr[2*edge], uvlinesize, bS,
6468 ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1);
6471 filter_mb_edgeh( h, &img_y[4*edge*linesize], linesize, bS, qp );
6472 if( (edge&1) == 0 ) {
6473 filter_mb_edgech( h, &img_cb[2*edge*uvlinesize], uvlinesize, bS,
6474 ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1);
6475 filter_mb_edgech( h, &img_cr[2*edge*uvlinesize], uvlinesize, bS,
6476 ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1);
6482 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize) {
6483 MpegEncContext * const s = &h->s;
6484 const int mb_xy= mb_x + mb_y*s->mb_stride;
6485 const int mb_type = s->current_picture.mb_type[mb_xy];
6486 const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4;
6487 int first_vertical_edge_done = 0;
6490 //for sufficiently low qp, filtering wouldn't do anything
6491 //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
6493 int qp_thresh = 15 - h->slice_alpha_c0_offset - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1]);
6494 int qp = s->current_picture.qscale_table[mb_xy];
6496 && (mb_x == 0 || ((qp + s->current_picture.qscale_table[mb_xy-1] + 1)>>1) <= qp_thresh)
6497 && (mb_y == 0 || ((qp + s->current_picture.qscale_table[h->top_mb_xy] + 1)>>1) <= qp_thresh)){
6502 // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
6503 if(!h->pps.cabac && h->pps.transform_8x8_mode){
6504 int top_type, left_type[2];
6505 top_type = s->current_picture.mb_type[h->top_mb_xy] ;
6506 left_type[0] = s->current_picture.mb_type[h->left_mb_xy[0]];
6507 left_type[1] = s->current_picture.mb_type[h->left_mb_xy[1]];
6509 if(IS_8x8DCT(top_type)){
6510 h->non_zero_count_cache[4+8*0]=
6511 h->non_zero_count_cache[5+8*0]= h->cbp_table[h->top_mb_xy] & 4;
6512 h->non_zero_count_cache[6+8*0]=
6513 h->non_zero_count_cache[7+8*0]= h->cbp_table[h->top_mb_xy] & 8;
6515 if(IS_8x8DCT(left_type[0])){
6516 h->non_zero_count_cache[3+8*1]=
6517 h->non_zero_count_cache[3+8*2]= h->cbp_table[h->left_mb_xy[0]]&2; //FIXME check MBAFF
6519 if(IS_8x8DCT(left_type[1])){
6520 h->non_zero_count_cache[3+8*3]=
6521 h->non_zero_count_cache[3+8*4]= h->cbp_table[h->left_mb_xy[1]]&8; //FIXME check MBAFF
6524 if(IS_8x8DCT(mb_type)){
6525 h->non_zero_count_cache[scan8[0 ]]= h->non_zero_count_cache[scan8[1 ]]=
6526 h->non_zero_count_cache[scan8[2 ]]= h->non_zero_count_cache[scan8[3 ]]= h->cbp & 1;
6528 h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
6529 h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
6531 h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
6532 h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
6534 h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
6535 h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
6540 // left mb is in picture
6541 && h->slice_table[mb_xy-1] != 0xFFFF
6542 // and current and left pair do not have the same interlaced type
6543 && (IS_INTERLACED(mb_type) != IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]))
6544 // and left mb is in the same slice if deblocking_filter == 2
6545 && (h->deblocking_filter!=2 || h->slice_table[mb_xy-1] == h->slice_table[mb_xy])) {
6546 /* First vertical edge is different in MBAFF frames
6547 * There are 8 different bS to compute and 2 different Qp
6549 const int pair_xy = mb_x + (mb_y&~1)*s->mb_stride;
6550 const int left_mb_xy[2] = { pair_xy-1, pair_xy-1+s->mb_stride };
6555 int mb_qp, mbn0_qp, mbn1_qp;
6557 first_vertical_edge_done = 1;
6559 if( IS_INTRA(mb_type) )
6560 bS[0] = bS[1] = bS[2] = bS[3] = bS[4] = bS[5] = bS[6] = bS[7] = 4;
6562 for( i = 0; i < 8; i++ ) {
6563 int mbn_xy = MB_FIELD ? left_mb_xy[i>>2] : left_mb_xy[i&1];
6565 if( IS_INTRA( s->current_picture.mb_type[mbn_xy] ) )
6567 else if( h->non_zero_count_cache[12+8*(i>>1)] != 0 ||
6568 ((!h->pps.cabac && IS_8x8DCT(s->current_picture.mb_type[mbn_xy])) ?
6569 (h->cbp_table[mbn_xy] & ((MB_FIELD ? (i&2) : (mb_y&1)) ? 8 : 2))
6571 h->non_zero_count[mbn_xy][MB_FIELD ? i&3 : (i>>2)+(mb_y&1)*2]))
6578 mb_qp = s->current_picture.qscale_table[mb_xy];
6579 mbn0_qp = s->current_picture.qscale_table[left_mb_xy[0]];
6580 mbn1_qp = s->current_picture.qscale_table[left_mb_xy[1]];
6581 qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1;
6582 bqp[0] = ( get_chroma_qp( h, 0, mb_qp ) +
6583 get_chroma_qp( h, 0, mbn0_qp ) + 1 ) >> 1;
6584 rqp[0] = ( get_chroma_qp( h, 1, mb_qp ) +
6585 get_chroma_qp( h, 1, mbn0_qp ) + 1 ) >> 1;
6586 qp[1] = ( mb_qp + mbn1_qp + 1 ) >> 1;
6587 bqp[1] = ( get_chroma_qp( h, 0, mb_qp ) +
6588 get_chroma_qp( h, 0, mbn1_qp ) + 1 ) >> 1;
6589 rqp[1] = ( get_chroma_qp( h, 1, mb_qp ) +
6590 get_chroma_qp( h, 1, mbn1_qp ) + 1 ) >> 1;
6593 tprintf(s->avctx, "filter mb:%d/%d MBAFF, QPy:%d/%d, QPb:%d/%d QPr:%d/%d ls:%d uvls:%d", mb_x, mb_y, qp[0], qp[1], bqp[0], bqp[1], rqp[0], rqp[1], linesize, uvlinesize);
6594 { int i; for (i = 0; i < 8; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
6595 filter_mb_mbaff_edgev ( h, &img_y [0], linesize, bS, qp );
6596 filter_mb_mbaff_edgecv( h, &img_cb[0], uvlinesize, bS, bqp );
6597 filter_mb_mbaff_edgecv( h, &img_cr[0], uvlinesize, bS, rqp );
6601 for( dir = 0; dir < 2; dir++ )
6602 filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, dir ? 0 : first_vertical_edge_done, dir);
6604 filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, first_vertical_edge_done, 0);
6605 filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, 0, 1);
6609 static int decode_slice(struct AVCodecContext *avctx, void *arg){
6610 H264Context *h = *(void**)arg;
6611 MpegEncContext * const s = &h->s;
6612 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
6616 h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
6617 (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
6619 if( h->pps.cabac ) {
6623 align_get_bits( &s->gb );
6626 ff_init_cabac_states( &h->cabac);
6627 ff_init_cabac_decoder( &h->cabac,
6628 s->gb.buffer + get_bits_count(&s->gb)/8,
6629 ( s->gb.size_in_bits - get_bits_count(&s->gb) + 7)/8);
6630 /* calculate pre-state */
6631 for( i= 0; i < 460; i++ ) {
6633 if( h->slice_type_nos == FF_I_TYPE )
6634 pre = av_clip( ((cabac_context_init_I[i][0] * s->qscale) >>4 ) + cabac_context_init_I[i][1], 1, 126 );
6636 pre = av_clip( ((cabac_context_init_PB[h->cabac_init_idc][i][0] * s->qscale) >>4 ) + cabac_context_init_PB[h->cabac_init_idc][i][1], 1, 126 );
6639 h->cabac_state[i] = 2 * ( 63 - pre ) + 0;
6641 h->cabac_state[i] = 2 * ( pre - 64 ) + 1;
6646 int ret = decode_mb_cabac(h);
6648 //STOP_TIMER("decode_mb_cabac")
6650 if(ret>=0) hl_decode_mb(h);
6652 if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
6655 ret = decode_mb_cabac(h);
6657 if(ret>=0) hl_decode_mb(h);
6660 eos = get_cabac_terminate( &h->cabac );
6662 if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
6663 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d, bytestream (%td)\n", s->mb_x, s->mb_y, h->cabac.bytestream_end - h->cabac.bytestream);
6664 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
6668 if( ++s->mb_x >= s->mb_width ) {
6670 ff_draw_horiz_band(s, 16*s->mb_y, 16);
6672 if(FIELD_OR_MBAFF_PICTURE) {
6677 if( eos || s->mb_y >= s->mb_height ) {
6678 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
6679 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
6686 int ret = decode_mb_cavlc(h);
6688 if(ret>=0) hl_decode_mb(h);
6690 if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
6692 ret = decode_mb_cavlc(h);
6694 if(ret>=0) hl_decode_mb(h);
6699 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
6700 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
6705 if(++s->mb_x >= s->mb_width){
6707 ff_draw_horiz_band(s, 16*s->mb_y, 16);
6709 if(FIELD_OR_MBAFF_PICTURE) {
6712 if(s->mb_y >= s->mb_height){
6713 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
6715 if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
6716 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
6720 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
6727 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
6728 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
6729 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
6730 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
6734 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
6743 for(;s->mb_y < s->mb_height; s->mb_y++){
6744 for(;s->mb_x < s->mb_width; s->mb_x++){
6745 int ret= decode_mb(h);
6750 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
6751 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
6756 if(++s->mb_x >= s->mb_width){
6758 if(++s->mb_y >= s->mb_height){
6759 if(get_bits_count(s->gb) == s->gb.size_in_bits){
6760 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
6764 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
6771 if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
6772 if(get_bits_count(s->gb) == s->gb.size_in_bits){
6773 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
6777 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
6784 ff_draw_horiz_band(s, 16*s->mb_y, 16);
6787 return -1; //not reached
6790 static int decode_picture_timing(H264Context *h){
6791 MpegEncContext * const s = &h->s;
6792 if(h->sps.nal_hrd_parameters_present_flag || h->sps.vcl_hrd_parameters_present_flag){
6793 h->sei_cpb_removal_delay = get_bits(&s->gb, h->sps.cpb_removal_delay_length);
6794 h->sei_dpb_output_delay = get_bits(&s->gb, h->sps.dpb_output_delay_length);
6796 if(h->sps.pic_struct_present_flag){
6797 unsigned int i, num_clock_ts;
6798 h->sei_pic_struct = get_bits(&s->gb, 4);
6800 if (h->sei_pic_struct > SEI_PIC_STRUCT_FRAME_TRIPLING)
6803 num_clock_ts = sei_num_clock_ts_table[h->sei_pic_struct];
6805 for (i = 0 ; i < num_clock_ts ; i++){
6806 if(get_bits(&s->gb, 1)){ /* clock_timestamp_flag */
6807 unsigned int full_timestamp_flag;
6808 skip_bits(&s->gb, 2); /* ct_type */
6809 skip_bits(&s->gb, 1); /* nuit_field_based_flag */
6810 skip_bits(&s->gb, 5); /* counting_type */
6811 full_timestamp_flag = get_bits(&s->gb, 1);
6812 skip_bits(&s->gb, 1); /* discontinuity_flag */
6813 skip_bits(&s->gb, 1); /* cnt_dropped_flag */
6814 skip_bits(&s->gb, 8); /* n_frames */
6815 if(full_timestamp_flag){
6816 skip_bits(&s->gb, 6); /* seconds_value 0..59 */
6817 skip_bits(&s->gb, 6); /* minutes_value 0..59 */
6818 skip_bits(&s->gb, 5); /* hours_value 0..23 */
6820 if(get_bits(&s->gb, 1)){ /* seconds_flag */
6821 skip_bits(&s->gb, 6); /* seconds_value range 0..59 */
6822 if(get_bits(&s->gb, 1)){ /* minutes_flag */
6823 skip_bits(&s->gb, 6); /* minutes_value 0..59 */
6824 if(get_bits(&s->gb, 1)) /* hours_flag */
6825 skip_bits(&s->gb, 5); /* hours_value 0..23 */
6829 if(h->sps.time_offset_length > 0)
6830 skip_bits(&s->gb, h->sps.time_offset_length); /* time_offset */
6837 static int decode_unregistered_user_data(H264Context *h, int size){
6838 MpegEncContext * const s = &h->s;
6839 uint8_t user_data[16+256];
6845 for(i=0; i<sizeof(user_data)-1 && i<size; i++){
6846 user_data[i]= get_bits(&s->gb, 8);
6850 e= sscanf(user_data+16, "x264 - core %d"/*%s - H.264/MPEG-4 AVC codec - Copyleft 2005 - http://www.videolan.org/x264.html*/, &build);
6851 if(e==1 && build>=0)
6852 h->x264_build= build;
6854 if(s->avctx->debug & FF_DEBUG_BUGS)
6855 av_log(s->avctx, AV_LOG_DEBUG, "user data:\"%s\"\n", user_data+16);
6858 skip_bits(&s->gb, 8);
6863 static int decode_recovery_point(H264Context *h){
6864 MpegEncContext * const s = &h->s;
6866 h->sei_recovery_frame_cnt = get_ue_golomb(&s->gb);
6867 skip_bits(&s->gb, 4); /* 1b exact_match_flag, 1b broken_link_flag, 2b changing_slice_group_idc */
6872 static int decode_buffering_period(H264Context *h){
6873 MpegEncContext * const s = &h->s;
6874 unsigned int sps_id;
6878 sps_id = get_ue_golomb_31(&s->gb);
6879 if(sps_id > 31 || !h->sps_buffers[sps_id]) {
6880 av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %d referenced in buffering period\n", sps_id);
6883 sps = h->sps_buffers[sps_id];
6885 // NOTE: This is really so duplicated in the standard... See H.264, D.1.1
6886 if (sps->nal_hrd_parameters_present_flag) {
6887 for (sched_sel_idx = 0; sched_sel_idx < sps->cpb_cnt; sched_sel_idx++) {
6888 h->initial_cpb_removal_delay[sched_sel_idx] = get_bits(&s->gb, sps->initial_cpb_removal_delay_length);
6889 skip_bits(&s->gb, sps->initial_cpb_removal_delay_length); // initial_cpb_removal_delay_offset
6892 if (sps->vcl_hrd_parameters_present_flag) {
6893 for (sched_sel_idx = 0; sched_sel_idx < sps->cpb_cnt; sched_sel_idx++) {
6894 h->initial_cpb_removal_delay[sched_sel_idx] = get_bits(&s->gb, sps->initial_cpb_removal_delay_length);
6895 skip_bits(&s->gb, sps->initial_cpb_removal_delay_length); // initial_cpb_removal_delay_offset
6899 h->sei_buffering_period_present = 1;
6903 int ff_h264_decode_sei(H264Context *h){
6904 MpegEncContext * const s = &h->s;
6906 while(get_bits_count(&s->gb) + 16 < s->gb.size_in_bits){
6911 type+= show_bits(&s->gb, 8);
6912 }while(get_bits(&s->gb, 8) == 255);
6916 size+= show_bits(&s->gb, 8);
6917 }while(get_bits(&s->gb, 8) == 255);
6920 case SEI_TYPE_PIC_TIMING: // Picture timing SEI
6921 if(decode_picture_timing(h) < 0)
6924 case SEI_TYPE_USER_DATA_UNREGISTERED:
6925 if(decode_unregistered_user_data(h, size) < 0)
6928 case SEI_TYPE_RECOVERY_POINT:
6929 if(decode_recovery_point(h) < 0)
6932 case SEI_BUFFERING_PERIOD:
6933 if(decode_buffering_period(h) < 0)
6937 skip_bits(&s->gb, 8*size);
6940 //FIXME check bits here
6941 align_get_bits(&s->gb);
6947 static inline int decode_hrd_parameters(H264Context *h, SPS *sps){
6948 MpegEncContext * const s = &h->s;
6950 cpb_count = get_ue_golomb_31(&s->gb) + 1;
6952 if(cpb_count > 32U){
6953 av_log(h->s.avctx, AV_LOG_ERROR, "cpb_count %d invalid\n", cpb_count);
6957 get_bits(&s->gb, 4); /* bit_rate_scale */
6958 get_bits(&s->gb, 4); /* cpb_size_scale */
6959 for(i=0; i<cpb_count; i++){
6960 get_ue_golomb(&s->gb); /* bit_rate_value_minus1 */
6961 get_ue_golomb(&s->gb); /* cpb_size_value_minus1 */
6962 get_bits1(&s->gb); /* cbr_flag */
6964 sps->initial_cpb_removal_delay_length = get_bits(&s->gb, 5) + 1;
6965 sps->cpb_removal_delay_length = get_bits(&s->gb, 5) + 1;
6966 sps->dpb_output_delay_length = get_bits(&s->gb, 5) + 1;
6967 sps->time_offset_length = get_bits(&s->gb, 5);
6968 sps->cpb_cnt = cpb_count;
6972 static inline int decode_vui_parameters(H264Context *h, SPS *sps){
6973 MpegEncContext * const s = &h->s;
6974 int aspect_ratio_info_present_flag;
6975 unsigned int aspect_ratio_idc;
6977 aspect_ratio_info_present_flag= get_bits1(&s->gb);
6979 if( aspect_ratio_info_present_flag ) {
6980 aspect_ratio_idc= get_bits(&s->gb, 8);
6981 if( aspect_ratio_idc == EXTENDED_SAR ) {
6982 sps->sar.num= get_bits(&s->gb, 16);
6983 sps->sar.den= get_bits(&s->gb, 16);
6984 }else if(aspect_ratio_idc < FF_ARRAY_ELEMS(pixel_aspect)){
6985 sps->sar= pixel_aspect[aspect_ratio_idc];
6987 av_log(h->s.avctx, AV_LOG_ERROR, "illegal aspect ratio\n");
6994 // s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height);
6996 if(get_bits1(&s->gb)){ /* overscan_info_present_flag */
6997 get_bits1(&s->gb); /* overscan_appropriate_flag */
7000 if(get_bits1(&s->gb)){ /* video_signal_type_present_flag */
7001 get_bits(&s->gb, 3); /* video_format */
7002 get_bits1(&s->gb); /* video_full_range_flag */
7003 if(get_bits1(&s->gb)){ /* colour_description_present_flag */
7004 get_bits(&s->gb, 8); /* colour_primaries */
7005 get_bits(&s->gb, 8); /* transfer_characteristics */
7006 get_bits(&s->gb, 8); /* matrix_coefficients */
7010 if(get_bits1(&s->gb)){ /* chroma_location_info_present_flag */
7011 get_ue_golomb(&s->gb); /* chroma_sample_location_type_top_field */
7012 get_ue_golomb(&s->gb); /* chroma_sample_location_type_bottom_field */
7015 sps->timing_info_present_flag = get_bits1(&s->gb);
7016 if(sps->timing_info_present_flag){
7017 sps->num_units_in_tick = get_bits_long(&s->gb, 32);
7018 sps->time_scale = get_bits_long(&s->gb, 32);
7019 sps->fixed_frame_rate_flag = get_bits1(&s->gb);
7022 sps->nal_hrd_parameters_present_flag = get_bits1(&s->gb);
7023 if(sps->nal_hrd_parameters_present_flag)
7024 if(decode_hrd_parameters(h, sps) < 0)
7026 sps->vcl_hrd_parameters_present_flag = get_bits1(&s->gb);
7027 if(sps->vcl_hrd_parameters_present_flag)
7028 if(decode_hrd_parameters(h, sps) < 0)
7030 if(sps->nal_hrd_parameters_present_flag || sps->vcl_hrd_parameters_present_flag)
7031 get_bits1(&s->gb); /* low_delay_hrd_flag */
7032 sps->pic_struct_present_flag = get_bits1(&s->gb);
7034 sps->bitstream_restriction_flag = get_bits1(&s->gb);
7035 if(sps->bitstream_restriction_flag){
7036 get_bits1(&s->gb); /* motion_vectors_over_pic_boundaries_flag */
7037 get_ue_golomb(&s->gb); /* max_bytes_per_pic_denom */
7038 get_ue_golomb(&s->gb); /* max_bits_per_mb_denom */
7039 get_ue_golomb(&s->gb); /* log2_max_mv_length_horizontal */
7040 get_ue_golomb(&s->gb); /* log2_max_mv_length_vertical */
7041 sps->num_reorder_frames= get_ue_golomb(&s->gb);
7042 get_ue_golomb(&s->gb); /*max_dec_frame_buffering*/
7044 if(sps->num_reorder_frames > 16U /*max_dec_frame_buffering || max_dec_frame_buffering > 16*/){
7045 av_log(h->s.avctx, AV_LOG_ERROR, "illegal num_reorder_frames %d\n", sps->num_reorder_frames);
7053 static void decode_scaling_list(H264Context *h, uint8_t *factors, int size,
7054 const uint8_t *jvt_list, const uint8_t *fallback_list){
7055 MpegEncContext * const s = &h->s;
7056 int i, last = 8, next = 8;
7057 const uint8_t *scan = size == 16 ? zigzag_scan : ff_zigzag_direct;
7058 if(!get_bits1(&s->gb)) /* matrix not written, we use the predicted one */
7059 memcpy(factors, fallback_list, size*sizeof(uint8_t));
7061 for(i=0;i<size;i++){
7063 next = (last + get_se_golomb(&s->gb)) & 0xff;
7064 if(!i && !next){ /* matrix not written, we use the preset one */
7065 memcpy(factors, jvt_list, size*sizeof(uint8_t));
7068 last = factors[scan[i]] = next ? next : last;
7072 static void decode_scaling_matrices(H264Context *h, SPS *sps, PPS *pps, int is_sps,
7073 uint8_t (*scaling_matrix4)[16], uint8_t (*scaling_matrix8)[64]){
7074 MpegEncContext * const s = &h->s;
7075 int fallback_sps = !is_sps && sps->scaling_matrix_present;
7076 const uint8_t *fallback[4] = {
7077 fallback_sps ? sps->scaling_matrix4[0] : default_scaling4[0],
7078 fallback_sps ? sps->scaling_matrix4[3] : default_scaling4[1],
7079 fallback_sps ? sps->scaling_matrix8[0] : default_scaling8[0],
7080 fallback_sps ? sps->scaling_matrix8[1] : default_scaling8[1]
7082 if(get_bits1(&s->gb)){
7083 sps->scaling_matrix_present |= is_sps;
7084 decode_scaling_list(h,scaling_matrix4[0],16,default_scaling4[0],fallback[0]); // Intra, Y
7085 decode_scaling_list(h,scaling_matrix4[1],16,default_scaling4[0],scaling_matrix4[0]); // Intra, Cr
7086 decode_scaling_list(h,scaling_matrix4[2],16,default_scaling4[0],scaling_matrix4[1]); // Intra, Cb
7087 decode_scaling_list(h,scaling_matrix4[3],16,default_scaling4[1],fallback[1]); // Inter, Y
7088 decode_scaling_list(h,scaling_matrix4[4],16,default_scaling4[1],scaling_matrix4[3]); // Inter, Cr
7089 decode_scaling_list(h,scaling_matrix4[5],16,default_scaling4[1],scaling_matrix4[4]); // Inter, Cb
7090 if(is_sps || pps->transform_8x8_mode){
7091 decode_scaling_list(h,scaling_matrix8[0],64,default_scaling8[0],fallback[2]); // Intra, Y
7092 decode_scaling_list(h,scaling_matrix8[1],64,default_scaling8[1],fallback[3]); // Inter, Y
7097 int ff_h264_decode_seq_parameter_set(H264Context *h){
7098 MpegEncContext * const s = &h->s;
7099 int profile_idc, level_idc;
7100 unsigned int sps_id;
7104 profile_idc= get_bits(&s->gb, 8);
7105 get_bits1(&s->gb); //constraint_set0_flag
7106 get_bits1(&s->gb); //constraint_set1_flag
7107 get_bits1(&s->gb); //constraint_set2_flag
7108 get_bits1(&s->gb); //constraint_set3_flag
7109 get_bits(&s->gb, 4); // reserved
7110 level_idc= get_bits(&s->gb, 8);
7111 sps_id= get_ue_golomb_31(&s->gb);
7113 if(sps_id >= MAX_SPS_COUNT) {
7114 av_log(h->s.avctx, AV_LOG_ERROR, "sps_id (%d) out of range\n", sps_id);
7117 sps= av_mallocz(sizeof(SPS));
7121 sps->profile_idc= profile_idc;
7122 sps->level_idc= level_idc;
7124 memset(sps->scaling_matrix4, 16, sizeof(sps->scaling_matrix4));
7125 memset(sps->scaling_matrix8, 16, sizeof(sps->scaling_matrix8));
7126 sps->scaling_matrix_present = 0;
7128 if(sps->profile_idc >= 100){ //high profile
7129 sps->chroma_format_idc= get_ue_golomb_31(&s->gb);
7130 if(sps->chroma_format_idc == 3)
7131 sps->residual_color_transform_flag = get_bits1(&s->gb);
7132 sps->bit_depth_luma = get_ue_golomb(&s->gb) + 8;
7133 sps->bit_depth_chroma = get_ue_golomb(&s->gb) + 8;
7134 sps->transform_bypass = get_bits1(&s->gb);
7135 decode_scaling_matrices(h, sps, NULL, 1, sps->scaling_matrix4, sps->scaling_matrix8);
7137 sps->chroma_format_idc= 1;
7140 sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4;
7141 sps->poc_type= get_ue_golomb_31(&s->gb);
7143 if(sps->poc_type == 0){ //FIXME #define
7144 sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4;
7145 } else if(sps->poc_type == 1){//FIXME #define
7146 sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb);
7147 sps->offset_for_non_ref_pic= get_se_golomb(&s->gb);
7148 sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb);
7149 sps->poc_cycle_length = get_ue_golomb(&s->gb);
7151 if((unsigned)sps->poc_cycle_length >= FF_ARRAY_ELEMS(sps->offset_for_ref_frame)){
7152 av_log(h->s.avctx, AV_LOG_ERROR, "poc_cycle_length overflow %u\n", sps->poc_cycle_length);
7156 for(i=0; i<sps->poc_cycle_length; i++)
7157 sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb);
7158 }else if(sps->poc_type != 2){
7159 av_log(h->s.avctx, AV_LOG_ERROR, "illegal POC type %d\n", sps->poc_type);
7163 sps->ref_frame_count= get_ue_golomb_31(&s->gb);
7164 if(sps->ref_frame_count > MAX_PICTURE_COUNT-2 || sps->ref_frame_count >= 32U){
7165 av_log(h->s.avctx, AV_LOG_ERROR, "too many reference frames\n");
7168 sps->gaps_in_frame_num_allowed_flag= get_bits1(&s->gb);
7169 sps->mb_width = get_ue_golomb(&s->gb) + 1;
7170 sps->mb_height= get_ue_golomb(&s->gb) + 1;
7171 if((unsigned)sps->mb_width >= INT_MAX/16 || (unsigned)sps->mb_height >= INT_MAX/16 ||
7172 avcodec_check_dimensions(NULL, 16*sps->mb_width, 16*sps->mb_height)){
7173 av_log(h->s.avctx, AV_LOG_ERROR, "mb_width/height overflow\n");
7177 sps->frame_mbs_only_flag= get_bits1(&s->gb);
7178 if(!sps->frame_mbs_only_flag)
7179 sps->mb_aff= get_bits1(&s->gb);
7183 sps->direct_8x8_inference_flag= get_bits1(&s->gb);
7185 #ifndef ALLOW_INTERLACE
7187 av_log(h->s.avctx, AV_LOG_ERROR, "MBAFF support not included; enable it at compile-time.\n");
7189 sps->crop= get_bits1(&s->gb);
7191 sps->crop_left = get_ue_golomb(&s->gb);
7192 sps->crop_right = get_ue_golomb(&s->gb);
7193 sps->crop_top = get_ue_golomb(&s->gb);
7194 sps->crop_bottom= get_ue_golomb(&s->gb);
7195 if(sps->crop_left || sps->crop_top){
7196 av_log(h->s.avctx, AV_LOG_ERROR, "insane cropping not completely supported, this could look slightly wrong ...\n");
7198 if(sps->crop_right >= 8 || sps->crop_bottom >= (8>> !sps->frame_mbs_only_flag)){
7199 av_log(h->s.avctx, AV_LOG_ERROR, "brainfart cropping not supported, this could look slightly wrong ...\n");
7205 sps->crop_bottom= 0;
7208 sps->vui_parameters_present_flag= get_bits1(&s->gb);
7209 if( sps->vui_parameters_present_flag )
7210 decode_vui_parameters(h, sps);
7212 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
7213 av_log(h->s.avctx, AV_LOG_DEBUG, "sps:%u profile:%d/%d poc:%d ref:%d %dx%d %s %s crop:%d/%d/%d/%d %s %s\n",
7214 sps_id, sps->profile_idc, sps->level_idc,
7216 sps->ref_frame_count,
7217 sps->mb_width, sps->mb_height,
7218 sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"),
7219 sps->direct_8x8_inference_flag ? "8B8" : "",
7220 sps->crop_left, sps->crop_right,
7221 sps->crop_top, sps->crop_bottom,
7222 sps->vui_parameters_present_flag ? "VUI" : "",
7223 ((const char*[]){"Gray","420","422","444"})[sps->chroma_format_idc]
7227 av_free(h->sps_buffers[sps_id]);
7228 h->sps_buffers[sps_id]= sps;
7237 build_qp_table(PPS *pps, int t, int index)
7240 for(i = 0; i < 52; i++)
7241 pps->chroma_qp_table[t][i] = chroma_qp[av_clip(i + index, 0, 51)];
7244 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length){
7245 MpegEncContext * const s = &h->s;
7246 unsigned int pps_id= get_ue_golomb(&s->gb);
7249 if(pps_id >= MAX_PPS_COUNT) {
7250 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id (%d) out of range\n", pps_id);
7254 pps= av_mallocz(sizeof(PPS));
7257 pps->sps_id= get_ue_golomb_31(&s->gb);
7258 if((unsigned)pps->sps_id>=MAX_SPS_COUNT || h->sps_buffers[pps->sps_id] == NULL){
7259 av_log(h->s.avctx, AV_LOG_ERROR, "sps_id out of range\n");
7263 pps->cabac= get_bits1(&s->gb);
7264 pps->pic_order_present= get_bits1(&s->gb);
7265 pps->slice_group_count= get_ue_golomb(&s->gb) + 1;
7266 if(pps->slice_group_count > 1 ){
7267 pps->mb_slice_group_map_type= get_ue_golomb(&s->gb);
7268 av_log(h->s.avctx, AV_LOG_ERROR, "FMO not supported\n");
7269 switch(pps->mb_slice_group_map_type){
7272 | for( i = 0; i <= num_slice_groups_minus1; i++ ) | | |
7273 | run_length[ i ] |1 |ue(v) |
7278 | for( i = 0; i < num_slice_groups_minus1; i++ ) | | |
7280 | top_left_mb[ i ] |1 |ue(v) |
7281 | bottom_right_mb[ i ] |1 |ue(v) |
7289 | slice_group_change_direction_flag |1 |u(1) |
7290 | slice_group_change_rate_minus1 |1 |ue(v) |
7295 | slice_group_id_cnt_minus1 |1 |ue(v) |
7296 | for( i = 0; i <= slice_group_id_cnt_minus1; i++ | | |
7298 | slice_group_id[ i ] |1 |u(v) |
7303 pps->ref_count[0]= get_ue_golomb(&s->gb) + 1;
7304 pps->ref_count[1]= get_ue_golomb(&s->gb) + 1;
7305 if(pps->ref_count[0]-1 > 32-1 || pps->ref_count[1]-1 > 32-1){
7306 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow (pps)\n");
7310 pps->weighted_pred= get_bits1(&s->gb);
7311 pps->weighted_bipred_idc= get_bits(&s->gb, 2);
7312 pps->init_qp= get_se_golomb(&s->gb) + 26;
7313 pps->init_qs= get_se_golomb(&s->gb) + 26;
7314 pps->chroma_qp_index_offset[0]= get_se_golomb(&s->gb);
7315 pps->deblocking_filter_parameters_present= get_bits1(&s->gb);
7316 pps->constrained_intra_pred= get_bits1(&s->gb);
7317 pps->redundant_pic_cnt_present = get_bits1(&s->gb);
7319 pps->transform_8x8_mode= 0;
7320 h->dequant_coeff_pps= -1; //contents of sps/pps can change even if id doesn't, so reinit
7321 memcpy(pps->scaling_matrix4, h->sps_buffers[pps->sps_id]->scaling_matrix4, sizeof(pps->scaling_matrix4));
7322 memcpy(pps->scaling_matrix8, h->sps_buffers[pps->sps_id]->scaling_matrix8, sizeof(pps->scaling_matrix8));
7324 if(get_bits_count(&s->gb) < bit_length){
7325 pps->transform_8x8_mode= get_bits1(&s->gb);
7326 decode_scaling_matrices(h, h->sps_buffers[pps->sps_id], pps, 0, pps->scaling_matrix4, pps->scaling_matrix8);
7327 pps->chroma_qp_index_offset[1]= get_se_golomb(&s->gb); //second_chroma_qp_index_offset
7329 pps->chroma_qp_index_offset[1]= pps->chroma_qp_index_offset[0];
7332 build_qp_table(pps, 0, pps->chroma_qp_index_offset[0]);
7333 build_qp_table(pps, 1, pps->chroma_qp_index_offset[1]);
7334 if(pps->chroma_qp_index_offset[0] != pps->chroma_qp_index_offset[1])
7335 h->pps.chroma_qp_diff= 1;
7337 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
7338 av_log(h->s.avctx, AV_LOG_DEBUG, "pps:%u sps:%u %s slice_groups:%d ref:%d/%d %s qp:%d/%d/%d/%d %s %s %s %s\n",
7339 pps_id, pps->sps_id,
7340 pps->cabac ? "CABAC" : "CAVLC",
7341 pps->slice_group_count,
7342 pps->ref_count[0], pps->ref_count[1],
7343 pps->weighted_pred ? "weighted" : "",
7344 pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset[0], pps->chroma_qp_index_offset[1],
7345 pps->deblocking_filter_parameters_present ? "LPAR" : "",
7346 pps->constrained_intra_pred ? "CONSTR" : "",
7347 pps->redundant_pic_cnt_present ? "REDU" : "",
7348 pps->transform_8x8_mode ? "8x8DCT" : ""
7352 av_free(h->pps_buffers[pps_id]);
7353 h->pps_buffers[pps_id]= pps;
7361 * Call decode_slice() for each context.
7363 * @param h h264 master context
7364 * @param context_count number of contexts to execute
7366 static void execute_decode_slices(H264Context *h, int context_count){
7367 MpegEncContext * const s = &h->s;
7368 AVCodecContext * const avctx= s->avctx;
7372 if (s->avctx->hwaccel)
7374 if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
7376 if(context_count == 1) {
7377 decode_slice(avctx, &h);
7379 for(i = 1; i < context_count; i++) {
7380 hx = h->thread_context[i];
7381 hx->s.error_recognition = avctx->error_recognition;
7382 hx->s.error_count = 0;
7385 avctx->execute(avctx, (void *)decode_slice,
7386 (void **)h->thread_context, NULL, context_count, sizeof(void*));
7388 /* pull back stuff from slices to master context */
7389 hx = h->thread_context[context_count - 1];
7390 s->mb_x = hx->s.mb_x;
7391 s->mb_y = hx->s.mb_y;
7392 s->dropable = hx->s.dropable;
7393 s->picture_structure = hx->s.picture_structure;
7394 for(i = 1; i < context_count; i++)
7395 h->s.error_count += h->thread_context[i]->s.error_count;
7400 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
7401 MpegEncContext * const s = &h->s;
7402 AVCodecContext * const avctx= s->avctx;
7404 H264Context *hx; ///< thread context
7405 int context_count = 0;
7407 h->max_contexts = avctx->thread_count;
7410 for(i=0; i<50; i++){
7411 av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
7414 if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
7415 h->current_slice = 0;
7416 if (!s->first_field)
7417 s->current_picture_ptr= NULL;
7430 if(buf_index >= buf_size) break;
7432 for(i = 0; i < h->nal_length_size; i++)
7433 nalsize = (nalsize << 8) | buf[buf_index++];
7434 if(nalsize <= 1 || (nalsize+buf_index > buf_size)){
7439 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
7444 // start code prefix search
7445 for(; buf_index + 3 < buf_size; buf_index++){
7446 // This should always succeed in the first iteration.
7447 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
7451 if(buf_index+3 >= buf_size) break;
7456 hx = h->thread_context[context_count];
7458 ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, h->is_avc ? nalsize : buf_size - buf_index);
7459 if (ptr==NULL || dst_length < 0){
7462 while(ptr[dst_length - 1] == 0 && dst_length > 0)
7464 bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
7466 if(s->avctx->debug&FF_DEBUG_STARTCODE){
7467 av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d/%d length %d\n", hx->nal_unit_type, buf_index, buf_size, dst_length);
7470 if (h->is_avc && (nalsize != consumed)){
7471 int i, debug_level = AV_LOG_DEBUG;
7472 for (i = consumed; i < nalsize; i++)
7473 if (buf[buf_index+i])
7474 debug_level = AV_LOG_ERROR;
7475 av_log(h->s.avctx, debug_level, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
7479 buf_index += consumed;
7481 if( (s->hurry_up == 1 && h->nal_ref_idc == 0) //FIXME do not discard SEI id
7482 ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
7487 switch(hx->nal_unit_type){
7489 if (h->nal_unit_type != NAL_IDR_SLICE) {
7490 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
7493 idr(h); //FIXME ensure we don't loose some frames if there is reordering
7495 init_get_bits(&hx->s.gb, ptr, bit_length);
7497 hx->inter_gb_ptr= &hx->s.gb;
7498 hx->s.data_partitioning = 0;
7500 if((err = decode_slice_header(hx, h)))
7503 if (s->avctx->hwaccel && h->current_slice == 1) {
7504 if (s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
7508 s->current_picture_ptr->key_frame |=
7509 (hx->nal_unit_type == NAL_IDR_SLICE) ||
7510 (h->sei_recovery_frame_cnt >= 0);
7511 if(hx->redundant_pic_count==0 && hx->s.hurry_up < 5
7512 && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
7513 && (avctx->skip_frame < AVDISCARD_BIDIR || hx->slice_type_nos!=FF_B_TYPE)
7514 && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
7515 && avctx->skip_frame < AVDISCARD_ALL){
7516 if(avctx->hwaccel) {
7517 if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
7520 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
7521 static const uint8_t start_code[] = {0x00, 0x00, 0x01};
7522 ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
7523 ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
7529 init_get_bits(&hx->s.gb, ptr, bit_length);
7531 hx->inter_gb_ptr= NULL;
7532 hx->s.data_partitioning = 1;
7534 err = decode_slice_header(hx, h);
7537 init_get_bits(&hx->intra_gb, ptr, bit_length);
7538 hx->intra_gb_ptr= &hx->intra_gb;
7541 init_get_bits(&hx->inter_gb, ptr, bit_length);
7542 hx->inter_gb_ptr= &hx->inter_gb;
7544 if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
7545 && s->context_initialized
7547 && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
7548 && (avctx->skip_frame < AVDISCARD_BIDIR || hx->slice_type_nos!=FF_B_TYPE)
7549 && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
7550 && avctx->skip_frame < AVDISCARD_ALL)
7554 init_get_bits(&s->gb, ptr, bit_length);
7555 ff_h264_decode_sei(h);
7558 init_get_bits(&s->gb, ptr, bit_length);
7559 ff_h264_decode_seq_parameter_set(h);
7561 if(s->flags& CODEC_FLAG_LOW_DELAY)
7564 if(avctx->has_b_frames < 2)
7565 avctx->has_b_frames= !s->low_delay;
7568 init_get_bits(&s->gb, ptr, bit_length);
7570 ff_h264_decode_picture_parameter_set(h, bit_length);
7574 case NAL_END_SEQUENCE:
7575 case NAL_END_STREAM:
7576 case NAL_FILLER_DATA:
7578 case NAL_AUXILIARY_SLICE:
7581 av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", h->nal_unit_type, bit_length);
7584 if(context_count == h->max_contexts) {
7585 execute_decode_slices(h, context_count);
7590 av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
7592 /* Slice could not be decoded in parallel mode, copy down
7593 * NAL unit stuff to context 0 and restart. Note that
7594 * rbsp_buffer is not transferred, but since we no longer
7595 * run in parallel mode this should not be an issue. */
7596 h->nal_unit_type = hx->nal_unit_type;
7597 h->nal_ref_idc = hx->nal_ref_idc;
7603 execute_decode_slices(h, context_count);
7608 * returns the number of bytes consumed for building the current frame
7610 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
7611 if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
7612 if(pos+10>buf_size) pos=buf_size; // oops ;)
7617 static int decode_frame(AVCodecContext *avctx,
7618 void *data, int *data_size,
7619 const uint8_t *buf, int buf_size)
7621 H264Context *h = avctx->priv_data;
7622 MpegEncContext *s = &h->s;
7623 AVFrame *pict = data;
7626 s->flags= avctx->flags;
7627 s->flags2= avctx->flags2;
7629 /* end of stream, output what is still in the buffers */
7630 if (buf_size == 0) {
7634 //FIXME factorize this with the output code below
7635 out = h->delayed_pic[0];
7637 for(i=1; h->delayed_pic[i] && (h->delayed_pic[i]->poc && !h->delayed_pic[i]->key_frame); i++)
7638 if(h->delayed_pic[i]->poc < out->poc){
7639 out = h->delayed_pic[i];
7643 for(i=out_idx; h->delayed_pic[i]; i++)
7644 h->delayed_pic[i] = h->delayed_pic[i+1];
7647 *data_size = sizeof(AVFrame);
7648 *pict= *(AVFrame*)out;
7654 if(h->is_avc && !h->got_avcC) {
7655 int i, cnt, nalsize;
7656 unsigned char *p = avctx->extradata;
7657 if(avctx->extradata_size < 7) {
7658 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
7662 av_log(avctx, AV_LOG_ERROR, "Unknown avcC version %d\n", *p);
7665 /* sps and pps in the avcC always have length coded with 2 bytes,
7666 so put a fake nal_length_size = 2 while parsing them */
7667 h->nal_length_size = 2;
7668 // Decode sps from avcC
7669 cnt = *(p+5) & 0x1f; // Number of sps
7671 for (i = 0; i < cnt; i++) {
7672 nalsize = AV_RB16(p) + 2;
7673 if(decode_nal_units(h, p, nalsize) < 0) {
7674 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
7679 // Decode pps from avcC
7680 cnt = *(p++); // Number of pps
7681 for (i = 0; i < cnt; i++) {
7682 nalsize = AV_RB16(p) + 2;
7683 if(decode_nal_units(h, p, nalsize) != nalsize) {
7684 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
7689 // Now store right nal length size, that will be use to parse all other nals
7690 h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
7691 // Do not reparse avcC
7695 if(!h->got_avcC && !h->is_avc && s->avctx->extradata_size){
7696 if(decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) < 0)
7701 buf_index=decode_nal_units(h, buf, buf_size);
7705 if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
7706 if (avctx->skip_frame >= AVDISCARD_NONREF || s->hurry_up) return 0;
7707 av_log(avctx, AV_LOG_ERROR, "no frame!\n");
7711 if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
7712 Picture *out = s->current_picture_ptr;
7713 Picture *cur = s->current_picture_ptr;
7714 int i, pics, cross_idr, out_of_order, out_idx;
7718 s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
7719 s->current_picture_ptr->pict_type= s->pict_type;
7721 if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
7722 ff_vdpau_h264_set_reference_frames(s);
7725 execute_ref_pic_marking(h, h->mmco, h->mmco_index);
7726 h->prev_poc_msb= h->poc_msb;
7727 h->prev_poc_lsb= h->poc_lsb;
7729 h->prev_frame_num_offset= h->frame_num_offset;
7730 h->prev_frame_num= h->frame_num;
7732 if (avctx->hwaccel) {
7733 if (avctx->hwaccel->end_frame(avctx) < 0)
7734 av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
7737 if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
7738 ff_vdpau_h264_picture_complete(s);
7741 * FIXME: Error handling code does not seem to support interlaced
7742 * when slices span multiple rows
7743 * The ff_er_add_slice calls don't work right for bottom
7744 * fields; they cause massive erroneous error concealing
7745 * Error marking covers both fields (top and bottom).
7746 * This causes a mismatched s->error_count
7747 * and a bad error table. Further, the error count goes to
7748 * INT_MAX when called for bottom field, because mb_y is
7749 * past end by one (callers fault) and resync_mb_y != 0
7750 * causes problems for the first MB line, too.
7757 if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
7758 /* Wait for second field. */
7762 cur->repeat_pict = 0;
7764 /* Signal interlacing information externally. */
7765 /* Prioritize picture timing SEI information over used decoding process if it exists. */
7766 if(h->sps.pic_struct_present_flag){
7767 switch (h->sei_pic_struct)
7769 case SEI_PIC_STRUCT_FRAME:
7770 cur->interlaced_frame = 0;
7772 case SEI_PIC_STRUCT_TOP_FIELD:
7773 case SEI_PIC_STRUCT_BOTTOM_FIELD:
7774 case SEI_PIC_STRUCT_TOP_BOTTOM:
7775 case SEI_PIC_STRUCT_BOTTOM_TOP:
7776 cur->interlaced_frame = 1;
7778 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
7779 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
7780 // Signal the possibility of telecined film externally (pic_struct 5,6)
7781 // From these hints, let the applications decide if they apply deinterlacing.
7782 cur->repeat_pict = 1;
7783 cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
7785 case SEI_PIC_STRUCT_FRAME_DOUBLING:
7786 // Force progressive here, as doubling interlaced frame is a bad idea.
7787 cur->interlaced_frame = 0;
7788 cur->repeat_pict = 2;
7790 case SEI_PIC_STRUCT_FRAME_TRIPLING:
7791 cur->interlaced_frame = 0;
7792 cur->repeat_pict = 4;
7796 /* Derive interlacing flag from used decoding process. */
7797 cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
7800 if (cur->field_poc[0] != cur->field_poc[1]){
7801 /* Derive top_field_first from field pocs. */
7802 cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
7804 if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
7805 /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
7806 if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
7807 || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
7808 cur->top_field_first = 1;
7810 cur->top_field_first = 0;
7812 /* Most likely progressive */
7813 cur->top_field_first = 0;
7817 //FIXME do something with unavailable reference frames
7819 /* Sort B-frames into display order */
7821 if(h->sps.bitstream_restriction_flag
7822 && s->avctx->has_b_frames < h->sps.num_reorder_frames){
7823 s->avctx->has_b_frames = h->sps.num_reorder_frames;
7827 if( s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
7828 && !h->sps.bitstream_restriction_flag){
7829 s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
7834 while(h->delayed_pic[pics]) pics++;
7836 assert(pics <= MAX_DELAYED_PIC_COUNT);
7838 h->delayed_pic[pics++] = cur;
7839 if(cur->reference == 0)
7840 cur->reference = DELAYED_PIC_REF;
7842 out = h->delayed_pic[0];
7844 for(i=1; h->delayed_pic[i] && (h->delayed_pic[i]->poc && !h->delayed_pic[i]->key_frame); i++)
7845 if(h->delayed_pic[i]->poc < out->poc){
7846 out = h->delayed_pic[i];
7849 cross_idr = !h->delayed_pic[0]->poc || !!h->delayed_pic[i] || h->delayed_pic[0]->key_frame;
7851 out_of_order = !cross_idr && out->poc < h->outputed_poc;
7853 if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
7855 else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
7857 ((!cross_idr && out->poc > h->outputed_poc + 2)
7858 || cur->pict_type == FF_B_TYPE)))
7861 s->avctx->has_b_frames++;
7864 if(out_of_order || pics > s->avctx->has_b_frames){
7865 out->reference &= ~DELAYED_PIC_REF;
7866 for(i=out_idx; h->delayed_pic[i]; i++)
7867 h->delayed_pic[i] = h->delayed_pic[i+1];
7869 if(!out_of_order && pics > s->avctx->has_b_frames){
7870 *data_size = sizeof(AVFrame);
7872 h->outputed_poc = out->poc;
7873 *pict= *(AVFrame*)out;
7875 av_log(avctx, AV_LOG_DEBUG, "no picture\n");
7880 assert(pict->data[0] || !*data_size);
7881 ff_print_debug_info(s, pict);
7882 //printf("out %d\n", (int)pict->data[0]);
7885 /* Return the Picture timestamp as the frame number */
7886 /* we subtract 1 because it is added on utils.c */
7887 avctx->frame_number = s->picture_number - 1;
7889 return get_consumed_bytes(s, buf_index, buf_size);
7892 static inline void fill_mb_avail(H264Context *h){
7893 MpegEncContext * const s = &h->s;
7894 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
7897 h->mb_avail[0]= s->mb_x && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
7898 h->mb_avail[1]= h->slice_table[mb_xy - s->mb_stride ] == h->slice_num;
7899 h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
7905 h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
7906 h->mb_avail[4]= 1; //FIXME move out
7907 h->mb_avail[5]= 0; //FIXME move out
7915 #define SIZE (COUNT*40)
7921 // int int_temp[10000];
7923 AVCodecContext avctx;
7925 dsputil_init(&dsp, &avctx);
7927 init_put_bits(&pb, temp, SIZE);
7928 printf("testing unsigned exp golomb\n");
7929 for(i=0; i<COUNT; i++){
7931 set_ue_golomb(&pb, i);
7932 STOP_TIMER("set_ue_golomb");
7934 flush_put_bits(&pb);
7936 init_get_bits(&gb, temp, 8*SIZE);
7937 for(i=0; i<COUNT; i++){
7940 s= show_bits(&gb, 24);
7943 j= get_ue_golomb(&gb);
7945 printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
7948 STOP_TIMER("get_ue_golomb");
7952 init_put_bits(&pb, temp, SIZE);
7953 printf("testing signed exp golomb\n");
7954 for(i=0; i<COUNT; i++){
7956 set_se_golomb(&pb, i - COUNT/2);
7957 STOP_TIMER("set_se_golomb");
7959 flush_put_bits(&pb);
7961 init_get_bits(&gb, temp, 8*SIZE);
7962 for(i=0; i<COUNT; i++){
7965 s= show_bits(&gb, 24);
7968 j= get_se_golomb(&gb);
7969 if(j != i - COUNT/2){
7970 printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
7973 STOP_TIMER("get_se_golomb");
7977 printf("testing 4x4 (I)DCT\n");
7980 uint8_t src[16], ref[16];
7981 uint64_t error= 0, max_error=0;
7983 for(i=0; i<COUNT; i++){
7985 // printf("%d %d %d\n", r1, r2, (r2-r1)*16);
7986 for(j=0; j<16; j++){
7987 ref[j]= random()%255;
7988 src[j]= random()%255;
7991 h264_diff_dct_c(block, src, ref, 4);
7994 for(j=0; j<16; j++){
7995 // printf("%d ", block[j]);
7996 block[j]= block[j]*4;
7997 if(j&1) block[j]= (block[j]*4 + 2)/5;
7998 if(j&4) block[j]= (block[j]*4 + 2)/5;
8002 s->dsp.h264_idct_add(ref, block, 4);
8003 /* for(j=0; j<16; j++){
8004 printf("%d ", ref[j]);
8008 for(j=0; j<16; j++){
8009 int diff= FFABS(src[j] - ref[j]);
8012 max_error= FFMAX(max_error, diff);
8015 printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
8016 printf("testing quantizer\n");
8017 for(qp=0; qp<52; qp++){
8019 src1_block[i]= src2_block[i]= random()%255;
8022 printf("Testing NAL layer\n");
8024 uint8_t bitstream[COUNT];
8025 uint8_t nal[COUNT*2];
8027 memset(&h, 0, sizeof(H264Context));
8029 for(i=0; i<COUNT; i++){
8037 for(j=0; j<COUNT; j++){
8038 bitstream[j]= (random() % 255) + 1;
8041 for(j=0; j<zeros; j++){
8042 int pos= random() % COUNT;
8043 while(bitstream[pos] == 0){
8052 nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
8054 printf("encoding failed\n");
8058 out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
8062 if(out_length != COUNT){
8063 printf("incorrect length %d %d\n", out_length, COUNT);
8067 if(consumed != nal_length){
8068 printf("incorrect consumed length %d %d\n", nal_length, consumed);
8072 if(memcmp(bitstream, out, COUNT)){
8073 printf("mismatch\n");
8079 printf("Testing RBSP\n");
8087 static av_cold int decode_end(AVCodecContext *avctx)
8089 H264Context *h = avctx->priv_data;
8090 MpegEncContext *s = &h->s;
8093 av_freep(&h->rbsp_buffer[0]);
8094 av_freep(&h->rbsp_buffer[1]);
8095 free_tables(h); //FIXME cleanup init stuff perhaps
8097 for(i = 0; i < MAX_SPS_COUNT; i++)
8098 av_freep(h->sps_buffers + i);
8100 for(i = 0; i < MAX_PPS_COUNT; i++)
8101 av_freep(h->pps_buffers + i);
8105 // memset(h, 0, sizeof(H264Context));
8111 AVCodec h264_decoder = {
8115 sizeof(H264Context),
8120 /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY,
8122 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
8123 .pix_fmts= ff_hwaccel_pixfmt_list_420,
8126 #if CONFIG_H264_VDPAU_DECODER
8127 AVCodec h264_vdpau_decoder = {
8131 sizeof(H264Context),
8136 CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
8138 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
8142 #if CONFIG_SVQ3_DECODER