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 < MAX_THREADS; 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;
3272 int i, av_uninit(j);
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 av_uninit(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 %u referenced\n", pps_id);
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 %u referenced\n", h->pps.sps_id);
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;