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>
30 #include "mpegvideo.h"
33 #include "h264_parser.h"
36 #include "rectangle.h"
37 #include "vdpau_internal.h"
41 #include "x86/h264_i386.h"
48 * Value of Picture.reference when Picture is not a reference picture, but
49 * is held for delayed output.
51 #define DELAYED_PIC_REF 4
53 static VLC coeff_token_vlc[4];
54 static VLC_TYPE coeff_token_vlc_tables[520+332+280+256][2];
55 static const int coeff_token_vlc_tables_size[4]={520,332,280,256};
57 static VLC chroma_dc_coeff_token_vlc;
58 static VLC_TYPE chroma_dc_coeff_token_vlc_table[256][2];
59 static const int chroma_dc_coeff_token_vlc_table_size = 256;
61 static VLC total_zeros_vlc[15];
62 static VLC_TYPE total_zeros_vlc_tables[15][512][2];
63 static const int total_zeros_vlc_tables_size = 512;
65 static VLC chroma_dc_total_zeros_vlc[3];
66 static VLC_TYPE chroma_dc_total_zeros_vlc_tables[3][8][2];
67 static const int chroma_dc_total_zeros_vlc_tables_size = 8;
69 static VLC run_vlc[6];
70 static VLC_TYPE run_vlc_tables[6][8][2];
71 static const int run_vlc_tables_size = 8;
74 static VLC_TYPE run7_vlc_table[96][2];
75 static const int run7_vlc_table_size = 96;
77 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
78 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
79 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);
80 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);
81 static Picture * remove_long(H264Context *h, int i, int ref_mask);
83 static av_always_inline uint32_t pack16to32(int a, int b){
84 #ifdef WORDS_BIGENDIAN
85 return (b&0xFFFF) + (a<<16);
87 return (a&0xFFFF) + (b<<16);
91 static const uint8_t rem6[52]={
92 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,
95 static const uint8_t div6[52]={
96 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,
99 static const uint8_t left_block_options[4][8]={
106 #define LEVEL_TAB_BITS 8
107 static int8_t cavlc_level_tab[7][1<<LEVEL_TAB_BITS][2];
109 static void fill_caches(H264Context *h, int mb_type, int for_deblock){
110 MpegEncContext * const s = &h->s;
111 const int mb_xy= h->mb_xy;
112 int topleft_xy, top_xy, topright_xy, left_xy[2];
113 int topleft_type, top_type, topright_type, left_type[2];
114 const uint8_t * left_block;
115 int topleft_partition= -1;
118 top_xy = mb_xy - (s->mb_stride << FIELD_PICTURE);
120 //FIXME deblocking could skip the intra and nnz parts.
121 if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[top_xy]) && !FRAME_MBAFF)
124 /* Wow, what a mess, why didn't they simplify the interlacing & intra
125 * stuff, I can't imagine that these complex rules are worth it. */
127 topleft_xy = top_xy - 1;
128 topright_xy= top_xy + 1;
129 left_xy[1] = left_xy[0] = mb_xy-1;
130 left_block = left_block_options[0];
132 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride;
133 const int top_pair_xy = pair_xy - s->mb_stride;
134 const int topleft_pair_xy = top_pair_xy - 1;
135 const int topright_pair_xy = top_pair_xy + 1;
136 const int topleft_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
137 const int top_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
138 const int topright_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
139 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
140 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
141 const int bottom = (s->mb_y & 1);
142 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);
144 if (curr_mb_field_flag && (bottom || top_mb_field_flag)){
145 top_xy -= s->mb_stride;
147 if (curr_mb_field_flag && (bottom || topleft_mb_field_flag)){
148 topleft_xy -= s->mb_stride;
149 } else if(bottom && !curr_mb_field_flag && left_mb_field_flag) {
150 topleft_xy += s->mb_stride;
151 // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
152 topleft_partition = 0;
154 if (curr_mb_field_flag && (bottom || topright_mb_field_flag)){
155 topright_xy -= s->mb_stride;
157 if (left_mb_field_flag != curr_mb_field_flag) {
158 left_xy[1] = left_xy[0] = pair_xy - 1;
159 if (curr_mb_field_flag) {
160 left_xy[1] += s->mb_stride;
161 left_block = left_block_options[3];
163 left_block= left_block_options[2 - bottom];
168 h->top_mb_xy = top_xy;
169 h->left_mb_xy[0] = left_xy[0];
170 h->left_mb_xy[1] = left_xy[1];
174 top_type = h->slice_table[top_xy ] < 0xFFFF ? s->current_picture.mb_type[top_xy] : 0;
175 left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0;
176 left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0;
178 if(MB_MBAFF && !IS_INTRA(mb_type)){
180 for(list=0; list<h->list_count; list++){
181 //These values where changed for ease of performing MC, we need to change them back
182 //FIXME maybe we can make MC and loop filter use the same values or prevent
183 //the MC code from changing ref_cache and rather use a temporary array.
184 if(USES_LIST(mb_type,list)){
185 int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
186 *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
187 *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
189 *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
190 *(uint32_t*)&h->ref_cache[list][scan8[10]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
195 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
196 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
197 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
198 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
199 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
201 if(IS_INTRA(mb_type)){
202 int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
203 h->topleft_samples_available=
204 h->top_samples_available=
205 h->left_samples_available= 0xFFFF;
206 h->topright_samples_available= 0xEEEA;
208 if(!(top_type & type_mask)){
209 h->topleft_samples_available= 0xB3FF;
210 h->top_samples_available= 0x33FF;
211 h->topright_samples_available= 0x26EA;
213 if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
214 if(IS_INTERLACED(mb_type)){
215 if(!(left_type[0] & type_mask)){
216 h->topleft_samples_available&= 0xDFFF;
217 h->left_samples_available&= 0x5FFF;
219 if(!(left_type[1] & type_mask)){
220 h->topleft_samples_available&= 0xFF5F;
221 h->left_samples_available&= 0xFF5F;
224 int left_typei = h->slice_table[left_xy[0] + s->mb_stride ] == h->slice_num
225 ? s->current_picture.mb_type[left_xy[0] + s->mb_stride] : 0;
226 assert(left_xy[0] == left_xy[1]);
227 if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
228 h->topleft_samples_available&= 0xDF5F;
229 h->left_samples_available&= 0x5F5F;
233 if(!(left_type[0] & type_mask)){
234 h->topleft_samples_available&= 0xDF5F;
235 h->left_samples_available&= 0x5F5F;
239 if(!(topleft_type & type_mask))
240 h->topleft_samples_available&= 0x7FFF;
242 if(!(topright_type & type_mask))
243 h->topright_samples_available&= 0xFBFF;
245 if(IS_INTRA4x4(mb_type)){
246 if(IS_INTRA4x4(top_type)){
247 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
248 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
249 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
250 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
253 if(!(top_type & type_mask))
258 h->intra4x4_pred_mode_cache[4+8*0]=
259 h->intra4x4_pred_mode_cache[5+8*0]=
260 h->intra4x4_pred_mode_cache[6+8*0]=
261 h->intra4x4_pred_mode_cache[7+8*0]= pred;
264 if(IS_INTRA4x4(left_type[i])){
265 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
266 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
269 if(!(left_type[i] & type_mask))
274 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
275 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
291 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
293 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
294 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
295 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
296 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
298 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
299 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
301 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
302 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
305 h->non_zero_count_cache[4+8*0]=
306 h->non_zero_count_cache[5+8*0]=
307 h->non_zero_count_cache[6+8*0]=
308 h->non_zero_count_cache[7+8*0]=
310 h->non_zero_count_cache[1+8*0]=
311 h->non_zero_count_cache[2+8*0]=
313 h->non_zero_count_cache[1+8*3]=
314 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
318 for (i=0; i<2; i++) {
320 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
321 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
322 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
323 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
325 h->non_zero_count_cache[3+8*1 + 2*8*i]=
326 h->non_zero_count_cache[3+8*2 + 2*8*i]=
327 h->non_zero_count_cache[0+8*1 + 8*i]=
328 h->non_zero_count_cache[0+8*4 + 8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
335 h->top_cbp = h->cbp_table[top_xy];
336 } else if(IS_INTRA(mb_type)) {
343 h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
344 } else if(IS_INTRA(mb_type)) {
350 h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
353 h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
358 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
360 for(list=0; list<h->list_count; list++){
361 if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
362 /*if(!h->mv_cache_clean[list]){
363 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
364 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
365 h->mv_cache_clean[list]= 1;
369 h->mv_cache_clean[list]= 0;
371 if(USES_LIST(top_type, list)){
372 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
373 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
374 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
375 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
376 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
377 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
378 h->ref_cache[list][scan8[0] + 0 - 1*8]=
379 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
380 h->ref_cache[list][scan8[0] + 2 - 1*8]=
381 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
383 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
384 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
385 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
386 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
387 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
391 int cache_idx = scan8[0] - 1 + i*2*8;
392 if(USES_LIST(left_type[i], list)){
393 const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
394 const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
395 *(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]];
396 *(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]];
397 h->ref_cache[list][cache_idx ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
398 h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
400 *(uint32_t*)h->mv_cache [list][cache_idx ]=
401 *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
402 h->ref_cache[list][cache_idx ]=
403 h->ref_cache[list][cache_idx+8]= left_type[i] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
407 if(for_deblock || ((IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred) && !FRAME_MBAFF))
410 if(USES_LIST(topleft_type, list)){
411 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + h->b_stride + (topleft_partition & 2*h->b_stride);
412 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + (topleft_partition & h->b8_stride);
413 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
414 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
416 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
417 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
420 if(USES_LIST(topright_type, list)){
421 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
422 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
423 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
424 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
426 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
427 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
430 if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
433 h->ref_cache[list][scan8[5 ]+1] =
434 h->ref_cache[list][scan8[7 ]+1] =
435 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewhere else)
436 h->ref_cache[list][scan8[4 ]] =
437 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
438 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
439 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
440 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
441 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
442 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
445 /* XXX beurk, Load mvd */
446 if(USES_LIST(top_type, list)){
447 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
448 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
449 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
450 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
451 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
453 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
454 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
455 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
456 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
458 if(USES_LIST(left_type[0], list)){
459 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
460 *(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]];
461 *(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]];
463 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
464 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
466 if(USES_LIST(left_type[1], list)){
467 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
468 *(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]];
469 *(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]];
471 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
472 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
474 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
475 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
476 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
477 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
478 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
480 if(h->slice_type_nos == FF_B_TYPE){
481 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
483 if(IS_DIRECT(top_type)){
484 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
485 }else if(IS_8X8(top_type)){
486 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
487 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
488 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
490 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
493 if(IS_DIRECT(left_type[0]))
494 h->direct_cache[scan8[0] - 1 + 0*8]= 1;
495 else if(IS_8X8(left_type[0]))
496 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)];
498 h->direct_cache[scan8[0] - 1 + 0*8]= 0;
500 if(IS_DIRECT(left_type[1]))
501 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
502 else if(IS_8X8(left_type[1]))
503 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)];
505 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
511 MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
512 MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
513 MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
514 MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
515 MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
516 MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
517 MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
518 MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
519 MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
520 MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
522 #define MAP_F2F(idx, mb_type)\
523 if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
524 h->ref_cache[list][idx] <<= 1;\
525 h->mv_cache[list][idx][1] /= 2;\
526 h->mvd_cache[list][idx][1] /= 2;\
531 #define MAP_F2F(idx, mb_type)\
532 if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
533 h->ref_cache[list][idx] >>= 1;\
534 h->mv_cache[list][idx][1] <<= 1;\
535 h->mvd_cache[list][idx][1] <<= 1;\
545 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
548 static inline void write_back_intra_pred_mode(H264Context *h){
549 const int mb_xy= h->mb_xy;
551 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
552 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
553 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
554 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
555 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
556 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
557 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
561 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
563 static inline int check_intra4x4_pred_mode(H264Context *h){
564 MpegEncContext * const s = &h->s;
565 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
566 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
569 if(!(h->top_samples_available&0x8000)){
571 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
573 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);
576 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
581 if((h->left_samples_available&0x8888)!=0x8888){
582 static const int mask[4]={0x8000,0x2000,0x80,0x20};
584 if(!(h->left_samples_available&mask[i])){
585 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
587 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);
590 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
597 } //FIXME cleanup like next
600 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
602 static inline int check_intra_pred_mode(H264Context *h, int mode){
603 MpegEncContext * const s = &h->s;
604 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
605 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
608 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);
612 if(!(h->top_samples_available&0x8000)){
615 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);
620 if((h->left_samples_available&0x8080) != 0x8080){
622 if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
623 mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
626 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);
635 * gets the predicted intra4x4 prediction mode.
637 static inline int pred_intra_mode(H264Context *h, int n){
638 const int index8= scan8[n];
639 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
640 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
641 const int min= FFMIN(left, top);
643 tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
645 if(min<0) return DC_PRED;
649 static inline void write_back_non_zero_count(H264Context *h){
650 const int mb_xy= h->mb_xy;
652 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
653 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
654 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
655 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
656 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
657 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
658 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
660 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
661 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
662 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
664 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
665 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
666 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
670 * gets the predicted number of non-zero coefficients.
671 * @param n block index
673 static inline int pred_non_zero_count(H264Context *h, int n){
674 const int index8= scan8[n];
675 const int left= h->non_zero_count_cache[index8 - 1];
676 const int top = h->non_zero_count_cache[index8 - 8];
679 if(i<64) i= (i+1)>>1;
681 tprintf(h->s.avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
686 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
687 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
688 MpegEncContext *s = &h->s;
690 /* there is no consistent mapping of mvs to neighboring locations that will
691 * make mbaff happy, so we can't move all this logic to fill_caches */
693 const uint32_t *mb_types = s->current_picture_ptr->mb_type;
695 *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
696 *C = h->mv_cache[list][scan8[0]-2];
699 && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
700 int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
701 if(IS_INTERLACED(mb_types[topright_xy])){
702 #define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
703 const int x4 = X4, y4 = Y4;\
704 const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
705 if(!USES_LIST(mb_type,list))\
706 return LIST_NOT_USED;\
707 mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
708 h->mv_cache[list][scan8[0]-2][0] = mv[0];\
709 h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
710 return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;
712 SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
715 if(topright_ref == PART_NOT_AVAILABLE
716 && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
717 && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
719 && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
720 SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
723 && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
725 // left shift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's OK.
726 SET_DIAG_MV(/2, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
732 if(topright_ref != PART_NOT_AVAILABLE){
733 *C= h->mv_cache[list][ i - 8 + part_width ];
736 tprintf(s->avctx, "topright MV not available\n");
738 *C= h->mv_cache[list][ i - 8 - 1 ];
739 return h->ref_cache[list][ i - 8 - 1 ];
744 * gets the predicted MV.
745 * @param n the block index
746 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
747 * @param mx the x component of the predicted motion vector
748 * @param my the y component of the predicted motion vector
750 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
751 const int index8= scan8[n];
752 const int top_ref= h->ref_cache[list][ index8 - 8 ];
753 const int left_ref= h->ref_cache[list][ index8 - 1 ];
754 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
755 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
757 int diagonal_ref, match_count;
759 assert(part_width==1 || part_width==2 || part_width==4);
769 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
770 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
771 tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count);
772 if(match_count > 1){ //most common
773 *mx= mid_pred(A[0], B[0], C[0]);
774 *my= mid_pred(A[1], B[1], C[1]);
775 }else if(match_count==1){
779 }else if(top_ref==ref){
787 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
791 *mx= mid_pred(A[0], B[0], C[0]);
792 *my= mid_pred(A[1], B[1], C[1]);
796 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);
800 * gets the directionally predicted 16x8 MV.
801 * @param n the block index
802 * @param mx the x component of the predicted motion vector
803 * @param my the y component of the predicted motion vector
805 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
807 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
808 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
810 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);
818 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
819 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
821 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);
831 pred_motion(h, n, 4, list, ref, mx, my);
835 * gets the directionally predicted 8x16 MV.
836 * @param n the block index
837 * @param mx the x component of the predicted motion vector
838 * @param my the y component of the predicted motion vector
840 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
842 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
843 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
845 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);
856 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
858 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);
860 if(diagonal_ref == ref){
868 pred_motion(h, n, 2, list, ref, mx, my);
871 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
872 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
873 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
875 tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
877 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
878 || !( top_ref | *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ])
879 || !(left_ref | *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ])){
885 pred_motion(h, 0, 4, 0, 0, mx, my);
890 static int get_scale_factor(H264Context * const h, int poc, int poc1, int i){
891 int poc0 = h->ref_list[0][i].poc;
892 int td = av_clip(poc1 - poc0, -128, 127);
893 if(td == 0 || h->ref_list[0][i].long_ref){
896 int tb = av_clip(poc - poc0, -128, 127);
897 int tx = (16384 + (FFABS(td) >> 1)) / td;
898 return av_clip((tb*tx + 32) >> 6, -1024, 1023);
902 static inline void direct_dist_scale_factor(H264Context * const h){
903 MpegEncContext * const s = &h->s;
904 const int poc = h->s.current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
905 const int poc1 = h->ref_list[1][0].poc;
907 for(field=0; field<2; field++){
908 const int poc = h->s.current_picture_ptr->field_poc[field];
909 const int poc1 = h->ref_list[1][0].field_poc[field];
910 for(i=0; i < 2*h->ref_count[0]; i++)
911 h->dist_scale_factor_field[field][i^field] = get_scale_factor(h, poc, poc1, i+16);
914 for(i=0; i<h->ref_count[0]; i++){
915 h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i);
919 static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field, int colfield, int mbafi){
920 MpegEncContext * const s = &h->s;
921 Picture * const ref1 = &h->ref_list[1][0];
922 int j, old_ref, rfield;
923 int start= mbafi ? 16 : 0;
924 int end = mbafi ? 16+2*h->ref_count[list] : h->ref_count[list];
925 int interl= mbafi || s->picture_structure != PICT_FRAME;
927 /* bogus; fills in for missing frames */
928 memset(map[list], 0, sizeof(map[list]));
930 for(rfield=0; rfield<2; rfield++){
931 for(old_ref=0; old_ref<ref1->ref_count[colfield][list]; old_ref++){
932 int poc = ref1->ref_poc[colfield][list][old_ref];
936 else if( interl && (poc&3) == 3) //FIXME store all MBAFF references so this isnt needed
937 poc= (poc&~3) + rfield + 1;
939 for(j=start; j<end; j++){
940 if(4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3) == poc){
941 int cur_ref= mbafi ? (j-16)^field : j;
942 map[list][2*old_ref + (rfield^field) + 16] = cur_ref;
944 map[list][old_ref] = cur_ref;
952 static inline void direct_ref_list_init(H264Context * const h){
953 MpegEncContext * const s = &h->s;
954 Picture * const ref1 = &h->ref_list[1][0];
955 Picture * const cur = s->current_picture_ptr;
957 int sidx= (s->picture_structure&1)^1;
958 int ref1sidx= (ref1->reference&1)^1;
960 for(list=0; list<2; list++){
961 cur->ref_count[sidx][list] = h->ref_count[list];
962 for(j=0; j<h->ref_count[list]; j++)
963 cur->ref_poc[sidx][list][j] = 4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3);
966 if(s->picture_structure == PICT_FRAME){
967 memcpy(cur->ref_count[1], cur->ref_count[0], sizeof(cur->ref_count[0]));
968 memcpy(cur->ref_poc [1], cur->ref_poc [0], sizeof(cur->ref_poc [0]));
971 cur->mbaff= FRAME_MBAFF;
973 if(cur->pict_type != FF_B_TYPE || h->direct_spatial_mv_pred)
976 for(list=0; list<2; list++){
977 fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0);
978 for(field=0; field<2; field++)
979 fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1);
983 static inline void pred_direct_motion(H264Context * const h, int *mb_type){
984 MpegEncContext * const s = &h->s;
985 int b8_stride = h->b8_stride;
986 int b4_stride = h->b_stride;
987 int mb_xy = h->mb_xy;
989 const int16_t (*l1mv0)[2], (*l1mv1)[2];
990 const int8_t *l1ref0, *l1ref1;
991 const int is_b8x8 = IS_8X8(*mb_type);
992 unsigned int sub_mb_type;
995 #define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
997 if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL
998 if(!IS_INTERLACED(*mb_type)){ // AFR/FR -> AFL/FL
999 int cur_poc = s->current_picture_ptr->poc;
1000 int *col_poc = h->ref_list[1]->field_poc;
1001 int col_parity = FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc);
1002 mb_xy= s->mb_x + ((s->mb_y&~1) + col_parity)*s->mb_stride;
1004 }else if(!(s->picture_structure & h->ref_list[1][0].reference) && !h->ref_list[1][0].mbaff){// FL -> FL & differ parity
1005 int fieldoff= 2*(h->ref_list[1][0].reference)-3;
1006 mb_xy += s->mb_stride*fieldoff;
1009 }else{ // AFL/AFR/FR/FL -> AFR/FR
1010 if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR
1011 mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
1012 mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
1013 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
1016 //FIXME IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag
1017 if( (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
1018 && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
1020 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1021 *mb_type |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */
1023 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1024 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
1026 }else{ // AFR/FR -> AFR/FR
1029 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
1030 if(IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag){
1031 /* FIXME save sub mb types from previous frames (or derive from MVs)
1032 * so we know exactly what block size to use */
1033 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1034 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
1035 }else if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
1036 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1037 *mb_type |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1039 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1040 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
1045 l1mv0 = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
1046 l1mv1 = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
1047 l1ref0 = &h->ref_list[1][0].ref_index [0][h->mb2b8_xy[mb_xy]];
1048 l1ref1 = &h->ref_list[1][0].ref_index [1][h->mb2b8_xy[mb_xy]];
1051 l1ref0 += h->b8_stride;
1052 l1ref1 += h->b8_stride;
1053 l1mv0 += 2*b4_stride;
1054 l1mv1 += 2*b4_stride;
1058 if(h->direct_spatial_mv_pred){
1063 /* FIXME interlacing + spatial direct uses wrong colocated block positions */
1065 /* ref = min(neighbors) */
1066 for(list=0; list<2; list++){
1067 int refa = h->ref_cache[list][scan8[0] - 1];
1068 int refb = h->ref_cache[list][scan8[0] - 8];
1069 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1070 if(refc == PART_NOT_AVAILABLE)
1071 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1072 ref[list] = FFMIN3((unsigned)refa, (unsigned)refb, (unsigned)refc);
1077 if(ref[0] < 0 && ref[1] < 0){
1078 ref[0] = ref[1] = 0;
1079 mv[0][0] = mv[0][1] =
1080 mv[1][0] = mv[1][1] = 0;
1082 for(list=0; list<2; list++){
1084 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1086 mv[list][0] = mv[list][1] = 0;
1092 *mb_type &= ~MB_TYPE_L1;
1093 sub_mb_type &= ~MB_TYPE_L1;
1094 }else if(ref[0] < 0){
1096 *mb_type &= ~MB_TYPE_L0;
1097 sub_mb_type &= ~MB_TYPE_L0;
1100 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
1101 for(i8=0; i8<4; i8++){
1104 int xy8 = x8+y8*b8_stride;
1105 int xy4 = 3*x8+y8*b4_stride;
1108 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1110 h->sub_mb_type[i8] = sub_mb_type;
1112 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1113 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1114 if(!IS_INTRA(mb_type_col[y8])
1115 && ( (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
1116 || (l1ref0[xy8] < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
1118 a= pack16to32(mv[0][0],mv[0][1]);
1120 b= pack16to32(mv[1][0],mv[1][1]);
1122 a= pack16to32(mv[0][0],mv[0][1]);
1123 b= pack16to32(mv[1][0],mv[1][1]);
1125 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
1126 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
1128 }else if(IS_16X16(*mb_type)){
1131 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1132 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
1133 if(!IS_INTRA(mb_type_col[0])
1134 && ( (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
1135 || (l1ref0[0] < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
1136 && (h->x264_build>33 || !h->x264_build)))){
1138 a= pack16to32(mv[0][0],mv[0][1]);
1140 b= pack16to32(mv[1][0],mv[1][1]);
1142 a= pack16to32(mv[0][0],mv[0][1]);
1143 b= pack16to32(mv[1][0],mv[1][1]);
1145 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
1146 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
1148 for(i8=0; i8<4; i8++){
1149 const int x8 = i8&1;
1150 const int y8 = i8>>1;
1152 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1154 h->sub_mb_type[i8] = sub_mb_type;
1156 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1157 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1158 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1159 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1162 if(!IS_INTRA(mb_type_col[0]) && ( l1ref0[x8 + y8*b8_stride] == 0
1163 || (l1ref0[x8 + y8*b8_stride] < 0 && l1ref1[x8 + y8*b8_stride] == 0
1164 && (h->x264_build>33 || !h->x264_build)))){
1165 const int16_t (*l1mv)[2]= l1ref0[x8 + y8*b8_stride] == 0 ? l1mv0 : l1mv1;
1166 if(IS_SUB_8X8(sub_mb_type)){
1167 const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
1168 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1170 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1172 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1175 for(i4=0; i4<4; i4++){
1176 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
1177 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1179 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1181 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1187 }else{ /* direct temporal mv pred */
1188 const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
1189 const int *dist_scale_factor = h->dist_scale_factor;
1192 if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){
1193 map_col_to_list0[0] = h->map_col_to_list0_field[s->mb_y&1][0];
1194 map_col_to_list0[1] = h->map_col_to_list0_field[s->mb_y&1][1];
1195 dist_scale_factor =h->dist_scale_factor_field[s->mb_y&1];
1197 if(h->ref_list[1][0].mbaff && IS_INTERLACED(mb_type_col[0]))
1200 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
1201 /* FIXME assumes direct_8x8_inference == 1 */
1202 int y_shift = 2*!IS_INTERLACED(*mb_type);
1204 for(i8=0; i8<4; i8++){
1205 const int x8 = i8&1;
1206 const int y8 = i8>>1;
1208 const int16_t (*l1mv)[2]= l1mv0;
1210 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1212 h->sub_mb_type[i8] = sub_mb_type;
1214 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1215 if(IS_INTRA(mb_type_col[y8])){
1216 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1217 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1218 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1222 ref0 = l1ref0[x8 + y8*b8_stride];
1224 ref0 = map_col_to_list0[0][ref0 + ref_offset];
1226 ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
1229 scale = dist_scale_factor[ref0];
1230 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1233 const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
1234 int my_col = (mv_col[1]<<y_shift)/2;
1235 int mx = (scale * mv_col[0] + 128) >> 8;
1236 int my = (scale * my_col + 128) >> 8;
1237 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1238 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
1244 /* one-to-one mv scaling */
1246 if(IS_16X16(*mb_type)){
1249 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1250 if(IS_INTRA(mb_type_col[0])){
1253 const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset]
1254 : map_col_to_list0[1][l1ref1[0] + ref_offset];
1255 const int scale = dist_scale_factor[ref0];
1256 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1258 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1259 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1261 mv0= pack16to32(mv_l0[0],mv_l0[1]);
1262 mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1264 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
1265 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
1266 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
1268 for(i8=0; i8<4; i8++){
1269 const int x8 = i8&1;
1270 const int y8 = i8>>1;
1272 const int16_t (*l1mv)[2]= l1mv0;
1274 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1276 h->sub_mb_type[i8] = sub_mb_type;
1277 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1278 if(IS_INTRA(mb_type_col[0])){
1279 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1280 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1281 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1285 ref0 = l1ref0[x8 + y8*b8_stride] + ref_offset;
1287 ref0 = map_col_to_list0[0][ref0];
1289 ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
1292 scale = dist_scale_factor[ref0];
1294 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1295 if(IS_SUB_8X8(sub_mb_type)){
1296 const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
1297 int mx = (scale * mv_col[0] + 128) >> 8;
1298 int my = (scale * mv_col[1] + 128) >> 8;
1299 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1300 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1302 for(i4=0; i4<4; i4++){
1303 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
1304 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1305 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1306 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1307 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1308 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1315 static inline void write_back_motion(H264Context *h, int mb_type){
1316 MpegEncContext * const s = &h->s;
1317 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1318 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1321 if(!USES_LIST(mb_type, 0))
1322 fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1324 for(list=0; list<h->list_count; list++){
1326 if(!USES_LIST(mb_type, list))
1330 *(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];
1331 *(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];
1333 if( h->pps.cabac ) {
1334 if(IS_SKIP(mb_type))
1335 fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
1338 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1339 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1344 int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1345 ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1346 ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1347 ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1348 ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1352 if(h->slice_type_nos == FF_B_TYPE && h->pps.cabac){
1353 if(IS_8X8(mb_type)){
1354 uint8_t *direct_table = &h->direct_table[b8_xy];
1355 direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1356 direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1357 direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1363 * Decodes a network abstraction layer unit.
1364 * @param consumed is the number of bytes used as input
1365 * @param length is the length of the array
1366 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1367 * @returns decoded bytes, might be src+1 if no escapes
1369 static const uint8_t *decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
1374 // src[0]&0x80; //forbidden bit
1375 h->nal_ref_idc= src[0]>>5;
1376 h->nal_unit_type= src[0]&0x1F;
1380 for(i=0; i<length; i++)
1381 printf("%2X ", src[i]);
1384 #if HAVE_FAST_UNALIGNED
1385 # if HAVE_FAST_64BIT
1387 for(i=0; i+1<length; i+=9){
1388 if(!((~*(uint64_t*)(src+i) & (*(uint64_t*)(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
1391 for(i=0; i+1<length; i+=5){
1392 if(!((~*(uint32_t*)(src+i) & (*(uint32_t*)(src+i) - 0x01000101U)) & 0x80008080U))
1395 if(i>0 && !src[i]) i--;
1399 for(i=0; i+1<length; i+=2){
1400 if(src[i]) continue;
1401 if(i>0 && src[i-1]==0) i--;
1403 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1405 /* startcode, so we must be past the end */
1413 if(i>=length-1){ //no escaped 0
1414 *dst_length= length;
1415 *consumed= length+1; //+1 for the header
1419 bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
1420 h->rbsp_buffer[bufidx]= av_fast_realloc(h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
1421 dst= h->rbsp_buffer[bufidx];
1427 //printf("decoding esc\n");
1428 memcpy(dst, src, i);
1431 //remove escapes (very rare 1:2^22)
1433 dst[di++]= src[si++];
1434 dst[di++]= src[si++];
1435 }else if(src[si]==0 && src[si+1]==0){
1436 if(src[si+2]==3){ //escape
1441 }else //next start code
1445 dst[di++]= src[si++];
1448 dst[di++]= src[si++];
1451 memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
1454 *consumed= si + 1;//+1 for the header
1455 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
1460 * identifies the exact end of the bitstream
1461 * @return the length of the trailing, or 0 if damaged
1463 static int decode_rbsp_trailing(H264Context *h, const uint8_t *src){
1467 tprintf(h->s.avctx, "rbsp trailing %X\n", v);
1477 * IDCT transforms the 16 dc values and dequantizes them.
1478 * @param qp quantization parameter
1480 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1483 int temp[16]; //FIXME check if this is a good idea
1484 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1485 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1487 //memset(block, 64, 2*256);
1490 const int offset= y_offset[i];
1491 const int z0= block[offset+stride*0] + block[offset+stride*4];
1492 const int z1= block[offset+stride*0] - block[offset+stride*4];
1493 const int z2= block[offset+stride*1] - block[offset+stride*5];
1494 const int z3= block[offset+stride*1] + block[offset+stride*5];
1503 const int offset= x_offset[i];
1504 const int z0= temp[4*0+i] + temp[4*2+i];
1505 const int z1= temp[4*0+i] - temp[4*2+i];
1506 const int z2= temp[4*1+i] - temp[4*3+i];
1507 const int z3= temp[4*1+i] + temp[4*3+i];
1509 block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
1510 block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1511 block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1512 block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1518 * DCT transforms the 16 dc values.
1519 * @param qp quantization parameter ??? FIXME
1521 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1522 // const int qmul= dequant_coeff[qp][0];
1524 int temp[16]; //FIXME check if this is a good idea
1525 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1526 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1529 const int offset= y_offset[i];
1530 const int z0= block[offset+stride*0] + block[offset+stride*4];
1531 const int z1= block[offset+stride*0] - block[offset+stride*4];
1532 const int z2= block[offset+stride*1] - block[offset+stride*5];
1533 const int z3= block[offset+stride*1] + block[offset+stride*5];
1542 const int offset= x_offset[i];
1543 const int z0= temp[4*0+i] + temp[4*2+i];
1544 const int z1= temp[4*0+i] - temp[4*2+i];
1545 const int z2= temp[4*1+i] - temp[4*3+i];
1546 const int z3= temp[4*1+i] + temp[4*3+i];
1548 block[stride*0 +offset]= (z0 + z3)>>1;
1549 block[stride*2 +offset]= (z1 + z2)>>1;
1550 block[stride*8 +offset]= (z1 - z2)>>1;
1551 block[stride*10+offset]= (z0 - z3)>>1;
1559 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1560 const int stride= 16*2;
1561 const int xStride= 16;
1564 a= block[stride*0 + xStride*0];
1565 b= block[stride*0 + xStride*1];
1566 c= block[stride*1 + xStride*0];
1567 d= block[stride*1 + xStride*1];
1574 block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1575 block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1576 block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1577 block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
1581 static void chroma_dc_dct_c(DCTELEM *block){
1582 const int stride= 16*2;
1583 const int xStride= 16;
1586 a= block[stride*0 + xStride*0];
1587 b= block[stride*0 + xStride*1];
1588 c= block[stride*1 + xStride*0];
1589 d= block[stride*1 + xStride*1];
1596 block[stride*0 + xStride*0]= (a+c);
1597 block[stride*0 + xStride*1]= (e+b);
1598 block[stride*1 + xStride*0]= (a-c);
1599 block[stride*1 + xStride*1]= (e-b);
1604 * gets the chroma qp.
1606 static inline int get_chroma_qp(H264Context *h, int t, int qscale){
1607 return h->pps.chroma_qp_table[t][qscale];
1610 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1611 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1612 int src_x_offset, int src_y_offset,
1613 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1614 MpegEncContext * const s = &h->s;
1615 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1616 int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1617 const int luma_xy= (mx&3) + ((my&3)<<2);
1618 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
1619 uint8_t * src_cb, * src_cr;
1620 int extra_width= h->emu_edge_width;
1621 int extra_height= h->emu_edge_height;
1623 const int full_mx= mx>>2;
1624 const int full_my= my>>2;
1625 const int pic_width = 16*s->mb_width;
1626 const int pic_height = 16*s->mb_height >> MB_FIELD;
1628 if(mx&7) extra_width -= 3;
1629 if(my&7) extra_height -= 3;
1631 if( full_mx < 0-extra_width
1632 || full_my < 0-extra_height
1633 || full_mx + 16/*FIXME*/ > pic_width + extra_width
1634 || full_my + 16/*FIXME*/ > pic_height + extra_height){
1635 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);
1636 src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
1640 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
1642 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
1645 if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
1648 // chroma offset when predicting from a field of opposite parity
1649 my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
1650 emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
1652 src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1653 src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1656 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);
1657 src_cb= s->edge_emu_buffer;
1659 chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1662 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);
1663 src_cr= s->edge_emu_buffer;
1665 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1668 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
1669 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1670 int x_offset, int y_offset,
1671 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1672 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1673 int list0, int list1){
1674 MpegEncContext * const s = &h->s;
1675 qpel_mc_func *qpix_op= qpix_put;
1676 h264_chroma_mc_func chroma_op= chroma_put;
1678 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize;
1679 dest_cb += x_offset + y_offset*h->mb_uvlinesize;
1680 dest_cr += x_offset + y_offset*h->mb_uvlinesize;
1681 x_offset += 8*s->mb_x;
1682 y_offset += 8*(s->mb_y >> MB_FIELD);
1685 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1686 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1687 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1688 qpix_op, chroma_op);
1691 chroma_op= chroma_avg;
1695 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
1696 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1697 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1698 qpix_op, chroma_op);
1702 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
1703 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1704 int x_offset, int y_offset,
1705 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1706 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
1707 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
1708 int list0, int list1){
1709 MpegEncContext * const s = &h->s;
1711 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize;
1712 dest_cb += x_offset + y_offset*h->mb_uvlinesize;
1713 dest_cr += x_offset + y_offset*h->mb_uvlinesize;
1714 x_offset += 8*s->mb_x;
1715 y_offset += 8*(s->mb_y >> MB_FIELD);
1718 /* don't optimize for luma-only case, since B-frames usually
1719 * use implicit weights => chroma too. */
1720 uint8_t *tmp_cb = s->obmc_scratchpad;
1721 uint8_t *tmp_cr = s->obmc_scratchpad + 8;
1722 uint8_t *tmp_y = s->obmc_scratchpad + 8*h->mb_uvlinesize;
1723 int refn0 = h->ref_cache[0][ scan8[n] ];
1724 int refn1 = h->ref_cache[1][ scan8[n] ];
1726 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
1727 dest_y, dest_cb, dest_cr,
1728 x_offset, y_offset, qpix_put, chroma_put);
1729 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
1730 tmp_y, tmp_cb, tmp_cr,
1731 x_offset, y_offset, qpix_put, chroma_put);
1733 if(h->use_weight == 2){
1734 int weight0 = h->implicit_weight[refn0][refn1];
1735 int weight1 = 64 - weight0;
1736 luma_weight_avg( dest_y, tmp_y, h-> mb_linesize, 5, weight0, weight1, 0);
1737 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
1738 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
1740 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
1741 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
1742 h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
1743 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1744 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
1745 h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
1746 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1747 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
1748 h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
1751 int list = list1 ? 1 : 0;
1752 int refn = h->ref_cache[list][ scan8[n] ];
1753 Picture *ref= &h->ref_list[list][refn];
1754 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
1755 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1756 qpix_put, chroma_put);
1758 luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
1759 h->luma_weight[list][refn], h->luma_offset[list][refn]);
1760 if(h->use_weight_chroma){
1761 chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1762 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
1763 chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1764 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
1769 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1770 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1771 int x_offset, int y_offset,
1772 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1773 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1774 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
1775 int list0, int list1){
1776 if((h->use_weight==2 && list0 && list1
1777 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
1778 || h->use_weight==1)
1779 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1780 x_offset, y_offset, qpix_put, chroma_put,
1781 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
1783 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1784 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
1787 static inline void prefetch_motion(H264Context *h, int list){
1788 /* fetch pixels for estimated mv 4 macroblocks ahead
1789 * optimized for 64byte cache lines */
1790 MpegEncContext * const s = &h->s;
1791 const int refn = h->ref_cache[list][scan8[0]];
1793 const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
1794 const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
1795 uint8_t **src= h->ref_list[list][refn].data;
1796 int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
1797 s->dsp.prefetch(src[0]+off, s->linesize, 4);
1798 off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
1799 s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
1803 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1804 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1805 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
1806 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
1807 MpegEncContext * const s = &h->s;
1808 const int mb_xy= h->mb_xy;
1809 const int mb_type= s->current_picture.mb_type[mb_xy];
1811 assert(IS_INTER(mb_type));
1813 prefetch_motion(h, 0);
1815 if(IS_16X16(mb_type)){
1816 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1817 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1818 &weight_op[0], &weight_avg[0],
1819 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1820 }else if(IS_16X8(mb_type)){
1821 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1822 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1823 &weight_op[1], &weight_avg[1],
1824 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1825 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1826 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1827 &weight_op[1], &weight_avg[1],
1828 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1829 }else if(IS_8X16(mb_type)){
1830 mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
1831 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1832 &weight_op[2], &weight_avg[2],
1833 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1834 mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
1835 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1836 &weight_op[2], &weight_avg[2],
1837 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1841 assert(IS_8X8(mb_type));
1844 const int sub_mb_type= h->sub_mb_type[i];
1846 int x_offset= (i&1)<<2;
1847 int y_offset= (i&2)<<1;
1849 if(IS_SUB_8X8(sub_mb_type)){
1850 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1851 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1852 &weight_op[3], &weight_avg[3],
1853 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1854 }else if(IS_SUB_8X4(sub_mb_type)){
1855 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1856 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1857 &weight_op[4], &weight_avg[4],
1858 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1859 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
1860 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1861 &weight_op[4], &weight_avg[4],
1862 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1863 }else if(IS_SUB_4X8(sub_mb_type)){
1864 mc_part(h, n , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1865 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1866 &weight_op[5], &weight_avg[5],
1867 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1868 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
1869 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1870 &weight_op[5], &weight_avg[5],
1871 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1874 assert(IS_SUB_4X4(sub_mb_type));
1876 int sub_x_offset= x_offset + 2*(j&1);
1877 int sub_y_offset= y_offset + (j&2);
1878 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
1879 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1880 &weight_op[6], &weight_avg[6],
1881 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1887 prefetch_motion(h, 1);
1890 static av_cold void init_cavlc_level_tab(void){
1891 int suffix_length, mask;
1894 for(suffix_length=0; suffix_length<7; suffix_length++){
1895 for(i=0; i<(1<<LEVEL_TAB_BITS); i++){
1896 int prefix= LEVEL_TAB_BITS - av_log2(2*i);
1897 int level_code= (prefix<<suffix_length) + (i>>(LEVEL_TAB_BITS-prefix-1-suffix_length)) - (1<<suffix_length);
1899 mask= -(level_code&1);
1900 level_code= (((2+level_code)>>1) ^ mask) - mask;
1901 if(prefix + 1 + suffix_length <= LEVEL_TAB_BITS){
1902 cavlc_level_tab[suffix_length][i][0]= level_code;
1903 cavlc_level_tab[suffix_length][i][1]= prefix + 1 + suffix_length;
1904 }else if(prefix + 1 <= LEVEL_TAB_BITS){
1905 cavlc_level_tab[suffix_length][i][0]= prefix+100;
1906 cavlc_level_tab[suffix_length][i][1]= prefix + 1;
1908 cavlc_level_tab[suffix_length][i][0]= LEVEL_TAB_BITS+100;
1909 cavlc_level_tab[suffix_length][i][1]= LEVEL_TAB_BITS;
1915 static av_cold void decode_init_vlc(void){
1916 static int done = 0;
1923 chroma_dc_coeff_token_vlc.table = chroma_dc_coeff_token_vlc_table;
1924 chroma_dc_coeff_token_vlc.table_allocated = chroma_dc_coeff_token_vlc_table_size;
1925 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
1926 &chroma_dc_coeff_token_len [0], 1, 1,
1927 &chroma_dc_coeff_token_bits[0], 1, 1,
1928 INIT_VLC_USE_NEW_STATIC);
1932 coeff_token_vlc[i].table = coeff_token_vlc_tables+offset;
1933 coeff_token_vlc[i].table_allocated = coeff_token_vlc_tables_size[i];
1934 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
1935 &coeff_token_len [i][0], 1, 1,
1936 &coeff_token_bits[i][0], 1, 1,
1937 INIT_VLC_USE_NEW_STATIC);
1938 offset += coeff_token_vlc_tables_size[i];
1941 * This is a one time safety check to make sure that
1942 * the packed static coeff_token_vlc table sizes
1943 * were initialized correctly.
1945 assert(offset == FF_ARRAY_ELEMS(coeff_token_vlc_tables));
1948 chroma_dc_total_zeros_vlc[i].table = chroma_dc_total_zeros_vlc_tables[i];
1949 chroma_dc_total_zeros_vlc[i].table_allocated = chroma_dc_total_zeros_vlc_tables_size;
1950 init_vlc(&chroma_dc_total_zeros_vlc[i],
1951 CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
1952 &chroma_dc_total_zeros_len [i][0], 1, 1,
1953 &chroma_dc_total_zeros_bits[i][0], 1, 1,
1954 INIT_VLC_USE_NEW_STATIC);
1956 for(i=0; i<15; i++){
1957 total_zeros_vlc[i].table = total_zeros_vlc_tables[i];
1958 total_zeros_vlc[i].table_allocated = total_zeros_vlc_tables_size;
1959 init_vlc(&total_zeros_vlc[i],
1960 TOTAL_ZEROS_VLC_BITS, 16,
1961 &total_zeros_len [i][0], 1, 1,
1962 &total_zeros_bits[i][0], 1, 1,
1963 INIT_VLC_USE_NEW_STATIC);
1967 run_vlc[i].table = run_vlc_tables[i];
1968 run_vlc[i].table_allocated = run_vlc_tables_size;
1969 init_vlc(&run_vlc[i],
1971 &run_len [i][0], 1, 1,
1972 &run_bits[i][0], 1, 1,
1973 INIT_VLC_USE_NEW_STATIC);
1975 run7_vlc.table = run7_vlc_table,
1976 run7_vlc.table_allocated = run7_vlc_table_size;
1977 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
1978 &run_len [6][0], 1, 1,
1979 &run_bits[6][0], 1, 1,
1980 INIT_VLC_USE_NEW_STATIC);
1982 init_cavlc_level_tab();
1986 static void free_tables(H264Context *h){
1989 av_freep(&h->intra4x4_pred_mode);
1990 av_freep(&h->chroma_pred_mode_table);
1991 av_freep(&h->cbp_table);
1992 av_freep(&h->mvd_table[0]);
1993 av_freep(&h->mvd_table[1]);
1994 av_freep(&h->direct_table);
1995 av_freep(&h->non_zero_count);
1996 av_freep(&h->slice_table_base);
1997 h->slice_table= NULL;
1999 av_freep(&h->mb2b_xy);
2000 av_freep(&h->mb2b8_xy);
2002 for(i = 0; i < h->s.avctx->thread_count; i++) {
2003 hx = h->thread_context[i];
2005 av_freep(&hx->top_borders[1]);
2006 av_freep(&hx->top_borders[0]);
2007 av_freep(&hx->s.obmc_scratchpad);
2011 static void init_dequant8_coeff_table(H264Context *h){
2013 const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
2014 h->dequant8_coeff[0] = h->dequant8_buffer[0];
2015 h->dequant8_coeff[1] = h->dequant8_buffer[1];
2017 for(i=0; i<2; i++ ){
2018 if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
2019 h->dequant8_coeff[1] = h->dequant8_buffer[0];
2023 for(q=0; q<52; q++){
2024 int shift = div6[q];
2027 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
2028 ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
2029 h->pps.scaling_matrix8[i][x]) << shift;
2034 static void init_dequant4_coeff_table(H264Context *h){
2036 const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
2037 for(i=0; i<6; i++ ){
2038 h->dequant4_coeff[i] = h->dequant4_buffer[i];
2040 if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
2041 h->dequant4_coeff[i] = h->dequant4_buffer[j];
2048 for(q=0; q<52; q++){
2049 int shift = div6[q] + 2;
2052 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
2053 ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
2054 h->pps.scaling_matrix4[i][x]) << shift;
2059 static void init_dequant_tables(H264Context *h){
2061 init_dequant4_coeff_table(h);
2062 if(h->pps.transform_8x8_mode)
2063 init_dequant8_coeff_table(h);
2064 if(h->sps.transform_bypass){
2067 h->dequant4_coeff[i][0][x] = 1<<6;
2068 if(h->pps.transform_8x8_mode)
2071 h->dequant8_coeff[i][0][x] = 1<<6;
2078 * needs width/height
2080 static int alloc_tables(H264Context *h){
2081 MpegEncContext * const s = &h->s;
2082 const int big_mb_num= s->mb_stride * (s->mb_height+1);
2085 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
2087 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
2088 CHECKED_ALLOCZ(h->slice_table_base , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base))
2089 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2091 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2092 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2093 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2094 CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2096 memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base));
2097 h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
2099 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint32_t));
2100 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
2101 for(y=0; y<s->mb_height; y++){
2102 for(x=0; x<s->mb_width; x++){
2103 const int mb_xy= x + y*s->mb_stride;
2104 const int b_xy = 4*x + 4*y*h->b_stride;
2105 const int b8_xy= 2*x + 2*y*h->b8_stride;
2107 h->mb2b_xy [mb_xy]= b_xy;
2108 h->mb2b8_xy[mb_xy]= b8_xy;
2112 s->obmc_scratchpad = NULL;
2114 if(!h->dequant4_coeff[0])
2115 init_dequant_tables(h);
2124 * Mimic alloc_tables(), but for every context thread.
2126 static void clone_tables(H264Context *dst, H264Context *src){
2127 dst->intra4x4_pred_mode = src->intra4x4_pred_mode;
2128 dst->non_zero_count = src->non_zero_count;
2129 dst->slice_table = src->slice_table;
2130 dst->cbp_table = src->cbp_table;
2131 dst->mb2b_xy = src->mb2b_xy;
2132 dst->mb2b8_xy = src->mb2b8_xy;
2133 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
2134 dst->mvd_table[0] = src->mvd_table[0];
2135 dst->mvd_table[1] = src->mvd_table[1];
2136 dst->direct_table = src->direct_table;
2138 dst->s.obmc_scratchpad = NULL;
2139 ff_h264_pred_init(&dst->hpc, src->s.codec_id);
2144 * Allocate buffers which are not shared amongst multiple threads.
2146 static int context_init(H264Context *h){
2147 CHECKED_ALLOCZ(h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2148 CHECKED_ALLOCZ(h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2152 return -1; // free_tables will clean up for us
2155 static av_cold void common_init(H264Context *h){
2156 MpegEncContext * const s = &h->s;
2158 s->width = s->avctx->width;
2159 s->height = s->avctx->height;
2160 s->codec_id= s->avctx->codec->id;
2162 ff_h264_pred_init(&h->hpc, s->codec_id);
2164 h->dequant_coeff_pps= -1;
2165 s->unrestricted_mv=1;
2166 s->decode=1; //FIXME
2168 dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
2170 memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
2171 memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
2174 static av_cold int decode_init(AVCodecContext *avctx){
2175 H264Context *h= avctx->priv_data;
2176 MpegEncContext * const s = &h->s;
2178 MPV_decode_defaults(s);
2183 s->out_format = FMT_H264;
2184 s->workaround_bugs= avctx->workaround_bugs;
2187 // s->decode_mb= ff_h263_decode_mb;
2188 s->quarter_sample = 1;
2191 if(avctx->codec_id == CODEC_ID_SVQ3)
2192 avctx->pix_fmt= PIX_FMT_YUVJ420P;
2193 else if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2194 avctx->pix_fmt= PIX_FMT_VDPAU_H264;
2196 avctx->pix_fmt= PIX_FMT_YUV420P;
2200 if(avctx->extradata_size > 0 && avctx->extradata &&
2201 *(char *)avctx->extradata == 1){
2208 h->thread_context[0] = h;
2209 h->outputed_poc = INT_MIN;
2210 h->prev_poc_msb= 1<<16;
2211 h->sei_recovery_frame_cnt = -1;
2212 h->sei_dpb_output_delay = 0;
2216 static int frame_start(H264Context *h){
2217 MpegEncContext * const s = &h->s;
2220 if(MPV_frame_start(s, s->avctx) < 0)
2222 ff_er_frame_start(s);
2224 * MPV_frame_start uses pict_type to derive key_frame.
2225 * This is incorrect for H.264; IDR markings must be used.
2226 * Zero here; IDR markings per slice in frame or fields are ORed in later.
2227 * See decode_nal_units().
2229 s->current_picture_ptr->key_frame= 0;
2231 assert(s->linesize && s->uvlinesize);
2233 for(i=0; i<16; i++){
2234 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2235 h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
2238 h->block_offset[16+i]=
2239 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2240 h->block_offset[24+16+i]=
2241 h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2244 /* can't be in alloc_tables because linesize isn't known there.
2245 * FIXME: redo bipred weight to not require extra buffer? */
2246 for(i = 0; i < s->avctx->thread_count; i++)
2247 if(!h->thread_context[i]->s.obmc_scratchpad)
2248 h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
2250 /* some macroblocks will be accessed before they're available */
2251 if(FRAME_MBAFF || s->avctx->thread_count > 1)
2252 memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
2254 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2256 // We mark the current picture as non-reference after allocating it, so
2257 // that if we break out due to an error it can be released automatically
2258 // in the next MPV_frame_start().
2259 // SVQ3 as well as most other codecs have only last/next/current and thus
2260 // get released even with set reference, besides SVQ3 and others do not
2261 // mark frames as reference later "naturally".
2262 if(s->codec_id != CODEC_ID_SVQ3)
2263 s->current_picture_ptr->reference= 0;
2265 s->current_picture_ptr->field_poc[0]=
2266 s->current_picture_ptr->field_poc[1]= INT_MAX;
2267 assert(s->current_picture_ptr->long_ref==0);
2272 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){
2273 MpegEncContext * const s = &h->s;
2282 src_cb -= uvlinesize;
2283 src_cr -= uvlinesize;
2285 if(!simple && FRAME_MBAFF){
2287 offset = MB_MBAFF ? 1 : 17;
2288 uvoffset= MB_MBAFF ? 1 : 9;
2290 *(uint64_t*)(h->top_borders[0][s->mb_x]+ 0)= *(uint64_t*)(src_y + 15*linesize);
2291 *(uint64_t*)(h->top_borders[0][s->mb_x]+ 8)= *(uint64_t*)(src_y +8+15*linesize);
2292 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2293 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+7*uvlinesize);
2294 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+7*uvlinesize);
2299 h->left_border[0]= h->top_borders[0][s->mb_x][15];
2300 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2301 h->left_border[34 ]= h->top_borders[0][s->mb_x][16+7 ];
2302 h->left_border[34+18]= h->top_borders[0][s->mb_x][16+8+7];
2308 top_idx = MB_MBAFF ? 0 : 1;
2310 step= MB_MBAFF ? 2 : 1;
2313 // There are two lines saved, the line above the the top macroblock of a pair,
2314 // and the line above the bottom macroblock
2315 h->left_border[offset]= h->top_borders[top_idx][s->mb_x][15];
2316 for(i=1; i<17 - skiplast; i++){
2317 h->left_border[offset+i*step]= src_y[15+i* linesize];
2320 *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
2321 *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2323 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2324 h->left_border[uvoffset+34 ]= h->top_borders[top_idx][s->mb_x][16+7];
2325 h->left_border[uvoffset+34+18]= h->top_borders[top_idx][s->mb_x][24+7];
2326 for(i=1; i<9 - skiplast; i++){
2327 h->left_border[uvoffset+34 +i*step]= src_cb[7+i*uvlinesize];
2328 h->left_border[uvoffset+34+18+i*step]= src_cr[7+i*uvlinesize];
2330 *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2331 *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2335 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){
2336 MpegEncContext * const s = &h->s;
2347 if(!simple && FRAME_MBAFF){
2349 offset = MB_MBAFF ? 1 : 17;
2350 uvoffset= MB_MBAFF ? 1 : 9;
2354 top_idx = MB_MBAFF ? 0 : 1;
2356 step= MB_MBAFF ? 2 : 1;
2359 if(h->deblocking_filter == 2) {
2361 deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
2362 deblock_top = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
2364 deblock_left = (s->mb_x > 0);
2365 deblock_top = (s->mb_y > !!MB_FIELD);
2368 src_y -= linesize + 1;
2369 src_cb -= uvlinesize + 1;
2370 src_cr -= uvlinesize + 1;
2372 #define XCHG(a,b,t,xchg)\
2379 for(i = !deblock_top; i<16; i++){
2380 XCHG(h->left_border[offset+i*step], src_y [i* linesize], temp8, xchg);
2382 XCHG(h->left_border[offset+i*step], src_y [i* linesize], temp8, 1);
2386 XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2387 XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2388 if(s->mb_x+1 < s->mb_width){
2389 XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
2393 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2395 for(i = !deblock_top; i<8; i++){
2396 XCHG(h->left_border[uvoffset+34 +i*step], src_cb[i*uvlinesize], temp8, xchg);
2397 XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, xchg);
2399 XCHG(h->left_border[uvoffset+34 +i*step], src_cb[i*uvlinesize], temp8, 1);
2400 XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, 1);
2403 XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2404 XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2409 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
2410 MpegEncContext * const s = &h->s;
2411 const int mb_x= s->mb_x;
2412 const int mb_y= s->mb_y;
2413 const int mb_xy= h->mb_xy;
2414 const int mb_type= s->current_picture.mb_type[mb_xy];
2415 uint8_t *dest_y, *dest_cb, *dest_cr;
2416 int linesize, uvlinesize /*dct_offset*/;
2418 int *block_offset = &h->block_offset[0];
2419 const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
2420 /* is_h264 should always be true if SVQ3 is disabled. */
2421 const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
2422 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
2423 void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
2425 dest_y = s->current_picture.data[0] + (mb_x + mb_y * s->linesize ) * 16;
2426 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2427 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2429 s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
2430 s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
2432 if (!simple && MB_FIELD) {
2433 linesize = h->mb_linesize = s->linesize * 2;
2434 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2435 block_offset = &h->block_offset[24];
2436 if(mb_y&1){ //FIXME move out of this function?
2437 dest_y -= s->linesize*15;
2438 dest_cb-= s->uvlinesize*7;
2439 dest_cr-= s->uvlinesize*7;
2443 for(list=0; list<h->list_count; list++){
2444 if(!USES_LIST(mb_type, list))
2446 if(IS_16X16(mb_type)){
2447 int8_t *ref = &h->ref_cache[list][scan8[0]];
2448 fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
2450 for(i=0; i<16; i+=4){
2451 int ref = h->ref_cache[list][scan8[i]];
2453 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
2459 linesize = h->mb_linesize = s->linesize;
2460 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2461 // dct_offset = s->linesize * 16;
2464 if (!simple && IS_INTRA_PCM(mb_type)) {
2465 for (i=0; i<16; i++) {
2466 memcpy(dest_y + i* linesize, h->mb + i*8, 16);
2468 for (i=0; i<8; i++) {
2469 memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4, 8);
2470 memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4, 8);
2473 if(IS_INTRA(mb_type)){
2474 if(h->deblocking_filter)
2475 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
2477 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2478 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2479 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2482 if(IS_INTRA4x4(mb_type)){
2483 if(simple || !s->encoding){
2484 if(IS_8x8DCT(mb_type)){
2485 if(transform_bypass){
2487 idct_add = s->dsp.add_pixels8;
2489 idct_dc_add = s->dsp.h264_idct8_dc_add;
2490 idct_add = s->dsp.h264_idct8_add;
2492 for(i=0; i<16; i+=4){
2493 uint8_t * const ptr= dest_y + block_offset[i];
2494 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2495 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
2496 h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
2498 const int nnz = h->non_zero_count_cache[ scan8[i] ];
2499 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
2500 (h->topright_samples_available<<i)&0x4000, linesize);
2502 if(nnz == 1 && h->mb[i*16])
2503 idct_dc_add(ptr, h->mb + i*16, linesize);
2505 idct_add (ptr, h->mb + i*16, linesize);
2510 if(transform_bypass){
2512 idct_add = s->dsp.add_pixels4;
2514 idct_dc_add = s->dsp.h264_idct_dc_add;
2515 idct_add = s->dsp.h264_idct_add;
2517 for(i=0; i<16; i++){
2518 uint8_t * const ptr= dest_y + block_offset[i];
2519 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2521 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
2522 h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
2526 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2527 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2528 assert(mb_y || linesize <= block_offset[i]);
2529 if(!topright_avail){
2530 tr= ptr[3 - linesize]*0x01010101;
2531 topright= (uint8_t*) &tr;
2533 topright= ptr + 4 - linesize;
2537 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
2538 nnz = h->non_zero_count_cache[ scan8[i] ];
2541 if(nnz == 1 && h->mb[i*16])
2542 idct_dc_add(ptr, h->mb + i*16, linesize);
2544 idct_add (ptr, h->mb + i*16, linesize);
2546 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2553 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2555 if(!transform_bypass)
2556 h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
2558 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2560 if(h->deblocking_filter)
2561 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
2563 hl_motion(h, dest_y, dest_cb, dest_cr,
2564 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2565 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2566 s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
2570 if(!IS_INTRA4x4(mb_type)){
2572 if(IS_INTRA16x16(mb_type)){
2573 if(transform_bypass){
2574 if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
2575 h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
2577 for(i=0; i<16; i++){
2578 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
2579 s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
2583 s->dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2585 }else if(h->cbp&15){
2586 if(transform_bypass){
2587 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
2588 idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
2589 for(i=0; i<16; i+=di){
2590 if(h->non_zero_count_cache[ scan8[i] ]){
2591 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2595 if(IS_8x8DCT(mb_type)){
2596 s->dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2598 s->dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2603 for(i=0; i<16; i++){
2604 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2605 uint8_t * const ptr= dest_y + block_offset[i];
2606 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2612 if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
2613 uint8_t *dest[2] = {dest_cb, dest_cr};
2614 if(transform_bypass){
2615 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
2616 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
2617 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
2619 idct_add = s->dsp.add_pixels4;
2620 for(i=16; i<16+8; i++){
2621 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
2622 idct_add (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2626 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]);
2627 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]);
2629 idct_add = s->dsp.h264_idct_add;
2630 idct_dc_add = s->dsp.h264_idct_dc_add;
2631 for(i=16; i<16+8; i++){
2632 if(h->non_zero_count_cache[ scan8[i] ])
2633 idct_add (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2634 else if(h->mb[i*16])
2635 idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2638 for(i=16; i<16+8; i++){
2639 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2640 uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
2641 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2648 if(h->cbp || IS_INTRA(mb_type))
2649 s->dsp.clear_blocks(h->mb);
2651 if(h->deblocking_filter) {
2652 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, simple);
2653 fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
2654 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2655 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2656 if (!simple && FRAME_MBAFF) {
2657 filter_mb (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2659 filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2665 * Process a macroblock; this case avoids checks for expensive uncommon cases.
2667 static void hl_decode_mb_simple(H264Context *h){
2668 hl_decode_mb_internal(h, 1);
2672 * Process a macroblock; this handles edge cases, such as interlacing.
2674 static void av_noinline hl_decode_mb_complex(H264Context *h){
2675 hl_decode_mb_internal(h, 0);
2678 static void hl_decode_mb(H264Context *h){
2679 MpegEncContext * const s = &h->s;
2680 const int mb_xy= h->mb_xy;
2681 const int mb_type= s->current_picture.mb_type[mb_xy];
2682 int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
2685 hl_decode_mb_complex(h);
2686 else hl_decode_mb_simple(h);
2689 static void pic_as_field(Picture *pic, const int parity){
2691 for (i = 0; i < 4; ++i) {
2692 if (parity == PICT_BOTTOM_FIELD)
2693 pic->data[i] += pic->linesize[i];
2694 pic->reference = parity;
2695 pic->linesize[i] *= 2;
2697 pic->poc= pic->field_poc[parity == PICT_BOTTOM_FIELD];
2700 static int split_field_copy(Picture *dest, Picture *src,
2701 int parity, int id_add){
2702 int match = !!(src->reference & parity);
2706 if(parity != PICT_FRAME){
2707 pic_as_field(dest, parity);
2709 dest->pic_id += id_add;
2716 static int build_def_list(Picture *def, Picture **in, int len, int is_long, int sel){
2720 while(i[0]<len || i[1]<len){
2721 while(i[0]<len && !(in[ i[0] ] && (in[ i[0] ]->reference & sel)))
2723 while(i[1]<len && !(in[ i[1] ] && (in[ i[1] ]->reference & (sel^3))))
2726 in[ i[0] ]->pic_id= is_long ? i[0] : in[ i[0] ]->frame_num;
2727 split_field_copy(&def[index++], in[ i[0]++ ], sel , 1);
2730 in[ i[1] ]->pic_id= is_long ? i[1] : in[ i[1] ]->frame_num;
2731 split_field_copy(&def[index++], in[ i[1]++ ], sel^3, 0);
2738 static int add_sorted(Picture **sorted, Picture **src, int len, int limit, int dir){
2743 best_poc= dir ? INT_MIN : INT_MAX;
2745 for(i=0; i<len; i++){
2746 const int poc= src[i]->poc;
2747 if(((poc > limit) ^ dir) && ((poc < best_poc) ^ dir)){
2749 sorted[out_i]= src[i];
2752 if(best_poc == (dir ? INT_MIN : INT_MAX))
2754 limit= sorted[out_i++]->poc - dir;
2760 * fills the default_ref_list.
2762 static int fill_default_ref_list(H264Context *h){
2763 MpegEncContext * const s = &h->s;
2766 if(h->slice_type_nos==FF_B_TYPE){
2767 Picture *sorted[32];
2772 cur_poc= s->current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
2774 cur_poc= s->current_picture_ptr->poc;
2776 for(list= 0; list<2; list++){
2777 len= add_sorted(sorted , h->short_ref, h->short_ref_count, cur_poc, 1^list);
2778 len+=add_sorted(sorted+len, h->short_ref, h->short_ref_count, cur_poc, 0^list);
2780 len= build_def_list(h->default_ref_list[list] , sorted , len, 0, s->picture_structure);
2781 len+=build_def_list(h->default_ref_list[list]+len, h->long_ref, 16 , 1, s->picture_structure);
2784 if(len < h->ref_count[list])
2785 memset(&h->default_ref_list[list][len], 0, sizeof(Picture)*(h->ref_count[list] - len));
2789 if(lens[0] == lens[1] && lens[1] > 1){
2790 for(i=0; h->default_ref_list[0][i].data[0] == h->default_ref_list[1][i].data[0] && i<lens[0]; i++);
2792 FFSWAP(Picture, h->default_ref_list[1][0], h->default_ref_list[1][1]);
2795 len = build_def_list(h->default_ref_list[0] , h->short_ref, h->short_ref_count, 0, s->picture_structure);
2796 len+= build_def_list(h->default_ref_list[0]+len, h-> long_ref, 16 , 1, s->picture_structure);
2798 if(len < h->ref_count[0])
2799 memset(&h->default_ref_list[0][len], 0, sizeof(Picture)*(h->ref_count[0] - len));
2802 for (i=0; i<h->ref_count[0]; i++) {
2803 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]);
2805 if(h->slice_type_nos==FF_B_TYPE){
2806 for (i=0; i<h->ref_count[1]; i++) {
2807 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]);
2814 static void print_short_term(H264Context *h);
2815 static void print_long_term(H264Context *h);
2818 * Extract structure information about the picture described by pic_num in
2819 * the current decoding context (frame or field). Note that pic_num is
2820 * picture number without wrapping (so, 0<=pic_num<max_pic_num).
2821 * @param pic_num picture number for which to extract structure information
2822 * @param structure one of PICT_XXX describing structure of picture
2824 * @return frame number (short term) or long term index of picture
2825 * described by pic_num
2827 static int pic_num_extract(H264Context *h, int pic_num, int *structure){
2828 MpegEncContext * const s = &h->s;
2830 *structure = s->picture_structure;
2833 /* opposite field */
2834 *structure ^= PICT_FRAME;
2841 static int decode_ref_pic_list_reordering(H264Context *h){
2842 MpegEncContext * const s = &h->s;
2843 int list, index, pic_structure;
2845 print_short_term(h);
2848 for(list=0; list<h->list_count; list++){
2849 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2851 if(get_bits1(&s->gb)){
2852 int pred= h->curr_pic_num;
2854 for(index=0; ; index++){
2855 unsigned int reordering_of_pic_nums_idc= get_ue_golomb_31(&s->gb);
2856 unsigned int pic_id;
2858 Picture *ref = NULL;
2860 if(reordering_of_pic_nums_idc==3)
2863 if(index >= h->ref_count[list]){
2864 av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
2868 if(reordering_of_pic_nums_idc<3){
2869 if(reordering_of_pic_nums_idc<2){
2870 const unsigned int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2873 if(abs_diff_pic_num > h->max_pic_num){
2874 av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
2878 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2879 else pred+= abs_diff_pic_num;
2880 pred &= h->max_pic_num - 1;
2882 frame_num = pic_num_extract(h, pred, &pic_structure);
2884 for(i= h->short_ref_count-1; i>=0; i--){
2885 ref = h->short_ref[i];
2886 assert(ref->reference);
2887 assert(!ref->long_ref);
2889 ref->frame_num == frame_num &&
2890 (ref->reference & pic_structure)
2898 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2900 long_idx= pic_num_extract(h, pic_id, &pic_structure);
2903 av_log(h->s.avctx, AV_LOG_ERROR, "long_term_pic_idx overflow\n");
2906 ref = h->long_ref[long_idx];
2907 assert(!(ref && !ref->reference));
2908 if(ref && (ref->reference & pic_structure)){
2909 ref->pic_id= pic_id;
2910 assert(ref->long_ref);
2918 av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
2919 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2921 for(i=index; i+1<h->ref_count[list]; i++){
2922 if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)
2925 for(; i > index; i--){
2926 h->ref_list[list][i]= h->ref_list[list][i-1];
2928 h->ref_list[list][index]= *ref;
2930 pic_as_field(&h->ref_list[list][index], pic_structure);
2934 av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
2940 for(list=0; list<h->list_count; list++){
2941 for(index= 0; index < h->ref_count[list]; index++){
2942 if(!h->ref_list[list][index].data[0]){
2943 av_log(h->s.avctx, AV_LOG_ERROR, "Missing reference picture\n");
2944 h->ref_list[list][index]= s->current_picture; //FIXME this is not a sensible solution
2952 static void fill_mbaff_ref_list(H264Context *h){
2954 for(list=0; list<2; list++){ //FIXME try list_count
2955 for(i=0; i<h->ref_count[list]; i++){
2956 Picture *frame = &h->ref_list[list][i];
2957 Picture *field = &h->ref_list[list][16+2*i];
2960 field[0].linesize[j] <<= 1;
2961 field[0].reference = PICT_TOP_FIELD;
2962 field[0].poc= field[0].field_poc[0];
2963 field[1] = field[0];
2965 field[1].data[j] += frame->linesize[j];
2966 field[1].reference = PICT_BOTTOM_FIELD;
2967 field[1].poc= field[1].field_poc[1];
2969 h->luma_weight[list][16+2*i] = h->luma_weight[list][16+2*i+1] = h->luma_weight[list][i];
2970 h->luma_offset[list][16+2*i] = h->luma_offset[list][16+2*i+1] = h->luma_offset[list][i];
2972 h->chroma_weight[list][16+2*i][j] = h->chroma_weight[list][16+2*i+1][j] = h->chroma_weight[list][i][j];
2973 h->chroma_offset[list][16+2*i][j] = h->chroma_offset[list][16+2*i+1][j] = h->chroma_offset[list][i][j];
2977 for(j=0; j<h->ref_count[1]; j++){
2978 for(i=0; i<h->ref_count[0]; i++)
2979 h->implicit_weight[j][16+2*i] = h->implicit_weight[j][16+2*i+1] = h->implicit_weight[j][i];
2980 memcpy(h->implicit_weight[16+2*j], h->implicit_weight[j], sizeof(*h->implicit_weight));
2981 memcpy(h->implicit_weight[16+2*j+1], h->implicit_weight[j], sizeof(*h->implicit_weight));
2985 static int pred_weight_table(H264Context *h){
2986 MpegEncContext * const s = &h->s;
2988 int luma_def, chroma_def;
2991 h->use_weight_chroma= 0;
2992 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2993 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2994 luma_def = 1<<h->luma_log2_weight_denom;
2995 chroma_def = 1<<h->chroma_log2_weight_denom;
2997 for(list=0; list<2; list++){
2998 h->luma_weight_flag[list] = 0;
2999 h->chroma_weight_flag[list] = 0;
3000 for(i=0; i<h->ref_count[list]; i++){
3001 int luma_weight_flag, chroma_weight_flag;
3003 luma_weight_flag= get_bits1(&s->gb);
3004 if(luma_weight_flag){
3005 h->luma_weight[list][i]= get_se_golomb(&s->gb);
3006 h->luma_offset[list][i]= get_se_golomb(&s->gb);
3007 if( h->luma_weight[list][i] != luma_def
3008 || h->luma_offset[list][i] != 0) {
3010 h->luma_weight_flag[list]= 1;
3013 h->luma_weight[list][i]= luma_def;
3014 h->luma_offset[list][i]= 0;
3018 chroma_weight_flag= get_bits1(&s->gb);
3019 if(chroma_weight_flag){
3022 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3023 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3024 if( h->chroma_weight[list][i][j] != chroma_def
3025 || h->chroma_offset[list][i][j] != 0) {
3026 h->use_weight_chroma= 1;
3027 h->chroma_weight_flag[list]= 1;
3033 h->chroma_weight[list][i][j]= chroma_def;
3034 h->chroma_offset[list][i][j]= 0;
3039 if(h->slice_type_nos != FF_B_TYPE) break;
3041 h->use_weight= h->use_weight || h->use_weight_chroma;
3045 static void implicit_weight_table(H264Context *h){
3046 MpegEncContext * const s = &h->s;
3048 int cur_poc = s->current_picture_ptr->poc;
3050 for (i = 0; i < 2; i++) {
3051 h->luma_weight_flag[i] = 0;
3052 h->chroma_weight_flag[i] = 0;
3055 if( h->ref_count[0] == 1 && h->ref_count[1] == 1
3056 && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3058 h->use_weight_chroma= 0;
3063 h->use_weight_chroma= 2;
3064 h->luma_log2_weight_denom= 5;
3065 h->chroma_log2_weight_denom= 5;
3067 for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3068 int poc0 = h->ref_list[0][ref0].poc;
3069 for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3070 int poc1 = h->ref_list[1][ref1].poc;
3071 int td = av_clip(poc1 - poc0, -128, 127);
3073 int tb = av_clip(cur_poc - poc0, -128, 127);
3074 int tx = (16384 + (FFABS(td) >> 1)) / td;
3075 int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3076 if(dist_scale_factor < -64 || dist_scale_factor > 128)
3077 h->implicit_weight[ref0][ref1] = 32;
3079 h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3081 h->implicit_weight[ref0][ref1] = 32;
3087 * Mark a picture as no longer needed for reference. The refmask
3088 * argument allows unreferencing of individual fields or the whole frame.
3089 * If the picture becomes entirely unreferenced, but is being held for
3090 * display purposes, it is marked as such.
3091 * @param refmask mask of fields to unreference; the mask is bitwise
3092 * anded with the reference marking of pic
3093 * @return non-zero if pic becomes entirely unreferenced (except possibly
3094 * for display purposes) zero if one of the fields remains in
3097 static inline int unreference_pic(H264Context *h, Picture *pic, int refmask){
3099 if (pic->reference &= refmask) {
3102 for(i = 0; h->delayed_pic[i]; i++)
3103 if(pic == h->delayed_pic[i]){
3104 pic->reference=DELAYED_PIC_REF;
3112 * instantaneous decoder refresh.
3114 static void idr(H264Context *h){
3117 for(i=0; i<16; i++){
3118 remove_long(h, i, 0);
3120 assert(h->long_ref_count==0);
3122 for(i=0; i<h->short_ref_count; i++){
3123 unreference_pic(h, h->short_ref[i], 0);
3124 h->short_ref[i]= NULL;
3126 h->short_ref_count=0;
3127 h->prev_frame_num= 0;
3128 h->prev_frame_num_offset= 0;
3133 /* forget old pics after a seek */
3134 static void flush_dpb(AVCodecContext *avctx){
3135 H264Context *h= avctx->priv_data;
3137 for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
3138 if(h->delayed_pic[i])
3139 h->delayed_pic[i]->reference= 0;
3140 h->delayed_pic[i]= NULL;
3142 h->outputed_poc= INT_MIN;
3144 if(h->s.current_picture_ptr)
3145 h->s.current_picture_ptr->reference= 0;
3146 h->s.first_field= 0;
3147 h->sei_recovery_frame_cnt = -1;
3148 h->sei_dpb_output_delay = 0;
3149 ff_mpeg_flush(avctx);
3153 * Find a Picture in the short term reference list by frame number.
3154 * @param frame_num frame number to search for
3155 * @param idx the index into h->short_ref where returned picture is found
3156 * undefined if no picture found.
3157 * @return pointer to the found picture, or NULL if no pic with the provided
3158 * frame number is found
3160 static Picture * find_short(H264Context *h, int frame_num, int *idx){
3161 MpegEncContext * const s = &h->s;
3164 for(i=0; i<h->short_ref_count; i++){
3165 Picture *pic= h->short_ref[i];
3166 if(s->avctx->debug&FF_DEBUG_MMCO)
3167 av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3168 if(pic->frame_num == frame_num) {
3177 * Remove a picture from the short term reference list by its index in
3178 * that list. This does no checking on the provided index; it is assumed
3179 * to be valid. Other list entries are shifted down.
3180 * @param i index into h->short_ref of picture to remove.
3182 static void remove_short_at_index(H264Context *h, int i){
3183 assert(i >= 0 && i < h->short_ref_count);
3184 h->short_ref[i]= NULL;
3185 if (--h->short_ref_count)
3186 memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i)*sizeof(Picture*));
3191 * @return the removed picture or NULL if an error occurs
3193 static Picture * remove_short(H264Context *h, int frame_num, int ref_mask){
3194 MpegEncContext * const s = &h->s;
3198 if(s->avctx->debug&FF_DEBUG_MMCO)
3199 av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3201 pic = find_short(h, frame_num, &i);
3203 if(unreference_pic(h, pic, ref_mask))
3204 remove_short_at_index(h, i);
3211 * Remove a picture from the long term reference list by its index in
3213 * @return the removed picture or NULL if an error occurs
3215 static Picture * remove_long(H264Context *h, int i, int ref_mask){
3218 pic= h->long_ref[i];
3220 if(unreference_pic(h, pic, ref_mask)){
3221 assert(h->long_ref[i]->long_ref == 1);
3222 h->long_ref[i]->long_ref= 0;
3223 h->long_ref[i]= NULL;
3224 h->long_ref_count--;
3232 * print short term list
3234 static void print_short_term(H264Context *h) {
3236 if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3237 av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3238 for(i=0; i<h->short_ref_count; i++){
3239 Picture *pic= h->short_ref[i];
3240 av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3246 * print long term list
3248 static void print_long_term(H264Context *h) {
3250 if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3251 av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3252 for(i = 0; i < 16; i++){
3253 Picture *pic= h->long_ref[i];
3255 av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3262 * Executes the reference picture marking (memory management control operations).
3264 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3265 MpegEncContext * const s = &h->s;
3267 int current_ref_assigned=0;
3270 if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3271 av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3273 for(i=0; i<mmco_count; i++){
3274 int structure, frame_num;
3275 if(s->avctx->debug&FF_DEBUG_MMCO)
3276 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);
3278 if( mmco[i].opcode == MMCO_SHORT2UNUSED
3279 || mmco[i].opcode == MMCO_SHORT2LONG){
3280 frame_num = pic_num_extract(h, mmco[i].short_pic_num, &structure);
3281 pic = find_short(h, frame_num, &j);
3283 if(mmco[i].opcode != MMCO_SHORT2LONG || !h->long_ref[mmco[i].long_arg]
3284 || h->long_ref[mmco[i].long_arg]->frame_num != frame_num)
3285 av_log(h->s.avctx, AV_LOG_ERROR, "mmco: unref short failure\n");
3290 switch(mmco[i].opcode){
3291 case MMCO_SHORT2UNUSED:
3292 if(s->avctx->debug&FF_DEBUG_MMCO)
3293 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);
3294 remove_short(h, frame_num, structure ^ PICT_FRAME);
3296 case MMCO_SHORT2LONG:
3297 if (h->long_ref[mmco[i].long_arg] != pic)
3298 remove_long(h, mmco[i].long_arg, 0);
3300 remove_short_at_index(h, j);
3301 h->long_ref[ mmco[i].long_arg ]= pic;
3302 if (h->long_ref[ mmco[i].long_arg ]){
3303 h->long_ref[ mmco[i].long_arg ]->long_ref=1;
3304 h->long_ref_count++;
3307 case MMCO_LONG2UNUSED:
3308 j = pic_num_extract(h, mmco[i].long_arg, &structure);
3309 pic = h->long_ref[j];
3311 remove_long(h, j, structure ^ PICT_FRAME);
3312 } else if(s->avctx->debug&FF_DEBUG_MMCO)
3313 av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: unref long failure\n");
3316 // Comment below left from previous code as it is an interresting note.
3317 /* First field in pair is in short term list or
3318 * at a different long term index.
3319 * This is not allowed; see 7.4.3.3, notes 2 and 3.
3320 * Report the problem and keep the pair where it is,
3321 * and mark this field valid.
3324 if (h->long_ref[mmco[i].long_arg] != s->current_picture_ptr) {
3325 remove_long(h, mmco[i].long_arg, 0);
3327 h->long_ref[ mmco[i].long_arg ]= s->current_picture_ptr;
3328 h->long_ref[ mmco[i].long_arg ]->long_ref=1;
3329 h->long_ref_count++;
3332 s->current_picture_ptr->reference |= s->picture_structure;
3333 current_ref_assigned=1;
3335 case MMCO_SET_MAX_LONG:
3336 assert(mmco[i].long_arg <= 16);
3337 // just remove the long term which index is greater than new max
3338 for(j = mmco[i].long_arg; j<16; j++){
3339 remove_long(h, j, 0);
3343 while(h->short_ref_count){
3344 remove_short(h, h->short_ref[0]->frame_num, 0);
3346 for(j = 0; j < 16; j++) {
3347 remove_long(h, j, 0);
3349 s->current_picture_ptr->poc=
3350 s->current_picture_ptr->field_poc[0]=
3351 s->current_picture_ptr->field_poc[1]=
3355 s->current_picture_ptr->frame_num= 0;
3361 if (!current_ref_assigned) {
3362 /* Second field of complementary field pair; the first field of
3363 * which is already referenced. If short referenced, it
3364 * should be first entry in short_ref. If not, it must exist
3365 * in long_ref; trying to put it on the short list here is an
3366 * error in the encoded bit stream (ref: 7.4.3.3, NOTE 2 and 3).
3368 if (h->short_ref_count && h->short_ref[0] == s->current_picture_ptr) {
3369 /* Just mark the second field valid */
3370 s->current_picture_ptr->reference = PICT_FRAME;
3371 } else if (s->current_picture_ptr->long_ref) {
3372 av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term reference "
3373 "assignment for second field "
3374 "in complementary field pair "
3375 "(first field is long term)\n");
3377 pic= remove_short(h, s->current_picture_ptr->frame_num, 0);
3379 av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3382 if(h->short_ref_count)
3383 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3385 h->short_ref[0]= s->current_picture_ptr;
3386 h->short_ref_count++;
3387 s->current_picture_ptr->reference |= s->picture_structure;
3391 if (h->long_ref_count + h->short_ref_count > h->sps.ref_frame_count){
3393 /* We have too many reference frames, probably due to corrupted
3394 * stream. Need to discard one frame. Prevents overrun of the
3395 * short_ref and long_ref buffers.
3397 av_log(h->s.avctx, AV_LOG_ERROR,
3398 "number of reference frames exceeds max (probably "
3399 "corrupt input), discarding one\n");
3401 if (h->long_ref_count && !h->short_ref_count) {
3402 for (i = 0; i < 16; ++i)
3407 remove_long(h, i, 0);
3409 pic = h->short_ref[h->short_ref_count - 1];
3410 remove_short(h, pic->frame_num, 0);
3414 print_short_term(h);
3419 static int decode_ref_pic_marking(H264Context *h, GetBitContext *gb){
3420 MpegEncContext * const s = &h->s;
3424 if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3425 s->broken_link= get_bits1(gb) -1;
3427 h->mmco[0].opcode= MMCO_LONG;
<