c3e26c5ce3a3eb6988d097f89ea3adb31940d175
[ffmpeg.git] / libavcodec / h264.c
1 /*
2  * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4  *
5  * This file is part of FFmpeg.
6  *
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.
11  *
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.
16  *
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
20  */
21
22 /**
23  * @file h264.c
24  * H.264 / AVC / MPEG4 part10 codec.
25  * @author Michael Niedermayer <michaelni@gmx.at>
26  */
27
28 #include "dsputil.h"
29 #include "avcodec.h"
30 #include "mpegvideo.h"
31 #include "h264.h"
32 #include "h264data.h"
33 #include "h264_parser.h"
34 #include "golomb.h"
35 #include "rectangle.h"
36
37 #include "cabac.h"
38 #ifdef ARCH_X86
39 #include "i386/h264_i386.h"
40 #endif
41
42 //#undef NDEBUG
43 #include <assert.h>
44
45 /**
46  * Value of Picture.reference when Picture is not a reference picture, but
47  * is held for delayed output.
48  */
49 #define DELAYED_PIC_REF 4
50
51 static VLC coeff_token_vlc[4];
52 static VLC_TYPE coeff_token_vlc_tables[520+332+280+256][2];
53 static const int coeff_token_vlc_tables_size[4]={520,332,280,256};
54
55 static VLC chroma_dc_coeff_token_vlc;
56 static VLC_TYPE chroma_dc_coeff_token_vlc_table[256][2];
57 static const int chroma_dc_coeff_token_vlc_table_size = 256;
58
59 static VLC total_zeros_vlc[15];
60 static VLC_TYPE total_zeros_vlc_tables[15][512][2];
61 static const int total_zeros_vlc_tables_size = 512;
62
63 static VLC chroma_dc_total_zeros_vlc[3];
64 static VLC_TYPE chroma_dc_total_zeros_vlc_tables[3][8][2];
65 static const int chroma_dc_total_zeros_vlc_tables_size = 8;
66
67 static VLC run_vlc[6];
68 static VLC_TYPE run_vlc_tables[6][8][2];
69 static const int run_vlc_tables_size = 8;
70
71 static VLC run7_vlc;
72 static VLC_TYPE run7_vlc_table[96][2];
73 static const int run7_vlc_table_size = 96;
74
75 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
76 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
77 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);
78 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);
79 static Picture * remove_long(H264Context *h, int i, int ref_mask);
80
81 static av_always_inline uint32_t pack16to32(int a, int b){
82 #ifdef WORDS_BIGENDIAN
83    return (b&0xFFFF) + (a<<16);
84 #else
85    return (a&0xFFFF) + (b<<16);
86 #endif
87 }
88
89 static const uint8_t rem6[52]={
90 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,
91 };
92
93 static const uint8_t div6[52]={
94 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,
95 };
96
97 static const int left_block_options[4][8]={
98     {0,1,2,3,7,10,8,11},
99     {2,2,3,3,8,11,8,11},
100     {0,0,1,1,7,10,7,10},
101     {0,2,0,2,7,10,7,10}
102 };
103
104 static void fill_caches(H264Context *h, int mb_type, int for_deblock){
105     MpegEncContext * const s = &h->s;
106     const int mb_xy= h->mb_xy;
107     int topleft_xy, top_xy, topright_xy, left_xy[2];
108     int topleft_type, top_type, topright_type, left_type[2];
109     const int * left_block;
110     int topleft_partition= -1;
111     int i;
112
113     top_xy     = mb_xy  - (s->mb_stride << FIELD_PICTURE);
114
115     //FIXME deblocking could skip the intra and nnz parts.
116     if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[top_xy]) && !FRAME_MBAFF)
117         return;
118
119     /* Wow, what a mess, why didn't they simplify the interlacing & intra
120      * stuff, I can't imagine that these complex rules are worth it. */
121
122     topleft_xy = top_xy - 1;
123     topright_xy= top_xy + 1;
124     left_xy[1] = left_xy[0] = mb_xy-1;
125     left_block = left_block_options[0];
126     if(FRAME_MBAFF){
127         const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
128         const int top_pair_xy      = pair_xy     - s->mb_stride;
129         const int topleft_pair_xy  = top_pair_xy - 1;
130         const int topright_pair_xy = top_pair_xy + 1;
131         const int topleft_mb_frame_flag  = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
132         const int top_mb_frame_flag      = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
133         const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
134         const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
135         const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
136         const int bottom = (s->mb_y & 1);
137         tprintf(s->avctx, "fill_caches: curr_mb_frame_flag:%d, left_mb_frame_flag:%d, topleft_mb_frame_flag:%d, top_mb_frame_flag:%d, topright_mb_frame_flag:%d\n", curr_mb_frame_flag, left_mb_frame_flag, topleft_mb_frame_flag, top_mb_frame_flag, topright_mb_frame_flag);
138         if (bottom
139                 ? !curr_mb_frame_flag // bottom macroblock
140                 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
141                 ) {
142             top_xy -= s->mb_stride;
143         }
144         if (bottom
145                 ? !curr_mb_frame_flag // bottom macroblock
146                 : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
147                 ) {
148             topleft_xy -= s->mb_stride;
149         } else if(bottom && curr_mb_frame_flag && !left_mb_frame_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;
153         }
154         if (bottom
155                 ? !curr_mb_frame_flag // bottom macroblock
156                 : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
157                 ) {
158             topright_xy -= s->mb_stride;
159         }
160         if (left_mb_frame_flag != curr_mb_frame_flag) {
161             left_xy[1] = left_xy[0] = pair_xy - 1;
162             if (curr_mb_frame_flag) {
163                 if (bottom) {
164                     left_block = left_block_options[1];
165                 } else {
166                     left_block= left_block_options[2];
167                 }
168             } else {
169                 left_xy[1] += s->mb_stride;
170                 left_block = left_block_options[3];
171             }
172         }
173     }
174
175     h->top_mb_xy = top_xy;
176     h->left_mb_xy[0] = left_xy[0];
177     h->left_mb_xy[1] = left_xy[1];
178     if(for_deblock){
179         topleft_type = 0;
180         topright_type = 0;
181         top_type     = h->slice_table[top_xy     ] < 0xFFFF ? s->current_picture.mb_type[top_xy]     : 0;
182         left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0;
183         left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0;
184
185         if(MB_MBAFF && !IS_INTRA(mb_type)){
186             int list;
187             for(list=0; list<h->list_count; list++){
188                 //These values where changed for ease of performing MC, we need to change them back
189                 //FIXME maybe we can make MC and loop filter use the same values or prevent
190                 //the MC code from changing ref_cache and rather use a temporary array.
191                 if(USES_LIST(mb_type,list)){
192                     int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
193                     *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
194                     *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
195                     ref += h->b8_stride;
196                     *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
197                     *(uint32_t*)&h->ref_cache[list][scan8[10]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
198                 }
199             }
200         }
201     }else{
202         topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
203         top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
204         topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
205         left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
206         left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
207
208     if(IS_INTRA(mb_type)){
209         int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
210         h->topleft_samples_available=
211         h->top_samples_available=
212         h->left_samples_available= 0xFFFF;
213         h->topright_samples_available= 0xEEEA;
214
215         if(!(top_type & type_mask)){
216             h->topleft_samples_available= 0xB3FF;
217             h->top_samples_available= 0x33FF;
218             h->topright_samples_available= 0x26EA;
219         }
220         if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
221             if(IS_INTERLACED(mb_type)){
222                 if(!(left_type[0] & type_mask)){
223                     h->topleft_samples_available&= 0xDFFF;
224                     h->left_samples_available&= 0x5FFF;
225                 }
226                 if(!(left_type[1] & type_mask)){
227                     h->topleft_samples_available&= 0xFF5F;
228                     h->left_samples_available&= 0xFF5F;
229                 }
230             }else{
231                 int left_typei = h->slice_table[left_xy[0] + s->mb_stride ] == h->slice_num
232                                 ? s->current_picture.mb_type[left_xy[0] + s->mb_stride] : 0;
233                 assert(left_xy[0] == left_xy[1]);
234                 if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
235                     h->topleft_samples_available&= 0xDF5F;
236                     h->left_samples_available&= 0x5F5F;
237                 }
238             }
239         }else{
240             if(!(left_type[0] & type_mask)){
241                 h->topleft_samples_available&= 0xDF5F;
242                 h->left_samples_available&= 0x5F5F;
243             }
244         }
245
246         if(!(topleft_type & type_mask))
247             h->topleft_samples_available&= 0x7FFF;
248
249         if(!(topright_type & type_mask))
250             h->topright_samples_available&= 0xFBFF;
251
252         if(IS_INTRA4x4(mb_type)){
253             if(IS_INTRA4x4(top_type)){
254                 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
255                 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
256                 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
257                 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
258             }else{
259                 int pred;
260                 if(!(top_type & type_mask))
261                     pred= -1;
262                 else{
263                     pred= 2;
264                 }
265                 h->intra4x4_pred_mode_cache[4+8*0]=
266                 h->intra4x4_pred_mode_cache[5+8*0]=
267                 h->intra4x4_pred_mode_cache[6+8*0]=
268                 h->intra4x4_pred_mode_cache[7+8*0]= pred;
269             }
270             for(i=0; i<2; i++){
271                 if(IS_INTRA4x4(left_type[i])){
272                     h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
273                     h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
274                 }else{
275                     int pred;
276                     if(!(left_type[i] & type_mask))
277                         pred= -1;
278                     else{
279                         pred= 2;
280                     }
281                     h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
282                     h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
283                 }
284             }
285         }
286     }
287     }
288
289
290 /*
291 0 . T T. T T T T
292 1 L . .L . . . .
293 2 L . .L . . . .
294 3 . T TL . . . .
295 4 L . .L . . . .
296 5 L . .. . . . .
297 */
298 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
299     if(top_type){
300         h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
301         h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
302         h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
303         h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
304
305         h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
306         h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
307
308         h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
309         h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
310
311     }else{
312         h->non_zero_count_cache[4+8*0]=
313         h->non_zero_count_cache[5+8*0]=
314         h->non_zero_count_cache[6+8*0]=
315         h->non_zero_count_cache[7+8*0]=
316
317         h->non_zero_count_cache[1+8*0]=
318         h->non_zero_count_cache[2+8*0]=
319
320         h->non_zero_count_cache[1+8*3]=
321         h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
322
323     }
324
325     for (i=0; i<2; i++) {
326         if(left_type[i]){
327             h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
328             h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
329             h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
330             h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
331         }else{
332             h->non_zero_count_cache[3+8*1 + 2*8*i]=
333             h->non_zero_count_cache[3+8*2 + 2*8*i]=
334             h->non_zero_count_cache[0+8*1 +   8*i]=
335             h->non_zero_count_cache[0+8*4 +   8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
336         }
337     }
338
339     if( h->pps.cabac ) {
340         // top_cbp
341         if(top_type) {
342             h->top_cbp = h->cbp_table[top_xy];
343         } else if(IS_INTRA(mb_type)) {
344             h->top_cbp = 0x1C0;
345         } else {
346             h->top_cbp = 0;
347         }
348         // left_cbp
349         if (left_type[0]) {
350             h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
351         } else if(IS_INTRA(mb_type)) {
352             h->left_cbp = 0x1C0;
353         } else {
354             h->left_cbp = 0;
355         }
356         if (left_type[0]) {
357             h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
358         }
359         if (left_type[1]) {
360             h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
361         }
362     }
363
364 #if 1
365     if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
366         int list;
367         for(list=0; list<h->list_count; list++){
368             if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
369                 /*if(!h->mv_cache_clean[list]){
370                     memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
371                     memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
372                     h->mv_cache_clean[list]= 1;
373                 }*/
374                 continue;
375             }
376             h->mv_cache_clean[list]= 0;
377
378             if(USES_LIST(top_type, list)){
379                 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
380                 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
381                 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
382                 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
383                 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
384                 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
385                 h->ref_cache[list][scan8[0] + 0 - 1*8]=
386                 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
387                 h->ref_cache[list][scan8[0] + 2 - 1*8]=
388                 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
389             }else{
390                 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
391                 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
392                 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
393                 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
394                 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
395             }
396
397             for(i=0; i<2; i++){
398                 int cache_idx = scan8[0] - 1 + i*2*8;
399                 if(USES_LIST(left_type[i], list)){
400                     const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
401                     const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
402                     *(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]];
403                     *(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]];
404                     h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
405                     h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
406                 }else{
407                     *(uint32_t*)h->mv_cache [list][cache_idx  ]=
408                     *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
409                     h->ref_cache[list][cache_idx  ]=
410                     h->ref_cache[list][cache_idx+8]= left_type[i] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
411                 }
412             }
413
414             if(for_deblock || ((IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred) && !FRAME_MBAFF))
415                 continue;
416
417             if(USES_LIST(topleft_type, list)){
418                 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + h->b_stride + (topleft_partition & 2*h->b_stride);
419                 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + (topleft_partition & h->b8_stride);
420                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
421                 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
422             }else{
423                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
424                 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
425             }
426
427             if(USES_LIST(topright_type, list)){
428                 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
429                 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
430                 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
431                 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
432             }else{
433                 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
434                 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
435             }
436
437             if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
438                 continue;
439
440             h->ref_cache[list][scan8[5 ]+1] =
441             h->ref_cache[list][scan8[7 ]+1] =
442             h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
443             h->ref_cache[list][scan8[4 ]] =
444             h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
445             *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
446             *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
447             *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
448             *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
449             *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
450
451             if( h->pps.cabac ) {
452                 /* XXX beurk, Load mvd */
453                 if(USES_LIST(top_type, list)){
454                     const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
455                     *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
456                     *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
457                     *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
458                     *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
459                 }else{
460                     *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
461                     *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
462                     *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
463                     *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
464                 }
465                 if(USES_LIST(left_type[0], list)){
466                     const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
467                     *(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]];
468                     *(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]];
469                 }else{
470                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
471                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
472                 }
473                 if(USES_LIST(left_type[1], list)){
474                     const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
475                     *(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]];
476                     *(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]];
477                 }else{
478                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
479                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
480                 }
481                 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
482                 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
483                 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
484                 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
485                 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
486
487                 if(h->slice_type_nos == FF_B_TYPE){
488                     fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
489
490                     if(IS_DIRECT(top_type)){
491                         *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
492                     }else if(IS_8X8(top_type)){
493                         int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
494                         h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
495                         h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
496                     }else{
497                         *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
498                     }
499
500                     if(IS_DIRECT(left_type[0]))
501                         h->direct_cache[scan8[0] - 1 + 0*8]= 1;
502                     else if(IS_8X8(left_type[0]))
503                         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)];
504                     else
505                         h->direct_cache[scan8[0] - 1 + 0*8]= 0;
506
507                     if(IS_DIRECT(left_type[1]))
508                         h->direct_cache[scan8[0] - 1 + 2*8]= 1;
509                     else if(IS_8X8(left_type[1]))
510                         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)];
511                     else
512                         h->direct_cache[scan8[0] - 1 + 2*8]= 0;
513                 }
514             }
515
516             if(FRAME_MBAFF){
517 #define MAP_MVS\
518                     MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
519                     MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
520                     MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
521                     MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
522                     MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
523                     MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
524                     MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
525                     MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
526                     MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
527                     MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
528                 if(MB_FIELD){
529 #define MAP_F2F(idx, mb_type)\
530                     if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
531                         h->ref_cache[list][idx] <<= 1;\
532                         h->mv_cache[list][idx][1] /= 2;\
533                         h->mvd_cache[list][idx][1] /= 2;\
534                     }
535                     MAP_MVS
536 #undef MAP_F2F
537                 }else{
538 #define MAP_F2F(idx, mb_type)\
539                     if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
540                         h->ref_cache[list][idx] >>= 1;\
541                         h->mv_cache[list][idx][1] <<= 1;\
542                         h->mvd_cache[list][idx][1] <<= 1;\
543                     }
544                     MAP_MVS
545 #undef MAP_F2F
546                 }
547             }
548         }
549     }
550 #endif
551
552     h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
553 }
554
555 static inline void write_back_intra_pred_mode(H264Context *h){
556     const int mb_xy= h->mb_xy;
557
558     h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
559     h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
560     h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
561     h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
562     h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
563     h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
564     h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
565 }
566
567 /**
568  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
569  */
570 static inline int check_intra4x4_pred_mode(H264Context *h){
571     MpegEncContext * const s = &h->s;
572     static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
573     static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
574     int i;
575
576     if(!(h->top_samples_available&0x8000)){
577         for(i=0; i<4; i++){
578             int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
579             if(status<0){
580                 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);
581                 return -1;
582             } else if(status){
583                 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
584             }
585         }
586     }
587
588     if((h->left_samples_available&0x8888)!=0x8888){
589         static const int mask[4]={0x8000,0x2000,0x80,0x20};
590         for(i=0; i<4; i++){
591             if(!(h->left_samples_available&mask[i])){
592             int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
593             if(status<0){
594                 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);
595                 return -1;
596             } else if(status){
597                 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
598             }
599             }
600         }
601     }
602
603     return 0;
604 } //FIXME cleanup like next
605
606 /**
607  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
608  */
609 static inline int check_intra_pred_mode(H264Context *h, int mode){
610     MpegEncContext * const s = &h->s;
611     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
612     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
613
614     if(mode > 6U) {
615         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);
616         return -1;
617     }
618
619     if(!(h->top_samples_available&0x8000)){
620         mode= top[ mode ];
621         if(mode<0){
622             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);
623             return -1;
624         }
625     }
626
627     if((h->left_samples_available&0x8080) != 0x8080){
628         mode= left[ mode ];
629         if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
630             mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
631         }
632         if(mode<0){
633             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);
634             return -1;
635         }
636     }
637
638     return mode;
639 }
640
641 /**
642  * gets the predicted intra4x4 prediction mode.
643  */
644 static inline int pred_intra_mode(H264Context *h, int n){
645     const int index8= scan8[n];
646     const int left= h->intra4x4_pred_mode_cache[index8 - 1];
647     const int top = h->intra4x4_pred_mode_cache[index8 - 8];
648     const int min= FFMIN(left, top);
649
650     tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
651
652     if(min<0) return DC_PRED;
653     else      return min;
654 }
655
656 static inline void write_back_non_zero_count(H264Context *h){
657     const int mb_xy= h->mb_xy;
658
659     h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
660     h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
661     h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
662     h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
663     h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
664     h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
665     h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
666
667     h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
668     h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
669     h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
670
671     h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
672     h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
673     h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
674 }
675
676 /**
677  * gets the predicted number of non-zero coefficients.
678  * @param n block index
679  */
680 static inline int pred_non_zero_count(H264Context *h, int n){
681     const int index8= scan8[n];
682     const int left= h->non_zero_count_cache[index8 - 1];
683     const int top = h->non_zero_count_cache[index8 - 8];
684     int i= left + top;
685
686     if(i<64) i= (i+1)>>1;
687
688     tprintf(h->s.avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
689
690     return i&31;
691 }
692
693 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
694     const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
695     MpegEncContext *s = &h->s;
696
697     /* there is no consistent mapping of mvs to neighboring locations that will
698      * make mbaff happy, so we can't move all this logic to fill_caches */
699     if(FRAME_MBAFF){
700         const uint32_t *mb_types = s->current_picture_ptr->mb_type;
701         const int16_t *mv;
702         *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
703         *C = h->mv_cache[list][scan8[0]-2];
704
705         if(!MB_FIELD
706            && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
707             int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
708             if(IS_INTERLACED(mb_types[topright_xy])){
709 #define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
710                 const int x4 = X4, y4 = Y4;\
711                 const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
712                 if(!USES_LIST(mb_type,list))\
713                     return LIST_NOT_USED;\
714                 mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
715                 h->mv_cache[list][scan8[0]-2][0] = mv[0];\
716                 h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
717                 return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;
718
719                 SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
720             }
721         }
722         if(topright_ref == PART_NOT_AVAILABLE
723            && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
724            && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
725             if(!MB_FIELD
726                && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
727                 SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
728             }
729             if(MB_FIELD
730                && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
731                && i >= scan8[0]+8){
732                 // left shift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's OK.
733                 SET_DIAG_MV(/2, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
734             }
735         }
736 #undef SET_DIAG_MV
737     }
738
739     if(topright_ref != PART_NOT_AVAILABLE){
740         *C= h->mv_cache[list][ i - 8 + part_width ];
741         return topright_ref;
742     }else{
743         tprintf(s->avctx, "topright MV not available\n");
744
745         *C= h->mv_cache[list][ i - 8 - 1 ];
746         return h->ref_cache[list][ i - 8 - 1 ];
747     }
748 }
749
750 /**
751  * gets the predicted MV.
752  * @param n the block index
753  * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
754  * @param mx the x component of the predicted motion vector
755  * @param my the y component of the predicted motion vector
756  */
757 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
758     const int index8= scan8[n];
759     const int top_ref=      h->ref_cache[list][ index8 - 8 ];
760     const int left_ref=     h->ref_cache[list][ index8 - 1 ];
761     const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
762     const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
763     const int16_t * C;
764     int diagonal_ref, match_count;
765
766     assert(part_width==1 || part_width==2 || part_width==4);
767
768 /* mv_cache
769   B . . A T T T T
770   U . . L . . , .
771   U . . L . . . .
772   U . . L . . , .
773   . . . L . . . .
774 */
775
776     diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
777     match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
778     tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count);
779     if(match_count > 1){ //most common
780         *mx= mid_pred(A[0], B[0], C[0]);
781         *my= mid_pred(A[1], B[1], C[1]);
782     }else if(match_count==1){
783         if(left_ref==ref){
784             *mx= A[0];
785             *my= A[1];
786         }else if(top_ref==ref){
787             *mx= B[0];
788             *my= B[1];
789         }else{
790             *mx= C[0];
791             *my= C[1];
792         }
793     }else{
794         if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
795             *mx= A[0];
796             *my= A[1];
797         }else{
798             *mx= mid_pred(A[0], B[0], C[0]);
799             *my= mid_pred(A[1], B[1], C[1]);
800         }
801     }
802
803     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);
804 }
805
806 /**
807  * gets the directionally predicted 16x8 MV.
808  * @param n the block index
809  * @param mx the x component of the predicted motion vector
810  * @param my the y component of the predicted motion vector
811  */
812 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
813     if(n==0){
814         const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
815         const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
816
817         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
819         if(top_ref == ref){
820             *mx= B[0];
821             *my= B[1];
822             return;
823         }
824     }else{
825         const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
826         const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
827
828         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);
829
830         if(left_ref == ref){
831             *mx= A[0];
832             *my= A[1];
833             return;
834         }
835     }
836
837     //RARE
838     pred_motion(h, n, 4, list, ref, mx, my);
839 }
840
841 /**
842  * gets the directionally predicted 8x16 MV.
843  * @param n the block index
844  * @param mx the x component of the predicted motion vector
845  * @param my the y component of the predicted motion vector
846  */
847 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
848     if(n==0){
849         const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
850         const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
851
852         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);
853
854         if(left_ref == ref){
855             *mx= A[0];
856             *my= A[1];
857             return;
858         }
859     }else{
860         const int16_t * C;
861         int diagonal_ref;
862
863         diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
864
865         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);
866
867         if(diagonal_ref == ref){
868             *mx= C[0];
869             *my= C[1];
870             return;
871         }
872     }
873
874     //RARE
875     pred_motion(h, n, 2, list, ref, mx, my);
876 }
877
878 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
879     const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
880     const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
881
882     tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
883
884     if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
885        || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
886        || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
887
888         *mx = *my = 0;
889         return;
890     }
891
892     pred_motion(h, 0, 4, 0, 0, mx, my);
893
894     return;
895 }
896
897 static int get_scale_factor(H264Context * const h, int poc, int poc1, int i){
898     int poc0 = h->ref_list[0][i].poc;
899     int td = av_clip(poc1 - poc0, -128, 127);
900     if(td == 0 || h->ref_list[0][i].long_ref){
901         return 256;
902     }else{
903         int tb = av_clip(poc - poc0, -128, 127);
904         int tx = (16384 + (FFABS(td) >> 1)) / td;
905         return av_clip((tb*tx + 32) >> 6, -1024, 1023);
906     }
907 }
908
909 static inline void direct_dist_scale_factor(H264Context * const h){
910     MpegEncContext * const s = &h->s;
911     const int poc = h->s.current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
912     const int poc1 = h->ref_list[1][0].poc;
913     int i, field;
914     for(field=0; field<2; field++){
915         const int poc  = h->s.current_picture_ptr->field_poc[field];
916         const int poc1 = h->ref_list[1][0].field_poc[field];
917         for(i=0; i < 2*h->ref_count[0]; i++)
918             h->dist_scale_factor_field[field][i^field] = get_scale_factor(h, poc, poc1, i+16);
919     }
920
921     for(i=0; i<h->ref_count[0]; i++){
922         h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i);
923     }
924 }
925
926 static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field, int colfield, int mbafi){
927     MpegEncContext * const s = &h->s;
928     Picture * const ref1 = &h->ref_list[1][0];
929     int j, old_ref, rfield;
930     int start= mbafi ? 16                      : 0;
931     int end  = mbafi ? 16+2*h->ref_count[list] : h->ref_count[list];
932     int interl= mbafi || s->picture_structure != PICT_FRAME;
933
934     /* bogus; fills in for missing frames */
935     memset(map[list], 0, sizeof(map[list]));
936
937     for(rfield=0; rfield<2; rfield++){
938         for(old_ref=0; old_ref<ref1->ref_count[colfield][list]; old_ref++){
939             int poc = ref1->ref_poc[colfield][list][old_ref];
940
941             if     (!interl)
942                 poc |= 3;
943             else if( interl && (poc&3) == 3) //FIXME store all MBAFF references so this isnt needed
944                 poc= (poc&~3) + rfield + 1;
945
946             for(j=start; j<end; j++){
947                 if(4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3) == poc){
948                     int cur_ref= mbafi ? (j-16)^field : j;
949                     map[list][2*old_ref + (rfield^field) + 16] = cur_ref;
950                     if(rfield == field)
951                         map[list][old_ref] = cur_ref;
952                     break;
953                 }
954             }
955         }
956     }
957 }
958
959 static inline void direct_ref_list_init(H264Context * const h){
960     MpegEncContext * const s = &h->s;
961     Picture * const ref1 = &h->ref_list[1][0];
962     Picture * const cur = s->current_picture_ptr;
963     int list, j, field;
964     int sidx= (s->picture_structure&1)^1;
965     int ref1sidx= (ref1->reference&1)^1;
966
967     for(list=0; list<2; list++){
968         cur->ref_count[sidx][list] = h->ref_count[list];
969         for(j=0; j<h->ref_count[list]; j++)
970             cur->ref_poc[sidx][list][j] = 4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3);
971     }
972
973     if(s->picture_structure == PICT_FRAME){
974         memcpy(cur->ref_count[1], cur->ref_count[0], sizeof(cur->ref_count[0]));
975         memcpy(cur->ref_poc  [1], cur->ref_poc  [0], sizeof(cur->ref_poc  [0]));
976     }
977
978     cur->mbaff= FRAME_MBAFF;
979
980     if(cur->pict_type != FF_B_TYPE || h->direct_spatial_mv_pred)
981         return;
982
983     for(list=0; list<2; list++){
984         fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0);
985         for(field=0; field<2; field++)
986             fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1);
987     }
988 }
989
990 static inline void pred_direct_motion(H264Context * const h, int *mb_type){
991     MpegEncContext * const s = &h->s;
992     int b8_stride = h->b8_stride;
993     int b4_stride = h->b_stride;
994     int mb_xy = h->mb_xy;
995     int mb_type_col[2];
996     const int16_t (*l1mv0)[2], (*l1mv1)[2];
997     const int8_t *l1ref0, *l1ref1;
998     const int is_b8x8 = IS_8X8(*mb_type);
999     unsigned int sub_mb_type;
1000     int i8, i4;
1001
1002 #define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
1003
1004     if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL
1005         if(!IS_INTERLACED(*mb_type)){                    //     AFR/FR    -> AFL/FL
1006             int cur_poc = s->current_picture_ptr->poc;
1007             int *col_poc = h->ref_list[1]->field_poc;
1008             int col_parity = FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc);
1009             mb_xy= s->mb_x + ((s->mb_y&~1) + col_parity)*s->mb_stride;
1010             b8_stride = 0;
1011         }else if(!(s->picture_structure & h->ref_list[1][0].reference) && !h->ref_list[1][0].mbaff){// FL -> FL & differ parity
1012             int fieldoff= 2*(h->ref_list[1][0].reference)-3;
1013             mb_xy += s->mb_stride*fieldoff;
1014         }
1015         goto single_col;
1016     }else{                                               // AFL/AFR/FR/FL -> AFR/FR
1017         if(IS_INTERLACED(*mb_type)){                     // AFL       /FL -> AFR/FR
1018             mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
1019             mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
1020             mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
1021             b8_stride *= 3;
1022             b4_stride *= 6;
1023             //FIXME IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag
1024             if(    (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
1025                 && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
1026                 && !is_b8x8){
1027                 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1028                 *mb_type   |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */
1029             }else{
1030                 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1031                 *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
1032             }
1033         }else{                                           //     AFR/FR    -> AFR/FR
1034 single_col:
1035             mb_type_col[0] =
1036             mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
1037             if(IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag){
1038                 /* FIXME save sub mb types from previous frames (or derive from MVs)
1039                 * so we know exactly what block size to use */
1040                 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1041                 *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
1042             }else if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
1043                 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1044                 *mb_type   |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1045             }else{
1046                 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1047                 *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
1048             }
1049         }
1050     }
1051
1052     l1mv0  = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
1053     l1mv1  = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
1054     l1ref0 = &h->ref_list[1][0].ref_index [0][h->mb2b8_xy[mb_xy]];
1055     l1ref1 = &h->ref_list[1][0].ref_index [1][h->mb2b8_xy[mb_xy]];
1056     if(!b8_stride){
1057         if(s->mb_y&1){
1058             l1ref0 += h->b8_stride;
1059             l1ref1 += h->b8_stride;
1060             l1mv0  +=  2*b4_stride;
1061             l1mv1  +=  2*b4_stride;
1062         }
1063     }
1064
1065     if(h->direct_spatial_mv_pred){
1066         int ref[2];
1067         int mv[2][2];
1068         int list;
1069
1070         /* FIXME interlacing + spatial direct uses wrong colocated block positions */
1071
1072         /* ref = min(neighbors) */
1073         for(list=0; list<2; list++){
1074             int refa = h->ref_cache[list][scan8[0] - 1];
1075             int refb = h->ref_cache[list][scan8[0] - 8];
1076             int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1077             if(refc == PART_NOT_AVAILABLE)
1078                 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1079             ref[list] = FFMIN3((unsigned)refa, (unsigned)refb, (unsigned)refc);
1080             if(ref[list] < 0)
1081                 ref[list] = -1;
1082         }
1083
1084         if(ref[0] < 0 && ref[1] < 0){
1085             ref[0] = ref[1] = 0;
1086             mv[0][0] = mv[0][1] =
1087             mv[1][0] = mv[1][1] = 0;
1088         }else{
1089             for(list=0; list<2; list++){
1090                 if(ref[list] >= 0)
1091                     pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1092                 else
1093                     mv[list][0] = mv[list][1] = 0;
1094             }
1095         }
1096
1097         if(ref[1] < 0){
1098             if(!is_b8x8)
1099                 *mb_type &= ~MB_TYPE_L1;
1100             sub_mb_type &= ~MB_TYPE_L1;
1101         }else if(ref[0] < 0){
1102             if(!is_b8x8)
1103                 *mb_type &= ~MB_TYPE_L0;
1104             sub_mb_type &= ~MB_TYPE_L0;
1105         }
1106
1107         if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
1108             for(i8=0; i8<4; i8++){
1109                 int x8 = i8&1;
1110                 int y8 = i8>>1;
1111                 int xy8 = x8+y8*b8_stride;
1112                 int xy4 = 3*x8+y8*b4_stride;
1113                 int a=0, b=0;
1114
1115                 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1116                     continue;
1117                 h->sub_mb_type[i8] = sub_mb_type;
1118
1119                 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1120                 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1121                 if(!IS_INTRA(mb_type_col[y8])
1122                    && (   (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
1123                        || (l1ref0[xy8]  < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
1124                     if(ref[0] > 0)
1125                         a= pack16to32(mv[0][0],mv[0][1]);
1126                     if(ref[1] > 0)
1127                         b= pack16to32(mv[1][0],mv[1][1]);
1128                 }else{
1129                     a= pack16to32(mv[0][0],mv[0][1]);
1130                     b= pack16to32(mv[1][0],mv[1][1]);
1131                 }
1132                 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
1133                 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
1134             }
1135         }else if(IS_16X16(*mb_type)){
1136             int a=0, b=0;
1137
1138             fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1139             fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
1140             if(!IS_INTRA(mb_type_col[0])
1141                && (   (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
1142                    || (l1ref0[0]  < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
1143                        && (h->x264_build>33 || !h->x264_build)))){
1144                 if(ref[0] > 0)
1145                     a= pack16to32(mv[0][0],mv[0][1]);
1146                 if(ref[1] > 0)
1147                     b= pack16to32(mv[1][0],mv[1][1]);
1148             }else{
1149                 a= pack16to32(mv[0][0],mv[0][1]);
1150                 b= pack16to32(mv[1][0],mv[1][1]);
1151             }
1152             fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
1153             fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
1154         }else{
1155             for(i8=0; i8<4; i8++){
1156                 const int x8 = i8&1;
1157                 const int y8 = i8>>1;
1158
1159                 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1160                     continue;
1161                 h->sub_mb_type[i8] = sub_mb_type;
1162
1163                 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1164                 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1165                 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1166                 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1167
1168                 /* col_zero_flag */
1169                 if(!IS_INTRA(mb_type_col[0]) && (   l1ref0[x8 + y8*b8_stride] == 0
1170                                               || (l1ref0[x8 + y8*b8_stride] < 0 && l1ref1[x8 + y8*b8_stride] == 0
1171                                                   && (h->x264_build>33 || !h->x264_build)))){
1172                     const int16_t (*l1mv)[2]= l1ref0[x8 + y8*b8_stride] == 0 ? l1mv0 : l1mv1;
1173                     if(IS_SUB_8X8(sub_mb_type)){
1174                         const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
1175                         if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1176                             if(ref[0] == 0)
1177                                 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1178                             if(ref[1] == 0)
1179                                 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1180                         }
1181                     }else
1182                     for(i4=0; i4<4; i4++){
1183                         const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
1184                         if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1185                             if(ref[0] == 0)
1186                                 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1187                             if(ref[1] == 0)
1188                                 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1189                         }
1190                     }
1191                 }
1192             }
1193         }
1194     }else{ /* direct temporal mv pred */
1195         const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
1196         const int *dist_scale_factor = h->dist_scale_factor;
1197         int ref_offset= 0;
1198
1199         if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){
1200             map_col_to_list0[0] = h->map_col_to_list0_field[s->mb_y&1][0];
1201             map_col_to_list0[1] = h->map_col_to_list0_field[s->mb_y&1][1];
1202             dist_scale_factor   =h->dist_scale_factor_field[s->mb_y&1];
1203         }
1204         if(h->ref_list[1][0].mbaff && IS_INTERLACED(mb_type_col[0]))
1205             ref_offset += 16;
1206
1207         if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
1208             /* FIXME assumes direct_8x8_inference == 1 */
1209             int y_shift  = 2*!IS_INTERLACED(*mb_type);
1210
1211             for(i8=0; i8<4; i8++){
1212                 const int x8 = i8&1;
1213                 const int y8 = i8>>1;
1214                 int ref0, scale;
1215                 const int16_t (*l1mv)[2]= l1mv0;
1216
1217                 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1218                     continue;
1219                 h->sub_mb_type[i8] = sub_mb_type;
1220
1221                 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1222                 if(IS_INTRA(mb_type_col[y8])){
1223                     fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1224                     fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1225                     fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1226                     continue;
1227                 }
1228
1229                 ref0 = l1ref0[x8 + y8*b8_stride];
1230                 if(ref0 >= 0)
1231                     ref0 = map_col_to_list0[0][ref0 + ref_offset];
1232                 else{
1233                     ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
1234                     l1mv= l1mv1;
1235                 }
1236                 scale = dist_scale_factor[ref0];
1237                 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1238
1239                 {
1240                     const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
1241                     int my_col = (mv_col[1]<<y_shift)/2;
1242                     int mx = (scale * mv_col[0] + 128) >> 8;
1243                     int my = (scale * my_col + 128) >> 8;
1244                     fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1245                     fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
1246                 }
1247             }
1248             return;
1249         }
1250
1251         /* one-to-one mv scaling */
1252
1253         if(IS_16X16(*mb_type)){
1254             int ref, mv0, mv1;
1255
1256             fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1257             if(IS_INTRA(mb_type_col[0])){
1258                 ref=mv0=mv1=0;
1259             }else{
1260                 const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset]
1261                                                 : map_col_to_list0[1][l1ref1[0] + ref_offset];
1262                 const int scale = dist_scale_factor[ref0];
1263                 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1264                 int mv_l0[2];
1265                 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1266                 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1267                 ref= ref0;
1268                 mv0= pack16to32(mv_l0[0],mv_l0[1]);
1269                 mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1270             }
1271             fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
1272             fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
1273             fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
1274         }else{
1275             for(i8=0; i8<4; i8++){
1276                 const int x8 = i8&1;
1277                 const int y8 = i8>>1;
1278                 int ref0, scale;
1279                 const int16_t (*l1mv)[2]= l1mv0;
1280
1281                 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1282                     continue;
1283                 h->sub_mb_type[i8] = sub_mb_type;
1284                 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1285                 if(IS_INTRA(mb_type_col[0])){
1286                     fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1287                     fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1288                     fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1289                     continue;
1290                 }
1291
1292                 ref0 = l1ref0[x8 + y8*b8_stride] + ref_offset;
1293                 if(ref0 >= 0)
1294                     ref0 = map_col_to_list0[0][ref0];
1295                 else{
1296                     ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
1297                     l1mv= l1mv1;
1298                 }
1299                 scale = dist_scale_factor[ref0];
1300
1301                 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1302                 if(IS_SUB_8X8(sub_mb_type)){
1303                     const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
1304                     int mx = (scale * mv_col[0] + 128) >> 8;
1305                     int my = (scale * mv_col[1] + 128) >> 8;
1306                     fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1307                     fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1308                 }else
1309                 for(i4=0; i4<4; i4++){
1310                     const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
1311                     int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1312                     mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1313                     mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1314                     *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1315                         pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1316                 }
1317             }
1318         }
1319     }
1320 }
1321
1322 static inline void write_back_motion(H264Context *h, int mb_type){
1323     MpegEncContext * const s = &h->s;
1324     const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1325     const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1326     int list;
1327
1328     if(!USES_LIST(mb_type, 0))
1329         fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1330
1331     for(list=0; list<h->list_count; list++){
1332         int y;
1333         if(!USES_LIST(mb_type, list))
1334             continue;
1335
1336         for(y=0; y<4; y++){
1337             *(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];
1338             *(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];
1339         }
1340         if( h->pps.cabac ) {
1341             if(IS_SKIP(mb_type))
1342                 fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
1343             else
1344             for(y=0; y<4; y++){
1345                 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1346                 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1347             }
1348         }
1349
1350         {
1351             int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1352             ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1353             ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1354             ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1355             ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1356         }
1357     }
1358
1359     if(h->slice_type_nos == FF_B_TYPE && h->pps.cabac){
1360         if(IS_8X8(mb_type)){
1361             uint8_t *direct_table = &h->direct_table[b8_xy];
1362             direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1363             direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1364             direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1365         }
1366     }
1367 }
1368
1369 /**
1370  * Decodes a network abstraction layer unit.
1371  * @param consumed is the number of bytes used as input
1372  * @param length is the length of the array
1373  * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1374  * @returns decoded bytes, might be src+1 if no escapes
1375  */
1376 static const uint8_t *decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
1377     int i, si, di;
1378     uint8_t *dst;
1379     int bufidx;
1380
1381 //    src[0]&0x80;                //forbidden bit
1382     h->nal_ref_idc= src[0]>>5;
1383     h->nal_unit_type= src[0]&0x1F;
1384
1385     src++; length--;
1386 #if 0
1387     for(i=0; i<length; i++)
1388         printf("%2X ", src[i]);
1389 #endif
1390     for(i=0; i+1<length; i+=2){
1391         if(src[i]) continue;
1392         if(i>0 && src[i-1]==0) i--;
1393         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1394             if(src[i+2]!=3){
1395                 /* startcode, so we must be past the end */
1396                 length=i;
1397             }
1398             break;
1399         }
1400     }
1401
1402     if(i>=length-1){ //no escaped 0
1403         *dst_length= length;
1404         *consumed= length+1; //+1 for the header
1405         return src;
1406     }
1407
1408     bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
1409     h->rbsp_buffer[bufidx]= av_fast_realloc(h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
1410     dst= h->rbsp_buffer[bufidx];
1411
1412     if (dst == NULL){
1413         return NULL;
1414     }
1415
1416 //printf("decoding esc\n");
1417     si=di=0;
1418     while(si<length){
1419         //remove escapes (very rare 1:2^22)
1420         if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1421             if(src[si+2]==3){ //escape
1422                 dst[di++]= 0;
1423                 dst[di++]= 0;
1424                 si+=3;
1425                 continue;
1426             }else //next start code
1427                 break;
1428         }
1429
1430         dst[di++]= src[si++];
1431     }
1432
1433     memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
1434
1435     *dst_length= di;
1436     *consumed= si + 1;//+1 for the header
1437 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
1438     return dst;
1439 }
1440
1441 /**
1442  * identifies the exact end of the bitstream
1443  * @return the length of the trailing, or 0 if damaged
1444  */
1445 static int decode_rbsp_trailing(H264Context *h, const uint8_t *src){
1446     int v= *src;
1447     int r;
1448
1449     tprintf(h->s.avctx, "rbsp trailing %X\n", v);
1450
1451     for(r=1; r<9; r++){
1452         if(v&1) return r;
1453         v>>=1;
1454     }
1455     return 0;
1456 }
1457
1458 /**
1459  * IDCT transforms the 16 dc values and dequantizes them.
1460  * @param qp quantization parameter
1461  */
1462 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1463 #define stride 16
1464     int i;
1465     int temp[16]; //FIXME check if this is a good idea
1466     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1467     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1468
1469 //memset(block, 64, 2*256);
1470 //return;
1471     for(i=0; i<4; i++){
1472         const int offset= y_offset[i];
1473         const int z0= block[offset+stride*0] + block[offset+stride*4];
1474         const int z1= block[offset+stride*0] - block[offset+stride*4];
1475         const int z2= block[offset+stride*1] - block[offset+stride*5];
1476         const int z3= block[offset+stride*1] + block[offset+stride*5];
1477
1478         temp[4*i+0]= z0+z3;
1479         temp[4*i+1]= z1+z2;
1480         temp[4*i+2]= z1-z2;
1481         temp[4*i+3]= z0-z3;
1482     }
1483
1484     for(i=0; i<4; i++){
1485         const int offset= x_offset[i];
1486         const int z0= temp[4*0+i] + temp[4*2+i];
1487         const int z1= temp[4*0+i] - temp[4*2+i];
1488         const int z2= temp[4*1+i] - temp[4*3+i];
1489         const int z3= temp[4*1+i] + temp[4*3+i];
1490
1491         block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
1492         block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1493         block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1494         block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1495     }
1496 }
1497
1498 #if 0
1499 /**
1500  * DCT transforms the 16 dc values.
1501  * @param qp quantization parameter ??? FIXME
1502  */
1503 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1504 //    const int qmul= dequant_coeff[qp][0];
1505     int i;
1506     int temp[16]; //FIXME check if this is a good idea
1507     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1508     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1509
1510     for(i=0; i<4; i++){
1511         const int offset= y_offset[i];
1512         const int z0= block[offset+stride*0] + block[offset+stride*4];
1513         const int z1= block[offset+stride*0] - block[offset+stride*4];
1514         const int z2= block[offset+stride*1] - block[offset+stride*5];
1515         const int z3= block[offset+stride*1] + block[offset+stride*5];
1516
1517         temp[4*i+0]= z0+z3;
1518         temp[4*i+1]= z1+z2;
1519         temp[4*i+2]= z1-z2;
1520         temp[4*i+3]= z0-z3;
1521     }
1522
1523     for(i=0; i<4; i++){
1524         const int offset= x_offset[i];
1525         const int z0= temp[4*0+i] + temp[4*2+i];
1526         const int z1= temp[4*0+i] - temp[4*2+i];
1527         const int z2= temp[4*1+i] - temp[4*3+i];
1528         const int z3= temp[4*1+i] + temp[4*3+i];
1529
1530         block[stride*0 +offset]= (z0 + z3)>>1;
1531         block[stride*2 +offset]= (z1 + z2)>>1;
1532         block[stride*8 +offset]= (z1 - z2)>>1;
1533         block[stride*10+offset]= (z0 - z3)>>1;
1534     }
1535 }
1536 #endif
1537
1538 #undef xStride
1539 #undef stride
1540
1541 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1542     const int stride= 16*2;
1543     const int xStride= 16;
1544     int a,b,c,d,e;
1545
1546     a= block[stride*0 + xStride*0];
1547     b= block[stride*0 + xStride*1];
1548     c= block[stride*1 + xStride*0];
1549     d= block[stride*1 + xStride*1];
1550
1551     e= a-b;
1552     a= a+b;
1553     b= c-d;
1554     c= c+d;
1555
1556     block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1557     block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1558     block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1559     block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
1560 }
1561
1562 #if 0
1563 static void chroma_dc_dct_c(DCTELEM *block){
1564     const int stride= 16*2;
1565     const int xStride= 16;
1566     int a,b,c,d,e;
1567
1568     a= block[stride*0 + xStride*0];
1569     b= block[stride*0 + xStride*1];
1570     c= block[stride*1 + xStride*0];
1571     d= block[stride*1 + xStride*1];
1572
1573     e= a-b;
1574     a= a+b;
1575     b= c-d;
1576     c= c+d;
1577
1578     block[stride*0 + xStride*0]= (a+c);
1579     block[stride*0 + xStride*1]= (e+b);
1580     block[stride*1 + xStride*0]= (a-c);
1581     block[stride*1 + xStride*1]= (e-b);
1582 }
1583 #endif
1584
1585 /**
1586  * gets the chroma qp.
1587  */
1588 static inline int get_chroma_qp(H264Context *h, int t, int qscale){
1589     return h->pps.chroma_qp_table[t][qscale];
1590 }
1591
1592 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1593                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1594                            int src_x_offset, int src_y_offset,
1595                            qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1596     MpegEncContext * const s = &h->s;
1597     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1598     int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1599     const int luma_xy= (mx&3) + ((my&3)<<2);
1600     uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
1601     uint8_t * src_cb, * src_cr;
1602     int extra_width= h->emu_edge_width;
1603     int extra_height= h->emu_edge_height;
1604     int emu=0;
1605     const int full_mx= mx>>2;
1606     const int full_my= my>>2;
1607     const int pic_width  = 16*s->mb_width;
1608     const int pic_height = 16*s->mb_height >> MB_FIELD;
1609
1610     if(mx&7) extra_width -= 3;
1611     if(my&7) extra_height -= 3;
1612
1613     if(   full_mx < 0-extra_width
1614        || full_my < 0-extra_height
1615        || full_mx + 16/*FIXME*/ > pic_width + extra_width
1616        || full_my + 16/*FIXME*/ > pic_height + extra_height){
1617         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);
1618             src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
1619         emu=1;
1620     }
1621
1622     qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
1623     if(!square){
1624         qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
1625     }
1626
1627     if(ENABLE_GRAY && s->flags&CODEC_FLAG_GRAY) return;
1628
1629     if(MB_FIELD){
1630         // chroma offset when predicting from a field of opposite parity
1631         my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
1632         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
1633     }
1634     src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1635     src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1636
1637     if(emu){
1638         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);
1639             src_cb= s->edge_emu_buffer;
1640     }
1641     chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1642
1643     if(emu){
1644         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);
1645             src_cr= s->edge_emu_buffer;
1646     }
1647     chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1648 }
1649
1650 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
1651                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1652                            int x_offset, int y_offset,
1653                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1654                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1655                            int list0, int list1){
1656     MpegEncContext * const s = &h->s;
1657     qpel_mc_func *qpix_op=  qpix_put;
1658     h264_chroma_mc_func chroma_op= chroma_put;
1659
1660     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
1661     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
1662     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
1663     x_offset += 8*s->mb_x;
1664     y_offset += 8*(s->mb_y >> MB_FIELD);
1665
1666     if(list0){
1667         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1668         mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1669                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
1670                            qpix_op, chroma_op);
1671
1672         qpix_op=  qpix_avg;
1673         chroma_op= chroma_avg;
1674     }
1675
1676     if(list1){
1677         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
1678         mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1679                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
1680                            qpix_op, chroma_op);
1681     }
1682 }
1683
1684 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
1685                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1686                            int x_offset, int y_offset,
1687                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1688                            h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
1689                            h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
1690                            int list0, int list1){
1691     MpegEncContext * const s = &h->s;
1692
1693     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
1694     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
1695     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
1696     x_offset += 8*s->mb_x;
1697     y_offset += 8*(s->mb_y >> MB_FIELD);
1698
1699     if(list0 && list1){
1700         /* don't optimize for luma-only case, since B-frames usually
1701          * use implicit weights => chroma too. */
1702         uint8_t *tmp_cb = s->obmc_scratchpad;
1703         uint8_t *tmp_cr = s->obmc_scratchpad + 8;
1704         uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
1705         int refn0 = h->ref_cache[0][ scan8[n] ];
1706         int refn1 = h->ref_cache[1][ scan8[n] ];
1707
1708         mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
1709                     dest_y, dest_cb, dest_cr,
1710                     x_offset, y_offset, qpix_put, chroma_put);
1711         mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
1712                     tmp_y, tmp_cb, tmp_cr,
1713                     x_offset, y_offset, qpix_put, chroma_put);
1714
1715         if(h->use_weight == 2){
1716             int weight0 = h->implicit_weight[refn0][refn1];
1717             int weight1 = 64 - weight0;
1718             luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
1719             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
1720             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
1721         }else{
1722             luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
1723                             h->luma_weight[0][refn0], h->luma_weight[1][refn1],
1724                             h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
1725             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1726                             h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
1727                             h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
1728             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1729                             h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
1730                             h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
1731         }
1732     }else{
1733         int list = list1 ? 1 : 0;
1734         int refn = h->ref_cache[list][ scan8[n] ];
1735         Picture *ref= &h->ref_list[list][refn];
1736         mc_dir_part(h, ref, n, square, chroma_height, delta, list,
1737                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
1738                     qpix_put, chroma_put);
1739
1740         luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
1741                        h->luma_weight[list][refn], h->luma_offset[list][refn]);
1742         if(h->use_weight_chroma){
1743             chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1744                              h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
1745             chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1746                              h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
1747         }
1748     }
1749 }
1750
1751 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1752                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1753                            int x_offset, int y_offset,
1754                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1755                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1756                            h264_weight_func *weight_op, h264_biweight_func *weight_avg,
1757                            int list0, int list1){
1758     if((h->use_weight==2 && list0 && list1
1759         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
1760        || h->use_weight==1)
1761         mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1762                          x_offset, y_offset, qpix_put, chroma_put,
1763                          weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
1764     else
1765         mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1766                     x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
1767 }
1768
1769 static inline void prefetch_motion(H264Context *h, int list){
1770     /* fetch pixels for estimated mv 4 macroblocks ahead
1771      * optimized for 64byte cache lines */
1772     MpegEncContext * const s = &h->s;
1773     const int refn = h->ref_cache[list][scan8[0]];
1774     if(refn >= 0){
1775         const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
1776         const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
1777         uint8_t **src= h->ref_list[list][refn].data;
1778         int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
1779         s->dsp.prefetch(src[0]+off, s->linesize, 4);
1780         off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
1781         s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
1782     }
1783 }
1784
1785 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1786                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1787                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
1788                       h264_weight_func *weight_op, h264_biweight_func *weight_avg){
1789     MpegEncContext * const s = &h->s;
1790     const int mb_xy= h->mb_xy;
1791     const int mb_type= s->current_picture.mb_type[mb_xy];
1792
1793     assert(IS_INTER(mb_type));
1794
1795     prefetch_motion(h, 0);
1796
1797     if(IS_16X16(mb_type)){
1798         mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1799                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1800                 &weight_op[0], &weight_avg[0],
1801                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1802     }else if(IS_16X8(mb_type)){
1803         mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1804                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1805                 &weight_op[1], &weight_avg[1],
1806                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1807         mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1808                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1809                 &weight_op[1], &weight_avg[1],
1810                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1811     }else if(IS_8X16(mb_type)){
1812         mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
1813                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1814                 &weight_op[2], &weight_avg[2],
1815                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1816         mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
1817                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1818                 &weight_op[2], &weight_avg[2],
1819                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1820     }else{
1821         int i;
1822
1823         assert(IS_8X8(mb_type));
1824
1825         for(i=0; i<4; i++){
1826             const int sub_mb_type= h->sub_mb_type[i];
1827             const int n= 4*i;
1828             int x_offset= (i&1)<<2;
1829             int y_offset= (i&2)<<1;
1830
1831             if(IS_SUB_8X8(sub_mb_type)){
1832                 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1833                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1834                     &weight_op[3], &weight_avg[3],
1835                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1836             }else if(IS_SUB_8X4(sub_mb_type)){
1837                 mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1838                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1839                     &weight_op[4], &weight_avg[4],
1840                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1841                 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
1842                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1843                     &weight_op[4], &weight_avg[4],
1844                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1845             }else if(IS_SUB_4X8(sub_mb_type)){
1846                 mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1847                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1848                     &weight_op[5], &weight_avg[5],
1849                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1850                 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
1851                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1852                     &weight_op[5], &weight_avg[5],
1853                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1854             }else{
1855                 int j;
1856                 assert(IS_SUB_4X4(sub_mb_type));
1857                 for(j=0; j<4; j++){
1858                     int sub_x_offset= x_offset + 2*(j&1);
1859                     int sub_y_offset= y_offset +   (j&2);
1860                     mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
1861                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1862                         &weight_op[6], &weight_avg[6],
1863                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1864                 }
1865             }
1866         }
1867     }
1868
1869     prefetch_motion(h, 1);
1870 }
1871
1872 static av_cold void decode_init_vlc(void){
1873     static int done = 0;
1874
1875     if (!done) {
1876         int i;
1877         int offset;
1878         done = 1;
1879
1880         chroma_dc_coeff_token_vlc.table = chroma_dc_coeff_token_vlc_table;
1881         chroma_dc_coeff_token_vlc.table_allocated = chroma_dc_coeff_token_vlc_table_size;
1882         init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
1883                  &chroma_dc_coeff_token_len [0], 1, 1,
1884                  &chroma_dc_coeff_token_bits[0], 1, 1,
1885                  INIT_VLC_USE_NEW_STATIC);
1886
1887         offset = 0;
1888         for(i=0; i<4; i++){
1889             coeff_token_vlc[i].table = coeff_token_vlc_tables+offset;
1890             coeff_token_vlc[i].table_allocated = coeff_token_vlc_tables_size[i];
1891             init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
1892                      &coeff_token_len [i][0], 1, 1,
1893                      &coeff_token_bits[i][0], 1, 1,
1894                      INIT_VLC_USE_NEW_STATIC);
1895             offset += coeff_token_vlc_tables_size[i];
1896         }
1897         /*
1898          * This is a one time safety check to make sure that
1899          * the packed static coeff_token_vlc table sizes
1900          * were initialized correctly.
1901          */
1902         assert(offset == FF_ARRAY_ELEMS(coeff_token_vlc_tables));
1903
1904         for(i=0; i<3; i++){
1905             chroma_dc_total_zeros_vlc[i].table = chroma_dc_total_zeros_vlc_tables[i];
1906             chroma_dc_total_zeros_vlc[i].table_allocated = chroma_dc_total_zeros_vlc_tables_size;
1907             init_vlc(&chroma_dc_total_zeros_vlc[i],
1908                      CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
1909                      &chroma_dc_total_zeros_len [i][0], 1, 1,
1910                      &chroma_dc_total_zeros_bits[i][0], 1, 1,
1911                      INIT_VLC_USE_NEW_STATIC);
1912         }
1913         for(i=0; i<15; i++){
1914             total_zeros_vlc[i].table = total_zeros_vlc_tables[i];
1915             total_zeros_vlc[i].table_allocated = total_zeros_vlc_tables_size;
1916             init_vlc(&total_zeros_vlc[i],
1917                      TOTAL_ZEROS_VLC_BITS, 16,
1918                      &total_zeros_len [i][0], 1, 1,
1919                      &total_zeros_bits[i][0], 1, 1,
1920                      INIT_VLC_USE_NEW_STATIC);
1921         }
1922
1923         for(i=0; i<6; i++){
1924             run_vlc[i].table = run_vlc_tables[i];
1925             run_vlc[i].table_allocated = run_vlc_tables_size;
1926             init_vlc(&run_vlc[i],
1927                      RUN_VLC_BITS, 7,
1928                      &run_len [i][0], 1, 1,
1929                      &run_bits[i][0], 1, 1,
1930                      INIT_VLC_USE_NEW_STATIC);
1931         }
1932         run7_vlc.table = run7_vlc_table,
1933         run7_vlc.table_allocated = run7_vlc_table_size;
1934         init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
1935                  &run_len [6][0], 1, 1,
1936                  &run_bits[6][0], 1, 1,
1937                  INIT_VLC_USE_NEW_STATIC);
1938     }
1939 }
1940
1941 static void free_tables(H264Context *h){
1942     int i;
1943     H264Context *hx;
1944     av_freep(&h->intra4x4_pred_mode);
1945     av_freep(&h->chroma_pred_mode_table);
1946     av_freep(&h->cbp_table);
1947     av_freep(&h->mvd_table[0]);
1948     av_freep(&h->mvd_table[1]);
1949     av_freep(&h->direct_table);
1950     av_freep(&h->non_zero_count);
1951     av_freep(&h->slice_table_base);
1952     h->slice_table= NULL;
1953
1954     av_freep(&h->mb2b_xy);
1955     av_freep(&h->mb2b8_xy);
1956
1957     for(i = 0; i < h->s.avctx->thread_count; i++) {
1958         hx = h->thread_context[i];
1959         if(!hx) continue;
1960         av_freep(&hx->top_borders[1]);
1961         av_freep(&hx->top_borders[0]);
1962         av_freep(&hx->s.obmc_scratchpad);
1963     }
1964 }
1965
1966 static void init_dequant8_coeff_table(H264Context *h){
1967     int i,q,x;
1968     const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
1969     h->dequant8_coeff[0] = h->dequant8_buffer[0];
1970     h->dequant8_coeff[1] = h->dequant8_buffer[1];
1971
1972     for(i=0; i<2; i++ ){
1973         if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
1974             h->dequant8_coeff[1] = h->dequant8_buffer[0];
1975             break;
1976         }
1977
1978         for(q=0; q<52; q++){
1979             int shift = div6[q];
1980             int idx = rem6[q];
1981             for(x=0; x<64; x++)
1982                 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
1983                     ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
1984                     h->pps.scaling_matrix8[i][x]) << shift;
1985         }
1986     }
1987 }
1988
1989 static void init_dequant4_coeff_table(H264Context *h){
1990     int i,j,q,x;
1991     const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
1992     for(i=0; i<6; i++ ){
1993         h->dequant4_coeff[i] = h->dequant4_buffer[i];
1994         for(j=0; j<i; j++){
1995             if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
1996                 h->dequant4_coeff[i] = h->dequant4_buffer[j];
1997                 break;
1998             }
1999         }
2000         if(j<i)
2001             continue;
2002
2003         for(q=0; q<52; q++){
2004             int shift = div6[q] + 2;
2005             int idx = rem6[q];
2006             for(x=0; x<16; x++)
2007                 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
2008                     ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
2009                     h->pps.scaling_matrix4[i][x]) << shift;
2010         }
2011     }
2012 }
2013
2014 static void init_dequant_tables(H264Context *h){
2015     int i,x;
2016     init_dequant4_coeff_table(h);
2017     if(h->pps.transform_8x8_mode)
2018         init_dequant8_coeff_table(h);
2019     if(h->sps.transform_bypass){
2020         for(i=0; i<6; i++)
2021             for(x=0; x<16; x++)
2022                 h->dequant4_coeff[i][0][x] = 1<<6;
2023         if(h->pps.transform_8x8_mode)
2024             for(i=0; i<2; i++)
2025                 for(x=0; x<64; x++)
2026                     h->dequant8_coeff[i][0][x] = 1<<6;
2027     }
2028 }
2029
2030
2031 /**
2032  * allocates tables.
2033  * needs width/height
2034  */
2035 static int alloc_tables(H264Context *h){
2036     MpegEncContext * const s = &h->s;
2037     const int big_mb_num= s->mb_stride * (s->mb_height+1);
2038     int x,y;
2039
2040     CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2041
2042     CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2043     CHECKED_ALLOCZ(h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base))
2044     CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2045
2046     CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2047     CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2048     CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2049     CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2050
2051     memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
2052     h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
2053
2054     CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint32_t));
2055     CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
2056     for(y=0; y<s->mb_height; y++){
2057         for(x=0; x<s->mb_width; x++){
2058             const int mb_xy= x + y*s->mb_stride;
2059             const int b_xy = 4*x + 4*y*h->b_stride;
2060             const int b8_xy= 2*x + 2*y*h->b8_stride;
2061
2062             h->mb2b_xy [mb_xy]= b_xy;
2063             h->mb2b8_xy[mb_xy]= b8_xy;
2064         }
2065     }
2066
2067     s->obmc_scratchpad = NULL;
2068
2069     if(!h->dequant4_coeff[0])
2070         init_dequant_tables(h);
2071
2072     return 0;
2073 fail:
2074     free_tables(h);
2075     return -1;
2076 }
2077
2078 /**
2079  * Mimic alloc_tables(), but for every context thread.
2080  */
2081 static void clone_tables(H264Context *dst, H264Context *src){
2082     dst->intra4x4_pred_mode       = src->intra4x4_pred_mode;
2083     dst->non_zero_count           = src->non_zero_count;
2084     dst->slice_table              = src->slice_table;
2085     dst->cbp_table                = src->cbp_table;
2086     dst->mb2b_xy                  = src->mb2b_xy;
2087     dst->mb2b8_xy                 = src->mb2b8_xy;
2088     dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
2089     dst->mvd_table[0]             = src->mvd_table[0];
2090     dst->mvd_table[1]             = src->mvd_table[1];
2091     dst->direct_table             = src->direct_table;
2092
2093     dst->s.obmc_scratchpad = NULL;
2094     ff_h264_pred_init(&dst->hpc, src->s.codec_id);
2095 }
2096
2097 /**
2098  * Init context
2099  * Allocate buffers which are not shared amongst multiple threads.
2100  */
2101 static int context_init(H264Context *h){
2102     CHECKED_ALLOCZ(h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2103     CHECKED_ALLOCZ(h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2104
2105     return 0;
2106 fail:
2107     return -1; // free_tables will clean up for us
2108 }
2109
2110 static av_cold void common_init(H264Context *h){
2111     MpegEncContext * const s = &h->s;
2112
2113     s->width = s->avctx->width;
2114     s->height = s->avctx->height;
2115     s->codec_id= s->avctx->codec->id;
2116
2117     ff_h264_pred_init(&h->hpc, s->codec_id);
2118
2119     h->dequant_coeff_pps= -1;
2120     s->unrestricted_mv=1;
2121     s->decode=1; //FIXME
2122
2123     dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
2124
2125     memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
2126     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
2127 }
2128
2129 static av_cold int decode_init(AVCodecContext *avctx){
2130     H264Context *h= avctx->priv_data;
2131     MpegEncContext * const s = &h->s;
2132
2133     MPV_decode_defaults(s);
2134
2135     s->avctx = avctx;
2136     common_init(h);
2137
2138     s->out_format = FMT_H264;
2139     s->workaround_bugs= avctx->workaround_bugs;
2140
2141     // set defaults
2142 //    s->decode_mb= ff_h263_decode_mb;
2143     s->quarter_sample = 1;
2144     s->low_delay= 1;
2145
2146     if(avctx->codec_id == CODEC_ID_SVQ3)
2147         avctx->pix_fmt= PIX_FMT_YUVJ420P;
2148     else
2149         avctx->pix_fmt= PIX_FMT_YUV420P;
2150
2151     decode_init_vlc();
2152
2153     if(avctx->extradata_size > 0 && avctx->extradata &&
2154        *(char *)avctx->extradata == 1){
2155         h->is_avc = 1;
2156         h->got_avcC = 0;
2157     } else {
2158         h->is_avc = 0;
2159     }
2160
2161     h->thread_context[0] = h;
2162     h->outputed_poc = INT_MIN;
2163     h->prev_poc_msb= 1<<16;
2164     return 0;
2165 }
2166
2167 static int frame_start(H264Context *h){
2168     MpegEncContext * const s = &h->s;
2169     int i;
2170
2171     if(MPV_frame_start(s, s->avctx) < 0)
2172         return -1;
2173     ff_er_frame_start(s);
2174     /*
2175      * MPV_frame_start uses pict_type to derive key_frame.
2176      * This is incorrect for H.264; IDR markings must be used.
2177      * Zero here; IDR markings per slice in frame or fields are ORed in later.
2178      * See decode_nal_units().
2179      */
2180     s->current_picture_ptr->key_frame= 0;
2181
2182     assert(s->linesize && s->uvlinesize);
2183
2184     for(i=0; i<16; i++){
2185         h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2186         h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
2187     }
2188     for(i=0; i<4; i++){
2189         h->block_offset[16+i]=
2190         h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2191         h->block_offset[24+16+i]=
2192         h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2193     }
2194
2195     /* can't be in alloc_tables because linesize isn't known there.
2196      * FIXME: redo bipred weight to not require extra buffer? */
2197     for(i = 0; i < s->avctx->thread_count; i++)
2198         if(!h->thread_context[i]->s.obmc_scratchpad)
2199             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
2200
2201     /* some macroblocks will be accessed before they're available */
2202     if(FRAME_MBAFF || s->avctx->thread_count > 1)
2203         memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
2204
2205 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2206
2207     // We mark the current picture as non-reference after allocating it, so
2208     // that if we break out due to an error it can be released automatically
2209     // in the next MPV_frame_start().
2210     // SVQ3 as well as most other codecs have only last/next/current and thus
2211     // get released even with set reference, besides SVQ3 and others do not
2212     // mark frames as reference later "naturally".
2213     if(s->codec_id != CODEC_ID_SVQ3)
2214         s->current_picture_ptr->reference= 0;
2215
2216     s->current_picture_ptr->field_poc[0]=
2217     s->current_picture_ptr->field_poc[1]= INT_MAX;
2218     assert(s->current_picture_ptr->long_ref==0);
2219
2220     return 0;
2221 }
2222
2223 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){
2224     MpegEncContext * const s = &h->s;
2225     int i;
2226     int step    = 1;
2227     int offset  = 1;
2228     int uvoffset= 1;
2229     int top_idx = 1;
2230     int skiplast= 0;
2231
2232     src_y  -=   linesize;
2233     src_cb -= uvlinesize;
2234     src_cr -= uvlinesize;
2235
2236     if(!simple && FRAME_MBAFF){
2237         if(s->mb_y&1){
2238             offset  = MB_MBAFF ? 1 : 17;
2239             uvoffset= MB_MBAFF ? 1 : 9;
2240             if(!MB_MBAFF){
2241                 *(uint64_t*)(h->top_borders[0][s->mb_x]+ 0)= *(uint64_t*)(src_y +  15*linesize);
2242                 *(uint64_t*)(h->top_borders[0][s->mb_x]+ 8)= *(uint64_t*)(src_y +8+15*linesize);
2243                 if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2244                     *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+7*uvlinesize);
2245                     *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+7*uvlinesize);
2246                 }
2247             }
2248         }else{
2249             if(!MB_MBAFF){
2250                 h->left_border[0]= h->top_borders[0][s->mb_x][15];
2251                 if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2252                     h->left_border[34   ]= h->top_borders[0][s->mb_x][16+7  ];
2253                     h->left_border[34+18]= h->top_borders[0][s->mb_x][16+8+7];
2254                 }
2255                 skiplast= 1;
2256             }
2257             offset  =
2258             uvoffset=
2259             top_idx = MB_MBAFF ? 0 : 1;
2260         }
2261         step= MB_MBAFF ? 2 : 1;
2262     }
2263
2264     // There are two lines saved, the line above the the top macroblock of a pair,
2265     // and the line above the bottom macroblock
2266     h->left_border[offset]= h->top_borders[top_idx][s->mb_x][15];
2267     for(i=1; i<17 - skiplast; i++){
2268         h->left_border[offset+i*step]= src_y[15+i*  linesize];
2269     }
2270
2271     *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
2272     *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2273
2274     if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2275         h->left_border[uvoffset+34   ]= h->top_borders[top_idx][s->mb_x][16+7];
2276         h->left_border[uvoffset+34+18]= h->top_borders[top_idx][s->mb_x][24+7];
2277         for(i=1; i<9 - skiplast; i++){
2278             h->left_border[uvoffset+34   +i*step]= src_cb[7+i*uvlinesize];
2279             h->left_border[uvoffset+34+18+i*step]= src_cr[7+i*uvlinesize];
2280         }
2281         *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2282         *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2283     }
2284 }
2285
2286 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){
2287     MpegEncContext * const s = &h->s;
2288     int temp8, i;
2289     uint64_t temp64;
2290     int deblock_left;
2291     int deblock_top;
2292     int mb_xy;
2293     int step    = 1;
2294     int offset  = 1;
2295     int uvoffset= 1;
2296     int top_idx = 1;
2297
2298     if(!simple && FRAME_MBAFF){
2299         if(s->mb_y&1){
2300             offset  = MB_MBAFF ? 1 : 17;
2301             uvoffset= MB_MBAFF ? 1 : 9;
2302         }else{
2303             offset  =
2304             uvoffset=
2305             top_idx = MB_MBAFF ? 0 : 1;
2306         }
2307         step= MB_MBAFF ? 2 : 1;
2308     }
2309
2310     if(h->deblocking_filter == 2) {
2311         mb_xy = h->mb_xy;
2312         deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
2313         deblock_top  = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
2314     } else {
2315         deblock_left = (s->mb_x > 0);
2316         deblock_top =  (s->mb_y > !!MB_FIELD);
2317     }
2318
2319     src_y  -=   linesize + 1;
2320     src_cb -= uvlinesize + 1;
2321     src_cr -= uvlinesize + 1;
2322
2323 #define XCHG(a,b,t,xchg)\
2324 t= a;\
2325 if(xchg)\
2326     a= b;\
2327 b= t;
2328
2329     if(deblock_left){
2330         for(i = !deblock_top; i<16; i++){
2331             XCHG(h->left_border[offset+i*step], src_y [i*  linesize], temp8, xchg);
2332         }
2333         XCHG(h->left_border[offset+i*step], src_y [i*  linesize], temp8, 1);
2334     }
2335
2336     if(deblock_top){
2337         XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2338         XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2339         if(s->mb_x+1 < s->mb_width){
2340             XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
2341         }
2342     }
2343
2344     if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2345         if(deblock_left){
2346             for(i = !deblock_top; i<8; i++){
2347                 XCHG(h->left_border[uvoffset+34   +i*step], src_cb[i*uvlinesize], temp8, xchg);
2348                 XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, xchg);
2349             }
2350             XCHG(h->left_border[uvoffset+34   +i*step], src_cb[i*uvlinesize], temp8, 1);
2351             XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, 1);
2352         }
2353         if(deblock_top){
2354             XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2355             XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2356         }
2357     }
2358 }
2359
2360 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
2361     MpegEncContext * const s = &h->s;
2362     const int mb_x= s->mb_x;
2363     const int mb_y= s->mb_y;
2364     const int mb_xy= h->mb_xy;
2365     const int mb_type= s->current_picture.mb_type[mb_xy];
2366     uint8_t  *dest_y, *dest_cb, *dest_cr;
2367     int linesize, uvlinesize /*dct_offset*/;
2368     int i;
2369     int *block_offset = &h->block_offset[0];
2370     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
2371     const int is_h264 = simple || s->codec_id == CODEC_ID_H264;
2372     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
2373     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
2374
2375     dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2376     dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2377     dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2378
2379     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
2380     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
2381
2382     if (!simple && MB_FIELD) {
2383         linesize   = h->mb_linesize   = s->linesize * 2;
2384         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2385         block_offset = &h->block_offset[24];
2386         if(mb_y&1){ //FIXME move out of this function?
2387             dest_y -= s->linesize*15;
2388             dest_cb-= s->uvlinesize*7;
2389             dest_cr-= s->uvlinesize*7;
2390         }
2391         if(FRAME_MBAFF) {
2392             int list;
2393             for(list=0; list<h->list_count; list++){
2394                 if(!USES_LIST(mb_type, list))
2395                     continue;
2396                 if(IS_16X16(mb_type)){
2397                     int8_t *ref = &h->ref_cache[list][scan8[0]];
2398                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
2399                 }else{
2400                     for(i=0; i<16; i+=4){
2401                         int ref = h->ref_cache[list][scan8[i]];
2402                         if(ref >= 0)
2403                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
2404                     }
2405                 }
2406             }
2407         }
2408     } else {
2409         linesize   = h->mb_linesize   = s->linesize;
2410         uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2411 //        dct_offset = s->linesize * 16;
2412     }
2413
2414     if (!simple && IS_INTRA_PCM(mb_type)) {
2415         for (i=0; i<16; i++) {
2416             memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
2417         }
2418         for (i=0; i<8; i++) {
2419             memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
2420             memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
2421         }
2422     } else {
2423         if(IS_INTRA(mb_type)){
2424             if(h->deblocking_filter)
2425                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
2426
2427             if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2428                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2429                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2430             }
2431
2432             if(IS_INTRA4x4(mb_type)){
2433                 if(simple || !s->encoding){
2434                     if(IS_8x8DCT(mb_type)){
2435                         if(transform_bypass){
2436                             idct_dc_add =
2437                             idct_add    = s->dsp.add_pixels8;
2438                         }else{
2439                             idct_dc_add = s->dsp.h264_idct8_dc_add;
2440                             idct_add    = s->dsp.h264_idct8_add;
2441                         }
2442                         for(i=0; i<16; i+=4){
2443                             uint8_t * const ptr= dest_y + block_offset[i];
2444                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2445                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
2446                                 h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
2447                             }else{
2448                                 const int nnz = h->non_zero_count_cache[ scan8[i] ];
2449                                 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
2450                                                             (h->topright_samples_available<<i)&0x4000, linesize);
2451                                 if(nnz){
2452                                     if(nnz == 1 && h->mb[i*16])
2453                                         idct_dc_add(ptr, h->mb + i*16, linesize);
2454                                     else
2455                                         idct_add   (ptr, h->mb + i*16, linesize);
2456                                 }
2457                             }
2458                         }
2459                     }else{
2460                         if(transform_bypass){
2461                             idct_dc_add =
2462                             idct_add    = s->dsp.add_pixels4;
2463                         }else{
2464                             idct_dc_add = s->dsp.h264_idct_dc_add;
2465                             idct_add    = s->dsp.h264_idct_add;
2466                         }
2467                         for(i=0; i<16; i++){
2468                             uint8_t * const ptr= dest_y + block_offset[i];
2469                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2470
2471                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
2472                                 h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
2473                             }else{
2474                                 uint8_t *topright;
2475                                 int nnz, tr;
2476                                 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2477                                     const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2478                                     assert(mb_y || linesize <= block_offset[i]);
2479                                     if(!topright_avail){
2480                                         tr= ptr[3 - linesize]*0x01010101;
2481                                         topright= (uint8_t*) &tr;
2482                                     }else
2483                                         topright= ptr + 4 - linesize;
2484                                 }else
2485                                     topright= NULL;
2486
2487                                 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
2488                                 nnz = h->non_zero_count_cache[ scan8[i] ];
2489                                 if(nnz){
2490                                     if(is_h264){
2491                                         if(nnz == 1 && h->mb[i*16])
2492                                             idct_dc_add(ptr, h->mb + i*16, linesize);
2493                                         else
2494                                             idct_add   (ptr, h->mb + i*16, linesize);
2495                                     }else
2496                                         svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2497                                 }
2498                             }
2499                         }
2500                     }
2501                 }
2502             }else{
2503                 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2504                 if(is_h264){
2505                     if(!transform_bypass)
2506                         h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
2507                 }else
2508                     svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2509             }
2510             if(h->deblocking_filter)
2511                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
2512         }else if(is_h264){
2513             hl_motion(h, dest_y, dest_cb, dest_cr,
2514                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2515                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2516                       s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
2517         }
2518
2519
2520         if(!IS_INTRA4x4(mb_type)){
2521             if(is_h264){
2522                 if(IS_INTRA16x16(mb_type)){
2523                     if(transform_bypass){
2524                         if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
2525                             h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
2526                         }else{
2527                             for(i=0; i<16; i++){
2528                                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
2529                                     s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
2530                             }
2531                         }
2532                     }else{
2533                          s->dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2534                     }
2535                 }else if(h->cbp&15){
2536                     if(transform_bypass){
2537                         const int di = IS_8x8DCT(mb_type) ? 4 : 1;
2538                         idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
2539                         for(i=0; i<16; i+=di){
2540                             if(h->non_zero_count_cache[ scan8[i] ]){
2541                                 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2542                             }
2543                         }
2544                     }else{
2545                         if(IS_8x8DCT(mb_type)){
2546                             s->dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2547                         }else{
2548                             s->dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2549                         }
2550                     }
2551                 }
2552             }else{
2553                 for(i=0; i<16; i++){
2554                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2555                         uint8_t * const ptr= dest_y + block_offset[i];
2556                         svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2557                     }
2558                 }
2559             }
2560         }
2561
2562         if((simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
2563             uint8_t *dest[2] = {dest_cb, dest_cr};
2564             if(transform_bypass){
2565                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
2566                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
2567                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
2568                 }else{
2569                     idct_add = s->dsp.add_pixels4;
2570                     for(i=16; i<16+8; i++){
2571                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
2572                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2573                     }
2574                 }
2575             }else{
2576                 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]);
2577                 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]);
2578                 if(is_h264){
2579                     idct_add = s->dsp.h264_idct_add;
2580                     idct_dc_add = s->dsp.h264_idct_dc_add;
2581                     for(i=16; i<16+8; i++){
2582                         if(h->non_zero_count_cache[ scan8[i] ])
2583                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2584                         else if(h->mb[i*16])
2585                             idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2586                     }
2587                 }else{
2588                     for(i=16; i<16+8; i++){
2589                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2590                             uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
2591                             svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2592                         }
2593                     }
2594                 }
2595             }
2596         }
2597     }
2598     if(h->cbp || IS_INTRA(mb_type))
2599         s->dsp.clear_blocks(h->mb);
2600
2601     if(h->deblocking_filter) {
2602         backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, simple);
2603         fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
2604         h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2605         h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2606         if (!simple && FRAME_MBAFF) {
2607             filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2608         } else {
2609             filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2610         }
2611     }
2612 }
2613
2614 /**
2615  * Process a macroblock; this case avoids checks for expensive uncommon cases.
2616  */
2617 static void hl_decode_mb_simple(H264Context *h){
2618     hl_decode_mb_internal(h, 1);
2619 }
2620
2621 /**
2622  * Process a macroblock; this handles edge cases, such as interlacing.
2623  */
2624 static void av_noinline hl_decode_mb_complex(H264Context *h){
2625     hl_decode_mb_internal(h, 0);
2626 }
2627
2628 static void hl_decode_mb(H264Context *h){
2629     MpegEncContext * const s = &h->s;
2630     const int mb_xy= h->mb_xy;
2631     const int mb_type= s->current_picture.mb_type[mb_xy];
2632     int is_complex = ENABLE_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
2633
2634     if(ENABLE_H264_ENCODER && !s->decode)
2635         return;
2636
2637     if (is_complex)
2638         hl_decode_mb_complex(h);
2639     else hl_decode_mb_simple(h);
2640 }
2641
2642 static void pic_as_field(Picture *pic, const int parity){
2643     int i;
2644     for (i = 0; i < 4; ++i) {
2645         if (parity == PICT_BOTTOM_FIELD)
2646             pic->data[i] += pic->linesize[i];
2647         pic->reference = parity;
2648         pic->linesize[i] *= 2;
2649     }
2650     pic->poc= pic->field_poc[parity == PICT_BOTTOM_FIELD];
2651 }
2652
2653 static int split_field_copy(Picture *dest, Picture *src,
2654                             int parity, int id_add){
2655     int match = !!(src->reference & parity);
2656
2657     if (match) {
2658         *dest = *src;
2659         if(parity != PICT_FRAME){
2660             pic_as_field(dest, parity);
2661             dest->pic_id *= 2;
2662             dest->pic_id += id_add;
2663         }
2664     }
2665
2666     return match;
2667 }
2668
2669 static int build_def_list(Picture *def, Picture **in, int len, int is_long, int sel){
2670     int i[2]={0};
2671     int index=0;
2672
2673     while(i[0]<len || i[1]<len){
2674         while(i[0]<len && !(in[ i[0] ] && (in[ i[0] ]->reference & sel)))
2675             i[0]++;
2676         while(i[1]<len && !(in[ i[1] ] && (in[ i[1] ]->reference & (sel^3))))
2677             i[1]++;
2678         if(i[0] < len){
2679             in[ i[0] ]->pic_id= is_long ? i[0] : in[ i[0] ]->frame_num;
2680             split_field_copy(&def[index++], in[ i[0]++ ], sel  , 1);
2681         }
2682         if(i[1] < len){
2683             in[ i[1] ]->pic_id= is_long ? i[1] : in[ i[1] ]->frame_num;
2684             split_field_copy(&def[index++], in[ i[1]++ ], sel^3, 0);
2685         }
2686     }
2687
2688     return index;
2689 }
2690
2691 static int add_sorted(Picture **sorted, Picture **src, int len, int limit, int dir){
2692     int i, best_poc;
2693     int out_i= 0;
2694
2695     for(;;){
2696         best_poc= dir ? INT_MIN : INT_MAX;
2697
2698         for(i=0; i<len; i++){
2699             const int poc= src[i]->poc;
2700             if(((poc > limit) ^ dir) && ((poc < best_poc) ^ dir)){
2701                 best_poc= poc;
2702                 sorted[out_i]= src[i];
2703             }
2704         }
2705         if(best_poc == (dir ? INT_MIN : INT_MAX))
2706             break;
2707         limit= sorted[out_i++]->poc - dir;
2708     }
2709     return out_i;
2710 }
2711
2712 /**
2713  * fills the default_ref_list.
2714  */
2715 static int fill_default_ref_list(H264Context *h){
2716     MpegEncContext * const s = &h->s;
2717     int i, len;
2718
2719     if(h->slice_type_nos==FF_B_TYPE){
2720         Picture *sorted[32];
2721         int cur_poc, list;
2722         int lens[2];
2723
2724         if(FIELD_PICTURE)
2725             cur_poc= s->current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
2726         else
2727             cur_poc= s->current_picture_ptr->poc;
2728
2729         for(list= 0; list<2; list++){
2730             len= add_sorted(sorted    , h->short_ref, h->short_ref_count, cur_poc, 1^list);
2731             len+=add_sorted(sorted+len, h->short_ref, h->short_ref_count, cur_poc, 0^list);
2732             assert(len<=32);
2733             len= build_def_list(h->default_ref_list[list]    , sorted     , len, 0, s->picture_structure);
2734             len+=build_def_list(h->default_ref_list[list]+len, h->long_ref, 16 , 1, s->picture_structure);
2735             assert(len<=32);
2736
2737             if(len < h->ref_count[list])
2738                 memset(&h->default_ref_list[list][len], 0, sizeof(Picture)*(h->ref_count[list] - len));
2739             lens[list]= len;
2740         }
2741
2742         if(lens[0] == lens[1] && lens[1] > 1){
2743             for(i=0; h->default_ref_list[0][i].data[0] == h->default_ref_list[1][i].data[0] && i<lens[0]; i++);
2744             if(i == lens[0])
2745                 FFSWAP(Picture, h->default_ref_list[1][0], h->default_ref_list[1][1]);
2746         }
2747     }else{
2748         len = build_def_list(h->default_ref_list[0]    , h->short_ref, h->short_ref_count, 0, s->picture_structure);
2749         len+= build_def_list(h->default_ref_list[0]+len, h-> long_ref, 16                , 1, s->picture_structure);
2750         assert(len <= 32);
2751         if(len < h->ref_count[0])
2752             memset(&h->default_ref_list[0][len], 0, sizeof(Picture)*(h->ref_count[0] - len));
2753     }
2754 #ifdef TRACE
2755     for (i=0; i<h->ref_count[0]; i++) {
2756         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]);
2757     }
2758     if(h->slice_type_nos==FF_B_TYPE){
2759         for (i=0; i<h->ref_count[1]; i++) {
2760             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]);
2761         }
2762     }
2763 #endif
2764     return 0;
2765 }
2766
2767 static void print_short_term(H264Context *h);
2768 static void print_long_term(H264Context *h);
2769
2770 /**
2771  * Extract structure information about the picture described by pic_num in
2772  * the current decoding context (frame or field). Note that pic_num is
2773  * picture number without wrapping (so, 0<=pic_num<max_pic_num).
2774  * @param pic_num picture number for which to extract structure information
2775  * @param structure one of PICT_XXX describing structure of picture
2776  *                      with pic_num
2777  * @return frame number (short term) or long term index of picture
2778  *         described by pic_num
2779  */
2780 static int pic_num_extract(H264Context *h, int pic_num, int *structure){
2781     MpegEncContext * const s = &h->s;
2782
2783     *structure = s->picture_structure;
2784     if(FIELD_PICTURE){
2785         if (!(pic_num & 1))
2786             /* opposite field */
2787             *structure ^= PICT_FRAME;
2788         pic_num >>= 1;
2789     }
2790
2791     return pic_num;
2792 }
2793
2794 static int decode_ref_pic_list_reordering(H264Context *h){
2795     MpegEncContext * const s = &h->s;
2796     int list, index, pic_structure;
2797
2798     print_short_term(h);
2799     print_long_term(h);
2800
2801     for(list=0; list<h->list_count; list++){
2802         memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2803
2804         if(get_bits1(&s->gb)){
2805             int pred= h->curr_pic_num;
2806
2807             for(index=0; ; index++){
2808                 unsigned int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2809                 unsigned int pic_id;
2810                 int i;
2811                 Picture *ref = NULL;
2812
2813                 if(reordering_of_pic_nums_idc==3)
2814                     break;
2815
2816                 if(index >= h->ref_count[list]){
2817                     av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
2818                     return -1;
2819                 }
2820
2821                 if(reordering_of_pic_nums_idc<3){
2822                     if(reordering_of_pic_nums_idc<2){
2823                         const unsigned int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2824                         int frame_num;
2825
2826                         if(abs_diff_pic_num > h->max_pic_num){
2827                             av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
2828                             return -1;
2829                         }
2830
2831                         if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2832                         else                                pred+= abs_diff_pic_num;
2833                         pred &= h->max_pic_num - 1;
2834
2835                         frame_num = pic_num_extract(h, pred, &pic_structure);
2836
2837                         for(i= h->short_ref_count-1; i>=0; i--){
2838                             ref = h->short_ref[i];
2839                             assert(ref->reference);
2840                             assert(!ref->long_ref);
2841                             if(
2842                                    ref->frame_num == frame_num &&
2843                                    (ref->reference & pic_structure)
2844                               )
2845                                 break;
2846                         }
2847                         if(i>=0)
2848                             ref->pic_id= pred;
2849                     }else{
2850                         int long_idx;
2851                         pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2852
2853                         long_idx= pic_num_extract(h, pic_id, &pic_structure);
2854
2855                         if(long_idx>31){
2856                             av_log(h->s.avctx, AV_LOG_ERROR, "long_term_pic_idx overflow\n");
2857                             return -1;
2858                         }
2859                         ref = h->long_ref[long_idx];
2860                         assert(!(ref && !ref->reference));
2861                         if(ref && (ref->reference & pic_structure)){
2862                             ref->pic_id= pic_id;
2863                             assert(ref->long_ref);
2864                             i=0;
2865                         }else{
2866                             i=-1;
2867                         }
2868                     }
2869
2870                     if (i < 0) {
2871                         av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
2872                         memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2873                     } else {
2874                         for(i=index; i+1<h->ref_count[list]; i++){
2875                             if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)
2876                                 break;
2877                         }
2878                         for(; i > index; i--){
2879                             h->ref_list[list][i]= h->ref_list[list][i-1];
2880                         }
2881                         h->ref_list[list][index]= *ref;
2882                         if (FIELD_PICTURE){
2883                             pic_as_field(&h->ref_list[list][index], pic_structure);
2884                         }
2885                     }
2886                 }else{
2887                     av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
2888                     return -1;
2889                 }
2890             }
2891         }
2892     }
2893     for(list=0; list<h->list_count; list++){
2894         for(index= 0; index < h->ref_count[list]; index++){
2895             if(!h->ref_list[list][index].data[0]){
2896                 av_log(h->s.avctx, AV_LOG_ERROR, "Missing reference picture\n");
2897                 h->ref_list[list][index]= s->current_picture; //FIXME this is not a sensible solution
2898             }
2899         }
2900     }
2901
2902     return 0;
2903 }
2904
2905 static void fill_mbaff_ref_list(H264Context *h){
2906     int list, i, j;
2907     for(list=0; list<2; list++){ //FIXME try list_count
2908         for(i=0; i<h->ref_count[list]; i++){
2909             Picture *frame = &h->ref_list[list][i];
2910             Picture *field = &h->ref_list[list][16+2*i];
2911             field[0] = *frame;
2912             for(j=0; j<3; j++)
2913                 field[0].linesize[j] <<= 1;
2914             field[0].reference = PICT_TOP_FIELD;
2915             field[0].poc= field[0].field_poc[0];
2916             field[1] = field[0];
2917             for(j=0; j<3; j++)
2918                 field[1].data[j] += frame->linesize[j];
2919             field[1].reference = PICT_BOTTOM_FIELD;
2920             field[1].poc= field[1].field_poc[1];
2921
2922             h->luma_weight[list][16+2*i] = h->luma_weight[list][16+2*i+1] = h->luma_weight[list][i];
2923             h->luma_offset[list][16+2*i] = h->luma_offset[list][16+2*i+1] = h->luma_offset[list][i];
2924             for(j=0; j<2; j++){
2925                 h->chroma_weight[list][16+2*i][j] = h->chroma_weight[list][16+2*i+1][j] = h->chroma_weight[list][i][j];
2926                 h->chroma_offset[list][16+2*i][j] = h->chroma_offset[list][16+2*i+1][j] = h->chroma_offset[list][i][j];
2927             }
2928         }
2929     }
2930     for(j=0; j<h->ref_count[1]; j++){
2931         for(i=0; i<h->ref_count[0]; i++)
2932             h->implicit_weight[j][16+2*i] = h->implicit_weight[j][16+2*i+1] = h->implicit_weight[j][i];
2933         memcpy(h->implicit_weight[16+2*j],   h->implicit_weight[j], sizeof(*h->implicit_weight));
2934         memcpy(h->implicit_weight[16+2*j+1], h->implicit_weight[j], sizeof(*h->implicit_weight));
2935     }
2936 }
2937
2938 static int pred_weight_table(H264Context *h){
2939     MpegEncContext * const s = &h->s;
2940     int list, i;
2941     int luma_def, chroma_def;
2942
2943     h->use_weight= 0;
2944     h->use_weight_chroma= 0;
2945     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2946     h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2947     luma_def = 1<<h->luma_log2_weight_denom;
2948     chroma_def = 1<<h->chroma_log2_weight_denom;
2949
2950     for(list=0; list<2; list++){
2951         for(i=0; i<h->ref_count[list]; i++){
2952             int luma_weight_flag, chroma_weight_flag;
2953
2954             luma_weight_flag= get_bits1(&s->gb);
2955             if(luma_weight_flag){
2956                 h->luma_weight[list][i]= get_se_golomb(&s->gb);
2957                 h->luma_offset[list][i]= get_se_golomb(&s->gb);
2958                 if(   h->luma_weight[list][i] != luma_def
2959                    || h->luma_offset[list][i] != 0)
2960                     h->use_weight= 1;
2961             }else{
2962                 h->luma_weight[list][i]= luma_def;
2963                 h->luma_offset[list][i]= 0;
2964             }
2965
2966             if(CHROMA){
2967                 chroma_weight_flag= get_bits1(&s->gb);
2968                 if(chroma_weight_flag){
2969                     int j;
2970                     for(j=0; j<2; j++){
2971                         h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
2972                         h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
2973                         if(   h->chroma_weight[list][i][j] != chroma_def
2974                         || h->chroma_offset[list][i][j] != 0)
2975                             h->use_weight_chroma= 1;
2976                     }
2977                 }else{
2978                     int j;
2979                     for(j=0; j<2; j++){
2980                         h->chroma_weight[list][i][j]= chroma_def;
2981                         h->chroma_offset[list][i][j]= 0;
2982                     }
2983                 }
2984             }
2985         }
2986         if(h->slice_type_nos != FF_B_TYPE) break;
2987     }
2988     h->use_weight= h->use_weight || h->use_weight_chroma;
2989     return 0;
2990 }
2991
2992 static void implicit_weight_table(H264Context *h){
2993     MpegEncContext * const s = &h->s;
2994     int ref0, ref1;
2995     int cur_poc = s->current_picture_ptr->poc;
2996
2997     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
2998        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
2999         h->use_weight= 0;
3000         h->use_weight_chroma= 0;
3001         return;
3002     }
3003
3004     h->use_weight= 2;
3005     h->use_weight_chroma= 2;
3006     h->luma_log2_weight_denom= 5;
3007     h->chroma_log2_weight_denom= 5;
3008
3009     for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3010         int poc0 = h->ref_list[0][ref0].poc;
3011         for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3012             int poc1 = h->ref_list[1][ref1].poc;
3013             int td = av_clip(poc1 - poc0, -128, 127);
3014             if(td){
3015                 int tb = av_clip(cur_poc - poc0, -128, 127);
3016                 int tx = (16384 + (FFABS(td) >> 1)) / td;
3017                 int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3018                 if(dist_scale_factor < -64 || dist_scale_factor > 128)
3019                     h->implicit_weight[ref0][ref1] = 32;
3020                 else
3021                     h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3022             }else
3023                 h->implicit_weight[ref0][ref1] = 32;
3024         }
3025     }
3026 }
3027
3028 /**
3029  * Mark a picture as no longer needed for reference. The refmask
3030  * argument allows unreferencing of individual fields or the whole frame.
3031  * If the picture becomes entirely unreferenced, but is being held for
3032  * display purposes, it is marked as such.
3033  * @param refmask mask of fields to unreference; the mask is bitwise
3034  *                anded with the reference marking of pic
3035  * @return non-zero if pic becomes entirely unreferenced (except possibly
3036  *         for display purposes) zero if one of the fields remains in
3037  *         reference
3038  */
3039 static inline int unreference_pic(H264Context *h, Picture *pic, int refmask){
3040     int i;
3041     if (pic->reference &= refmask) {
3042         return 0;
3043     } else {
3044         for(i = 0; h->delayed_pic[i]; i++)
3045             if(pic == h->delayed_pic[i]){
3046                 pic->reference=DELAYED_PIC_REF;
3047                 break;
3048             }
3049         return 1;
3050     }
3051 }
3052
3053 /**
3054  * instantaneous decoder refresh.
3055  */
3056 static void idr(H264Context *h){
3057     int i;
3058
3059     for(i=0; i<16; i++){
3060         remove_long(h, i, 0);
3061     }
3062     assert(h->long_ref_count==0);
3063
3064     for(i=0; i<h->short_ref_count; i++){
3065         unreference_pic(h, h->short_ref[i], 0);
3066         h->short_ref[i]= NULL;
3067     }
3068     h->short_ref_count=0;
3069     h->prev_frame_num= 0;
3070     h->prev_frame_num_offset= 0;
3071     h->prev_poc_msb=
3072     h->prev_poc_lsb= 0;
3073 }
3074
3075 /* forget old pics after a seek */
3076 static void flush_dpb(AVCodecContext *avctx){
3077     H264Context *h= avctx->priv_data;
3078     int i;
3079     for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
3080         if(h->delayed_pic[i])
3081             h->delayed_pic[i]->reference= 0;
3082         h->delayed_pic[i]= NULL;
3083     }
3084     h->outputed_poc= INT_MIN;
3085     idr(h);
3086     if(h->s.current_picture_ptr)
3087         h->s.current_picture_ptr->reference= 0;
3088     h->s.first_field= 0;
3089     ff_mpeg_flush(avctx);
3090 }
3091
3092 /**
3093  * Find a Picture in the short term reference list by frame number.
3094  * @param frame_num frame number to search for
3095  * @param idx the index into h->short_ref where returned picture is found
3096  *            undefined if no picture found.
3097  * @return pointer to the found picture, or NULL if no pic with the provided
3098  *                 frame number is found
3099  */
3100 static Picture * find_short(H264Context *h, int frame_num, int *idx){
3101     MpegEncContext * const s = &h->s;
3102     int i;
3103
3104     for(i=0; i<h->short_ref_count; i++){
3105         Picture *pic= h->short_ref[i];
3106         if(s->avctx->debug&FF_DEBUG_MMCO)
3107             av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3108         if(pic->frame_num == frame_num) {
3109             *idx = i;
3110             return pic;
3111         }
3112     }
3113     return NULL;
3114 }
3115
3116 /**
3117  * Remove a picture from the short term reference list by its index in
3118  * that list.  This does no checking on the provided index; it is assumed
3119  * to be valid. Other list entries are shifted down.
3120  * @param i index into h->short_ref of picture to remove.
3121  */
3122 static void remove_short_at_index(H264Context *h, int i){
3123     assert(i >= 0 && i < h->short_ref_count);
3124     h->short_ref[i]= NULL;
3125     if (--h->short_ref_count)
3126         memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i)*sizeof(Picture*));
3127 }
3128
3129 /**
3130  *
3131  * @return the removed picture or NULL if an error occurs
3132  */
3133 static Picture * remove_short(H264Context *h, int frame_num, int ref_mask){
3134     MpegEncContext * const s = &h->s;
3135     Picture *pic;
3136     int i;
3137
3138     if(s->avctx->debug&FF_DEBUG_MMCO)
3139         av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3140
3141     pic = find_short(h, frame_num, &i);
3142     if (pic){
3143         if(unreference_pic(h, pic, ref_mask))
3144         remove_short_at_index(h, i);
3145     }
3146
3147     return pic;
3148 }
3149
3150 /**
3151  * Remove a picture from the long term reference list by its index in
3152  * that list.
3153  * @return the removed picture or NULL if an error occurs
3154  */
3155 static Picture * remove_long(H264Context *h, int i, int ref_mask){
3156     Picture *pic;
3157
3158     pic= h->long_ref[i];
3159     if (pic){
3160         if(unreference_pic(h, pic, ref_mask)){
3161             assert(h->long_ref[i]->long_ref == 1);
3162             h->long_ref[i]->long_ref= 0;
3163             h->long_ref[i]= NULL;
3164             h->long_ref_count--;
3165         }
3166     }
3167
3168     return pic;
3169 }
3170
3171 /**
3172  * print short term list
3173  */
3174 static void print_short_term(H264Context *h) {
3175     uint32_t i;
3176     if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3177         av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3178         for(i=0; i<h->short_ref_count; i++){
3179             Picture *pic= h->short_ref[i];
3180             av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3181         }
3182     }
3183 }
3184
3185 /**
3186  * print long term list
3187  */
3188 static void print_long_term(H264Context *h) {
3189     uint32_t i;
3190     if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3191         av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3192         for(i = 0; i < 16; i++){
3193             Picture *pic= h->long_ref[i];
3194             if (pic) {
3195                 av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3196             }
3197         }
3198     }
3199 }
3200
3201 /**
3202  * Executes the reference picture marking (memory management control operations).
3203  */
3204 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3205     MpegEncContext * const s = &h->s;
3206     int i, j;
3207     int current_ref_assigned=0;
3208     Picture *pic;
3209
3210     if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3211         av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3212
3213     for(i=0; i<mmco_count; i++){
3214         int structure, frame_num;
3215         if(s->avctx->debug&FF_DEBUG_MMCO)
3216             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);
3217
3218         if(   mmco[i].opcode == MMCO_SHORT2UNUSED
3219            || mmco[i].opcode == MMCO_SHORT2LONG){
3220             frame_num = pic_num_extract(h, mmco[i].short_pic_num, &structure);
3221             pic = find_short(h, frame_num, &j);
3222             if(!pic){
3223                 if(mmco[i].opcode != MMCO_SHORT2LONG || !h->long_ref[mmco[i].long_arg]
3224                    || h->long_ref[mmco[i].long_arg]->frame_num != frame_num)
3225                 av_log(h->s.avctx, AV_LOG_ERROR, "mmco: unref short failure\n");
3226                 continue;
3227             }
3228         }
3229
3230         switch(mmco[i].opcode){
3231         case MMCO_SHORT2UNUSED:
3232             if(s->avctx->debug&FF_DEBUG_MMCO)
3233                 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);
3234             remove_short(h, frame_num, structure ^ PICT_FRAME);
3235             break;
3236         case MMCO_SHORT2LONG:
3237                 if (h->long_ref[mmco[i].long_arg] != pic)
3238                     remove_long(h, mmco[i].long_arg, 0);
3239
3240                 remove_short_at_index(h, j);
3241                 h->long_ref[ mmco[i].long_arg ]= pic;
3242                 if (h->long_ref[ mmco[i].long_arg ]){
3243                     h->long_ref[ mmco[i].long_arg ]->long_ref=1;
3244                     h->long_ref_count++;
3245                 }
3246             break;
3247         case MMCO_LONG2UNUSED:
3248             j = pic_num_extract(h, mmco[i].long_arg, &structure);
3249             pic = h->long_ref[j];
3250             if (pic) {
3251                 remove_long(h, j, structure ^ PICT_FRAME);
3252             } else if(s->avctx->debug&FF_DEBUG_MMCO)
3253                 av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: unref long failure\n");
3254             break;
3255         case MMCO_LONG:
3256                     // Comment below left from previous code as it is an interresting note.
3257                     /* First field in pair is in short term list or
3258                      * at a different long term index.
3259                      * This is not allowed; see 7.4.3.3, notes 2 and 3.
3260                      * Report the problem and keep the pair where it is,
3261                      * and mark this field valid.
3262                      */
3263
3264             if (h->long_ref[mmco[i].long_arg] != s->current_picture_ptr) {
3265                 remove_long(h, mmco[i].long_arg, 0);
3266
3267                 h->long_ref[ mmco[i].long_arg ]= s->current_picture_ptr;
3268                 h->long_ref[ mmco[i].long_arg ]->long_ref=1;
3269                 h->long_ref_count++;
3270             }
3271
3272             s->current_picture_ptr->reference |= s->picture_structure;
3273             current_ref_assigned=1;
3274             break;
3275         case MMCO_SET_MAX_LONG:
3276             assert(mmco[i].long_arg <= 16);
3277             // just remove the long term which index is greater than new max
3278             for(j = mmco[i].long_arg; j<16; j++){
3279                 remove_long(h, j, 0);
3280             }
3281             break;
3282         case MMCO_RESET:
3283             while(h->short_ref_count){
3284                 remove_short(h, h->short_ref[0]->frame_num, 0);
3285             }
3286             for(j = 0; j < 16; j++) {
3287                 remove_long(h, j, 0);
3288             }
3289             s->current_picture_ptr->poc=
3290             s->current_picture_ptr->field_poc[0]=
3291             s->current_picture_ptr->field_poc[1]=
3292             h->poc_lsb=
3293             h->poc_msb=
3294             h->frame_num=
3295             s->current_picture_ptr->frame_num= 0;
3296             break;
3297         default: assert(0);
3298         }
3299     }
3300
3301     if (!current_ref_assigned) {
3302         /* Second field of complementary field pair; the first field of
3303          * which is already referenced. If short referenced, it
3304          * should be first entry in short_ref. If not, it must exist
3305          * in long_ref; trying to put it on the short list here is an
3306          * error in the encoded bit stream (ref: 7.4.3.3, NOTE 2 and 3).
3307          */
3308         if (h->short_ref_count && h->short_ref[0] == s->current_picture_ptr) {
3309             /* Just mark the second field valid */
3310             s->current_picture_ptr->reference = PICT_FRAME;
3311         } else if (s->current_picture_ptr->long_ref) {
3312             av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term reference "
3313                                              "assignment for second field "
3314                                              "in complementary field pair "
3315                                              "(first field is long term)\n");
3316         } else {
3317             pic= remove_short(h, s->current_picture_ptr->frame_num, 0);
3318             if(pic){
3319                 av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3320             }
3321
3322             if(h->short_ref_count)
3323                 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3324
3325             h->short_ref[0]= s->current_picture_ptr;
3326             h->short_ref_count++;
3327             s->current_picture_ptr->reference |= s->picture_structure;
3328         }
3329     }
3330
3331     if (h->long_ref_count + h->short_ref_count > h->sps.ref_frame_count){
3332
3333         /* We have too many reference frames, probably due to corrupted
3334          * stream. Need to discard one frame. Prevents overrun of the
3335          * short_ref and long_ref buffers.
3336          */
3337         av_log(h->s.avctx, AV_LOG_ERROR,
3338                "number of reference frames exceeds max (probably "
3339                "corrupt input), discarding one\n");
3340
3341         if (h->long_ref_count && !h->short_ref_count) {
3342             for (i = 0; i < 16; ++i)
3343                 if (h->long_ref[i])
3344                     break;
3345
3346             assert(i < 16);
3347             remove_long(h, i, 0);
3348         } else {
3349             pic = h->short_ref[h->short_ref_count - 1];
3350             remove_short(h, pic->frame_num, 0);
3351         }
3352     }
3353
3354     print_short_term(h);
3355     print_long_term(h);
3356     return 0;
3357 }
3358
3359 static int decode_ref_pic_marking(H264Context *h, GetBitContext *gb){
3360     MpegEncContext * const s = &h->s;
3361     int i;
3362
3363     h->mmco_index= 0;
3364     if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3365         s->broken_link= get_bits1(gb) -1;
3366         if(get_bits1(gb)){
3367             h->mmco[0].opcode= MMCO_LONG;
3368             h->mmco[0].long_arg= 0;
3369             h->mmco_index= 1;
3370         }
3371     }else{
3372         if(get_bits1(gb)){ // adaptive_ref_pic_marking_mode_flag
3373             for(i= 0; i<MAX_MMCO_COUNT; i++) {
3374                 MMCOOpcode opcode= get_ue_golomb(gb);
3375
3376                 h->mmco[i].opcode= opcode;
3377                 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3378                     h->mmco[i].short_pic_num= (h->curr_pic_num - get_ue_golomb(gb) - 1) & (h->max_pic_num - 1);
3379 /*                    if(h->mmco[i].short_pic_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_pic_num ] == NULL){
3380                         av_log(s->avctx, AV_LOG_ERROR, "illegal short ref in memory management control operation %d\n", mmco);
3381                         return -1;
3382                     }*/
3383                 }
3384                 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3385                     unsigned int long_arg= get_ue_golomb(gb);
3386                     if(long_arg >= 32 || (long_arg >= 16 && !(opcode == MMCO_LONG2UNUSED && FIELD_PICTURE))){
3387                         av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
3388                         return -1;
3389                     }
3390                     h->mmco[i].long_arg= long_arg;
3391                 }
3392
3393                 if(opcode > (unsigned)MMCO_LONG){
3394                     av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
3395                     return -1;
3396                 }
3397                 if(opcode == MMCO_END)
3398                     break;
3399             }
3400             h->mmco_index= i;
3401         }else{
3402             assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
3403
3404             if(h->short_ref_count && h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count &&
3405                     !(FIELD_PICTURE && !s->first_field && s->current_picture_ptr->reference)) {
3406                 h->mmco[0].opcode= MMCO_SHORT2UNUSED;
3407                 h->mmco[0].short_pic_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
3408                 h->mmco_index= 1;
3409                 if (FIELD_PICTURE) {
3410                     h->mmco[0].short_pic_num *= 2;
3411                     h->mmco[1].opcode= MMCO_SHORT2UNUSED;
3412                     h->mmco[1].short_pic_num= h->mmco[0].short_pic_num + 1;
3413                     h->mmco_index= 2;
3414                 }
3415             }
3416         }
3417     }
3418
3419     return 0;
3420 }
3421
3422 static int init_poc(H264Context *h){
3423     MpegEncContext * const s = &h->s;
3424     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
3425     int field_poc[2];
3426     Picture *cur = s->current_picture_ptr;
3427
3428     h->frame_num_offset= h->prev_frame_num_offset;
3429     if(h->frame_num < h->prev_frame_num)
3430         h->frame_num_offset += max_frame_num;
3431
3432     if(h->sps.poc_type==0){
3433         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
3434
3435         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
3436             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
3437         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
3438             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
3439         else
3440             h->poc_msb = h->prev_poc_msb;
3441 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
3442         field_poc[0] =
3443         field_poc[1] = h->poc_msb + h->poc_lsb;
3444         if(s->picture_structure == PICT_FRAME)
3445             field_poc[1] += h->delta_poc_bottom;
3446     }else if(h->sps.poc_type==1){
3447         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
3448         int i;
3449
3450         if(h->sps.poc_cycle_length != 0)
3451             abs_frame_num = h->frame_num_offset + h->frame_num;
3452         else
3453             abs_frame_num = 0;
3454
3455         if(h->nal_ref_idc==0 && abs_frame_num > 0)
3456             abs_frame_num--;
3457
3458         expected_delta_per_poc_cycle = 0;
3459         for(i=0; i < h->sps.poc_cycle_length; i++)
3460             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
3461
3462         if(abs_frame_num > 0){
3463             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
3464             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
3465
3466             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
3467             for(i = 0; i <= frame_num_in_poc_cycle; i++)
3468                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
3469         } else
3470             expectedpoc = 0;
3471
3472         if(h->nal_ref_idc == 0)
3473             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
3474
3475         field_poc[0] = expectedpoc + h->delta_poc[0];
3476         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
3477
3478         if(s->picture_structure == PICT_FRAME)
3479             field_poc[1] += h->delta_poc[1];
3480     }else{
3481         int poc= 2*(h->frame_num_offset + h->frame_num);
3482
3483         if(!h->nal_ref_idc)
3484             poc--;
3485
3486         field_poc[0]= poc;
3487         field_poc[1]= poc;
3488     }
3489
3490     if(s->picture_structure != PICT_BOTTOM_FIELD)
3491         s->current_picture_ptr->field_poc[0]= field_poc[0];
3492     if(s->picture_structure != PICT_TOP_FIELD)
3493         s->current_picture_ptr->field_poc[1]= field_poc[1];
3494     cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
3495
3496     return 0;
3497 }
3498
3499
3500 /**
3501  * initialize scan tables
3502  */
3503 static void init_scan_tables(H264Context *h){
3504     MpegEncContext * const s = &h->s;
3505     int i;
3506     if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
3507         memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
3508         memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
3509     }else{
3510         for(i=0; i<16; i++){
3511 #define T(x) (x>>2) | ((x<<2) & 0xF)
3512             h->zigzag_scan[i] = T(zigzag_scan[i]);
3513             h-> field_scan[i] = T( field_scan[i]);
3514 #undef T
3515         }
3516     }
3517     if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
3518         memcpy(h->zigzag_scan8x8,       zigzag_scan8x8,       64*sizeof(uint8_t));
3519         memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
3520         memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
3521         memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
3522     }else{
3523         for(i=0; i<64; i++){
3524 #define T(x) (x>>3) | ((x&7)<<3)
3525             h->zigzag_scan8x8[i]       = T(zigzag_scan8x8[i]);
3526             h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
3527             h->field_scan8x8[i]        = T(field_scan8x8[i]);
3528             h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
3529 #undef T
3530         }
3531     }
3532     if(h->sps.transform_bypass){ //FIXME same ugly
3533         h->zigzag_scan_q0          = zigzag_scan;
3534         h->zigzag_scan8x8_q0       = zigzag_scan8x8;
3535         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
3536         h->field_scan_q0           = field_scan;
3537         h->field_scan8x8_q0        = field_scan8x8;
3538         h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
3539     }else{
3540         h->zigzag_scan_q0          = h->zigzag_scan;
3541         h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
3542         h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
3543         h->field_scan_q0           = h->field_scan;
3544         h->field_scan8x8_q0        = h->field_scan8x8;
3545         h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
3546     }
3547 }
3548
3549 /**
3550  * Replicates H264 "master" context to thread contexts.
3551  */
3552 static void clone_slice(H264Context *dst, H264Context *src)
3553 {
3554     memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
3555     dst->s.current_picture_ptr  = src->s.current_picture_ptr;
3556     dst->s.current_picture      = src->s.current_picture;
3557     dst->s.linesize             = src->s.linesize;
3558     dst->s.uvlinesize           = src->s.uvlinesize;
3559     dst->s.first_field          = src->s.first_field;
3560
3561     dst->prev_poc_msb           = src->prev_poc_msb;
3562     dst->prev_poc_lsb           = src->prev_poc_lsb;
3563     dst->prev_frame_num_offset  = src->prev_frame_num_offset;
3564     dst->prev_frame_num         = src->prev_frame_num;
3565     dst->short_ref_count        = src->short_ref_count;
3566
3567     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
3568     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
3569     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
3570     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
3571
3572     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
3573     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
3574 }
3575
3576 /**
3577  * decodes a slice header.
3578  * This will also call MPV_common_init() and frame_start() as needed.
3579  *
3580  * @param h h264context
3581  * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
3582  *
3583  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
3584  */
3585 static int decode_slice_header(H264Context *h, H264Context *h0){
3586     MpegEncContext * const s = &h->s;
3587     MpegEncContext * const s0 = &h0->s;
3588     unsigned int first_mb_in_slice;
3589     unsigned int pps_id;
3590     int num_ref_idx_active_override_flag;
3591     unsigned int slice_type, tmp, i, j;
3592     int default_ref_list_done = 0;
3593     int last_pic_structure;
3594
3595     s->dropable= h->nal_ref_idc == 0;
3596
3597     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
3598         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
3599         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
3600     }else{
3601         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
3602         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
3603     }
3604
3605     first_mb_in_slice= get_ue_golomb(&s->gb);
3606
3607     if((s->flags2 & CODEC_FLAG2_CHUNKS) && first_mb_in_slice == 0){
3608         h0->current_slice = 0;
3609         if (!s0->first_field)
3610             s->current_picture_ptr= NULL;
3611     }
3612
3613     slice_type= get_ue_golomb(&s->gb);
3614     if(slice_type > 9){
3615         av_log(h->s.avctx, AV_LOG_ERROR, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);
3616         return -1;
3617     }
3618     if(slice_type > 4){
3619         slice_type -= 5;
3620         h->slice_type_fixed=1;
3621     }else
3622         h->slice_type_fixed=0;
3623
3624     slice_type= golomb_to_pict_type[ slice_type ];
3625     if (slice_type == FF_I_TYPE
3626         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
3627         default_ref_list_done = 1;
3628     }
3629     h->slice_type= slice_type;
3630     h->slice_type_nos= slice_type & 3;
3631
3632     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
3633     if (s->pict_type == FF_B_TYPE && s0->last_picture_ptr == NULL) {
3634         av_log(h->s.avctx, AV_LOG_ERROR,
3635                "B picture before any references, skipping\n");
3636         return -1;
3637     }
3638
3639     pps_id= get_ue_golomb(&s->gb);
3640     if(pps_id>=MAX_PPS_COUNT){
3641         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3642         return -1;
3643     }
3644     if(!h0->pps_buffers[pps_id]) {
3645         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS referenced\n");
3646         return -1;
3647     }
3648     h->pps= *h0->pps_buffers[pps_id];
3649
3650     if(!h0->sps_buffers[h->pps.sps_id]) {
3651         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS referenced\n");
3652         return -1;
3653     }
3654     h->sps = *h0->sps_buffers[h->pps.sps_id];
3655
3656     if(h == h0 && h->dequant_coeff_pps != pps_id){
3657         h->dequant_coeff_pps = pps_id;
3658         init_dequant_tables(h);
3659     }
3660
3661     s->mb_width= h->sps.mb_width;
3662     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
3663
3664     h->b_stride=  s->mb_width*4;
3665     h->b8_stride= s->mb_width*2;
3666
3667     s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
3668     if(h->sps.frame_mbs_only_flag)
3669         s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
3670     else
3671         s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
3672
3673     if (s->context_initialized
3674         && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
3675         if(h != h0)
3676             return -1;   // width / height changed during parallelized decoding
3677         free_tables(h);
3678         flush_dpb(s->avctx);
3679         MPV_common_end(s);
3680     }
3681     if (!s->context_initialized) {
3682         if(h != h0)
3683             return -1;  // we cant (re-)initialize context during parallel decoding
3684         if (MPV_common_init(s) < 0)
3685             return -1;
3686         s->first_field = 0;
3687
3688         init_scan_tables(h);
3689         alloc_tables(h);
3690
3691         for(i = 1; i < s->avctx->thread_count; i++) {
3692             H264Context *c;
3693             c = h->thread_context[i] = av_malloc(sizeof(H264Context));
3694             memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
3695             memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
3696             c->sps = h->sps;
3697             c->pps = h->pps;
3698             init_scan_tables(c);
3699             clone_tables(c, h);
3700         }
3701
3702         for(i = 0; i < s->avctx->thread_count; i++)
3703             if(context_init(h->thread_context[i]) < 0)
3704                 return -1;
3705
3706         s->avctx->width = s->width;
3707         s->avctx->height = s->height;
3708         s->avctx->sample_aspect_ratio= h->sps.sar;
3709         if(!s->avctx->sample_aspect_ratio.den)
3710             s->avctx->sample_aspect_ratio.den = 1;
3711
3712         if(h->sps.timing_info_present_flag){
3713             s->avctx->time_base= (AVRational){h->sps.num_units_in_tick * 2, h->sps.time_scale};
3714             if(h->x264_build > 0 && h->x264_build < 44)
3715                 s->avctx->time_base.den *= 2;
3716             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
3717                       s->avctx->time_base.num, s->avctx->time_base.den, 1<<30);
3718         }
3719     }
3720
3721     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3722
3723     h->mb_mbaff = 0;
3724     h->mb_aff_frame = 0;
3725     last_pic_structure = s0->picture_structure;
3726     if(h->sps.frame_mbs_only_flag){
3727         s->picture_structure= PICT_FRAME;
3728     }else{
3729         if(get_bits1(&s->gb)) { //field_pic_flag
3730             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3731         } else {
3732             s->picture_structure= PICT_FRAME;
3733             h->mb_aff_frame = h->sps.mb_aff;
3734         }
3735     }
3736     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
3737
3738     if(h0->current_slice == 0){
3739         while(h->frame_num !=  h->prev_frame_num &&
3740               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
3741             av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
3742             frame_start(h);
3743             h->prev_frame_num++;
3744             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
3745             s->current_picture_ptr->frame_num= h->prev_frame_num;
3746             execute_ref_pic_marking(h, NULL, 0);
3747         }
3748
3749         /* See if we have a decoded first field looking for a pair... */
3750         if (s0->first_field) {
3751             assert(s0->current_picture_ptr);
3752             assert(s0->current_picture_ptr->data[0]);
3753             assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
3754
3755             /* figure out if we have a complementary field pair */
3756             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
3757                 /*
3758                  * Previous field is unmatched. Don't display it, but let it
3759                  * remain for reference if marked as such.
3760                  */
3761                 s0->current_picture_ptr = NULL;
3762                 s0->first_field = FIELD_PICTURE;
3763
3764             } else {
3765                 if (h->nal_ref_idc &&
3766                         s0->current_picture_ptr->reference &&
3767                         s0->current_picture_ptr->frame_num != h->frame_num) {
3768                     /*
3769                      * This and previous field were reference, but had
3770                      * different frame_nums. Consider this field first in
3771                      * pair. Throw away previous field except for reference
3772                      * purposes.
3773                      */
3774                     s0->first_field = 1;
3775                     s0->current_picture_ptr = NULL;
3776
3777                 } else {
3778                     /* Second field in complementary pair */
3779                     s0->first_field = 0;
3780                 }
3781             }
3782
3783         } else {
3784             /* Frame or first field in a potentially complementary pair */
3785             assert(!s0->current_picture_ptr);
3786             s0->first_field = FIELD_PICTURE;
3787         }
3788
3789         if((!FIELD_PICTURE || s0->first_field) && frame_start(h) < 0) {
3790             s0->first_field = 0;
3791             return -1;
3792         }
3793     }
3794     if(h != h0)
3795         clone_slice(h, h0);
3796
3797     s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
3798
3799     assert(s->mb_num == s->mb_width * s->mb_height);
3800     if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
3801        first_mb_in_slice                    >= s->mb_num){
3802         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
3803         return -1;
3804     }
3805     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
3806     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
3807     if (s->picture_structure == PICT_BOTTOM_FIELD)
3808         s->resync_mb_y = s->mb_y = s->mb_y + 1;
3809     assert(s->mb_y < s->mb_height);
3810
3811     if(s->picture_structure==PICT_FRAME){
3812         h->curr_pic_num=   h->frame_num;
3813         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3814     }else{
3815         h->curr_pic_num= 2*h->frame_num + 1;
3816         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3817     }
3818
3819     if(h->nal_unit_type == NAL_IDR_SLICE){
3820         get_ue_golomb(&s->gb); /* idr_pic_id */
3821     }
3822
3823     if(h->sps.poc_type==0){
3824         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3825
3826         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3827             h->delta_poc_bottom= get_se_golomb(&s->gb);
3828         }
3829     }
3830
3831     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3832         h->delta_poc[0]= get_se_golomb(&s->gb);
3833
3834         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3835             h->delta_poc[1]= get_se_golomb(&s->gb);
3836     }
3837
3838     init_poc(h);
3839
3840     if(h->pps.redundant_pic_cnt_present){
3841         h->redundant_pic_count= get_ue_golomb(&s->gb);
3842     }
3843
3844     //set defaults, might be overridden a few lines later
3845     h->ref_count[0]= h->pps.ref_count[0];
3846     h->ref_count[1]= h->pps.ref_count[1];
3847
3848     if(h->slice_type_nos != FF_I_TYPE){
3849         if(h->slice_type_nos == FF_B_TYPE){
3850             h->direct_spatial_mv_pred= get_bits1(&s->gb);
3851         }
3852         num_ref_idx_active_override_flag= get_bits1(&s->gb);
3853
3854         if(num_ref_idx_active_override_flag){
3855             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3856             if(h->slice_type_nos==FF_B_TYPE)
3857                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3858
3859             if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
3860                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3861                 h->ref_count[0]= h->ref_count[1]= 1;
3862                 return -1;
3863             }
3864         }
3865         if(h->slice_type_nos == FF_B_TYPE)
3866             h->list_count= 2;
3867         else
3868             h->list_count= 1;
3869     }else
3870         h->list_count= 0;
3871
3872     if(!default_ref_list_done){
3873         fill_default_ref_list(h);
3874     }
3875
3876     if(h->slice_type_nos!=FF_I_TYPE && decode_ref_pic_list_reordering(h) < 0)
3877         return -1;
3878
3879     if(h->slice_type_nos!=FF_I_TYPE){
3880         s->last_picture_ptr= &h->ref_list[0][0];
3881         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
3882     }
3883     if(h->slice_type_nos==FF_B_TYPE){
3884         s->next_picture_ptr= &h->ref_list[1][0];
3885         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
3886     }
3887
3888     if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
3889        ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
3890         pred_weight_table(h);
3891     else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE)
3892         implicit_weight_table(h);
3893     else
3894         h->use_weight = 0;
3895
3896     if(h->nal_ref_idc)
3897         decode_ref_pic_marking(h0, &s->gb);
3898
3899     if(FRAME_MBAFF)
3900         fill_mbaff_ref_list(h);
3901
3902     if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
3903         direct_dist_scale_factor(h);
3904     direct_ref_list_init(h);
3905
3906     if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
3907         tmp = get_ue_golomb(&s->gb);
3908         if(tmp > 2){
3909             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3910             return -1;
3911         }
3912         h->cabac_init_idc= tmp;
3913     }
3914
3915     h->last_qscale_diff = 0;
3916     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
3917     if(tmp>51){
3918         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3919         return -1;
3920     }
3921     s->qscale= tmp;
3922     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3923     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3924     //FIXME qscale / qp ... stuff
3925     if(h->slice_type == FF_SP_TYPE){
3926         get_bits1(&s->gb); /* sp_for_switch_flag */
3927     }
3928     if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
3929         get_se_golomb(&s->gb); /* slice_qs_delta */
3930     }
3931
3932     h->deblocking_filter = 1;
3933     h->slice_alpha_c0_offset = 0;
3934     h->slice_beta_offset = 0;
3935     if( h->pps.deblocking_filter_parameters_present ) {
3936         tmp= get_ue_golomb(&s->gb);
3937         if(tmp > 2){
3938             av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
3939             return -1;
3940         }
3941         h->deblocking_filter= tmp;
3942         if(h->deblocking_filter < 2)
3943             h->deblocking_filter^= 1; // 1<->0
3944
3945         if( h->deblocking_filter ) {
3946             h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
3947             h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
3948         }
3949     }
3950
3951     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
3952        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
3953        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
3954        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
3955         h->deblocking_filter= 0;
3956
3957     if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
3958         if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
3959             /* Cheat slightly for speed:
3960                Do not bother to deblock across slices. */
3961             h->deblocking_filter = 2;
3962         } else {
3963             h0->max_contexts = 1;
3964             if(!h0->single_decode_warning) {
3965                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
3966                 h0->single_decode_warning = 1;
3967             }
3968             if(h != h0)
3969                 return 1; // deblocking switched inside frame
3970         }
3971     }
3972
3973 #if 0 //FMO
3974     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3975         slice_group_change_cycle= get_bits(&s->gb, ?);
3976 #endif
3977
3978     h0->last_slice_type = slice_type;
3979     h->slice_num = ++h0->current_slice;
3980     if(h->slice_num >= MAX_SLICES){
3981         av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
3982     }
3983
3984     for(j=0; j<2; j++){
3985         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
3986         ref2frm[0]=
3987         ref2frm[1]= -1;
3988         for(i=0; i<16; i++)
3989             ref2frm[i+2]= 4*h->ref_list[j][i].frame_num
3990                           +(h->ref_list[j][i].reference&3);
3991         ref2frm[18+0]=
3992         ref2frm[18+1]= -1;
3993         for(i=16; i<48; i++)
3994             ref2frm[i+4]= 4*h->ref_list[j][i].frame_num
3995                           +(h->ref_list[j][i].reference&3);
3996     }
3997
3998     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
3999     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
4000
4001     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
4002         av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d %s mb:%d %c%s%s pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
4003                h->slice_num,
4004                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
4005                first_mb_in_slice,
4006                av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
4007                pps_id, h->frame_num,
4008                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
4009                h->ref_count[0], h->ref_count[1],
4010                s->qscale,
4011                h->deblocking_filter, h->slice_alpha_c0_offset/2, h->slice_beta_offset/2,
4012                h->use_weight,
4013                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
4014                h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
4015                );
4016     }
4017
4018     return 0;
4019 }
4020
4021 /**
4022  *
4023  */
4024 static inline int get_level_prefix(GetBitContext *gb){
4025     unsigned int buf;
4026     int log;
4027
4028     OPEN_READER(re, gb);
4029     UPDATE_CACHE(re, gb);
4030     buf=GET_CACHE(re, gb);
4031
4032     log= 32 - av_log2(buf);
4033 #ifdef TRACE
4034     print_bin(buf>>(32-log), log);