H264: have hl_motion() and its callees take a chroma_idc argument.
[ffmpeg.git] / libavcodec / h264.c
1 /*
2  * H.26L/H.264/AVC/JVT/14496-10/... decoder
3  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4  *
5  * This file is part of Libav.
6  *
7  * Libav 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  * Libav 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 Libav; 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
24  * H.264 / AVC / MPEG4 part10 codec.
25  * @author Michael Niedermayer <michaelni@gmx.at>
26  */
27
28 #include "libavutil/imgutils.h"
29 #include "internal.h"
30 #include "dsputil.h"
31 #include "avcodec.h"
32 #include "mpegvideo.h"
33 #include "h264.h"
34 #include "h264data.h"
35 #include "h264_mvpred.h"
36 #include "golomb.h"
37 #include "mathops.h"
38 #include "rectangle.h"
39 #include "thread.h"
40 #include "vdpau_internal.h"
41 #include "libavutil/avassert.h"
42
43 #include "cabac.h"
44
45 //#undef NDEBUG
46 #include <assert.h>
47
48 static const uint8_t rem6[QP_MAX_NUM+1]={
49 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, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
50 };
51
52 static const uint8_t div6[QP_MAX_NUM+1]={
53 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, 8, 8, 9, 9, 9, 9, 9, 9,10,10,10,10,
54 };
55
56 static const enum PixelFormat hwaccel_pixfmt_list_h264_jpeg_420[] = {
57     PIX_FMT_DXVA2_VLD,
58     PIX_FMT_VAAPI_VLD,
59     PIX_FMT_YUVJ420P,
60     PIX_FMT_NONE
61 };
62
63 /**
64  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
65  */
66 int ff_h264_check_intra4x4_pred_mode(H264Context *h){
67     MpegEncContext * const s = &h->s;
68     static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
69     static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
70     int i;
71
72     if(!(h->top_samples_available&0x8000)){
73         for(i=0; i<4; i++){
74             int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
75             if(status<0){
76                 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);
77                 return -1;
78             } else if(status){
79                 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
80             }
81         }
82     }
83
84     if((h->left_samples_available&0x8888)!=0x8888){
85         static const int mask[4]={0x8000,0x2000,0x80,0x20};
86         for(i=0; i<4; i++){
87             if(!(h->left_samples_available&mask[i])){
88                 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
89                 if(status<0){
90                     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);
91                     return -1;
92                 } else if(status){
93                     h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
94                 }
95             }
96         }
97     }
98
99     return 0;
100 } //FIXME cleanup like ff_h264_check_intra_pred_mode
101
102 /**
103  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
104  */
105 int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
106     MpegEncContext * const s = &h->s;
107     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
108     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
109
110     if(mode > 6U) {
111         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);
112         return -1;
113     }
114
115     if(!(h->top_samples_available&0x8000)){
116         mode= top[ mode ];
117         if(mode<0){
118             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);
119             return -1;
120         }
121     }
122
123     if((h->left_samples_available&0x8080) != 0x8080){
124         mode= left[ mode ];
125         if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
126             mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
127         }
128         if(mode<0){
129             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);
130             return -1;
131         }
132     }
133
134     return mode;
135 }
136
137 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
138     int i, si, di;
139     uint8_t *dst;
140     int bufidx;
141
142 //    src[0]&0x80;                //forbidden bit
143     h->nal_ref_idc= src[0]>>5;
144     h->nal_unit_type= src[0]&0x1F;
145
146     src++; length--;
147
148 #if HAVE_FAST_UNALIGNED
149 # if HAVE_FAST_64BIT
150 #   define RS 7
151     for(i=0; i+1<length; i+=9){
152         if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
153 # else
154 #   define RS 3
155     for(i=0; i+1<length; i+=5){
156         if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
157 # endif
158             continue;
159         if(i>0 && !src[i]) i--;
160         while(src[i]) i++;
161 #else
162 #   define RS 0
163     for(i=0; i+1<length; i+=2){
164         if(src[i]) continue;
165         if(i>0 && src[i-1]==0) i--;
166 #endif
167         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
168             if(src[i+2]!=3){
169                 /* startcode, so we must be past the end */
170                 length=i;
171             }
172             break;
173         }
174         i-= RS;
175     }
176
177     if(i>=length-1){ //no escaped 0
178         *dst_length= length;
179         *consumed= length+1; //+1 for the header
180         return src;
181     }
182
183     bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
184     av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
185     dst= h->rbsp_buffer[bufidx];
186
187     if (dst == NULL){
188         return NULL;
189     }
190
191 //printf("decoding esc\n");
192     memcpy(dst, src, i);
193     si=di=i;
194     while(si+2<length){
195         //remove escapes (very rare 1:2^22)
196         if(src[si+2]>3){
197             dst[di++]= src[si++];
198             dst[di++]= src[si++];
199         }else if(src[si]==0 && src[si+1]==0){
200             if(src[si+2]==3){ //escape
201                 dst[di++]= 0;
202                 dst[di++]= 0;
203                 si+=3;
204                 continue;
205             }else //next start code
206                 goto nsc;
207         }
208
209         dst[di++]= src[si++];
210     }
211     while(si<length)
212         dst[di++]= src[si++];
213 nsc:
214
215     memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
216
217     *dst_length= di;
218     *consumed= si + 1;//+1 for the header
219 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
220     return dst;
221 }
222
223 /**
224  * Identify the exact end of the bitstream
225  * @return the length of the trailing, or 0 if damaged
226  */
227 static int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
228     int v= *src;
229     int r;
230
231     tprintf(h->s.avctx, "rbsp trailing %X\n", v);
232
233     for(r=1; r<9; r++){
234         if(v&1) return r;
235         v>>=1;
236     }
237     return 0;
238 }
239
240 static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n, int height,
241                                  int y_offset, int list){
242     int raw_my= h->mv_cache[list][ scan8[n] ][1];
243     int filter_height= (raw_my&3) ? 2 : 0;
244     int full_my= (raw_my>>2) + y_offset;
245     int top = full_my - filter_height, bottom = full_my + height + filter_height;
246
247     return FFMAX(abs(top), bottom);
248 }
249
250 static inline void get_lowest_part_y(H264Context *h, int refs[2][48], int n, int height,
251                                int y_offset, int list0, int list1, int *nrefs){
252     MpegEncContext * const s = &h->s;
253     int my;
254
255     y_offset += 16*(s->mb_y >> MB_FIELD);
256
257     if(list0){
258         int ref_n = h->ref_cache[0][ scan8[n] ];
259         Picture *ref= &h->ref_list[0][ref_n];
260
261         // Error resilience puts the current picture in the ref list.
262         // Don't try to wait on these as it will cause a deadlock.
263         // Fields can wait on each other, though.
264         if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
265            (ref->f.reference & 3) != s->picture_structure) {
266             my = get_lowest_part_list_y(h, ref, n, height, y_offset, 0);
267             if (refs[0][ref_n] < 0) nrefs[0] += 1;
268             refs[0][ref_n] = FFMAX(refs[0][ref_n], my);
269         }
270     }
271
272     if(list1){
273         int ref_n = h->ref_cache[1][ scan8[n] ];
274         Picture *ref= &h->ref_list[1][ref_n];
275
276         if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
277            (ref->f.reference & 3) != s->picture_structure) {
278             my = get_lowest_part_list_y(h, ref, n, height, y_offset, 1);
279             if (refs[1][ref_n] < 0) nrefs[1] += 1;
280             refs[1][ref_n] = FFMAX(refs[1][ref_n], my);
281         }
282     }
283 }
284
285 /**
286  * Wait until all reference frames are available for MC operations.
287  *
288  * @param h the H264 context
289  */
290 static void await_references(H264Context *h){
291     MpegEncContext * const s = &h->s;
292     const int mb_xy= h->mb_xy;
293     const int mb_type = s->current_picture.f.mb_type[mb_xy];
294     int refs[2][48];
295     int nrefs[2] = {0};
296     int ref, list;
297
298     memset(refs, -1, sizeof(refs));
299
300     if(IS_16X16(mb_type)){
301         get_lowest_part_y(h, refs, 0, 16, 0,
302                   IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
303     }else if(IS_16X8(mb_type)){
304         get_lowest_part_y(h, refs, 0, 8, 0,
305                   IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
306         get_lowest_part_y(h, refs, 8, 8, 8,
307                   IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
308     }else if(IS_8X16(mb_type)){
309         get_lowest_part_y(h, refs, 0, 16, 0,
310                   IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
311         get_lowest_part_y(h, refs, 4, 16, 0,
312                   IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
313     }else{
314         int i;
315
316         assert(IS_8X8(mb_type));
317
318         for(i=0; i<4; i++){
319             const int sub_mb_type= h->sub_mb_type[i];
320             const int n= 4*i;
321             int y_offset= (i&2)<<2;
322
323             if(IS_SUB_8X8(sub_mb_type)){
324                 get_lowest_part_y(h, refs, n  , 8, y_offset,
325                           IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
326             }else if(IS_SUB_8X4(sub_mb_type)){
327                 get_lowest_part_y(h, refs, n  , 4, y_offset,
328                           IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
329                 get_lowest_part_y(h, refs, n+2, 4, y_offset+4,
330                           IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
331             }else if(IS_SUB_4X8(sub_mb_type)){
332                 get_lowest_part_y(h, refs, n  , 8, y_offset,
333                           IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
334                 get_lowest_part_y(h, refs, n+1, 8, y_offset,
335                           IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
336             }else{
337                 int j;
338                 assert(IS_SUB_4X4(sub_mb_type));
339                 for(j=0; j<4; j++){
340                     int sub_y_offset= y_offset + 2*(j&2);
341                     get_lowest_part_y(h, refs, n+j, 4, sub_y_offset,
342                               IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
343                 }
344             }
345         }
346     }
347
348     for(list=h->list_count-1; list>=0; list--){
349         for(ref=0; ref<48 && nrefs[list]; ref++){
350             int row = refs[list][ref];
351             if(row >= 0){
352                 Picture *ref_pic = &h->ref_list[list][ref];
353                 int ref_field = ref_pic->f.reference - 1;
354                 int ref_field_picture = ref_pic->field_picture;
355                 int pic_height = 16*s->mb_height >> ref_field_picture;
356
357                 row <<= MB_MBAFF;
358                 nrefs[list]--;
359
360                 if(!FIELD_PICTURE && ref_field_picture){ // frame referencing two fields
361                     ff_thread_await_progress((AVFrame*)ref_pic, FFMIN((row >> 1) - !(row&1), pic_height-1), 1);
362                     ff_thread_await_progress((AVFrame*)ref_pic, FFMIN((row >> 1)           , pic_height-1), 0);
363                 }else if(FIELD_PICTURE && !ref_field_picture){ // field referencing one field of a frame
364                     ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row*2 + ref_field    , pic_height-1), 0);
365                 }else if(FIELD_PICTURE){
366                     ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row, pic_height-1), ref_field);
367                 }else{
368                     ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row, pic_height-1), 0);
369                 }
370             }
371         }
372     }
373 }
374
375 #if 0
376 /**
377  * DCT transforms the 16 dc values.
378  * @param qp quantization parameter ??? FIXME
379  */
380 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
381 //    const int qmul= dequant_coeff[qp][0];
382     int i;
383     int temp[16]; //FIXME check if this is a good idea
384     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
385     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
386
387     for(i=0; i<4; i++){
388         const int offset= y_offset[i];
389         const int z0= block[offset+stride*0] + block[offset+stride*4];
390         const int z1= block[offset+stride*0] - block[offset+stride*4];
391         const int z2= block[offset+stride*1] - block[offset+stride*5];
392         const int z3= block[offset+stride*1] + block[offset+stride*5];
393
394         temp[4*i+0]= z0+z3;
395         temp[4*i+1]= z1+z2;
396         temp[4*i+2]= z1-z2;
397         temp[4*i+3]= z0-z3;
398     }
399
400     for(i=0; i<4; i++){
401         const int offset= x_offset[i];
402         const int z0= temp[4*0+i] + temp[4*2+i];
403         const int z1= temp[4*0+i] - temp[4*2+i];
404         const int z2= temp[4*1+i] - temp[4*3+i];
405         const int z3= temp[4*1+i] + temp[4*3+i];
406
407         block[stride*0 +offset]= (z0 + z3)>>1;
408         block[stride*2 +offset]= (z1 + z2)>>1;
409         block[stride*8 +offset]= (z1 - z2)>>1;
410         block[stride*10+offset]= (z0 - z3)>>1;
411     }
412 }
413 #endif
414
415 #undef xStride
416 #undef stride
417
418 #if 0
419 static void chroma_dc_dct_c(DCTELEM *block){
420     const int stride= 16*2;
421     const int xStride= 16;
422     int a,b,c,d,e;
423
424     a= block[stride*0 + xStride*0];
425     b= block[stride*0 + xStride*1];
426     c= block[stride*1 + xStride*0];
427     d= block[stride*1 + xStride*1];
428
429     e= a-b;
430     a= a+b;
431     b= c-d;
432     c= c+d;
433
434     block[stride*0 + xStride*0]= (a+c);
435     block[stride*0 + xStride*1]= (e+b);
436     block[stride*1 + xStride*0]= (a-c);
437     block[stride*1 + xStride*1]= (e-b);
438 }
439 #endif
440
441 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square,
442                                int height, int delta, int list,
443                                uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
444                                int src_x_offset, int src_y_offset,
445                                qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op,
446                                int pixel_shift, int chroma_idc)
447 {
448     MpegEncContext * const s = &h->s;
449     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
450     int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
451     const int luma_xy= (mx&3) + ((my&3)<<2);
452     int offset = ((mx>>2) << pixel_shift) + (my>>2)*h->mb_linesize;
453     uint8_t * src_y = pic->f.data[0] + offset;
454     uint8_t * src_cb, * src_cr;
455     int extra_width= h->emu_edge_width;
456     int extra_height= h->emu_edge_height;
457     int emu=0;
458     const int full_mx= mx>>2;
459     const int full_my= my>>2;
460     const int pic_width  = 16*s->mb_width;
461     const int pic_height = 16*s->mb_height >> MB_FIELD;
462     int ysh;
463
464     if(mx&7) extra_width -= 3;
465     if(my&7) extra_height -= 3;
466
467     if(   full_mx < 0-extra_width
468        || full_my < 0-extra_height
469        || full_mx + 16/*FIXME*/ > pic_width + extra_width
470        || full_my + 16/*FIXME*/ > pic_height + extra_height){
471         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_y - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
472                                 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
473             src_y= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
474         emu=1;
475     }
476
477     qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
478     if(!square){
479         qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
480     }
481
482     if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
483
484     if(chroma_idc == 3 /* yuv444 */){
485         src_cb = pic->f.data[1] + offset;
486         if(emu){
487             s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
488                                     16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
489             src_cb= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
490         }
491         qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); //FIXME try variable height perhaps?
492         if(!square){
493             qpix_op[luma_xy](dest_cb + delta, src_cb + delta, h->mb_linesize);
494         }
495
496         src_cr = pic->f.data[2] + offset;
497         if(emu){
498             s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
499                                     16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
500             src_cr= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
501         }
502         qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); //FIXME try variable height perhaps?
503         if(!square){
504             qpix_op[luma_xy](dest_cr + delta, src_cr + delta, h->mb_linesize);
505         }
506         return;
507     }
508
509     ysh = 3 - (chroma_idc == 2 /* yuv422 */);
510     if(chroma_idc == 1 /* yuv420 */ && MB_FIELD){
511         // chroma offset when predicting from a field of opposite parity
512         my += 2 * ((s->mb_y & 1) - (pic->f.reference - 1));
513         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
514     }
515
516     src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) + (my >> ysh) * h->mb_uvlinesize;
517     src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) + (my >> ysh) * h->mb_uvlinesize;
518
519     if(emu){
520         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize,
521                                 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
522                                 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
523             src_cb= s->edge_emu_buffer;
524     }
525     chroma_op(dest_cb, src_cb, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
526               mx&7, (my << (chroma_idc == 2 /* yuv422 */)) &7);
527
528     if(emu){
529         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize,
530                                 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
531                                 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
532             src_cr= s->edge_emu_buffer;
533     }
534     chroma_op(dest_cr, src_cr, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
535               mx&7, (my << (chroma_idc == 2 /* yuv422 */)) &7);
536 }
537
538 static inline void mc_part_std(H264Context *h, int n, int square, int height, int delta,
539                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
540                            int x_offset, int y_offset,
541                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
542                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
543                            int list0, int list1, int pixel_shift, int chroma_idc)
544 {
545     MpegEncContext * const s = &h->s;
546     qpel_mc_func *qpix_op=  qpix_put;
547     h264_chroma_mc_func chroma_op= chroma_put;
548
549     dest_y  += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
550     if (chroma_idc == 3 /* yuv444 */) {
551         dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
552         dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
553     } else if (chroma_idc == 2 /* yuv422 */) {
554         dest_cb += (  x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
555         dest_cr += (  x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
556     } else /* yuv420 */ {
557         dest_cb += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
558         dest_cr += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
559     }
560     x_offset += 8*s->mb_x;
561     y_offset += 8*(s->mb_y >> MB_FIELD);
562
563     if(list0){
564         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
565         mc_dir_part(h, ref, n, square, height, delta, 0,
566                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
567                            qpix_op, chroma_op, pixel_shift, chroma_idc);
568
569         qpix_op=  qpix_avg;
570         chroma_op= chroma_avg;
571     }
572
573     if(list1){
574         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
575         mc_dir_part(h, ref, n, square, height, delta, 1,
576                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
577                            qpix_op, chroma_op, pixel_shift, chroma_idc);
578     }
579 }
580
581 static inline void mc_part_weighted(H264Context *h, int n, int square, int height, int delta,
582                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
583                            int x_offset, int y_offset,
584                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
585                            h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
586                            h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
587                            int list0, int list1, int pixel_shift, int chroma_idc){
588     MpegEncContext * const s = &h->s;
589     int chroma_height;
590
591     dest_y += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
592     if (chroma_idc == 3 /* yuv444 */) {
593         chroma_height = height;
594         chroma_weight_avg = luma_weight_avg;
595         chroma_weight_op = luma_weight_op;
596         dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
597         dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
598     } else if (chroma_idc == 2 /* yuv422 */) {
599         chroma_height = height;
600         dest_cb += (  x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
601         dest_cr += (  x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
602     } else /* yuv420 */ {
603         chroma_height = height >> 1;
604         dest_cb += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
605         dest_cr += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
606     }
607     x_offset += 8*s->mb_x;
608     y_offset += 8*(s->mb_y >> MB_FIELD);
609
610     if(list0 && list1){
611         /* don't optimize for luma-only case, since B-frames usually
612          * use implicit weights => chroma too. */
613         uint8_t *tmp_cb = s->obmc_scratchpad;
614         uint8_t *tmp_cr = s->obmc_scratchpad + (16 << pixel_shift);
615         uint8_t *tmp_y  = s->obmc_scratchpad + 16*h->mb_uvlinesize;
616         int refn0 = h->ref_cache[0][ scan8[n] ];
617         int refn1 = h->ref_cache[1][ scan8[n] ];
618
619         mc_dir_part(h, &h->ref_list[0][refn0], n, square, height, delta, 0,
620                     dest_y, dest_cb, dest_cr,
621                     x_offset, y_offset, qpix_put, chroma_put,
622                     pixel_shift, chroma_idc);
623         mc_dir_part(h, &h->ref_list[1][refn1], n, square, height, delta, 1,
624                     tmp_y, tmp_cb, tmp_cr,
625                     x_offset, y_offset, qpix_put, chroma_put,
626                     pixel_shift, chroma_idc);
627
628         if(h->use_weight == 2){
629             int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
630             int weight1 = 64 - weight0;
631             luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize,
632                               height,        5, weight0, weight1, 0);
633             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize,
634                               chroma_height, 5, weight0, weight1, 0);
635             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize,
636                               chroma_height, 5, weight0, weight1, 0);
637         }else{
638             luma_weight_avg(dest_y, tmp_y, h->mb_linesize, height, h->luma_log2_weight_denom,
639                             h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
640                             h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
641             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
642                             h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
643                             h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
644             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
645                             h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
646                             h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
647         }
648     }else{
649         int list = list1 ? 1 : 0;
650         int refn = h->ref_cache[list][ scan8[n] ];
651         Picture *ref= &h->ref_list[list][refn];
652         mc_dir_part(h, ref, n, square, height, delta, list,
653                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
654                     qpix_put, chroma_put, pixel_shift, chroma_idc);
655
656         luma_weight_op(dest_y, h->mb_linesize, height, h->luma_log2_weight_denom,
657                        h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
658         if(h->use_weight_chroma){
659             chroma_weight_op(dest_cb, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
660                              h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
661             chroma_weight_op(dest_cr, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
662                              h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
663         }
664     }
665 }
666
667 static inline void mc_part(H264Context *h, int n, int square, int height, int delta,
668                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
669                            int x_offset, int y_offset,
670                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
671                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
672                            h264_weight_func *weight_op, h264_biweight_func *weight_avg,
673                            int list0, int list1, int pixel_shift, int chroma_idc)
674 {
675     if((h->use_weight==2 && list0 && list1
676         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
677        || h->use_weight==1)
678         mc_part_weighted(h, n, square, height, delta, dest_y, dest_cb, dest_cr,
679                          x_offset, y_offset, qpix_put, chroma_put,
680                          weight_op[0], weight_op[1], weight_avg[0],
681                          weight_avg[1], list0, list1, pixel_shift, chroma_idc);
682     else
683         mc_part_std(h, n, square, height, delta, dest_y, dest_cb, dest_cr,
684                     x_offset, y_offset, qpix_put, chroma_put, qpix_avg,
685                     chroma_avg, list0, list1, pixel_shift, chroma_idc);
686 }
687
688 static inline void prefetch_motion(H264Context *h, int list, int pixel_shift, int chroma_idc){
689     /* fetch pixels for estimated mv 4 macroblocks ahead
690      * optimized for 64byte cache lines */
691     MpegEncContext * const s = &h->s;
692     const int refn = h->ref_cache[list][scan8[0]];
693     if(refn >= 0){
694         const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
695         const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
696         uint8_t **src = h->ref_list[list][refn].f.data;
697         int off= (mx << pixel_shift) + (my + (s->mb_x&3)*4)*h->mb_linesize + (64 << pixel_shift);
698         s->dsp.prefetch(src[0]+off, s->linesize, 4);
699         if (chroma_idc == 3 /* yuv444 */) {
700             s->dsp.prefetch(src[1]+off, s->linesize, 4);
701             s->dsp.prefetch(src[2]+off, s->linesize, 4);
702         }else{
703             off= ((mx>>1) << pixel_shift) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + (64 << pixel_shift);
704             s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
705         }
706     }
707 }
708
709 static av_always_inline void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
710                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
711                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
712                       h264_weight_func *weight_op, h264_biweight_func *weight_avg,
713                       int pixel_shift, int chroma_idc)
714 {
715     MpegEncContext * const s = &h->s;
716     const int mb_xy= h->mb_xy;
717     const int mb_type = s->current_picture.f.mb_type[mb_xy];
718
719     assert(IS_INTER(mb_type));
720
721     if(HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
722         await_references(h);
723     prefetch_motion(h, 0, pixel_shift, chroma_idc);
724
725     if(IS_16X16(mb_type)){
726         mc_part(h, 0, 1, 16, 0, dest_y, dest_cb, dest_cr, 0, 0,
727                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
728                 weight_op, weight_avg,
729                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
730                 pixel_shift, chroma_idc);
731     }else if(IS_16X8(mb_type)){
732         mc_part(h, 0, 0, 8, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 0,
733                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
734                 weight_op, weight_avg,
735                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
736                 pixel_shift, chroma_idc);
737         mc_part(h, 8, 0, 8, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 4,
738                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
739                 weight_op, weight_avg,
740                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
741                 pixel_shift, chroma_idc);
742     }else if(IS_8X16(mb_type)){
743         mc_part(h, 0, 0, 16, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
744                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
745                 &weight_op[1], &weight_avg[1],
746                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
747                 pixel_shift, chroma_idc);
748         mc_part(h, 4, 0, 16, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
749                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
750                 &weight_op[1], &weight_avg[1],
751                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
752                 pixel_shift, chroma_idc);
753     }else{
754         int i;
755
756         assert(IS_8X8(mb_type));
757
758         for(i=0; i<4; i++){
759             const int sub_mb_type= h->sub_mb_type[i];
760             const int n= 4*i;
761             int x_offset= (i&1)<<2;
762             int y_offset= (i&2)<<1;
763
764             if(IS_SUB_8X8(sub_mb_type)){
765                 mc_part(h, n, 1, 8, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
766                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
767                     &weight_op[1], &weight_avg[1],
768                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
769                     pixel_shift, chroma_idc);
770             }else if(IS_SUB_8X4(sub_mb_type)){
771                 mc_part(h, n  , 0, 4, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset,
772                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
773                     &weight_op[1], &weight_avg[1],
774                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
775                     pixel_shift, chroma_idc);
776                 mc_part(h, n+2, 0, 4, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
777                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
778                     &weight_op[1], &weight_avg[1],
779                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
780                     pixel_shift, chroma_idc);
781             }else if(IS_SUB_4X8(sub_mb_type)){
782                 mc_part(h, n  , 0, 8, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
783                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
784                     &weight_op[2], &weight_avg[2],
785                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
786                     pixel_shift, chroma_idc);
787                 mc_part(h, n+1, 0, 8, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
788                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
789                     &weight_op[2], &weight_avg[2],
790                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
791                     pixel_shift, chroma_idc);
792             }else{
793                 int j;
794                 assert(IS_SUB_4X4(sub_mb_type));
795                 for(j=0; j<4; j++){
796                     int sub_x_offset= x_offset + 2*(j&1);
797                     int sub_y_offset= y_offset +   (j&2);
798                     mc_part(h, n+j, 1, 4, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
799                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
800                         &weight_op[2], &weight_avg[2],
801                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
802                         pixel_shift, chroma_idc);
803                 }
804             }
805         }
806     }
807
808     prefetch_motion(h, 1, pixel_shift, chroma_idc);
809 }
810
811 static av_always_inline void
812 hl_motion_420(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
813               qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
814               qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
815               h264_weight_func *weight_op, h264_biweight_func *weight_avg,
816               int pixel_shift)
817 {
818     hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put,
819               qpix_avg, chroma_avg, weight_op, weight_avg, pixel_shift, 1);
820 }
821
822 static av_always_inline void
823 hl_motion_422(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
824               qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
825               qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
826               h264_weight_func *weight_op, h264_biweight_func *weight_avg,
827               int pixel_shift)
828 {
829     hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put,
830               qpix_avg, chroma_avg, weight_op, weight_avg, pixel_shift, 2);
831 }
832
833 static void free_tables(H264Context *h, int free_rbsp){
834     int i;
835     H264Context *hx;
836
837     av_freep(&h->intra4x4_pred_mode);
838     av_freep(&h->chroma_pred_mode_table);
839     av_freep(&h->cbp_table);
840     av_freep(&h->mvd_table[0]);
841     av_freep(&h->mvd_table[1]);
842     av_freep(&h->direct_table);
843     av_freep(&h->non_zero_count);
844     av_freep(&h->slice_table_base);
845     h->slice_table= NULL;
846     av_freep(&h->list_counts);
847
848     av_freep(&h->mb2b_xy);
849     av_freep(&h->mb2br_xy);
850
851     for(i = 0; i < MAX_THREADS; i++) {
852         hx = h->thread_context[i];
853         if(!hx) continue;
854         av_freep(&hx->top_borders[1]);
855         av_freep(&hx->top_borders[0]);
856         av_freep(&hx->s.obmc_scratchpad);
857         if (free_rbsp){
858             av_freep(&hx->rbsp_buffer[1]);
859             av_freep(&hx->rbsp_buffer[0]);
860             hx->rbsp_buffer_size[0] = 0;
861             hx->rbsp_buffer_size[1] = 0;
862         }
863         if (i) av_freep(&h->thread_context[i]);
864     }
865 }
866
867 static void init_dequant8_coeff_table(H264Context *h){
868     int i,j,q,x;
869     const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
870
871     for(i=0; i<6; i++ ){
872         h->dequant8_coeff[i] = h->dequant8_buffer[i];
873         for(j=0; j<i; j++){
874             if(!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i], 64*sizeof(uint8_t))){
875                 h->dequant8_coeff[i] = h->dequant8_buffer[j];
876                 break;
877             }
878         }
879         if(j<i)
880             continue;
881
882         for(q=0; q<max_qp+1; q++){
883             int shift = div6[q];
884             int idx = rem6[q];
885             for(x=0; x<64; x++)
886                 h->dequant8_coeff[i][q][(x>>3)|((x&7)<<3)] =
887                     ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
888                     h->pps.scaling_matrix8[i][x]) << shift;
889         }
890     }
891 }
892
893 static void init_dequant4_coeff_table(H264Context *h){
894     int i,j,q,x;
895     const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
896     for(i=0; i<6; i++ ){
897         h->dequant4_coeff[i] = h->dequant4_buffer[i];
898         for(j=0; j<i; j++){
899             if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
900                 h->dequant4_coeff[i] = h->dequant4_buffer[j];
901                 break;
902             }
903         }
904         if(j<i)
905             continue;
906
907         for(q=0; q<max_qp+1; q++){
908             int shift = div6[q] + 2;
909             int idx = rem6[q];
910             for(x=0; x<16; x++)
911                 h->dequant4_coeff[i][q][(x>>2)|((x<<2)&0xF)] =
912                     ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
913                     h->pps.scaling_matrix4[i][x]) << shift;
914         }
915     }
916 }
917
918 static void init_dequant_tables(H264Context *h){
919     int i,x;
920     init_dequant4_coeff_table(h);
921     if(h->pps.transform_8x8_mode)
922         init_dequant8_coeff_table(h);
923     if(h->sps.transform_bypass){
924         for(i=0; i<6; i++)
925             for(x=0; x<16; x++)
926                 h->dequant4_coeff[i][0][x] = 1<<6;
927         if(h->pps.transform_8x8_mode)
928             for(i=0; i<6; i++)
929                 for(x=0; x<64; x++)
930                     h->dequant8_coeff[i][0][x] = 1<<6;
931     }
932 }
933
934
935 int ff_h264_alloc_tables(H264Context *h){
936     MpegEncContext * const s = &h->s;
937     const int big_mb_num= s->mb_stride * (s->mb_height+1);
938     const int row_mb_num= 2*s->mb_stride*s->avctx->thread_count;
939     int x,y;
940
941     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8  * sizeof(uint8_t), fail)
942
943     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 48 * sizeof(uint8_t), fail)
944     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
945     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
946
947     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
948     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
949     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
950     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
951     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
952
953     memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
954     h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
955
956     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy  , big_mb_num * sizeof(uint32_t), fail);
957     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
958     for(y=0; y<s->mb_height; y++){
959         for(x=0; x<s->mb_width; x++){
960             const int mb_xy= x + y*s->mb_stride;
961             const int b_xy = 4*x + 4*y*h->b_stride;
962
963             h->mb2b_xy [mb_xy]= b_xy;
964             h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
965         }
966     }
967
968     s->obmc_scratchpad = NULL;
969
970     if(!h->dequant4_coeff[0])
971         init_dequant_tables(h);
972
973     return 0;
974 fail:
975     free_tables(h, 1);
976     return -1;
977 }
978
979 /**
980  * Mimic alloc_tables(), but for every context thread.
981  */
982 static void clone_tables(H264Context *dst, H264Context *src, int i){
983     MpegEncContext * const s = &src->s;
984     dst->intra4x4_pred_mode       = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
985     dst->non_zero_count           = src->non_zero_count;
986     dst->slice_table              = src->slice_table;
987     dst->cbp_table                = src->cbp_table;
988     dst->mb2b_xy                  = src->mb2b_xy;
989     dst->mb2br_xy                 = src->mb2br_xy;
990     dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
991     dst->mvd_table[0]             = src->mvd_table[0] + i*8*2*s->mb_stride;
992     dst->mvd_table[1]             = src->mvd_table[1] + i*8*2*s->mb_stride;
993     dst->direct_table             = src->direct_table;
994     dst->list_counts              = src->list_counts;
995
996     dst->s.obmc_scratchpad = NULL;
997     ff_h264_pred_init(&dst->hpc, src->s.codec_id, src->sps.bit_depth_luma, src->sps.chroma_format_idc);
998 }
999
1000 /**
1001  * Init context
1002  * Allocate buffers which are not shared amongst multiple threads.
1003  */
1004 static int context_init(H264Context *h){
1005     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
1006     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
1007
1008     h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
1009     h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
1010
1011     return 0;
1012 fail:
1013     return -1; // free_tables will clean up for us
1014 }
1015
1016 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
1017
1018 static av_cold void common_init(H264Context *h){
1019     MpegEncContext * const s = &h->s;
1020
1021     s->width = s->avctx->width;
1022     s->height = s->avctx->height;
1023     s->codec_id= s->avctx->codec->id;
1024
1025     ff_h264dsp_init(&h->h264dsp, 8, 1);
1026     ff_h264_pred_init(&h->hpc, s->codec_id, 8, 1);
1027
1028     h->dequant_coeff_pps= -1;
1029     s->unrestricted_mv=1;
1030     s->decode=1; //FIXME
1031
1032     dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
1033
1034     memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
1035     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
1036 }
1037
1038 int ff_h264_decode_extradata(H264Context *h)
1039 {
1040     AVCodecContext *avctx = h->s.avctx;
1041
1042     if(avctx->extradata[0] == 1){
1043         int i, cnt, nalsize;
1044         unsigned char *p = avctx->extradata;
1045
1046         h->is_avc = 1;
1047
1048         if(avctx->extradata_size < 7) {
1049             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
1050             return -1;
1051         }
1052         /* sps and pps in the avcC always have length coded with 2 bytes,
1053            so put a fake nal_length_size = 2 while parsing them */
1054         h->nal_length_size = 2;
1055         // Decode sps from avcC
1056         cnt = *(p+5) & 0x1f; // Number of sps
1057         p += 6;
1058         for (i = 0; i < cnt; i++) {
1059             nalsize = AV_RB16(p) + 2;
1060             if (p - avctx->extradata + nalsize > avctx->extradata_size)
1061                 return -1;
1062             if(decode_nal_units(h, p, nalsize) < 0) {
1063                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
1064                 return -1;
1065             }
1066             p += nalsize;
1067         }
1068         // Decode pps from avcC
1069         cnt = *(p++); // Number of pps
1070         for (i = 0; i < cnt; i++) {
1071             nalsize = AV_RB16(p) + 2;
1072             if (p - avctx->extradata + nalsize > avctx->extradata_size)
1073                 return -1;
1074             if (decode_nal_units(h, p, nalsize) < 0) {
1075                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
1076                 return -1;
1077             }
1078             p += nalsize;
1079         }
1080         // Now store right nal length size, that will be use to parse all other nals
1081         h->nal_length_size = (avctx->extradata[4] & 0x03) + 1;
1082     } else {
1083         h->is_avc = 0;
1084         if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
1085             return -1;
1086     }
1087     return 0;
1088 }
1089
1090 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
1091     H264Context *h= avctx->priv_data;
1092     MpegEncContext * const s = &h->s;
1093
1094     MPV_decode_defaults(s);
1095
1096     s->avctx = avctx;
1097     common_init(h);
1098
1099     s->out_format = FMT_H264;
1100     s->workaround_bugs= avctx->workaround_bugs;
1101
1102     // set defaults
1103 //    s->decode_mb= ff_h263_decode_mb;
1104     s->quarter_sample = 1;
1105     if(!avctx->has_b_frames)
1106     s->low_delay= 1;
1107
1108     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
1109
1110     ff_h264_decode_init_vlc();
1111
1112     h->pixel_shift = 0;
1113     h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
1114
1115     h->thread_context[0] = h;
1116     h->outputed_poc = h->next_outputed_poc = INT_MIN;
1117     h->prev_poc_msb= 1<<16;
1118     h->x264_build = -1;
1119     ff_h264_reset_sei(h);
1120     if(avctx->codec_id == CODEC_ID_H264){
1121         if(avctx->ticks_per_frame == 1){
1122             s->avctx->time_base.den *=2;
1123         }
1124         avctx->ticks_per_frame = 2;
1125     }
1126
1127     if(avctx->extradata_size > 0 && avctx->extradata &&
1128         ff_h264_decode_extradata(h))
1129         return -1;
1130
1131     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
1132         s->avctx->has_b_frames = h->sps.num_reorder_frames;
1133         s->low_delay = 0;
1134     }
1135
1136     return 0;
1137 }
1138
1139 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b)+(size))))
1140 static void copy_picture_range(Picture **to, Picture **from, int count, MpegEncContext *new_base, MpegEncContext *old_base)
1141 {
1142     int i;
1143
1144     for (i=0; i<count; i++){
1145         assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
1146                 IN_RANGE(from[i], old_base->picture, sizeof(Picture) * old_base->picture_count) ||
1147                 !from[i]));
1148         to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1149     }
1150 }
1151
1152 static void copy_parameter_set(void **to, void **from, int count, int size)
1153 {
1154     int i;
1155
1156     for (i=0; i<count; i++){
1157         if (to[i] && !from[i]) av_freep(&to[i]);
1158         else if (from[i] && !to[i]) to[i] = av_malloc(size);
1159
1160         if (from[i]) memcpy(to[i], from[i], size);
1161     }
1162 }
1163
1164 static int decode_init_thread_copy(AVCodecContext *avctx){
1165     H264Context *h= avctx->priv_data;
1166
1167     if (!avctx->is_copy) return 0;
1168     memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1169     memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1170
1171     return 0;
1172 }
1173
1174 #define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
1175 static int decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src){
1176     H264Context *h= dst->priv_data, *h1= src->priv_data;
1177     MpegEncContext * const s = &h->s, * const s1 = &h1->s;
1178     int inited = s->context_initialized, err;
1179     int i;
1180
1181     if(dst == src || !s1->context_initialized) return 0;
1182
1183     err = ff_mpeg_update_thread_context(dst, src);
1184     if(err) return err;
1185
1186     //FIXME handle width/height changing
1187     if(!inited){
1188         for(i = 0; i < MAX_SPS_COUNT; i++)
1189             av_freep(h->sps_buffers + i);
1190
1191         for(i = 0; i < MAX_PPS_COUNT; i++)
1192             av_freep(h->pps_buffers + i);
1193
1194         memcpy(&h->s + 1, &h1->s + 1, sizeof(H264Context) - sizeof(MpegEncContext)); //copy all fields after MpegEnc
1195         memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1196         memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1197         if (ff_h264_alloc_tables(h) < 0) {
1198             av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
1199             return AVERROR(ENOMEM);
1200         }
1201         context_init(h);
1202
1203         for(i=0; i<2; i++){
1204             h->rbsp_buffer[i] = NULL;
1205             h->rbsp_buffer_size[i] = 0;
1206         }
1207
1208         h->thread_context[0] = h;
1209
1210         // frame_start may not be called for the next thread (if it's decoding a bottom field)
1211         // so this has to be allocated here
1212         h->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
1213
1214         s->dsp.clear_blocks(h->mb);
1215         s->dsp.clear_blocks(h->mb+(24*16<<h->pixel_shift));
1216     }
1217
1218     //extradata/NAL handling
1219     h->is_avc          = h1->is_avc;
1220
1221     //SPS/PPS
1222     copy_parameter_set((void**)h->sps_buffers, (void**)h1->sps_buffers, MAX_SPS_COUNT, sizeof(SPS));
1223     h->sps             = h1->sps;
1224     copy_parameter_set((void**)h->pps_buffers, (void**)h1->pps_buffers, MAX_PPS_COUNT, sizeof(PPS));
1225     h->pps             = h1->pps;
1226
1227     //Dequantization matrices
1228     //FIXME these are big - can they be only copied when PPS changes?
1229     copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1230
1231     for(i=0; i<6; i++)
1232         h->dequant4_coeff[i] = h->dequant4_buffer[0] + (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
1233
1234     for(i=0; i<6; i++)
1235         h->dequant8_coeff[i] = h->dequant8_buffer[0] + (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
1236
1237     h->dequant_coeff_pps = h1->dequant_coeff_pps;
1238
1239     //POC timing
1240     copy_fields(h, h1, poc_lsb, redundant_pic_count);
1241
1242     //reference lists
1243     copy_fields(h, h1, ref_count, list_count);
1244     copy_fields(h, h1, ref_list,  intra_gb);
1245     copy_fields(h, h1, short_ref, cabac_init_idc);
1246
1247     copy_picture_range(h->short_ref,   h1->short_ref,   32, s, s1);
1248     copy_picture_range(h->long_ref,    h1->long_ref,    32, s, s1);
1249     copy_picture_range(h->delayed_pic, h1->delayed_pic, MAX_DELAYED_PIC_COUNT+2, s, s1);
1250
1251     h->last_slice_type = h1->last_slice_type;
1252
1253     if(!s->current_picture_ptr) return 0;
1254
1255     if(!s->dropable) {
1256         err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1257         h->prev_poc_msb     = h->poc_msb;
1258         h->prev_poc_lsb     = h->poc_lsb;
1259     }
1260     h->prev_frame_num_offset= h->frame_num_offset;
1261     h->prev_frame_num       = h->frame_num;
1262     h->outputed_poc         = h->next_outputed_poc;
1263
1264     return err;
1265 }
1266
1267 int ff_h264_frame_start(H264Context *h){
1268     MpegEncContext * const s = &h->s;
1269     int i;
1270     const int pixel_shift = h->pixel_shift;
1271     int thread_count = (s->avctx->active_thread_type & FF_THREAD_SLICE) ? s->avctx->thread_count : 1;
1272
1273     if(MPV_frame_start(s, s->avctx) < 0)
1274         return -1;
1275     ff_er_frame_start(s);
1276     /*
1277      * MPV_frame_start uses pict_type to derive key_frame.
1278      * This is incorrect for H.264; IDR markings must be used.
1279      * Zero here; IDR markings per slice in frame or fields are ORed in later.
1280      * See decode_nal_units().
1281      */
1282     s->current_picture_ptr->f.key_frame = 0;
1283     s->current_picture_ptr->mmco_reset= 0;
1284
1285     assert(s->linesize && s->uvlinesize);
1286
1287     for(i=0; i<16; i++){
1288         h->block_offset[i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
1289         h->block_offset[48+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
1290     }
1291     for(i=0; i<16; i++){
1292         h->block_offset[16+i]=
1293         h->block_offset[32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1294         h->block_offset[48+16+i]=
1295         h->block_offset[48+32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1296     }
1297
1298     /* can't be in alloc_tables because linesize isn't known there.
1299      * FIXME: redo bipred weight to not require extra buffer? */
1300     for(i = 0; i < thread_count; i++)
1301         if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
1302             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
1303
1304     /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
1305     memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
1306
1307 //    s->decode = (s->flags & CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.f.reference /*|| h->contains_intra*/ || 1;
1308
1309     // We mark the current picture as non-reference after allocating it, so
1310     // that if we break out due to an error it can be released automatically
1311     // in the next MPV_frame_start().
1312     // SVQ3 as well as most other codecs have only last/next/current and thus
1313     // get released even with set reference, besides SVQ3 and others do not
1314     // mark frames as reference later "naturally".
1315     if(s->codec_id != CODEC_ID_SVQ3)
1316         s->current_picture_ptr->f.reference = 0;
1317
1318     s->current_picture_ptr->field_poc[0]=
1319     s->current_picture_ptr->field_poc[1]= INT_MAX;
1320
1321     h->next_output_pic = NULL;
1322
1323     assert(s->current_picture_ptr->long_ref==0);
1324
1325     return 0;
1326 }
1327
1328 /**
1329   * Run setup operations that must be run after slice header decoding.
1330   * This includes finding the next displayed frame.
1331   *
1332   * @param h h264 master context
1333   * @param setup_finished enough NALs have been read that we can call
1334   * ff_thread_finish_setup()
1335   */
1336 static void decode_postinit(H264Context *h, int setup_finished){
1337     MpegEncContext * const s = &h->s;
1338     Picture *out = s->current_picture_ptr;
1339     Picture *cur = s->current_picture_ptr;
1340     int i, pics, out_of_order, out_idx;
1341
1342     s->current_picture_ptr->f.qscale_type = FF_QSCALE_TYPE_H264;
1343     s->current_picture_ptr->f.pict_type   = s->pict_type;
1344
1345     if (h->next_output_pic) return;
1346
1347     if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
1348         //FIXME: if we have two PAFF fields in one packet, we can't start the next thread here.
1349         //If we have one field per packet, we can. The check in decode_nal_units() is not good enough
1350         //to find this yet, so we assume the worst for now.
1351         //if (setup_finished)
1352         //    ff_thread_finish_setup(s->avctx);
1353         return;
1354     }
1355
1356     cur->f.interlaced_frame = 0;
1357     cur->f.repeat_pict      = 0;
1358
1359     /* Signal interlacing information externally. */
1360     /* Prioritize picture timing SEI information over used decoding process if it exists. */
1361
1362     if(h->sps.pic_struct_present_flag){
1363         switch (h->sei_pic_struct)
1364         {
1365         case SEI_PIC_STRUCT_FRAME:
1366             break;
1367         case SEI_PIC_STRUCT_TOP_FIELD:
1368         case SEI_PIC_STRUCT_BOTTOM_FIELD:
1369             cur->f.interlaced_frame = 1;
1370             break;
1371         case SEI_PIC_STRUCT_TOP_BOTTOM:
1372         case SEI_PIC_STRUCT_BOTTOM_TOP:
1373             if (FIELD_OR_MBAFF_PICTURE)
1374                 cur->f.interlaced_frame = 1;
1375             else
1376                 // try to flag soft telecine progressive
1377                 cur->f.interlaced_frame = h->prev_interlaced_frame;
1378             break;
1379         case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1380         case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
1381             // Signal the possibility of telecined film externally (pic_struct 5,6)
1382             // From these hints, let the applications decide if they apply deinterlacing.
1383             cur->f.repeat_pict = 1;
1384             break;
1385         case SEI_PIC_STRUCT_FRAME_DOUBLING:
1386             // Force progressive here, as doubling interlaced frame is a bad idea.
1387             cur->f.repeat_pict = 2;
1388             break;
1389         case SEI_PIC_STRUCT_FRAME_TRIPLING:
1390             cur->f.repeat_pict = 4;
1391             break;
1392         }
1393
1394         if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
1395             cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
1396     }else{
1397         /* Derive interlacing flag from used decoding process. */
1398         cur->f.interlaced_frame = FIELD_OR_MBAFF_PICTURE;
1399     }
1400     h->prev_interlaced_frame = cur->f.interlaced_frame;
1401
1402     if (cur->field_poc[0] != cur->field_poc[1]){
1403         /* Derive top_field_first from field pocs. */
1404         cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1];
1405     }else{
1406         if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) {
1407             /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
1408             if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
1409               || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
1410                 cur->f.top_field_first = 1;
1411             else
1412                 cur->f.top_field_first = 0;
1413         }else{
1414             /* Most likely progressive */
1415             cur->f.top_field_first = 0;
1416         }
1417     }
1418
1419     //FIXME do something with unavailable reference frames
1420
1421     /* Sort B-frames into display order */
1422
1423     if(h->sps.bitstream_restriction_flag
1424        && s->avctx->has_b_frames < h->sps.num_reorder_frames){
1425         s->avctx->has_b_frames = h->sps.num_reorder_frames;
1426         s->low_delay = 0;
1427     }
1428
1429     if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
1430        && !h->sps.bitstream_restriction_flag){
1431         s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
1432         s->low_delay= 0;
1433     }
1434
1435     pics = 0;
1436     while(h->delayed_pic[pics]) pics++;
1437
1438     assert(pics <= MAX_DELAYED_PIC_COUNT);
1439
1440     h->delayed_pic[pics++] = cur;
1441     if (cur->f.reference == 0)
1442         cur->f.reference = DELAYED_PIC_REF;
1443
1444     out = h->delayed_pic[0];
1445     out_idx = 0;
1446     for (i = 1; h->delayed_pic[i] && !h->delayed_pic[i]->f.key_frame && !h->delayed_pic[i]->mmco_reset; i++)
1447         if(h->delayed_pic[i]->poc < out->poc){
1448             out = h->delayed_pic[i];
1449             out_idx = i;
1450         }
1451     if (s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset))
1452         h->next_outputed_poc= INT_MIN;
1453     out_of_order = out->poc < h->next_outputed_poc;
1454
1455     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
1456         { }
1457     else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
1458        || (s->low_delay &&
1459         ((h->next_outputed_poc != INT_MIN && out->poc > h->next_outputed_poc + 2)
1460          || cur->f.pict_type == AV_PICTURE_TYPE_B)))
1461     {
1462         s->low_delay = 0;
1463         s->avctx->has_b_frames++;
1464     }
1465
1466     if(out_of_order || pics > s->avctx->has_b_frames){
1467         out->f.reference &= ~DELAYED_PIC_REF;
1468         out->owner2 = s; // for frame threading, the owner must be the second field's thread
1469                          // or else the first thread can release the picture and reuse it unsafely
1470         for(i=out_idx; h->delayed_pic[i]; i++)
1471             h->delayed_pic[i] = h->delayed_pic[i+1];
1472     }
1473     if(!out_of_order && pics > s->avctx->has_b_frames){
1474         h->next_output_pic = out;
1475         if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset)) {
1476             h->next_outputed_poc = INT_MIN;
1477         } else
1478             h->next_outputed_poc = out->poc;
1479     }else{
1480         av_log(s->avctx, AV_LOG_DEBUG, "no picture\n");
1481     }
1482
1483     if (setup_finished)
1484         ff_thread_finish_setup(s->avctx);
1485 }
1486
1487 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
1488                                               uint8_t *src_cb, uint8_t *src_cr,
1489                                               int linesize, int uvlinesize, int simple)
1490 {
1491     MpegEncContext * const s = &h->s;
1492     uint8_t *top_border;
1493     int top_idx = 1;
1494     const int pixel_shift = h->pixel_shift;
1495     int chroma444 = CHROMA444;
1496     int chroma422 = CHROMA422;
1497
1498     src_y  -=   linesize;
1499     src_cb -= uvlinesize;
1500     src_cr -= uvlinesize;
1501
1502     if(!simple && FRAME_MBAFF){
1503         if(s->mb_y&1){
1504             if(!MB_MBAFF){
1505                 top_border = h->top_borders[0][s->mb_x];
1506                 AV_COPY128(top_border, src_y + 15*linesize);
1507                 if (pixel_shift)
1508                     AV_COPY128(top_border+16, src_y+15*linesize+16);
1509                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1510                     if(chroma444){
1511                         if (pixel_shift){
1512                             AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
1513                             AV_COPY128(top_border+48, src_cb + 15*uvlinesize+16);
1514                             AV_COPY128(top_border+64, src_cr + 15*uvlinesize);
1515                             AV_COPY128(top_border+80, src_cr + 15*uvlinesize+16);
1516                         } else {
1517                             AV_COPY128(top_border+16, src_cb + 15*uvlinesize);
1518                             AV_COPY128(top_border+32, src_cr + 15*uvlinesize);
1519                         }
1520                     } else if(chroma422) {
1521                         if (pixel_shift) {
1522                             AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
1523                             AV_COPY128(top_border+48, src_cr + 15*uvlinesize);
1524                         } else {
1525                             AV_COPY64(top_border+16, src_cb +  15*uvlinesize);
1526                             AV_COPY64(top_border+24, src_cr +  15*uvlinesize);
1527                         }
1528                     } else {
1529                         if (pixel_shift) {
1530                             AV_COPY128(top_border+32, src_cb+7*uvlinesize);
1531                             AV_COPY128(top_border+48, src_cr+7*uvlinesize);
1532                         } else {
1533                             AV_COPY64(top_border+16, src_cb+7*uvlinesize);
1534                             AV_COPY64(top_border+24, src_cr+7*uvlinesize);
1535                         }
1536                     }
1537                 }
1538             }
1539         }else if(MB_MBAFF){
1540             top_idx = 0;
1541         }else
1542             return;
1543     }
1544
1545     top_border = h->top_borders[top_idx][s->mb_x];
1546     // There are two lines saved, the line above the the top macroblock of a pair,
1547     // and the line above the bottom macroblock
1548     AV_COPY128(top_border, src_y + 16*linesize);
1549     if (pixel_shift)
1550         AV_COPY128(top_border+16, src_y+16*linesize+16);
1551
1552     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1553         if(chroma444){
1554             if (pixel_shift){
1555                 AV_COPY128(top_border+32, src_cb + 16*linesize);
1556                 AV_COPY128(top_border+48, src_cb + 16*linesize+16);
1557                 AV_COPY128(top_border+64, src_cr + 16*linesize);
1558                 AV_COPY128(top_border+80, src_cr + 16*linesize+16);
1559             } else {
1560                 AV_COPY128(top_border+16, src_cb + 16*linesize);
1561                 AV_COPY128(top_border+32, src_cr + 16*linesize);
1562             }
1563         } else if(chroma422) {
1564             if (pixel_shift) {
1565                 AV_COPY128(top_border+32, src_cb+16*uvlinesize);
1566                 AV_COPY128(top_border+48, src_cr+16*uvlinesize);
1567             } else {
1568                 AV_COPY64(top_border+16, src_cb+16*uvlinesize);
1569                 AV_COPY64(top_border+24, src_cr+16*uvlinesize);
1570             }
1571         } else {
1572             if (pixel_shift) {
1573                 AV_COPY128(top_border+32, src_cb+8*uvlinesize);
1574                 AV_COPY128(top_border+48, src_cr+8*uvlinesize);
1575             } else {
1576                 AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1577                 AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1578             }
1579         }
1580     }
1581 }
1582
1583 static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
1584                                   uint8_t *src_cb, uint8_t *src_cr,
1585                                   int linesize, int uvlinesize,
1586                                   int xchg, int chroma444,
1587                                   int simple, int pixel_shift){
1588     MpegEncContext * const s = &h->s;
1589     int deblock_topleft;
1590     int deblock_top;
1591     int top_idx = 1;
1592     uint8_t *top_border_m1;
1593     uint8_t *top_border;
1594
1595     if(!simple && FRAME_MBAFF){
1596         if(s->mb_y&1){
1597             if(!MB_MBAFF)
1598                 return;
1599         }else{
1600             top_idx = MB_MBAFF ? 0 : 1;
1601         }
1602     }
1603
1604     if(h->deblocking_filter == 2) {
1605         deblock_topleft = h->slice_table[h->mb_xy - 1 - s->mb_stride] == h->slice_num;
1606         deblock_top     = h->top_type;
1607     } else {
1608         deblock_topleft = (s->mb_x > 0);
1609         deblock_top     = (s->mb_y > !!MB_FIELD);
1610     }
1611
1612     src_y  -=   linesize + 1 + pixel_shift;
1613     src_cb -= uvlinesize + 1 + pixel_shift;
1614     src_cr -= uvlinesize + 1 + pixel_shift;
1615
1616     top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1617     top_border    = h->top_borders[top_idx][s->mb_x];
1618
1619 #define XCHG(a,b,xchg)\
1620     if (pixel_shift) {\
1621         if (xchg) {\
1622             AV_SWAP64(b+0,a+0);\
1623             AV_SWAP64(b+8,a+8);\
1624         } else {\
1625             AV_COPY128(b,a); \
1626         }\
1627     } else \
1628 if (xchg) AV_SWAP64(b,a);\
1629 else      AV_COPY64(b,a);
1630
1631     if(deblock_top){
1632         if(deblock_topleft){
1633             XCHG(top_border_m1 + (8 << pixel_shift), src_y - (7 << pixel_shift), 1);
1634         }
1635         XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
1636         XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
1637         if(s->mb_x+1 < s->mb_width){
1638             XCHG(h->top_borders[top_idx][s->mb_x+1], src_y + (17 << pixel_shift), 1);
1639         }
1640     }
1641     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1642         if(chroma444){
1643             if(deblock_topleft){
1644                 XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1645                 XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1646             }
1647             XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
1648             XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
1649             XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
1650             XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
1651             if(s->mb_x+1 < s->mb_width){
1652                 XCHG(h->top_borders[top_idx][s->mb_x+1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
1653                 XCHG(h->top_borders[top_idx][s->mb_x+1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
1654             }
1655         } else {
1656             if(deblock_top){
1657                 if(deblock_topleft){
1658                     XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1659                     XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1660                 }
1661                 XCHG(top_border + (16 << pixel_shift), src_cb+1+pixel_shift, 1);
1662                 XCHG(top_border + (24 << pixel_shift), src_cr+1+pixel_shift, 1);
1663             }
1664         }
1665     }
1666 }
1667
1668 static av_always_inline int dctcoef_get(DCTELEM *mb, int high_bit_depth, int index) {
1669     if (high_bit_depth) {
1670         return AV_RN32A(((int32_t*)mb) + index);
1671     } else
1672         return AV_RN16A(mb + index);
1673 }
1674
1675 static av_always_inline void dctcoef_set(DCTELEM *mb, int high_bit_depth, int index, int value) {
1676     if (high_bit_depth) {
1677         AV_WN32A(((int32_t*)mb) + index, value);
1678     } else
1679         AV_WN16A(mb + index, value);
1680 }
1681
1682 static av_always_inline void hl_decode_mb_predict_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1683                                                        int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1684 {
1685     MpegEncContext * const s = &h->s;
1686     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1687     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1688     int i;
1689     int qscale = p == 0 ? s->qscale : h->chroma_qp[p-1];
1690     block_offset += 16*p;
1691     if(IS_INTRA4x4(mb_type)){
1692         if(simple || !s->encoding){
1693             if(IS_8x8DCT(mb_type)){
1694                 if(transform_bypass){
1695                     idct_dc_add =
1696                     idct_add    = s->dsp.add_pixels8;
1697                 }else{
1698                     idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1699                     idct_add    = h->h264dsp.h264_idct8_add;
1700                 }
1701                 for(i=0; i<16; i+=4){
1702                     uint8_t * const ptr= dest_y + block_offset[i];
1703                     const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1704                     if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1705                         h->hpc.pred8x8l_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1706                     }else{
1707                         const int nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1708                         h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1709                                                     (h->topright_samples_available<<i)&0x4000, linesize);
1710                         if(nnz){
1711                             if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1712                                 idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1713                             else
1714                                 idct_add   (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1715                         }
1716                     }
1717                 }
1718             }else{
1719                 if(transform_bypass){
1720                     idct_dc_add =
1721                     idct_add    = s->dsp.add_pixels4;
1722                 }else{
1723                     idct_dc_add = h->h264dsp.h264_idct_dc_add;
1724                     idct_add    = h->h264dsp.h264_idct_add;
1725                 }
1726                 for(i=0; i<16; i++){
1727                     uint8_t * const ptr= dest_y + block_offset[i];
1728                     const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1729
1730                     if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1731                         h->hpc.pred4x4_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1732                     }else{
1733                         uint8_t *topright;
1734                         int nnz, tr;
1735                         uint64_t tr_high;
1736                         if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1737                             const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1738                             assert(s->mb_y || linesize <= block_offset[i]);
1739                             if(!topright_avail){
1740                                 if (pixel_shift) {
1741                                     tr_high= ((uint16_t*)ptr)[3 - linesize/2]*0x0001000100010001ULL;
1742                                     topright= (uint8_t*) &tr_high;
1743                                 } else {
1744                                     tr= ptr[3 - linesize]*0x01010101u;
1745                                     topright= (uint8_t*) &tr;
1746                                 }
1747                             }else
1748                                 topright= ptr + (4 << pixel_shift) - linesize;
1749                         }else
1750                             topright= NULL;
1751
1752                         h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1753                         nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1754                         if(nnz){
1755                             if(is_h264){
1756                                 if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1757                                     idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1758                                 else
1759                                     idct_add   (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1760                             }else
1761                                 ff_svq3_add_idct_c(ptr, h->mb + i*16+p*256, linesize, qscale, 0);
1762                         }
1763                     }
1764                 }
1765             }
1766         }
1767     }else{
1768         h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1769         if(is_h264){
1770             if(h->non_zero_count_cache[ scan8[LUMA_DC_BLOCK_INDEX+p] ]){
1771                 if(!transform_bypass)
1772                     h->h264dsp.h264_luma_dc_dequant_idct(h->mb+(p*256 << pixel_shift), h->mb_luma_dc[p], h->dequant4_coeff[p][qscale][0]);
1773                 else{
1774                     static const uint8_t dc_mapping[16] = { 0*16, 1*16, 4*16, 5*16, 2*16, 3*16, 6*16, 7*16,
1775                                                             8*16, 9*16,12*16,13*16,10*16,11*16,14*16,15*16};
1776                     for(i = 0; i < 16; i++)
1777                         dctcoef_set(h->mb+p*256, pixel_shift, dc_mapping[i], dctcoef_get(h->mb_luma_dc[p], pixel_shift, i));
1778                 }
1779             }
1780         }else
1781             ff_svq3_luma_dc_dequant_idct_c(h->mb+p*256, h->mb_luma_dc[p], qscale);
1782     }
1783 }
1784
1785 static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1786                                                     int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1787 {
1788     MpegEncContext * const s = &h->s;
1789     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1790     int i;
1791     block_offset += 16*p;
1792     if(!IS_INTRA4x4(mb_type)){
1793         if(is_h264){
1794             if(IS_INTRA16x16(mb_type)){
1795                 if(transform_bypass){
1796                     if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1797                         h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize);
1798                     }else{
1799                         for(i=0; i<16; i++){
1800                             if(h->non_zero_count_cache[ scan8[i+p*16] ] || dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1801                                 s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1802                         }
1803                     }
1804                 }else{
1805                     h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1806                 }
1807             }else if(h->cbp&15){
1808                 if(transform_bypass){
1809                     const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1810                     idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1811                     for(i=0; i<16; i+=di){
1812                         if(h->non_zero_count_cache[ scan8[i+p*16] ]){
1813                             idct_add(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1814                         }
1815                     }
1816                 }else{
1817                     if(IS_8x8DCT(mb_type)){
1818                         h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1819                     }else{
1820                         h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1821                     }
1822                 }
1823             }
1824         }else{
1825             for(i=0; i<16; i++){
1826                 if(h->non_zero_count_cache[ scan8[i+p*16] ] || h->mb[i*16+p*256]){ //FIXME benchmark weird rule, & below
1827                     uint8_t * const ptr= dest_y + block_offset[i];
1828                     ff_svq3_add_idct_c(ptr, h->mb + i*16 + p*256, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1829                 }
1830             }
1831         }
1832     }
1833 }
1834
1835 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, int pixel_shift)
1836 {
1837     MpegEncContext * const s = &h->s;
1838     const int mb_x= s->mb_x;
1839     const int mb_y= s->mb_y;
1840     const int mb_xy= h->mb_xy;
1841     const int mb_type = s->current_picture.f.mb_type[mb_xy];
1842     uint8_t  *dest_y, *dest_cb, *dest_cr;
1843     int linesize, uvlinesize /*dct_offset*/;
1844     int i, j;
1845     int *block_offset = &h->block_offset[0];
1846     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1847     /* is_h264 should always be true if SVQ3 is disabled. */
1848     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1849     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1850     const int block_h = 16 >> s->chroma_y_shift;
1851     const int chroma422 = CHROMA422;
1852
1853     dest_y  = s->current_picture.f.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize  ) * 16;
1854     dest_cb = s->current_picture.f.data[1] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
1855     dest_cr = s->current_picture.f.data[2] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
1856
1857     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
1858     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + (64 << pixel_shift), dest_cr - dest_cb, 2);
1859
1860     h->list_counts[mb_xy]= h->list_count;
1861
1862     if (!simple && MB_FIELD) {
1863         linesize   = h->mb_linesize   = s->linesize * 2;
1864         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1865         block_offset = &h->block_offset[48];
1866         if(mb_y&1){ //FIXME move out of this function?
1867             dest_y -= s->linesize*15;
1868             dest_cb-= s->uvlinesize * (block_h - 1);
1869             dest_cr-= s->uvlinesize * (block_h - 1);
1870         }
1871         if(FRAME_MBAFF) {
1872             int list;
1873             for(list=0; list<h->list_count; list++){
1874                 if(!USES_LIST(mb_type, list))
1875                     continue;
1876                 if(IS_16X16(mb_type)){
1877                     int8_t *ref = &h->ref_cache[list][scan8[0]];
1878                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1879                 }else{
1880                     for(i=0; i<16; i+=4){
1881                         int ref = h->ref_cache[list][scan8[i]];
1882                         if(ref >= 0)
1883                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1884                     }
1885                 }
1886             }
1887         }
1888     } else {
1889         linesize   = h->mb_linesize   = s->linesize;
1890         uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1891 //        dct_offset = s->linesize * 16;
1892     }
1893
1894     if (!simple && IS_INTRA_PCM(mb_type)) {
1895         if (pixel_shift) {
1896             const int bit_depth = h->sps.bit_depth_luma;
1897             int j;
1898             GetBitContext gb;
1899             init_get_bits(&gb, (uint8_t*)h->mb, 384*bit_depth);
1900
1901             for (i = 0; i < 16; i++) {
1902                 uint16_t *tmp_y  = (uint16_t*)(dest_y  + i*linesize);
1903                 for (j = 0; j < 16; j++)
1904                     tmp_y[j] = get_bits(&gb, bit_depth);
1905             }
1906             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1907                 if (!h->sps.chroma_format_idc) {
1908                     for (i = 0; i < block_h; i++) {
1909                         uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1910                         for (j = 0; j < 8; j++) {
1911                             tmp_cb[j] = 1 << (bit_depth - 1);
1912                         }
1913                     }
1914                     for (i = 0; i < block_h; i++) {
1915                         uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1916                         for (j = 0; j < 8; j++) {
1917                             tmp_cr[j] = 1 << (bit_depth - 1);
1918                         }
1919                     }
1920                 } else {
1921                     for (i = 0; i < block_h; i++) {
1922                         uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1923                         for (j = 0; j < 8; j++)
1924                             tmp_cb[j] = get_bits(&gb, bit_depth);
1925                     }
1926                     for (i = 0; i < block_h; i++) {
1927                         uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1928                         for (j = 0; j < 8; j++)
1929                             tmp_cr[j] = get_bits(&gb, bit_depth);
1930                     }
1931                 }
1932             }
1933         } else {
1934             for (i=0; i<16; i++) {
1935                 memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1936             }
1937             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1938                 if (!h->sps.chroma_format_idc) {
1939                     for (i = 0; i < block_h; i++) {
1940                         memset(dest_cb + i*uvlinesize, 128, 8);
1941                         memset(dest_cr + i*uvlinesize, 128, 8);
1942                     }
1943                 } else {
1944                     for (i = 0; i < block_h; i++) {
1945                         memcpy(dest_cb + i*uvlinesize, h->mb + 128 + i*4,  8);
1946                         memcpy(dest_cr + i*uvlinesize, h->mb + 160 + i*4,  8);
1947                     }
1948                 }
1949             }
1950         }
1951     } else {
1952         if(IS_INTRA(mb_type)){
1953             if(h->deblocking_filter)
1954                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, 0, simple, pixel_shift);
1955
1956             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1957                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1958                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1959             }
1960
1961             hl_decode_mb_predict_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
1962
1963             if(h->deblocking_filter)
1964                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, 0, simple, pixel_shift);
1965         }else if(is_h264){
1966             if (chroma422) {
1967                 hl_motion_422(h, dest_y, dest_cb, dest_cr,
1968                               s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1969                               s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1970                               h->h264dsp.weight_h264_pixels_tab,
1971                               h->h264dsp.biweight_h264_pixels_tab,
1972                               pixel_shift);
1973             } else {
1974                 hl_motion_420(h, dest_y, dest_cb, dest_cr,
1975                               s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1976                               s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1977                               h->h264dsp.weight_h264_pixels_tab,
1978                               h->h264dsp.biweight_h264_pixels_tab,
1979                               pixel_shift);
1980             }
1981         }
1982
1983         hl_decode_mb_idct_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
1984
1985         if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1986             uint8_t *dest[2] = {dest_cb, dest_cr};
1987             if(transform_bypass){
1988                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1989                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + (16*16*1 << pixel_shift), uvlinesize);
1990                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 32, h->mb + (16*16*2 << pixel_shift), uvlinesize);
1991                 }else{
1992                     idct_add = s->dsp.add_pixels4;
1993                     for(j=1; j<3; j++){
1994                         for(i=j*16; i<j*16+4; i++){
1995                             if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
1996                                 idct_add   (dest[j-1] + block_offset[i], h->mb + (i*16 << pixel_shift), uvlinesize);
1997                         }
1998                         if (chroma422) {
1999                             for(i=j*16+4; i<j*16+8; i++){
2000                                 if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
2001                                     idct_add   (dest[j-1] + block_offset[i+4], h->mb + (i*16 << pixel_shift), uvlinesize);
2002                             }
2003                         }
2004                     }
2005                 }
2006             }else{
2007                 if(is_h264){
2008                     int qp[2];
2009                     if (chroma422) {
2010                         qp[0] = h->chroma_qp[0] + 3;
2011                         qp[1] = h->chroma_qp[1] + 3;
2012                     } else {
2013                         qp[0] = h->chroma_qp[0];
2014                         qp[1] = h->chroma_qp[1];
2015                     }
2016                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+0] ])
2017                         h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16*1 << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][qp[0]][0]);
2018                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+1] ])
2019                         h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16*2 << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][qp[1]][0]);
2020                     h->h264dsp.h264_idct_add8(dest, block_offset,
2021                                               h->mb, uvlinesize,
2022                                               h->non_zero_count_cache);
2023                 }else{
2024                     h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16*1, h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
2025                     h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16*2, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
2026                     for(j=1; j<3; j++){
2027                         for(i=j*16; i<j*16+4; i++){
2028                             if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2029                                 uint8_t * const ptr= dest[j-1] + block_offset[i];
2030                                 ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[0][s->qscale + 12] - 12, 2);
2031                             }
2032                         }
2033                     }
2034                 }
2035             }
2036         }
2037     }
2038     if(h->cbp || IS_INTRA(mb_type))
2039     {
2040         s->dsp.clear_blocks(h->mb);
2041         s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
2042     }
2043 }
2044
2045 static av_always_inline void hl_decode_mb_444_internal(H264Context *h, int simple, int pixel_shift){
2046     MpegEncContext * const s = &h->s;
2047     const int mb_x= s->mb_x;
2048     const int mb_y= s->mb_y;
2049     const int mb_xy= h->mb_xy;
2050     const int mb_type = s->current_picture.f.mb_type[mb_xy];
2051     uint8_t  *dest[3];
2052     int linesize;
2053     int i, j, p;
2054     int *block_offset = &h->block_offset[0];
2055     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
2056     const int plane_count = (simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) ? 3 : 1;
2057
2058     for (p = 0; p < plane_count; p++)
2059     {
2060         dest[p] = s->current_picture.f.data[p] + ((mb_x << pixel_shift) + mb_y * s->linesize) * 16;
2061         s->dsp.prefetch(dest[p] + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
2062     }
2063
2064     h->list_counts[mb_xy]= h->list_count;
2065
2066     if (!simple && MB_FIELD) {
2067         linesize   = h->mb_linesize = h->mb_uvlinesize = s->linesize * 2;
2068         block_offset = &h->block_offset[48];
2069         if(mb_y&1) //FIXME move out of this function?
2070             for (p = 0; p < 3; p++)
2071                 dest[p] -= s->linesize*15;
2072         if(FRAME_MBAFF) {
2073             int list;
2074             for(list=0; list<h->list_count; list++){
2075                 if(!USES_LIST(mb_type, list))
2076                     continue;
2077                 if(IS_16X16(mb_type)){
2078                     int8_t *ref = &h->ref_cache[list][scan8[0]];
2079                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
2080                 }else{
2081                     for(i=0; i<16; i+=4){
2082                         int ref = h->ref_cache[list][scan8[i]];
2083                         if(ref >= 0)
2084                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
2085                     }
2086                 }
2087             }
2088         }
2089     } else {
2090         linesize   = h->mb_linesize = h->mb_uvlinesize = s->linesize;
2091     }
2092
2093     if (!simple && IS_INTRA_PCM(mb_type)) {
2094         if (pixel_shift) {
2095             const int bit_depth = h->sps.bit_depth_luma;
2096             GetBitContext gb;
2097             init_get_bits(&gb, (uint8_t*)h->mb, 768*bit_depth);
2098
2099             for (p = 0; p < plane_count; p++) {
2100                 for (i = 0; i < 16; i++) {
2101                     uint16_t *tmp = (uint16_t*)(dest[p] + i*linesize);
2102                     for (j = 0; j < 16; j++)
2103                         tmp[j] = get_bits(&gb, bit_depth);
2104                 }
2105             }
2106         } else {
2107             for (p = 0; p < plane_count; p++) {
2108                 for (i = 0; i < 16; i++) {
2109                     memcpy(dest[p] + i*linesize, h->mb + p*128 + i*8, 16);
2110                 }
2111             }
2112         }
2113     } else {
2114         if(IS_INTRA(mb_type)){
2115             if(h->deblocking_filter)
2116                 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 1, 1, simple, pixel_shift);
2117
2118             for (p = 0; p < plane_count; p++)
2119                 hl_decode_mb_predict_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2120
2121             if(h->deblocking_filter)
2122                 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 0, 1, simple, pixel_shift);
2123         }else{
2124             hl_motion(h, dest[0], dest[1], dest[2],
2125                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2126                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2127                       h->h264dsp.weight_h264_pixels_tab,
2128                       h->h264dsp.biweight_h264_pixels_tab, pixel_shift, 3);
2129         }
2130
2131         for (p = 0; p < plane_count; p++)
2132             hl_decode_mb_idct_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2133     }
2134     if(h->cbp || IS_INTRA(mb_type))
2135     {
2136         s->dsp.clear_blocks(h->mb);
2137         s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
2138     }
2139 }
2140
2141 /**
2142  * Process a macroblock; this case avoids checks for expensive uncommon cases.
2143  */
2144 #define hl_decode_mb_simple(sh, bits) \
2145 static void hl_decode_mb_simple_ ## bits(H264Context *h){ \
2146     hl_decode_mb_internal(h, 1, sh); \
2147 }
2148 hl_decode_mb_simple(0, 8);
2149 hl_decode_mb_simple(1, 16);
2150
2151 /**
2152  * Process a macroblock; this handles edge cases, such as interlacing.
2153  */
2154 static void av_noinline hl_decode_mb_complex(H264Context *h){
2155     hl_decode_mb_internal(h, 0, h->pixel_shift);
2156 }
2157
2158 static void av_noinline hl_decode_mb_444_complex(H264Context *h){
2159     hl_decode_mb_444_internal(h, 0, h->pixel_shift);
2160 }
2161
2162 static void av_noinline hl_decode_mb_444_simple(H264Context *h){
2163     hl_decode_mb_444_internal(h, 1, 0);
2164 }
2165
2166 void ff_h264_hl_decode_mb(H264Context *h){
2167     MpegEncContext * const s = &h->s;
2168     const int mb_xy= h->mb_xy;
2169     const int mb_type = s->current_picture.f.mb_type[mb_xy];
2170     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
2171
2172     if (CHROMA444) {
2173         if(is_complex || h->pixel_shift)
2174             hl_decode_mb_444_complex(h);
2175         else
2176             hl_decode_mb_444_simple(h);
2177     } else if (is_complex) {
2178         hl_decode_mb_complex(h);
2179     } else if (h->pixel_shift) {
2180         hl_decode_mb_simple_16(h);
2181     } else
2182         hl_decode_mb_simple_8(h);
2183 }
2184
2185 static int pred_weight_table(H264Context *h){
2186     MpegEncContext * const s = &h->s;
2187     int list, i;
2188     int luma_def, chroma_def;
2189
2190     h->use_weight= 0;
2191     h->use_weight_chroma= 0;
2192     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2193     if(h->sps.chroma_format_idc)
2194         h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2195     luma_def = 1<<h->luma_log2_weight_denom;
2196     chroma_def = 1<<h->chroma_log2_weight_denom;
2197
2198     for(list=0; list<2; list++){
2199         h->luma_weight_flag[list]   = 0;
2200         h->chroma_weight_flag[list] = 0;
2201         for(i=0; i<h->ref_count[list]; i++){
2202             int luma_weight_flag, chroma_weight_flag;
2203
2204             luma_weight_flag= get_bits1(&s->gb);
2205             if(luma_weight_flag){
2206                 h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
2207                 h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
2208                 if(   h->luma_weight[i][list][0] != luma_def
2209                    || h->luma_weight[i][list][1] != 0) {
2210                     h->use_weight= 1;
2211                     h->luma_weight_flag[list]= 1;
2212                 }
2213             }else{
2214                 h->luma_weight[i][list][0]= luma_def;
2215                 h->luma_weight[i][list][1]= 0;
2216             }
2217
2218             if(h->sps.chroma_format_idc){
2219                 chroma_weight_flag= get_bits1(&s->gb);
2220                 if(chroma_weight_flag){
2221                     int j;
2222                     for(j=0; j<2; j++){
2223                         h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
2224                         h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
2225                         if(   h->chroma_weight[i][list][j][0] != chroma_def
2226                            || h->chroma_weight[i][list][j][1] != 0) {
2227                             h->use_weight_chroma= 1;
2228                             h->chroma_weight_flag[list]= 1;
2229                         }
2230                     }
2231                 }else{
2232                     int j;
2233                     for(j=0; j<2; j++){
2234                         h->chroma_weight[i][list][j][0]= chroma_def;
2235                         h->chroma_weight[i][list][j][1]= 0;
2236                     }
2237                 }
2238             }
2239         }
2240         if(h->slice_type_nos != AV_PICTURE_TYPE_B) break;
2241     }
2242     h->use_weight= h->use_weight || h->use_weight_chroma;
2243     return 0;
2244 }
2245
2246 /**
2247  * Initialize implicit_weight table.
2248  * @param field  0/1 initialize the weight for interlaced MBAFF
2249  *                -1 initializes the rest
2250  */
2251 static void implicit_weight_table(H264Context *h, int field){
2252     MpegEncContext * const s = &h->s;
2253     int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
2254
2255     for (i = 0; i < 2; i++) {
2256         h->luma_weight_flag[i]   = 0;
2257         h->chroma_weight_flag[i] = 0;
2258     }
2259
2260     if(field < 0){
2261         if (s->picture_structure == PICT_FRAME) {
2262             cur_poc = s->current_picture_ptr->poc;
2263         } else {
2264             cur_poc = s->current_picture_ptr->field_poc[s->picture_structure - 1];
2265         }
2266     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
2267        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
2268         h->use_weight= 0;
2269         h->use_weight_chroma= 0;
2270         return;
2271     }
2272         ref_start= 0;
2273         ref_count0= h->ref_count[0];
2274         ref_count1= h->ref_count[1];
2275     }else{
2276         cur_poc = s->current_picture_ptr->field_poc[field];
2277         ref_start= 16;
2278         ref_count0= 16+2*h->ref_count[0];
2279         ref_count1= 16+2*h->ref_count[1];
2280     }
2281
2282     h->use_weight= 2;
2283     h->use_weight_chroma= 2;
2284     h->luma_log2_weight_denom= 5;
2285     h->chroma_log2_weight_denom= 5;
2286
2287     for(ref0=ref_start; ref0 < ref_count0; ref0++){
2288         int poc0 = h->ref_list[0][ref0].poc;
2289         for(ref1=ref_start; ref1 < ref_count1; ref1++){
2290             int w = 32;
2291             if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
2292                 int poc1 = h->ref_list[1][ref1].poc;
2293                 int td = av_clip(poc1 - poc0, -128, 127);
2294                 if(td){
2295                     int tb = av_clip(cur_poc - poc0, -128, 127);
2296                     int tx = (16384 + (FFABS(td) >> 1)) / td;
2297                     int dist_scale_factor = (tb*tx + 32) >> 8;
2298                     if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
2299                         w = 64 - dist_scale_factor;
2300                 }
2301             }
2302             if(field<0){
2303                 h->implicit_weight[ref0][ref1][0]=
2304                 h->implicit_weight[ref0][ref1][1]= w;
2305             }else{
2306                 h->implicit_weight[ref0][ref1][field]=w;
2307             }
2308         }
2309     }
2310 }
2311
2312 /**
2313  * instantaneous decoder refresh.
2314  */
2315 static void idr(H264Context *h){
2316     ff_h264_remove_all_refs(h);
2317     h->prev_frame_num= 0;
2318     h->prev_frame_num_offset= 0;
2319     h->prev_poc_msb=
2320     h->prev_poc_lsb= 0;
2321 }
2322
2323 /* forget old pics after a seek */
2324 static void flush_dpb(AVCodecContext *avctx){
2325     H264Context *h= avctx->priv_data;
2326     int i;
2327     for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
2328         if(h->delayed_pic[i])
2329             h->delayed_pic[i]->f.reference = 0;
2330         h->delayed_pic[i]= NULL;
2331     }
2332     h->outputed_poc=h->next_outputed_poc= INT_MIN;
2333     h->prev_interlaced_frame = 1;
2334     idr(h);
2335     if(h->s.current_picture_ptr)
2336         h->s.current_picture_ptr->f.reference = 0;
2337     h->s.first_field= 0;
2338     ff_h264_reset_sei(h);
2339     ff_mpeg_flush(avctx);
2340 }
2341
2342 static int init_poc(H264Context *h){
2343     MpegEncContext * const s = &h->s;
2344     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2345     int field_poc[2];
2346     Picture *cur = s->current_picture_ptr;
2347
2348     h->frame_num_offset= h->prev_frame_num_offset;
2349     if(h->frame_num < h->prev_frame_num)
2350         h->frame_num_offset += max_frame_num;
2351
2352     if(h->sps.poc_type==0){
2353         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2354
2355         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2356             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2357         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2358             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2359         else
2360             h->poc_msb = h->prev_poc_msb;
2361 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2362         field_poc[0] =
2363         field_poc[1] = h->poc_msb + h->poc_lsb;
2364         if(s->picture_structure == PICT_FRAME)
2365             field_poc[1] += h->delta_poc_bottom;
2366     }else if(h->sps.poc_type==1){
2367         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2368         int i;
2369
2370         if(h->sps.poc_cycle_length != 0)
2371             abs_frame_num = h->frame_num_offset + h->frame_num;
2372         else
2373             abs_frame_num = 0;
2374
2375         if(h->nal_ref_idc==0 && abs_frame_num > 0)
2376             abs_frame_num--;
2377
2378         expected_delta_per_poc_cycle = 0;
2379         for(i=0; i < h->sps.poc_cycle_length; i++)
2380             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2381
2382         if(abs_frame_num > 0){
2383             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2384             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2385
2386             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2387             for(i = 0; i <= frame_num_in_poc_cycle; i++)
2388                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2389         } else
2390             expectedpoc = 0;
2391
2392         if(h->nal_ref_idc == 0)
2393             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2394
2395         field_poc[0] = expectedpoc + h->delta_poc[0];
2396         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2397
2398         if(s->picture_structure == PICT_FRAME)
2399             field_poc[1] += h->delta_poc[1];
2400     }else{
2401         int poc= 2*(h->frame_num_offset + h->frame_num);
2402
2403         if(!h->nal_ref_idc)
2404             poc--;
2405
2406         field_poc[0]= poc;
2407         field_poc[1]= poc;
2408     }
2409
2410     if(s->picture_structure != PICT_BOTTOM_FIELD)
2411         s->current_picture_ptr->field_poc[0]= field_poc[0];
2412     if(s->picture_structure != PICT_TOP_FIELD)
2413         s->current_picture_ptr->field_poc[1]= field_poc[1];
2414     cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
2415
2416     return 0;
2417 }
2418
2419
2420 /**
2421  * initialize scan tables
2422  */
2423 static void init_scan_tables(H264Context *h){
2424     int i;
2425     for(i=0; i<16; i++){
2426 #define T(x) (x>>2) | ((x<<2) & 0xF)
2427         h->zigzag_scan[i] = T(zigzag_scan[i]);
2428         h-> field_scan[i] = T( field_scan[i]);
2429 #undef T
2430     }
2431     for(i=0; i<64; i++){
2432 #define T(x) (x>>3) | ((x&7)<<3)
2433         h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
2434         h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
2435         h->field_scan8x8[i]        = T(field_scan8x8[i]);
2436         h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
2437 #undef T
2438     }
2439     if(h->sps.transform_bypass){ //FIXME same ugly
2440         h->zigzag_scan_q0          = zigzag_scan;
2441         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
2442         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
2443         h->field_scan_q0           = field_scan;
2444         h->field_scan8x8_q0        = field_scan8x8;
2445         h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
2446     }else{
2447         h->zigzag_scan_q0          = h->zigzag_scan;
2448         h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
2449         h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
2450         h->field_scan_q0           = h->field_scan;
2451         h->field_scan8x8_q0        = h->field_scan8x8;
2452         h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
2453     }
2454 }
2455
2456 static int field_end(H264Context *h, int in_setup){
2457     MpegEncContext * const s = &h->s;
2458     AVCodecContext * const avctx= s->avctx;
2459     int err = 0;
2460     s->mb_y= 0;
2461
2462     if (!in_setup && !s->dropable)
2463         ff_thread_report_progress((AVFrame*)s->current_picture_ptr, (16*s->mb_height >> FIELD_PICTURE) - 1,
2464                                  s->picture_structure==PICT_BOTTOM_FIELD);
2465
2466     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2467         ff_vdpau_h264_set_reference_frames(s);
2468
2469     if(in_setup || !(avctx->active_thread_type&FF_THREAD_FRAME)){
2470         if(!s->dropable) {
2471             err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
2472             h->prev_poc_msb= h->poc_msb;
2473             h->prev_poc_lsb= h->poc_lsb;
2474         }
2475         h->prev_frame_num_offset= h->frame_num_offset;
2476         h->prev_frame_num= h->frame_num;
2477         h->outputed_poc = h->next_outputed_poc;
2478     }
2479
2480     if (avctx->hwaccel) {
2481         if (avctx->hwaccel->end_frame(avctx) < 0)
2482             av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
2483     }
2484
2485     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2486         ff_vdpau_h264_picture_complete(s);
2487
2488     /*
2489      * FIXME: Error handling code does not seem to support interlaced
2490      * when slices span multiple rows
2491      * The ff_er_add_slice calls don't work right for bottom
2492      * fields; they cause massive erroneous error concealing
2493      * Error marking covers both fields (top and bottom).
2494      * This causes a mismatched s->error_count
2495      * and a bad error table. Further, the error count goes to
2496      * INT_MAX when called for bottom field, because mb_y is
2497      * past end by one (callers fault) and resync_mb_y != 0
2498      * causes problems for the first MB line, too.
2499      */
2500     if (!FIELD_PICTURE)
2501         ff_er_frame_end(s);
2502
2503     MPV_frame_end(s);
2504
2505     h->current_slice=0;
2506
2507     return err;
2508 }
2509
2510 /**
2511  * Replicate H264 "master" context to thread contexts.
2512  */
2513 static void clone_slice(H264Context *dst, H264Context *src)
2514 {
2515     memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
2516     dst->s.current_picture_ptr  = src->s.current_picture_ptr;
2517     dst->s.current_picture      = src->s.current_picture;
2518     dst->s.linesize             = src->s.linesize;
2519     dst->s.uvlinesize           = src->s.uvlinesize;
2520     dst->s.first_field          = src->s.first_field;
2521
2522     dst->prev_poc_msb           = src->prev_poc_msb;
2523     dst->prev_poc_lsb           = src->prev_poc_lsb;
2524     dst->prev_frame_num_offset  = src->prev_frame_num_offset;
2525     dst->prev_frame_num         = src->prev_frame_num;
2526     dst->short_ref_count        = src->short_ref_count;
2527
2528     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
2529     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
2530     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
2531     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
2532
2533     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
2534     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
2535 }
2536
2537 /**
2538  * computes profile from profile_idc and constraint_set?_flags
2539  *
2540  * @param sps SPS
2541  *
2542  * @return profile as defined by FF_PROFILE_H264_*
2543  */
2544 int ff_h264_get_profile(SPS *sps)
2545 {
2546     int profile = sps->profile_idc;
2547
2548     switch(sps->profile_idc) {
2549     case FF_PROFILE_H264_BASELINE:
2550         // constraint_set1_flag set to 1
2551         profile |= (sps->constraint_set_flags & 1<<1) ? FF_PROFILE_H264_CONSTRAINED : 0;
2552         break;
2553     case FF_PROFILE_H264_HIGH_10:
2554     case FF_PROFILE_H264_HIGH_422:
2555     case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
2556         // constraint_set3_flag set to 1
2557         profile |= (sps->constraint_set_flags & 1<<3) ? FF_PROFILE_H264_INTRA : 0;
2558         break;
2559     }
2560
2561     return profile;
2562 }
2563
2564 /**
2565  * decodes a slice header.
2566  * This will also call MPV_common_init() and frame_start() as needed.
2567  *
2568  * @param h h264context
2569  * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
2570  *
2571  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
2572  */
2573 static int decode_slice_header(H264Context *h, H264Context *h0){
2574     MpegEncContext * const s = &h->s;
2575     MpegEncContext * const s0 = &h0->s;
2576     unsigned int first_mb_in_slice;
2577     unsigned int pps_id;
2578     int num_ref_idx_active_override_flag;
2579     unsigned int slice_type, tmp, i, j;
2580     int default_ref_list_done = 0;
2581     int last_pic_structure;
2582
2583     s->dropable= h->nal_ref_idc == 0;
2584
2585     /* FIXME: 2tap qpel isn't implemented for high bit depth. */
2586     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc && !h->pixel_shift){
2587         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
2588         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
2589     }else{
2590         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
2591         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
2592     }
2593
2594     first_mb_in_slice= get_ue_golomb(&s->gb);
2595
2596     if(first_mb_in_slice == 0){ //FIXME better field boundary detection
2597         if(h0->current_slice && FIELD_PICTURE){
2598             field_end(h, 1);
2599         }
2600
2601         h0->current_slice = 0;
2602         if (!s0->first_field)
2603             s->current_picture_ptr= NULL;
2604     }
2605
2606     slice_type= get_ue_golomb_31(&s->gb);
2607     if(slice_type > 9){
2608         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);
2609         return -1;
2610     }
2611     if(slice_type > 4){
2612         slice_type -= 5;
2613         h->slice_type_fixed=1;
2614     }else
2615         h->slice_type_fixed=0;
2616
2617     slice_type= golomb_to_pict_type[ slice_type ];
2618     if (slice_type == AV_PICTURE_TYPE_I
2619         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
2620         default_ref_list_done = 1;
2621     }
2622     h->slice_type= slice_type;
2623     h->slice_type_nos= slice_type & 3;
2624
2625     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
2626
2627     pps_id= get_ue_golomb(&s->gb);
2628     if(pps_id>=MAX_PPS_COUNT){
2629         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
2630         return -1;
2631     }
2632     if(!h0->pps_buffers[pps_id]) {
2633         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
2634         return -1;
2635     }
2636     h->pps= *h0->pps_buffers[pps_id];
2637
2638     if(!h0->sps_buffers[h->pps.sps_id]) {
2639         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
2640         return -1;
2641     }
2642     h->sps = *h0->sps_buffers[h->pps.sps_id];
2643
2644     s->avctx->profile = ff_h264_get_profile(&h->sps);
2645     s->avctx->level   = h->sps.level_idc;
2646     s->avctx->refs    = h->sps.ref_frame_count;
2647
2648     if(h == h0 && h->dequant_coeff_pps != pps_id){
2649         h->dequant_coeff_pps = pps_id;
2650         init_dequant_tables(h);
2651     }
2652
2653     s->mb_width= h->sps.mb_width;
2654     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
2655
2656     h->b_stride=  s->mb_width*4;
2657
2658     s->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
2659
2660     s->width = 16*s->mb_width - (2>>CHROMA444)*FFMIN(h->sps.crop_right, (8<<CHROMA444)-1);
2661     if(h->sps.frame_mbs_only_flag)
2662         s->height= 16*s->mb_height - (1<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1);
2663     else
2664         s->height= 16*s->mb_height - (2<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1);
2665
2666     if (s->context_initialized
2667         && (   s->width != s->avctx->width || s->height != s->avctx->height
2668             || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
2669         if(h != h0) {
2670             av_log_missing_feature(s->avctx, "Width/height changing with threads is", 0);
2671             return -1;   // width / height changed during parallelized decoding
2672         }
2673         free_tables(h, 0);
2674         flush_dpb(s->avctx);
2675         MPV_common_end(s);
2676     }
2677     if (!s->context_initialized) {
2678         if (h != h0) {
2679             av_log(h->s.avctx, AV_LOG_ERROR, "Cannot (re-)initialize context during parallel decoding.\n");
2680             return -1;
2681         }
2682
2683         avcodec_set_dimensions(s->avctx, s->width, s->height);
2684         s->avctx->sample_aspect_ratio= h->sps.sar;
2685         av_assert0(s->avctx->sample_aspect_ratio.den);
2686
2687         if(h->sps.video_signal_type_present_flag){
2688             s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
2689             if(h->sps.colour_description_present_flag){
2690                 s->avctx->color_primaries = h->sps.color_primaries;
2691                 s->avctx->color_trc       = h->sps.color_trc;
2692                 s->avctx->colorspace      = h->sps.colorspace;
2693             }
2694         }
2695
2696         if(h->sps.timing_info_present_flag){
2697             int64_t den= h->sps.time_scale;
2698             if(h->x264_build < 44U)
2699                 den *= 2;
2700             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
2701                       h->sps.num_units_in_tick, den, 1<<30);
2702         }
2703
2704         switch (h->sps.bit_depth_luma) {
2705             case 9 :
2706                 if (CHROMA444)
2707                     s->avctx->pix_fmt = PIX_FMT_YUV444P9;
2708                 else if (CHROMA422)
2709                     s->avctx->pix_fmt = PIX_FMT_YUV422P9;
2710                 else
2711                     s->avctx->pix_fmt = PIX_FMT_YUV420P9;
2712                 break;
2713             case 10 :
2714                 if (CHROMA444)
2715                     s->avctx->pix_fmt = PIX_FMT_YUV444P10;
2716                 else if (CHROMA422)
2717                     s->avctx->pix_fmt = PIX_FMT_YUV422P10;
2718                 else
2719                     s->avctx->pix_fmt = PIX_FMT_YUV420P10;
2720                 break;
2721             default:
2722                 if (CHROMA444){
2723                     s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ444P : PIX_FMT_YUV444P;
2724                 } else if (CHROMA422) {
2725                     s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ422P : PIX_FMT_YUV422P;
2726                 }else{
2727                     s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
2728                                                              s->avctx->codec->pix_fmts ?
2729                                                              s->avctx->codec->pix_fmts :
2730                                                              s->avctx->color_range == AVCOL_RANGE_JPEG ?
2731                                                              hwaccel_pixfmt_list_h264_jpeg_420 :
2732                                                              ff_hwaccel_pixfmt_list_420);
2733                 }
2734         }
2735
2736         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
2737
2738         if (MPV_common_init(s) < 0) {
2739             av_log(h->s.avctx, AV_LOG_ERROR, "MPV_common_init() failed.\n");
2740             return -1;
2741         }
2742         s->first_field = 0;
2743         h->prev_interlaced_frame = 1;
2744
2745         init_scan_tables(h);
2746         if (ff_h264_alloc_tables(h) < 0) {
2747             av_log(h->s.avctx, AV_LOG_ERROR, "Could not allocate memory for h264\n");
2748             return AVERROR(ENOMEM);
2749         }
2750
2751         if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_SLICE)) {
2752             if (context_init(h) < 0) {
2753                 av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2754                 return -1;
2755             }
2756         } else {
2757             for(i = 1; i < s->avctx->thread_count; i++) {
2758                 H264Context *c;
2759                 c = h->thread_context[i] = av_malloc(sizeof(H264Context));
2760                 memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
2761                 memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
2762                 c->h264dsp = h->h264dsp;
2763                 c->sps = h->sps;
2764                 c->pps = h->pps;
2765                 c->pixel_shift = h->pixel_shift;
2766                 init_scan_tables(c);
2767                 clone_tables(c, h, i);
2768             }
2769
2770             for(i = 0; i < s->avctx->thread_count; i++)
2771                 if (context_init(h->thread_context[i]) < 0) {
2772                     av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2773                     return -1;
2774                 }
2775         }
2776     }
2777
2778     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
2779
2780     h->mb_mbaff = 0;
2781     h->mb_aff_frame = 0;
2782     last_pic_structure = s0->picture_structure;
2783     if(h->sps.frame_mbs_only_flag){
2784         s->picture_structure= PICT_FRAME;
2785     }else{
2786         if(get_bits1(&s->gb)) { //field_pic_flag
2787             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
2788         } else {
2789             s->picture_structure= PICT_FRAME;
2790             h->mb_aff_frame = h->sps.mb_aff;
2791         }
2792     }
2793     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
2794
2795     if(h0->current_slice == 0){
2796         // Shorten frame num gaps so we don't have to allocate reference frames just to throw them away
2797         if(h->frame_num != h->prev_frame_num) {
2798             int unwrap_prev_frame_num = h->prev_frame_num, max_frame_num = 1<<h->sps.log2_max_frame_num;
2799
2800             if (unwrap_prev_frame_num > h->frame_num) unwrap_prev_frame_num -= max_frame_num;
2801
2802             if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
2803                 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
2804                 if (unwrap_prev_frame_num < 0)
2805                     unwrap_prev_frame_num += max_frame_num;
2806
2807                 h->prev_frame_num = unwrap_prev_frame_num;
2808             }
2809         }
2810
2811         while(h->frame_num !=  h->prev_frame_num &&
2812               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
2813             Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
2814             av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
2815             if (ff_h264_frame_start(h) < 0)
2816                 return -1;
2817             h->prev_frame_num++;
2818             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
2819             s->current_picture_ptr->frame_num= h->prev_frame_num;
2820             ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 0);
2821             ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 1);
2822             ff_generate_sliding_window_mmcos(h);
2823             if (ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index) < 0 &&
2824                 s->avctx->error_recognition >= FF_ER_EXPLODE)
2825                 return AVERROR_INVALIDDATA;
2826             /* Error concealment: if a ref is missing, copy the previous ref in its place.
2827              * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
2828              * about there being no actual duplicates.
2829              * FIXME: this doesn't copy padding for out-of-frame motion vectors.  Given we're
2830              * concealing a lost frame, this probably isn't noticable by comparison, but it should
2831              * be fixed. */
2832             if (h->short_ref_count) {
2833                 if (prev) {
2834                     av_image_copy(h->short_ref[0]->f.data, h->short_ref[0]->f.linesize,
2835                                   (const uint8_t**)prev->f.data, prev->f.linesize,
2836                                   s->avctx->pix_fmt, s->mb_width*16, s->mb_height*16);
2837                     h->short_ref[0]->poc = prev->poc+2;
2838                 }
2839                 h->short_ref[0]->frame_num = h->prev_frame_num;
2840             }
2841         }
2842
2843         /* See if we have a decoded first field looking for a pair... */
2844         if (s0->first_field) {
2845             assert(s0->current_picture_ptr);
2846             assert(s0->current_picture_ptr->f.data[0]);
2847             assert(s0->current_picture_ptr->f.reference != DELAYED_PIC_REF);
2848
2849             /* figure out if we have a complementary field pair */
2850             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2851                 /*
2852                  * Previous field is unmatched. Don't display it, but let it
2853                  * remain for reference if marked as such.
2854                  */
2855                 s0->current_picture_ptr = NULL;
2856                 s0->first_field = FIELD_PICTURE;
2857
2858             } else {
2859                 if (h->nal_ref_idc &&
2860                         s0->current_picture_ptr->f.reference &&
2861                         s0->current_picture_ptr->frame_num != h->frame_num) {
2862                     /*
2863                      * This and previous field were reference, but had
2864                      * different frame_nums. Consider this field first in
2865                      * pair. Throw away previous field except for reference
2866                      * purposes.
2867                      */
2868                     s0->first_field = 1;
2869                     s0->current_picture_ptr = NULL;
2870
2871                 } else {
2872                     /* Second field in complementary pair */
2873                     s0->first_field = 0;
2874                 }
2875             }
2876
2877         } else {
2878             /* Frame or first field in a potentially complementary pair */
2879             assert(!s0->current_picture_ptr);
2880             s0->first_field = FIELD_PICTURE;
2881         }
2882
2883         if(!FIELD_PICTURE || s0->first_field) {
2884             if (ff_h264_frame_start(h) < 0) {
2885                 s0->first_field = 0;
2886                 return -1;
2887             }
2888         } else {
2889             ff_release_unused_pictures(s, 0);
2890         }
2891     }
2892     if(h != h0)
2893         clone_slice(h, h0);
2894
2895     s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
2896
2897     assert(s->mb_num == s->mb_width * s->mb_height);
2898     if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
2899        first_mb_in_slice                    >= s->mb_num){
2900         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
2901         return -1;
2902     }
2903     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
2904     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
2905     if (s->picture_structure == PICT_BOTTOM_FIELD)
2906         s->resync_mb_y = s->mb_y = s->mb_y + 1;
2907     assert(s->mb_y < s->mb_height);
2908
2909     if(s->picture_structure==PICT_FRAME){
2910         h->curr_pic_num=   h->frame_num;
2911         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
2912     }else{
2913         h->curr_pic_num= 2*h->frame_num + 1;
2914         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
2915     }
2916
2917     if(h->nal_unit_type == NAL_IDR_SLICE){
2918         get_ue_golomb(&s->gb); /* idr_pic_id */
2919     }
2920
2921     if(h->sps.poc_type==0){
2922         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2923
2924         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
2925             h->delta_poc_bottom= get_se_golomb(&s->gb);
2926         }
2927     }
2928
2929     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
2930         h->delta_poc[0]= get_se_golomb(&s->gb);
2931
2932         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2933             h->delta_poc[1]= get_se_golomb(&s->gb);
2934     }
2935
2936     init_poc(h);
2937
2938     if(h->pps.redundant_pic_cnt_present){
2939         h->redundant_pic_count= get_ue_golomb(&s->gb);
2940     }
2941
2942     //set defaults, might be overridden a few lines later
2943     h->ref_count[0]= h->pps.ref_count[0];
2944     h->ref_count[1]= h->pps.ref_count[1];
2945
2946     if(h->slice_type_nos != AV_PICTURE_TYPE_I){
2947         if(h->slice_type_nos == AV_PICTURE_TYPE_B){
2948             h->direct_spatial_mv_pred= get_bits1(&s->gb);
2949         }
2950         num_ref_idx_active_override_flag= get_bits1(&s->gb);
2951
2952         if(num_ref_idx_active_override_flag){
2953             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2954             if(h->slice_type_nos==AV_PICTURE_TYPE_B)
2955                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2956
2957             if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
2958                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
2959                 h->ref_count[0]= h->ref_count[1]= 1;
2960                 return -1;
2961             }
2962         }
2963         if(h->slice_type_nos == AV_PICTURE_TYPE_B)
2964             h->list_count= 2;
2965         else
2966             h->list_count= 1;
2967     }else
2968         h->list_count= 0;
2969
2970     if(!default_ref_list_done){
2971         ff_h264_fill_default_ref_list(h);
2972     }
2973
2974     if(h->slice_type_nos!=AV_PICTURE_TYPE_I && ff_h264_decode_ref_pic_list_reordering(h) < 0) {
2975         h->ref_count[1]= h->ref_count[0]= 0;
2976         return -1;
2977     }
2978
2979     if(h->slice_type_nos!=AV_PICTURE_TYPE_I){
2980         s->last_picture_ptr= &h->ref_list[0][0];
2981         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
2982     }
2983     if(h->slice_type_nos==AV_PICTURE_TYPE_B){
2984         s->next_picture_ptr= &h->ref_list[1][0];
2985         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
2986     }
2987
2988     if(   (h->pps.weighted_pred          && h->slice_type_nos == AV_PICTURE_TYPE_P )
2989        ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== AV_PICTURE_TYPE_B ) )
2990         pred_weight_table(h);
2991     else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
2992         implicit_weight_table(h, -1);
2993     }else {
2994         h->use_weight = 0;
2995         for (i = 0; i < 2; i++) {
2996             h->luma_weight_flag[i]   = 0;
2997             h->chroma_weight_flag[i] = 0;
2998         }
2999     }
3000
3001     if(h->nal_ref_idc && ff_h264_decode_ref_pic_marking(h0, &s->gb) < 0 &&
3002        s->avctx->error_recognition >= FF_ER_EXPLODE)
3003         return AVERROR_INVALIDDATA;
3004
3005     if(FRAME_MBAFF){
3006         ff_h264_fill_mbaff_ref_list(h);
3007
3008         if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
3009             implicit_weight_table(h, 0);
3010             implicit_weight_table(h, 1);
3011         }
3012     }
3013
3014     if(h->slice_type_nos==AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
3015         ff_h264_direct_dist_scale_factor(h);
3016     ff_h264_direct_ref_list_init(h);
3017
3018     if( h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac ){
3019         tmp = get_ue_golomb_31(&s->gb);
3020         if(tmp > 2){
3021             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3022             return -1;
3023         }
3024         h->cabac_init_idc= tmp;
3025     }
3026
3027     h->last_qscale_diff = 0;
3028     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
3029     if(tmp>51+6*(h->sps.bit_depth_luma-8)){
3030         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3031         return -1;
3032     }
3033     s->qscale= tmp;
3034     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3035     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3036     //FIXME qscale / qp ... stuff
3037     if(h->slice_type == AV_PICTURE_TYPE_SP){
3038         get_bits1(&s->gb); /* sp_for_switch_flag */
3039     }
3040     if(h->slice_type==AV_PICTURE_TYPE_SP || h->slice_type == AV_PICTURE_TYPE_SI){
3041         get_se_golomb(&s->gb); /* slice_qs_delta */
3042     }
3043
3044     h->deblocking_filter = 1;
3045     h->slice_alpha_c0_offset = 52;
3046     h->slice_beta_offset = 52;
3047     if( h->pps.deblocking_filter_parameters_present ) {
3048         tmp= get_ue_golomb_31(&s->gb);
3049         if(tmp > 2){
3050             av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
3051             return -1;
3052         }
3053         h->deblocking_filter= tmp;
3054         if(h->deblocking_filter < 2)
3055             h->deblocking_filter^= 1; // 1<->0
3056
3057         if( h->deblocking_filter ) {
3058             h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
3059             h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
3060             if(   h->slice_alpha_c0_offset > 104U
3061                || h->slice_beta_offset     > 104U){
3062                 av_log(s->avctx, AV_LOG_ERROR, "deblocking filter parameters %d %d out of range\n", h->slice_alpha_c0_offset, h->slice_beta_offset);
3063                 return -1;
3064             }
3065         }
3066     }
3067
3068     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
3069        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != AV_PICTURE_TYPE_I)
3070        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == AV_PICTURE_TYPE_B)
3071        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
3072         h->deblocking_filter= 0;
3073
3074     if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
3075         if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
3076             /* Cheat slightly for speed:
3077                Do not bother to deblock across slices. */
3078             h->deblocking_filter = 2;
3079         } else {
3080             h0->max_contexts = 1;
3081             if(!h0->single_decode_warning) {
3082                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
3083                 h0->single_decode_warning = 1;
3084             }
3085             if (h != h0) {
3086                 av_log(h->s.avctx, AV_LOG_ERROR, "Deblocking switched inside frame.\n");
3087                 return 1;
3088             }
3089         }
3090     }
3091     h->qp_thresh = 15 + 52 - FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset)
3092                  - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1])
3093                  + 6 * (h->sps.bit_depth_luma - 8);
3094
3095 #if 0 //FMO
3096     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3097         slice_group_change_cycle= get_bits(&s->gb, ?);
3098 #endif
3099
3100     h0->last_slice_type = slice_type;
3101     h->slice_num = ++h0->current_slice;
3102     if(h->slice_num >= MAX_SLICES){
3103         av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
3104     }
3105
3106     for(j=0; j<2; j++){
3107         int id_list[16];
3108         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
3109         for(i=0; i<16; i++){
3110             id_list[i]= 60;
3111             if (h->ref_list[j][i].f.data[0]) {
3112                 int k;
3113                 uint8_t *base = h->ref_list[j][i].f.base[0];
3114                 for(k=0; k<h->short_ref_count; k++)
3115                     if (h->short_ref[k]->f.base[0] == base) {
3116                         id_list[i]= k;
3117                         break;
3118                     }
3119                 for(k=0; k<h->long_ref_count; k++)
3120                     if (h->long_ref[k] && h->long_ref[k]->f.base[0] == base) {
3121                         id_list[i]= h->short_ref_count + k;
3122                         break;
3123                     }
3124             }
3125         }
3126
3127         ref2frm[0]=
3128         ref2frm[1]= -1;
3129         for(i=0; i<16; i++)
3130             ref2frm[i+2]= 4*id_list[i]
3131                           + (h->ref_list[j][i].f.reference & 3);
3132         ref2frm[18+0]=
3133         ref2frm[18+1]= -1;
3134         for(i=16; i<48; i++)
3135             ref2frm[i+4]= 4*id_list[(i-16)>>1]
3136                           + (h->ref_list[j][i].f.reference & 3);
3137     }
3138
3139     //FIXME: fix draw_edges+PAFF+frame threads
3140     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE || (!h->sps.frame_mbs_only_flag && s->avctx->active_thread_type)) ? 0 : 16;
3141     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
3142
3143     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3144         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",
3145                h->slice_num,
3146                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
3147                first_mb_in_slice,
3148                av_get_picture_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
3149                pps_id, h->frame_num,
3150                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3151                h->ref_count[0], h->ref_count[1],
3152                s->qscale,
3153                h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
3154                h->use_weight,
3155                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
3156                h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
3157                );
3158     }
3159
3160     return 0;
3161 }
3162
3163 int ff_h264_get_slice_type(const H264Context *h)
3164 {
3165     switch (h->slice_type) {
3166     case AV_PICTURE_TYPE_P:  return 0;
3167     case AV_PICTURE_TYPE_B:  return 1;
3168     case AV_PICTURE_TYPE_I:  return 2;
3169     case AV_PICTURE_TYPE_SP: return 3;
3170     case AV_PICTURE_TYPE_SI: return 4;
3171     default:         return -1;
3172     }
3173 }
3174
3175 static av_always_inline void fill_filter_caches_inter(H264Context *h, MpegEncContext * const s, int mb_type, int top_xy,
3176                                                       int left_xy[LEFT_MBS], int top_type, int left_type[LEFT_MBS], int mb_xy, int list)
3177 {
3178     int b_stride = h->b_stride;
3179     int16_t (*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
3180     int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
3181     if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
3182         if(USES_LIST(top_type, list)){
3183             const int b_xy= h->mb2b_xy[top_xy] + 3*b_stride;
3184             const int b8_xy= 4*top_xy + 2;
3185             int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3186             AV_COPY128(mv_dst - 1*8, s->current_picture.f.motion_val[list][b_xy + 0]);
3187             ref_cache[0 - 1*8]=
3188             ref_cache[1 - 1*8]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 0]];
3189             ref_cache[2 - 1*8]=
3190             ref_cache[3 - 1*8]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 1]];
3191         }else{
3192             AV_ZERO128(mv_dst - 1*8);
3193             AV_WN32A(&ref_cache[0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3194         }
3195
3196         if(!IS_INTERLACED(mb_type^left_type[LTOP])){
3197             if(USES_LIST(left_type[LTOP], list)){
3198                 const int b_xy= h->mb2b_xy[left_xy[LTOP]] + 3;
3199                 const int b8_xy= 4*left_xy[LTOP] + 1;
3200                 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[LTOP]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3201                 AV_COPY32(mv_dst - 1 +  0, s->current_picture.f.motion_val[list][b_xy + b_stride*0]);
3202                 AV_COPY32(mv_dst - 1 +  8, s->current_picture.f.motion_val[list][b_xy + b_stride*1]);
3203                 AV_COPY32(mv_dst - 1 + 16, s->current_picture.f.motion_val[list][b_xy + b_stride*2]);
3204                 AV_COPY32(mv_dst - 1 + 24, s->current_picture.f.motion_val[list][b_xy + b_stride*3]);
3205                 ref_cache[-1 +  0]=
3206                 ref_cache[-1 +  8]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2*0]];
3207                 ref_cache[-1 + 16]=
3208                 ref_cache[-1 + 24]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2*1]];
3209             }else{
3210                 AV_ZERO32(mv_dst - 1 + 0);
3211                 AV_ZERO32(mv_dst - 1 + 8);
3212                 AV_ZERO32(mv_dst - 1 +16);
3213                 AV_ZERO32(mv_dst - 1 +24);
3214                 ref_cache[-1 +  0]=
3215                 ref_cache[-1 +  8]=
3216                 ref_cache[-1 + 16]=
3217                 ref_cache[-1 + 24]= LIST_NOT_USED;
3218             }
3219         }
3220     }
3221
3222     if(!USES_LIST(mb_type, list)){
3223         fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0,0), 4);
3224         AV_WN32A(&ref_cache[0*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3225         AV_WN32A(&ref_cache[1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3226         AV_WN32A(&ref_cache[2*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3227         AV_WN32A(&ref_cache[3*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3228         return;
3229     }
3230
3231     {
3232         int8_t *ref = &s->current_picture.f.ref_index[list][4*mb_xy];
3233         int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3234         uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101;
3235         uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]],ref2frm[list][ref[3]])&0x00FF00FF)*0x0101;
3236         AV_WN32A(&ref_cache[0*8], ref01);
3237         AV_WN32A(&ref_cache[1*8], ref01);
3238         AV_WN32A(&ref_cache[2*8], ref23);
3239         AV_WN32A(&ref_cache[3*8], ref23);
3240     }
3241
3242     {
3243         int16_t (*mv_src)[2] = &s->current_picture.f.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
3244         AV_COPY128(mv_dst + 8*0, mv_src + 0*b_stride);
3245         AV_COPY128(mv_dst + 8*1, mv_src + 1*b_stride);
3246         AV_COPY128(mv_dst + 8*2, mv_src + 2*b_stride);
3247         AV_COPY128(mv_dst + 8*3, mv_src + 3*b_stride);
3248     }
3249 }
3250
3251 /**
3252  *
3253  * @return non zero if the loop filter can be skiped
3254  */
3255 static int fill_filter_caches(H264Context *h, int mb_type){
3256     MpegEncContext * const s = &h->s;
3257     const int mb_xy= h->mb_xy;
3258     int top_xy, left_xy[LEFT_MBS];
3259     int top_type, left_type[LEFT_MBS];
3260     uint8_t *nnz;
3261     uint8_t *nnz_cache;
3262
3263     top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
3264
3265     /* Wow, what a mess, why didn't they simplify the interlacing & intra
3266      * stuff, I can't imagine that these complex rules are worth it. */
3267
3268     left_xy[LBOT] = left_xy[LTOP] = mb_xy-1;
3269     if(FRAME_MBAFF){
3270         const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.f.mb_type[mb_xy - 1]);
3271         const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
3272         if(s->mb_y&1){
3273             if (left_mb_field_flag != curr_mb_field_flag) {
3274                 left_xy[LTOP] -= s->mb_stride;
3275             }
3276         }else{
3277             if(curr_mb_field_flag){
3278                 top_xy += s->mb_stride & (((s->current_picture.f.mb_type[top_xy] >> 7) & 1) - 1);
3279             }
3280             if (left_mb_field_flag != curr_mb_field_flag) {
3281                 left_xy[LBOT] += s->mb_stride;
3282             }
3283         }
3284     }
3285
3286     h->top_mb_xy = top_xy;
3287     h->left_mb_xy[LTOP] = left_xy[LTOP];
3288     h->left_mb_xy[LBOT] = left_xy[LBOT];
3289     {
3290         //for sufficiently low qp, filtering wouldn't do anything
3291         //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
3292         int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
3293         int qp = s->current_picture.f.qscale_table[mb_xy];
3294         if(qp <= qp_thresh
3295            && (left_xy[LTOP] < 0 || ((qp + s->current_picture.f.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh)
3296            && (top_xy        < 0 || ((qp + s->current_picture.f.qscale_table[top_xy       ] + 1) >> 1) <= qp_thresh)) {
3297             if(!FRAME_MBAFF)
3298                 return 1;
3299             if ((left_xy[LTOP] < 0            || ((qp + s->current_picture.f.qscale_table[left_xy[LBOT]        ] + 1) >> 1) <= qp_thresh) &&
3300                 (top_xy        < s->mb_stride || ((qp + s->current_picture.f.qscale_table[top_xy - s->mb_stride] + 1) >> 1) <= qp_thresh))
3301                 return 1;
3302         }
3303     }
3304
3305     top_type        = s->current_picture.f.mb_type[top_xy];
3306     left_type[LTOP] = s->current_picture.f.mb_type[left_xy[LTOP]];
3307     left_type[LBOT] = s->current_picture.f.mb_type[left_xy[LBOT]];
3308     if(h->deblocking_filter == 2){
3309         if(h->slice_table[top_xy       ] != h->slice_num) top_type= 0;
3310         if(h->slice_table[left_xy[LBOT]] != h->slice_num) left_type[LTOP]= left_type[LBOT]= 0;
3311     }else{
3312         if(h->slice_table[top_xy       ] == 0xFFFF) top_type= 0;
3313         if(h->slice_table[left_xy[LBOT]] == 0xFFFF) left_type[LTOP]= left_type[LBOT] =0;
3314     }
3315     h->top_type       = top_type;
3316     h->left_type[LTOP]= left_type[LTOP];
3317     h->left_type[LBOT]= left_type[LBOT];
3318
3319     if(IS_INTRA(mb_type))
3320         return 0;
3321
3322     fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy, top_type, left_type, mb_xy, 0);
3323     if(h->list_count == 2)
3324         fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy, top_type, left_type, mb_xy, 1);
3325
3326     nnz = h->non_zero_count[mb_xy];
3327     nnz_cache = h->non_zero_count_cache;
3328     AV_COPY32(&nnz_cache[4+8*1], &nnz[ 0]);
3329     AV_COPY32(&nnz_cache[4+8*2], &nnz[ 4]);
3330     AV_COPY32(&nnz_cache[4+8*3], &nnz[ 8]);
3331     AV_COPY32(&nnz_cache[4+8*4], &nnz[12]);
3332     h->cbp= h->cbp_table[mb_xy];
3333
3334     if(top_type){
3335         nnz = h->non_zero_count[top_xy];
3336         AV_COPY32(&nnz_cache[4+8*0], &nnz[3*4]);
3337     }
3338
3339     if(left_type[LTOP]){
3340         nnz = h->non_zero_count[left_xy[LTOP]];
3341         nnz_cache[3+8*1]= nnz[3+0*4];
3342         nnz_cache[3+8*2]= nnz[3+1*4];
3343         nnz_cache[3+8*3]= nnz[3+2*4];
3344         nnz_cache[3+8*4]= nnz[3+3*4];
3345     }
3346
3347     // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
3348     if(!CABAC && h->pps.transform_8x8_mode){
3349         if(IS_8x8DCT(top_type)){
3350             nnz_cache[4+8*0]=
3351             nnz_cache[5+8*0]= (h->cbp_table[top_xy] & 0x4000) >> 12;
3352             nnz_cache[6+8*0]=
3353             nnz_cache[7+8*0]= (h->cbp_table[top_xy] & 0x8000) >> 12;
3354         }
3355         if(IS_8x8DCT(left_type[LTOP])){
3356             nnz_cache[3+8*1]=
3357             nnz_cache[3+8*2]= (h->cbp_table[left_xy[LTOP]]&0x2000) >> 12; //FIXME check MBAFF
3358         }
3359         if(IS_8x8DCT(left_type[LBOT])){
3360             nnz_cache[3+8*3]=
3361             nnz_cache[3+8*4]= (h->cbp_table[left_xy[LBOT]]&0x8000) >> 12; //FIXME check MBAFF
3362         }
3363
3364         if(IS_8x8DCT(mb_type)){
3365             nnz_cache[scan8[0   ]]= nnz_cache[scan8[1   ]]=
3366             nnz_cache[scan8[2   ]]= nnz_cache[scan8[3   ]]= (h->cbp & 0x1000) >> 12;
3367
3368             nnz_cache[scan8[0+ 4]]= nnz_cache[scan8[1+ 4]]=
3369             nnz_cache[scan8[2+ 4]]= nnz_cache[scan8[3+ 4]]= (h->cbp & 0x2000) >> 12;
3370
3371             nnz_cache[scan8[0+ 8]]= nnz_cache[scan8[1+ 8]]=
3372             nnz_cache[scan8[2+ 8]]= nnz_cache[scan8[3+ 8]]= (h->cbp & 0x4000) >> 12;
3373
3374             nnz_cache[scan8[0+12]]= nnz_cache[scan8[1+12]]=
3375             nnz_cache[scan8[2+12]]= nnz_cache[scan8[3+12]]= (h->cbp & 0x8000) >> 12;
3376         }
3377     }
3378
3379     return 0;
3380 }
3381
3382 static void loop_filter(H264Context *h, int start_x, int end_x){
3383     MpegEncContext * const s = &h->s;
3384     uint8_t  *dest_y, *dest_cb, *dest_cr;
3385     int linesize, uvlinesize, mb_x, mb_y;
3386     const int end_mb_y= s->mb_y + FRAME_MBAFF;
3387     const int old_slice_type= h->slice_type;
3388     const int pixel_shift = h->pixel_shift;
3389     const int block_h = 16 >> s->chroma_y_shift;
3390
3391     if(h->deblocking_filter) {
3392         for(mb_x= start_x; mb_x<end_x; mb_x++){
3393             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
3394                 int mb_xy, mb_type;
3395                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
3396                 h->slice_num= h->slice_table[mb_xy];
3397                 mb_type = s->current_picture.f.mb_type[mb_xy];
3398                 h->list_count= h->list_counts[mb_xy];
3399
3400                 if(FRAME_MBAFF)
3401                     h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
3402
3403                 s->mb_x= mb_x;
3404                 s->mb_y= mb_y;
3405                 dest_y  = s->current_picture.f.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize  ) * 16;
3406                 dest_cb = s->current_picture.f.data[1] + (mb_x << pixel_shift) * (8 << CHROMA444) + mb_y * s->uvlinesize * block_h;
3407                 dest_cr = s->current_picture.f.data[2] + (mb_x << pixel_shift) * (8 << CHROMA444) + mb_y * s->uvlinesize * block_h;
3408                     //FIXME simplify above
3409
3410                 if (MB_FIELD) {
3411                     linesize   = h->mb_linesize   = s->linesize * 2;
3412                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
3413                     if(mb_y&1){ //FIXME move out of this function?
3414                         dest_y -= s->linesize*15;
3415                         dest_cb-= s->uvlinesize * (block_h - 1);
3416                         dest_cr-= s->uvlinesize * (block_h - 1);
3417                     }
3418                 } else {
3419                     linesize   = h->mb_linesize   = s->linesize;
3420                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
3421                 }
3422                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
3423                 if(fill_filter_caches(h, mb_type))
3424                     continue;
3425                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.f.qscale_table[mb_xy]);
3426                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.f.qscale_table[mb_xy]);
3427
3428                 if (FRAME_MBAFF) {
3429                     ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3430                 } else {
3431                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3432                 }
3433             }
3434         }
3435     }
3436     h->slice_type= old_slice_type;
3437     s->mb_x= end_x;
3438     s->mb_y= end_mb_y - FRAME_MBAFF;
3439     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3440     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3441 }
3442
3443 static void predict_field_decoding_flag(H264Context *h){
3444     MpegEncContext * const s = &h->s;
3445     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3446     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
3447                 ? s->current_picture.f.mb_type[mb_xy - 1]
3448                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
3449                 ? s->current_picture.f.mb_type[mb_xy - s->mb_stride]
3450                 : 0;
3451     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
3452 }
3453
3454 /**
3455  * Draw edges and report progress for the last MB row.
3456  */
3457 static void decode_finish_row(H264Context *h){
3458     MpegEncContext * const s = &h->s;
3459     int top = 16*(s->mb_y >> FIELD_PICTURE);
3460     int height = 16 << FRAME_MBAFF;
3461     int deblock_border = (16 + 4) << FRAME_MBAFF;
3462     int pic_height = 16*s->mb_height >> FIELD_PICTURE;
3463
3464     if (h->deblocking_filter) {
3465         if((top + height) >= pic_height)
3466             height += deblock_border;
3467
3468         top -= deblock_border;
3469     }
3470
3471     if (top >= pic_height || (top + height) < h->emu_edge_height)
3472         return;
3473
3474     height = FFMIN(height, pic_height - top);
3475     if (top < h->emu_edge_height) {
3476         height = top+height;
3477         top = 0;
3478     }
3479
3480     ff_draw_horiz_band(s, top, height);
3481
3482     if (s->dropable) return;
3483
3484     ff_thread_report_progress((AVFrame*)s->current_picture_ptr, top + height - 1,
3485                              s->picture_structure==PICT_BOTTOM_FIELD);
3486 }
3487
3488 static int decode_slice(struct AVCodecContext *avctx, void *arg){
3489     H264Context *h = *(void**)arg;
3490     MpegEncContext * const s = &h->s;
3491     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
3492     int lf_x_start = s->mb_x;
3493
3494     s->mb_skip_run= -1;
3495
3496     h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
3497                     (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
3498
3499     if( h->pps.cabac ) {
3500         /* realign */
3501         align_get_bits( &s->gb );
3502
3503         /* init cabac */
3504         ff_init_cabac_states( &h->cabac);
3505         ff_init_cabac_decoder( &h->cabac,
3506                                s->gb.buffer + get_bits_count(&s->gb)/8,
3507                                (get_bits_left(&s->gb) + 7)/8);
3508
3509         ff_h264_init_cabac_states(h);
3510
3511         for(;;){
3512 //START_TIMER
3513             int ret = ff_h264_decode_mb_cabac(h);
3514             int eos;
3515 //STOP_TIMER("decode_mb_cabac")
3516
3517             if(ret>=0) ff_h264_hl_decode_mb(h);
3518
3519             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
3520                 s->mb_y++;
3521
3522                 ret = ff_h264_decode_mb_cabac(h);
3523
3524                 if(ret>=0) ff_h264_hl_decode_mb(h);
3525                 s->mb_y--;
3526             }
3527             eos = get_cabac_terminate( &h->cabac );
3528
3529             if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
3530                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3531                 if (s->mb_x >= lf_x_start) loop_filter(h, lf_x_start, s->mb_x + 1);
3532                 return 0;
3533             }
3534             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
3535                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d, bytestream (%td)\n", s->mb_x, s->mb_y, h->cabac.bytestream_end - h->cabac.bytestream);
3536                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
3537                 return -1;
3538             }
3539
3540             if( ++s->mb_x >= s->mb_width ) {
3541                 loop_filter(h, lf_x_start, s->mb_x);
3542                 s->mb_x = lf_x_start = 0;
3543                 decode_finish_row(h);
3544                 ++s->mb_y;
3545                 if(FIELD_OR_MBAFF_PICTURE) {
3546                     ++s->mb_y;
3547                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
3548                         predict_field_decoding_flag(h);
3549                 }
3550             }
3551
3552             if( eos || s->mb_y >= s->mb_height ) {
3553                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3554                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3555                 if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
3556                 return 0;
3557             }
3558         }
3559
3560     } else {
3561         for(;;){
3562             int ret = ff_h264_decode_mb_cavlc(h);
3563
3564             if(ret>=0) ff_h264_hl_decode_mb(h);
3565
3566             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
3567                 s->mb_y++;
3568                 ret = ff_h264_decode_mb_cavlc(h);
3569
3570                 if(ret>=0) ff_h264_hl_decode_mb(h);
3571                 s->mb_y--;
3572             }
3573
3574             if(ret<0){
3575                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3576                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
3577                 return -1;
3578             }
3579
3580             if(++s->mb_x >= s->mb_width){
3581                 loop_filter(h, lf_x_start, s->mb_x);
3582                 s->mb_x = lf_x_start = 0;
3583                 decode_finish_row(h);
3584                 ++s->mb_y;
3585                 if(FIELD_OR_MBAFF_PICTURE) {
3586                     ++s->mb_y;
3587                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
3588                         predict_field_decoding_flag(h);
3589                 }
3590                 if(s->mb_y >= s->mb_height){
3591                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3592
3593                     if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
3594                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3595
3596                         return 0;
3597                     }else{
3598                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3599
3600                         return -1;
3601                     }
3602                 }
3603             }
3604
3605             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
3606                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3607                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
3608                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3609                     if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
3610
3611                     return 0;
3612                 }else{
3613                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
3614
3615                     return -1;
3616                 }
3617             }
3618         }
3619     }
3620 }
3621
3622 /**
3623  * Call decode_slice() for each context.
3624  *
3625  * @param h h264 master context
3626  * @param context_count number of contexts to execute
3627  */
3628 static int execute_decode_slices(H264Context *h, int context_count){
3629     MpegEncContext * const s = &h->s;
3630     AVCodecContext * const avctx= s->avctx;
3631     H264Context *hx;
3632     int i;
3633
3634     if (s->avctx->hwaccel || s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
3635         return 0;
3636     if(context_count == 1) {
3637         return decode_slice(avctx, &h);
3638     } else {
3639         for(i = 1; i < context_count; i++) {
3640             hx = h->thread_context[i];
3641             hx->s.error_recognition = avctx->error_recognition;
3642             hx->s.error_count = 0;
3643         }
3644
3645         avctx->execute(avctx, (void *)decode_slice,
3646                        h->thread_context, NULL, context_count, sizeof(void*));
3647
3648         /* pull back stuff from slices to master context */
3649         hx = h->thread_context[context_count - 1];
3650         s->mb_x = hx->s.mb_x;
3651         s->mb_y = hx->s.mb_y;
3652         s->dropable = hx->s.dropable;
3653         s->picture_structure = hx->s.picture_structure;
3654         for(i = 1; i < context_count; i++)
3655             h->s.error_count += h->thread_context[i]->s.error_count;
3656     }
3657
3658     return 0;
3659 }
3660
3661
3662 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
3663     MpegEncContext * const s = &h->s;
3664     AVCodecContext * const avctx= s->avctx;
3665     H264Context *hx; ///< thread context
3666     int buf_index;
3667     int context_count;
3668     int next_avc;
3669     int pass = !(avctx->active_thread_type & FF_THREAD_FRAME);
3670     int nals_needed=0; ///< number of NALs that need decoding before the next frame thread starts
3671     int nal_index;
3672
3673     h->max_contexts = (HAVE_THREADS && (s->avctx->active_thread_type&FF_THREAD_SLICE)) ? avctx->thread_count : 1;
3674     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
3675         h->current_slice = 0;
3676         if (!s->first_field)
3677             s->current_picture_ptr= NULL;
3678         ff_h264_reset_sei(h);
3679     }
3680
3681     for(;pass <= 1;pass++){
3682         buf_index = 0;
3683         context_count = 0;
3684         next_avc = h->is_avc ? 0 : buf_size;
3685         nal_index = 0;
3686     for(;;){
3687         int consumed;
3688         int dst_length;
3689         int bit_length;
3690         const uint8_t *ptr;
3691         int i, nalsize = 0;
3692         int err;
3693
3694         if(buf_index >= next_avc) {
3695             if(buf_index >= buf_size) break;
3696             nalsize = 0;
3697             for(i = 0; i < h->nal_length_size; i++)
3698                 nalsize = (nalsize << 8) | buf[buf_index++];
3699             if(nalsize <= 0 || nalsize > buf_size - buf_index){
3700                 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
3701                 break;
3702             }
3703             next_avc= buf_index + nalsize;
3704         } else {
3705             // start code prefix search
3706             for(; buf_index + 3 < next_avc; buf_index++){
3707                 // This should always succeed in the first iteration.
3708                 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
3709                     break;
3710             }
3711
3712             if(buf_index+3 >= buf_size) break;
3713
3714             buf_index+=3;
3715             if(buf_index >= next_avc) continue;
3716         }
3717
3718         hx = h->thread_context[context_count];
3719
3720         ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
3721         if (ptr==NULL || dst_length < 0){
3722             return -1;
3723         }
3724         i= buf_index + consumed;
3725         if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
3726            buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
3727             s->workaround_bugs |= FF_BUG_TRUNCATED;
3728
3729         if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
3730         while(ptr[dst_length - 1] == 0 && dst_length > 0)
3731             dst_length--;
3732         }
3733         bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
3734
3735         if(s->avctx->debug&FF_DEBUG_STARTCODE){
3736             av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d/%d length %d\n", hx->nal_unit_type, buf_index, buf_size, dst_length);
3737         }
3738
3739         if (h->is_avc && (nalsize != consumed) && nalsize){
3740             av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
3741         }
3742
3743         buf_index += consumed;
3744         nal_index++;
3745
3746         if(pass == 0) {
3747             // packets can sometimes contain multiple PPS/SPS
3748             // e.g. two PAFF field pictures in one packet, or a demuxer which splits NALs strangely
3749             // if so, when frame threading we can't start the next thread until we've read all of them
3750             switch (hx->nal_unit_type) {
3751                 case NAL_SPS:
3752                 case NAL_PPS:
3753                     nals_needed = nal_index;
3754                     break;
3755                 case NAL_IDR_SLICE:
3756                 case NAL_SLICE:
3757                     init_get_bits(&hx->s.gb, ptr, bit_length);
3758                     if (!get_ue_golomb(&hx->s.gb))
3759                         nals_needed = nal_index;
3760             }
3761             continue;
3762         }
3763
3764         //FIXME do not discard SEI id
3765         if(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0)
3766             continue;
3767
3768       again:
3769         err = 0;
3770         switch(hx->nal_unit_type){
3771         case NAL_IDR_SLICE:
3772             if (h->nal_unit_type != NAL_IDR_SLICE) {
3773                 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
3774                 return -1;
3775             }
3776             idr(h); //FIXME ensure we don't loose some frames if there is reordering
3777         case NAL_SLICE:
3778             init_get_bits(&hx->s.gb, ptr, bit_length);
3779             hx->intra_gb_ptr=
3780             hx->inter_gb_ptr= &hx->s.gb;
3781             hx->s.data_partitioning = 0;
3782
3783             if((err = decode_slice_header(hx, h)))
3784                break;
3785
3786             s->current_picture_ptr->f.key_frame |=
3787                     (hx->nal_unit_type == NAL_IDR_SLICE) ||
3788                     (h->sei_recovery_frame_cnt >= 0);
3789
3790             if (h->current_slice == 1) {
3791                 if(!(s->flags2 & CODEC_FLAG2_CHUNKS)) {
3792                     decode_postinit(h, nal_index >= nals_needed);
3793                 }
3794
3795                 if (s->avctx->hwaccel && s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
3796                     return -1;
3797                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
3798                     ff_vdpau_h264_picture_start(s);
3799             }
3800
3801             if(hx->redundant_pic_count==0
3802                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
3803                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
3804                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
3805                && avctx->skip_frame < AVDISCARD_ALL){
3806                 if(avctx->hwaccel) {
3807                     if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
3808                         return -1;
3809                 }else
3810                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
3811                     static const uint8_t start_code[] = {0x00, 0x00, 0x01};
3812                     ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
3813                     ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
3814                 }else
3815                     context_count++;
3816             }
3817             break;
3818         case NAL_DPA:
3819             init_get_bits(&hx->s.gb, ptr, bit_length);
3820             hx->intra_gb_ptr=
3821             hx->inter_gb_ptr= NULL;
3822
3823             if ((err = decode_slice_header(hx, h)) < 0)
3824                 break;
3825
3826             hx->s.data_partitioning = 1;
3827
3828             break;
3829         case NAL_DPB:
3830             init_get_bits(&hx->intra_gb, ptr, bit_length);
3831             hx->intra_gb_ptr= &hx->intra_gb;
3832             break;
3833         case NAL_DPC:
3834             init_get_bits(&hx->inter_gb, ptr, bit_length);
3835             hx->inter_gb_ptr= &hx->inter_gb;
3836
3837             if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
3838                && s->context_initialized
3839                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
3840                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
3841                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
3842                && avctx->skip_frame < AVDISCARD_ALL)
3843                 context_count++;
3844             break;
3845         case NAL_SEI:
3846             init_get_bits(&s->gb, ptr, bit_length);
3847             ff_h264_decode_sei(h);
3848             break;
3849         case NAL_SPS:
3850             init_get_bits(&s->gb, ptr, bit_length);
3851             ff_h264_decode_seq_parameter_set(h);
3852
3853             if (s->flags& CODEC_FLAG_LOW_DELAY ||
3854                 (h->sps.bitstream_restriction_flag && !h->sps.num_reorder_frames))
3855                 s->low_delay=1;
3856
3857             if(avctx->has_b_frames < 2)
3858                 avctx->has_b_frames= !s->low_delay;
3859
3860             if (avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
3861                 h->cur_chroma_format_idc != h->sps.chroma_format_idc) {
3862                 if (h->sps.bit_depth_luma >= 8 && h->sps.bit_depth_luma <= 10) {
3863                     avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
3864                     h->cur_chroma_format_idc = h->sps.chroma_format_idc;
3865                     h->pixel_shift = h->sps.bit_depth_luma > 8;
3866
3867                     ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma, h->sps.chroma_format_idc);
3868                     ff_h264_pred_init(&h->hpc, s->codec_id, h->sps.bit_depth_luma, h->sps.chroma_format_idc);
3869                     s->dsp.dct_bits = h->sps.bit_depth_luma > 8 ? 32 : 16;
3870                     dsputil_init(&s->dsp, s->avctx);
3871                 } else {
3872                     av_log(avctx, AV_LOG_ERROR, "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
3873                     return -1;
3874                 }
3875             }
3876             break;
3877         case NAL_PPS:
3878             init_get_bits(&s->gb, ptr, bit_length);
3879
3880             ff_h264_decode_picture_parameter_set(h, bit_length);
3881
3882             break;
3883         case NAL_AUD:
3884         case NAL_END_SEQUENCE:
3885         case NAL_END_STREAM:
3886         case NAL_FILLER_DATA:
3887         case NAL_SPS_EXT:
3888         case NAL_AUXILIARY_SLICE:
3889             break;
3890         default:
3891             av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
3892         }
3893
3894         if(context_count == h->max_contexts) {
3895             execute_decode_slices(h, context_count);
3896             context_count = 0;
3897         }
3898
3899         if (err < 0)
3900             av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
3901         else if(err == 1) {
3902             /* Slice could not be decoded in parallel mode, copy down
3903              * NAL unit stuff to context 0 and restart. Note that
3904              * rbsp_buffer is not transferred, but since we no longer
3905              * run in parallel mode this should not be an issue. */
3906             h->nal_unit_type = hx->nal_unit_type;
3907             h->nal_ref_idc   = hx->nal_ref_idc;
3908             hx = h;
3909             goto again;
3910         }
3911     }
3912     }
3913     if(context_count)
3914         execute_decode_slices(h, context_count);
3915     return buf_index;
3916 }
3917
3918 /**
3919  * returns the number of bytes consumed for building the current frame
3920  */
3921 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
3922         if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
3923         if(pos+10>buf_size) pos=buf_size; // oops ;)
3924
3925         return pos;
3926 }
3927
3928 static int decode_frame(AVCodecContext *avctx,
3929                              void *data, int *data_size,
3930                              AVPacket *avpkt)
3931 {
3932     const uint8_t *buf = avpkt->data;
3933     int buf_size = avpkt->size;
3934     H264Context *h = avctx->priv_data;
3935     MpegEncContext *s = &h->s;
3936     AVFrame *pict = data;
3937     int buf_index;
3938
3939     s->flags= avctx->flags;
3940     s->flags2= avctx->flags2;
3941
3942    /* end of stream, output what is still in the buffers */
3943  out:
3944     if (buf_size == 0) {
3945         Picture *out;
3946         int i, out_idx;
3947
3948         s->current_picture_ptr = NULL;
3949
3950 //FIXME factorize this with the output code below
3951         out = h->delayed_pic[0];
3952         out_idx = 0;
3953         for (i = 1; h->delayed_pic[i] && !h->delayed_pic[i]->f.key_frame && !h->delayed_pic[i]->mmco_reset; i++)
3954             if(h->delayed_pic[i]->poc < out->poc){
3955                 out = h->delayed_pic[i];
3956                 out_idx = i;
3957             }
3958
3959         for(i=out_idx; h->delayed_pic[i]; i++)
3960             h->delayed_pic[i] = h->delayed_pic[i+1];
3961
3962         if(out){
3963             *data_size = sizeof(AVFrame);
3964             *pict= *(AVFrame*)out;
3965         }
3966
3967         return 0;
3968     }
3969
3970     buf_index=decode_nal_units(h, buf, buf_size);
3971     if(buf_index < 0)
3972         return -1;
3973
3974     if (!s->current_picture_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
3975         buf_size = 0;
3976         goto out;
3977     }
3978
3979     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
3980         if (avctx->skip_frame >= AVDISCARD_NONREF)
3981             return 0;
3982         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
3983         return -1;
3984     }
3985
3986     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
3987
3988         if(s->flags2 & CODEC_FLAG2_CHUNKS) decode_postinit(h, 1);
3989
3990         field_end(h, 0);
3991
3992         if (!h->next_output_pic) {
3993             /* Wait for second field. */
3994             *data_size = 0;
3995
3996         } else {
3997             *data_size = sizeof(AVFrame);
3998             *pict = *(AVFrame*)h->next_output_pic;
3999         }
4000     }
4001
4002     assert(pict->data[0] || !*data_size);
4003     ff_print_debug_info(s, pict);
4004 //printf("out %d\n", (int)pict->data[0]);
4005
4006     return get_consumed_bytes(s, buf_index, buf_size);
4007 }
4008 #if 0
4009 static inline void fill_mb_avail(H264Context *h){
4010     MpegEncContext * const s = &h->s;
4011     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4012
4013     if(s->mb_y){
4014         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
4015         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
4016         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
4017     }else{
4018         h->mb_avail[0]=
4019         h->mb_avail[1]=
4020         h->mb_avail[2]= 0;
4021     }
4022     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
4023     h->mb_avail[4]= 1; //FIXME move out
4024     h->mb_avail[5]= 0; //FIXME