Fix decoding of lossless 4:2:2 H.264
[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 av_always_inline void
442 mc_dir_part(H264Context *h, Picture *pic, int n, int square,
443             int height, int delta, int list,
444             uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
445             int src_x_offset, int src_y_offset,
446             qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op,
447             int pixel_shift, int chroma_idc)
448 {
449     MpegEncContext * const s = &h->s;
450     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
451     int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
452     const int luma_xy= (mx&3) + ((my&3)<<2);
453     int offset = ((mx>>2) << pixel_shift) + (my>>2)*h->mb_linesize;
454     uint8_t * src_y = pic->f.data[0] + offset;
455     uint8_t * src_cb, * src_cr;
456     int extra_width= h->emu_edge_width;
457     int extra_height= h->emu_edge_height;
458     int emu=0;
459     const int full_mx= mx>>2;
460     const int full_my= my>>2;
461     const int pic_width  = 16*s->mb_width;
462     const int pic_height = 16*s->mb_height >> MB_FIELD;
463     int ysh;
464
465     if(mx&7) extra_width -= 3;
466     if(my&7) extra_height -= 3;
467
468     if(   full_mx < 0-extra_width
469        || full_my < 0-extra_height
470        || full_mx + 16/*FIXME*/ > pic_width + extra_width
471        || full_my + 16/*FIXME*/ > pic_height + extra_height){
472         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_y - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
473                                 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
474             src_y= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
475         emu=1;
476     }
477
478     qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
479     if(!square){
480         qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
481     }
482
483     if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
484
485     if(chroma_idc == 3 /* yuv444 */){
486         src_cb = pic->f.data[1] + offset;
487         if(emu){
488             s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
489                                     16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
490             src_cb= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
491         }
492         qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); //FIXME try variable height perhaps?
493         if(!square){
494             qpix_op[luma_xy](dest_cb + delta, src_cb + delta, h->mb_linesize);
495         }
496
497         src_cr = pic->f.data[2] + offset;
498         if(emu){
499             s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
500                                     16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
501             src_cr= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
502         }
503         qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); //FIXME try variable height perhaps?
504         if(!square){
505             qpix_op[luma_xy](dest_cr + delta, src_cr + delta, h->mb_linesize);
506         }
507         return;
508     }
509
510     ysh = 3 - (chroma_idc == 2 /* yuv422 */);
511     if(chroma_idc == 1 /* yuv420 */ && MB_FIELD){
512         // chroma offset when predicting from a field of opposite parity
513         my += 2 * ((s->mb_y & 1) - (pic->f.reference - 1));
514         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
515     }
516
517     src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) + (my >> ysh) * h->mb_uvlinesize;
518     src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) + (my >> ysh) * h->mb_uvlinesize;
519
520     if(emu){
521         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize,
522                                 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
523                                 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
524             src_cb= s->edge_emu_buffer;
525     }
526     chroma_op(dest_cb, src_cb, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
527               mx&7, (my << (chroma_idc == 2 /* yuv422 */)) &7);
528
529     if(emu){
530         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize,
531                                 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
532                                 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
533             src_cr= s->edge_emu_buffer;
534     }
535     chroma_op(dest_cr, src_cr, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
536               mx&7, (my << (chroma_idc == 2 /* yuv422 */)) &7);
537 }
538
539 static av_always_inline void
540 mc_part_std(H264Context *h, int n, int square, int height, int delta,
541             uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
542             int x_offset, int y_offset,
543             qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
544             qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
545             int list0, int list1, int pixel_shift, int chroma_idc)
546 {
547     MpegEncContext * const s = &h->s;
548     qpel_mc_func *qpix_op=  qpix_put;
549     h264_chroma_mc_func chroma_op= chroma_put;
550
551     dest_y  += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
552     if (chroma_idc == 3 /* yuv444 */) {
553         dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
554         dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
555     } else if (chroma_idc == 2 /* yuv422 */) {
556         dest_cb += (  x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
557         dest_cr += (  x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
558     } else /* yuv420 */ {
559         dest_cb += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
560         dest_cr += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
561     }
562     x_offset += 8*s->mb_x;
563     y_offset += 8*(s->mb_y >> MB_FIELD);
564
565     if(list0){
566         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
567         mc_dir_part(h, ref, n, square, height, delta, 0,
568                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
569                            qpix_op, chroma_op, pixel_shift, chroma_idc);
570
571         qpix_op=  qpix_avg;
572         chroma_op= chroma_avg;
573     }
574
575     if(list1){
576         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
577         mc_dir_part(h, ref, n, square, height, delta, 1,
578                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
579                            qpix_op, chroma_op, pixel_shift, chroma_idc);
580     }
581 }
582
583 static av_always_inline void
584 mc_part_weighted(H264Context *h, int n, int square, int height, int delta,
585                  uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
586                  int x_offset, int y_offset,
587                  qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
588                  h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
589                  h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
590                  int list0, int list1, int pixel_shift, int chroma_idc){
591     MpegEncContext * const s = &h->s;
592     int chroma_height;
593
594     dest_y += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
595     if (chroma_idc == 3 /* yuv444 */) {
596         chroma_height = height;
597         chroma_weight_avg = luma_weight_avg;
598         chroma_weight_op = luma_weight_op;
599         dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
600         dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
601     } else if (chroma_idc == 2 /* yuv422 */) {
602         chroma_height = height;
603         dest_cb += (  x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
604         dest_cr += (  x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
605     } else /* yuv420 */ {
606         chroma_height = height >> 1;
607         dest_cb += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
608         dest_cr += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
609     }
610     x_offset += 8*s->mb_x;
611     y_offset += 8*(s->mb_y >> MB_FIELD);
612
613     if(list0 && list1){
614         /* don't optimize for luma-only case, since B-frames usually
615          * use implicit weights => chroma too. */
616         uint8_t *tmp_cb = s->obmc_scratchpad;
617         uint8_t *tmp_cr = s->obmc_scratchpad + (16 << pixel_shift);
618         uint8_t *tmp_y  = s->obmc_scratchpad + 16*h->mb_uvlinesize;
619         int refn0 = h->ref_cache[0][ scan8[n] ];
620         int refn1 = h->ref_cache[1][ scan8[n] ];
621
622         mc_dir_part(h, &h->ref_list[0][refn0], n, square, height, delta, 0,
623                     dest_y, dest_cb, dest_cr,
624                     x_offset, y_offset, qpix_put, chroma_put,
625                     pixel_shift, chroma_idc);
626         mc_dir_part(h, &h->ref_list[1][refn1], n, square, height, delta, 1,
627                     tmp_y, tmp_cb, tmp_cr,
628                     x_offset, y_offset, qpix_put, chroma_put,
629                     pixel_shift, chroma_idc);
630
631         if(h->use_weight == 2){
632             int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
633             int weight1 = 64 - weight0;
634             luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize,
635                               height,        5, weight0, weight1, 0);
636             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize,
637                               chroma_height, 5, weight0, weight1, 0);
638             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize,
639                               chroma_height, 5, weight0, weight1, 0);
640         }else{
641             luma_weight_avg(dest_y, tmp_y, h->mb_linesize, height, h->luma_log2_weight_denom,
642                             h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
643                             h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
644             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
645                             h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
646                             h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
647             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
648                             h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
649                             h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
650         }
651     }else{
652         int list = list1 ? 1 : 0;
653         int refn = h->ref_cache[list][ scan8[n] ];
654         Picture *ref= &h->ref_list[list][refn];
655         mc_dir_part(h, ref, n, square, height, delta, list,
656                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
657                     qpix_put, chroma_put, pixel_shift, chroma_idc);
658
659         luma_weight_op(dest_y, h->mb_linesize, height, h->luma_log2_weight_denom,
660                        h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
661         if(h->use_weight_chroma){
662             chroma_weight_op(dest_cb, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
663                              h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
664             chroma_weight_op(dest_cr, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
665                              h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
666         }
667     }
668 }
669
670 static av_always_inline void
671 mc_part(H264Context *h, int n, int square, int height, int delta,
672         uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
673         int x_offset, int y_offset,
674         qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
675         qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
676         h264_weight_func *weight_op, h264_biweight_func *weight_avg,
677         int list0, int list1, int pixel_shift, int chroma_idc)
678 {
679     if((h->use_weight==2 && list0 && list1
680         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
681        || h->use_weight==1)
682         mc_part_weighted(h, n, square, height, delta, dest_y, dest_cb, dest_cr,
683                          x_offset, y_offset, qpix_put, chroma_put,
684                          weight_op[0], weight_op[1], weight_avg[0],
685                          weight_avg[1], list0, list1, pixel_shift, chroma_idc);
686     else
687         mc_part_std(h, n, square, height, delta, dest_y, dest_cb, dest_cr,
688                     x_offset, y_offset, qpix_put, chroma_put, qpix_avg,
689                     chroma_avg, list0, list1, pixel_shift, chroma_idc);
690 }
691
692 static av_always_inline void
693 prefetch_motion(H264Context *h, int list, int pixel_shift, int chroma_idc)
694 {
695     /* fetch pixels for estimated mv 4 macroblocks ahead
696      * optimized for 64byte cache lines */
697     MpegEncContext * const s = &h->s;
698     const int refn = h->ref_cache[list][scan8[0]];
699     if(refn >= 0){
700         const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
701         const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
702         uint8_t **src = h->ref_list[list][refn].f.data;
703         int off= (mx << pixel_shift) + (my + (s->mb_x&3)*4)*h->mb_linesize + (64 << pixel_shift);
704         s->dsp.prefetch(src[0]+off, s->linesize, 4);
705         if (chroma_idc == 3 /* yuv444 */) {
706             s->dsp.prefetch(src[1]+off, s->linesize, 4);
707             s->dsp.prefetch(src[2]+off, s->linesize, 4);
708         }else{
709             off= ((mx>>1) << pixel_shift) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + (64 << pixel_shift);
710             s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
711         }
712     }
713 }
714
715 static av_always_inline void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
716                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
717                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
718                       h264_weight_func *weight_op, h264_biweight_func *weight_avg,
719                       int pixel_shift, int chroma_idc)
720 {
721     MpegEncContext * const s = &h->s;
722     const int mb_xy= h->mb_xy;
723     const int mb_type = s->current_picture.f.mb_type[mb_xy];
724
725     assert(IS_INTER(mb_type));
726
727     if(HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
728         await_references(h);
729     prefetch_motion(h, 0, pixel_shift, chroma_idc);
730
731     if(IS_16X16(mb_type)){
732         mc_part(h, 0, 1, 16, 0, dest_y, dest_cb, dest_cr, 0, 0,
733                 qpix_put[0], chroma_put[0], qpix_avg[0], 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     }else if(IS_16X8(mb_type)){
738         mc_part(h, 0, 0, 8, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 0,
739                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
740                 weight_op, weight_avg,
741                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
742                 pixel_shift, chroma_idc);
743         mc_part(h, 8, 0, 8, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 4,
744                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
745                 weight_op, weight_avg,
746                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
747                 pixel_shift, chroma_idc);
748     }else if(IS_8X16(mb_type)){
749         mc_part(h, 0, 0, 16, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
750                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
751                 &weight_op[1], &weight_avg[1],
752                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
753                 pixel_shift, chroma_idc);
754         mc_part(h, 4, 0, 16, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
755                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
756                 &weight_op[1], &weight_avg[1],
757                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
758                 pixel_shift, chroma_idc);
759     }else{
760         int i;
761
762         assert(IS_8X8(mb_type));
763
764         for(i=0; i<4; i++){
765             const int sub_mb_type= h->sub_mb_type[i];
766             const int n= 4*i;
767             int x_offset= (i&1)<<2;
768             int y_offset= (i&2)<<1;
769
770             if(IS_SUB_8X8(sub_mb_type)){
771                 mc_part(h, n, 1, 8, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
772                     qpix_put[1], chroma_put[1], qpix_avg[1], 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             }else if(IS_SUB_8X4(sub_mb_type)){
777                 mc_part(h, n  , 0, 4, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset,
778                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
779                     &weight_op[1], &weight_avg[1],
780                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
781                     pixel_shift, chroma_idc);
782                 mc_part(h, n+2, 0, 4, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
783                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
784                     &weight_op[1], &weight_avg[1],
785                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
786                     pixel_shift, chroma_idc);
787             }else if(IS_SUB_4X8(sub_mb_type)){
788                 mc_part(h, n  , 0, 8, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
789                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
790                     &weight_op[2], &weight_avg[2],
791                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
792                     pixel_shift, chroma_idc);
793                 mc_part(h, n+1, 0, 8, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
794                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
795                     &weight_op[2], &weight_avg[2],
796                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
797                     pixel_shift, chroma_idc);
798             }else{
799                 int j;
800                 assert(IS_SUB_4X4(sub_mb_type));
801                 for(j=0; j<4; j++){
802                     int sub_x_offset= x_offset + 2*(j&1);
803                     int sub_y_offset= y_offset +   (j&2);
804                     mc_part(h, n+j, 1, 4, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
805                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
806                         &weight_op[2], &weight_avg[2],
807                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
808                         pixel_shift, chroma_idc);
809                 }
810             }
811         }
812     }
813
814     prefetch_motion(h, 1, pixel_shift, chroma_idc);
815 }
816
817 static av_always_inline void
818 hl_motion_420(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
819               qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
820               qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
821               h264_weight_func *weight_op, h264_biweight_func *weight_avg,
822               int pixel_shift)
823 {
824     hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put,
825               qpix_avg, chroma_avg, weight_op, weight_avg, pixel_shift, 1);
826 }
827
828 static av_always_inline void
829 hl_motion_422(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
830               qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
831               qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
832               h264_weight_func *weight_op, h264_biweight_func *weight_avg,
833               int pixel_shift)
834 {
835     hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put,
836               qpix_avg, chroma_avg, weight_op, weight_avg, pixel_shift, 2);
837 }
838
839 static void free_tables(H264Context *h, int free_rbsp){
840     int i;
841     H264Context *hx;
842
843     av_freep(&h->intra4x4_pred_mode);
844     av_freep(&h->chroma_pred_mode_table);
845     av_freep(&h->cbp_table);
846     av_freep(&h->mvd_table[0]);
847     av_freep(&h->mvd_table[1]);
848     av_freep(&h->direct_table);
849     av_freep(&h->non_zero_count);
850     av_freep(&h->slice_table_base);
851     h->slice_table= NULL;
852     av_freep(&h->list_counts);
853
854     av_freep(&h->mb2b_xy);
855     av_freep(&h->mb2br_xy);
856
857     for(i = 0; i < MAX_THREADS; i++) {
858         hx = h->thread_context[i];
859         if(!hx) continue;
860         av_freep(&hx->top_borders[1]);
861         av_freep(&hx->top_borders[0]);
862         av_freep(&hx->s.obmc_scratchpad);
863         if (free_rbsp){
864             av_freep(&hx->rbsp_buffer[1]);
865             av_freep(&hx->rbsp_buffer[0]);
866             hx->rbsp_buffer_size[0] = 0;
867             hx->rbsp_buffer_size[1] = 0;
868         }
869         if (i) av_freep(&h->thread_context[i]);
870     }
871 }
872
873 static void init_dequant8_coeff_table(H264Context *h){
874     int i,j,q,x;
875     const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
876
877     for(i=0; i<6; i++ ){
878         h->dequant8_coeff[i] = h->dequant8_buffer[i];
879         for(j=0; j<i; j++){
880             if(!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i], 64*sizeof(uint8_t))){
881                 h->dequant8_coeff[i] = h->dequant8_buffer[j];
882                 break;
883             }
884         }
885         if(j<i)
886             continue;
887
888         for(q=0; q<max_qp+1; q++){
889             int shift = div6[q];
890             int idx = rem6[q];
891             for(x=0; x<64; x++)
892                 h->dequant8_coeff[i][q][(x>>3)|((x&7)<<3)] =
893                     ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
894                     h->pps.scaling_matrix8[i][x]) << shift;
895         }
896     }
897 }
898
899 static void init_dequant4_coeff_table(H264Context *h){
900     int i,j,q,x;
901     const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
902     for(i=0; i<6; i++ ){
903         h->dequant4_coeff[i] = h->dequant4_buffer[i];
904         for(j=0; j<i; j++){
905             if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
906                 h->dequant4_coeff[i] = h->dequant4_buffer[j];
907                 break;
908             }
909         }
910         if(j<i)
911             continue;
912
913         for(q=0; q<max_qp+1; q++){
914             int shift = div6[q] + 2;
915             int idx = rem6[q];
916             for(x=0; x<16; x++)
917                 h->dequant4_coeff[i][q][(x>>2)|((x<<2)&0xF)] =
918                     ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
919                     h->pps.scaling_matrix4[i][x]) << shift;
920         }
921     }
922 }
923
924 static void init_dequant_tables(H264Context *h){
925     int i,x;
926     init_dequant4_coeff_table(h);
927     if(h->pps.transform_8x8_mode)
928         init_dequant8_coeff_table(h);
929     if(h->sps.transform_bypass){
930         for(i=0; i<6; i++)
931             for(x=0; x<16; x++)
932                 h->dequant4_coeff[i][0][x] = 1<<6;
933         if(h->pps.transform_8x8_mode)
934             for(i=0; i<6; i++)
935                 for(x=0; x<64; x++)
936                     h->dequant8_coeff[i][0][x] = 1<<6;
937     }
938 }
939
940
941 int ff_h264_alloc_tables(H264Context *h){
942     MpegEncContext * const s = &h->s;
943     const int big_mb_num= s->mb_stride * (s->mb_height+1);
944     const int row_mb_num= 2*s->mb_stride*s->avctx->thread_count;
945     int x,y;
946
947     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8  * sizeof(uint8_t), fail)
948
949     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 48 * sizeof(uint8_t), fail)
950     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
951     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
952
953     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
954     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
955     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
956     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
957     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
958
959     memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
960     h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
961
962     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy  , big_mb_num * sizeof(uint32_t), fail);
963     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
964     for(y=0; y<s->mb_height; y++){
965         for(x=0; x<s->mb_width; x++){
966             const int mb_xy= x + y*s->mb_stride;
967             const int b_xy = 4*x + 4*y*h->b_stride;
968
969             h->mb2b_xy [mb_xy]= b_xy;
970             h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
971         }
972     }
973
974     s->obmc_scratchpad = NULL;
975
976     if(!h->dequant4_coeff[0])
977         init_dequant_tables(h);
978
979     return 0;
980 fail:
981     free_tables(h, 1);
982     return -1;
983 }
984
985 /**
986  * Mimic alloc_tables(), but for every context thread.
987  */
988 static void clone_tables(H264Context *dst, H264Context *src, int i){
989     MpegEncContext * const s = &src->s;
990     dst->intra4x4_pred_mode       = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
991     dst->non_zero_count           = src->non_zero_count;
992     dst->slice_table              = src->slice_table;
993     dst->cbp_table                = src->cbp_table;
994     dst->mb2b_xy                  = src->mb2b_xy;
995     dst->mb2br_xy                 = src->mb2br_xy;
996     dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
997     dst->mvd_table[0]             = src->mvd_table[0] + i*8*2*s->mb_stride;
998     dst->mvd_table[1]             = src->mvd_table[1] + i*8*2*s->mb_stride;
999     dst->direct_table             = src->direct_table;
1000     dst->list_counts              = src->list_counts;
1001
1002     dst->s.obmc_scratchpad = NULL;
1003     ff_h264_pred_init(&dst->hpc, src->s.codec_id, src->sps.bit_depth_luma, src->sps.chroma_format_idc);
1004 }
1005
1006 /**
1007  * Init context
1008  * Allocate buffers which are not shared amongst multiple threads.
1009  */
1010 static int context_init(H264Context *h){
1011     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
1012     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
1013
1014     h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
1015     h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
1016
1017     return 0;
1018 fail:
1019     return -1; // free_tables will clean up for us
1020 }
1021
1022 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
1023
1024 static av_cold void common_init(H264Context *h){
1025     MpegEncContext * const s = &h->s;
1026
1027     s->width = s->avctx->width;
1028     s->height = s->avctx->height;
1029     s->codec_id= s->avctx->codec->id;
1030
1031     ff_h264dsp_init(&h->h264dsp, 8, 1);
1032     ff_h264_pred_init(&h->hpc, s->codec_id, 8, 1);
1033
1034     h->dequant_coeff_pps= -1;
1035     s->unrestricted_mv=1;
1036
1037     dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
1038
1039     memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
1040     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
1041 }
1042
1043 int ff_h264_decode_extradata(H264Context *h)
1044 {
1045     AVCodecContext *avctx = h->s.avctx;
1046
1047     if(avctx->extradata[0] == 1){
1048         int i, cnt, nalsize;
1049         unsigned char *p = avctx->extradata;
1050
1051         h->is_avc = 1;
1052
1053         if(avctx->extradata_size < 7) {
1054             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
1055             return -1;
1056         }
1057         /* sps and pps in the avcC always have length coded with 2 bytes,
1058            so put a fake nal_length_size = 2 while parsing them */
1059         h->nal_length_size = 2;
1060         // Decode sps from avcC
1061         cnt = *(p+5) & 0x1f; // Number of sps
1062         p += 6;
1063         for (i = 0; i < cnt; i++) {
1064             nalsize = AV_RB16(p) + 2;
1065             if (p - avctx->extradata + nalsize > avctx->extradata_size)
1066                 return -1;
1067             if(decode_nal_units(h, p, nalsize) < 0) {
1068                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
1069                 return -1;
1070             }
1071             p += nalsize;
1072         }
1073         // Decode pps from avcC
1074         cnt = *(p++); // Number of pps
1075         for (i = 0; i < cnt; i++) {
1076             nalsize = AV_RB16(p) + 2;
1077             if (p - avctx->extradata + nalsize > avctx->extradata_size)
1078                 return -1;
1079             if (decode_nal_units(h, p, nalsize) < 0) {
1080                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
1081                 return -1;
1082             }
1083             p += nalsize;
1084         }
1085         // Now store right nal length size, that will be use to parse all other nals
1086         h->nal_length_size = (avctx->extradata[4] & 0x03) + 1;
1087     } else {
1088         h->is_avc = 0;
1089         if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
1090             return -1;
1091     }
1092     return 0;
1093 }
1094
1095 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
1096     H264Context *h= avctx->priv_data;
1097     MpegEncContext * const s = &h->s;
1098
1099     MPV_decode_defaults(s);
1100
1101     s->avctx = avctx;
1102     common_init(h);
1103
1104     s->out_format = FMT_H264;
1105     s->workaround_bugs= avctx->workaround_bugs;
1106
1107     // set defaults
1108 //    s->decode_mb= ff_h263_decode_mb;
1109     s->quarter_sample = 1;
1110     if(!avctx->has_b_frames)
1111     s->low_delay= 1;
1112
1113     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
1114
1115     ff_h264_decode_init_vlc();
1116
1117     h->pixel_shift = 0;
1118     h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
1119
1120     h->thread_context[0] = h;
1121     h->outputed_poc = h->next_outputed_poc = INT_MIN;
1122     h->prev_poc_msb= 1<<16;
1123     h->x264_build = -1;
1124     ff_h264_reset_sei(h);
1125     if(avctx->codec_id == CODEC_ID_H264){
1126         if(avctx->ticks_per_frame == 1){
1127             s->avctx->time_base.den *=2;
1128         }
1129         avctx->ticks_per_frame = 2;
1130     }
1131
1132     if(avctx->extradata_size > 0 && avctx->extradata &&
1133         ff_h264_decode_extradata(h))
1134         return -1;
1135
1136     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
1137         s->avctx->has_b_frames = h->sps.num_reorder_frames;
1138         s->low_delay = 0;
1139     }
1140
1141     return 0;
1142 }
1143
1144 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b)+(size))))
1145 static void copy_picture_range(Picture **to, Picture **from, int count, MpegEncContext *new_base, MpegEncContext *old_base)
1146 {
1147     int i;
1148
1149     for (i=0; i<count; i++){
1150         assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
1151                 IN_RANGE(from[i], old_base->picture, sizeof(Picture) * old_base->picture_count) ||
1152                 !from[i]));
1153         to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1154     }
1155 }
1156
1157 static void copy_parameter_set(void **to, void **from, int count, int size)
1158 {
1159     int i;
1160
1161     for (i=0; i<count; i++){
1162         if (to[i] && !from[i]) av_freep(&to[i]);
1163         else if (from[i] && !to[i]) to[i] = av_malloc(size);
1164
1165         if (from[i]) memcpy(to[i], from[i], size);
1166     }
1167 }
1168
1169 static int decode_init_thread_copy(AVCodecContext *avctx){
1170     H264Context *h= avctx->priv_data;
1171
1172     if (!avctx->is_copy) return 0;
1173     memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1174     memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1175
1176     return 0;
1177 }
1178
1179 #define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
1180 static int decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src){
1181     H264Context *h= dst->priv_data, *h1= src->priv_data;
1182     MpegEncContext * const s = &h->s, * const s1 = &h1->s;
1183     int inited = s->context_initialized, err;
1184     int i;
1185
1186     if(dst == src || !s1->context_initialized) return 0;
1187
1188     err = ff_mpeg_update_thread_context(dst, src);
1189     if(err) return err;
1190
1191     //FIXME handle width/height changing
1192     if(!inited){
1193         for(i = 0; i < MAX_SPS_COUNT; i++)
1194             av_freep(h->sps_buffers + i);
1195
1196         for(i = 0; i < MAX_PPS_COUNT; i++)
1197             av_freep(h->pps_buffers + i);
1198
1199         memcpy(&h->s + 1, &h1->s + 1, sizeof(H264Context) - sizeof(MpegEncContext)); //copy all fields after MpegEnc
1200         memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1201         memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1202         if (ff_h264_alloc_tables(h) < 0) {
1203             av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
1204             return AVERROR(ENOMEM);
1205         }
1206         context_init(h);
1207
1208         for(i=0; i<2; i++){
1209             h->rbsp_buffer[i] = NULL;
1210             h->rbsp_buffer_size[i] = 0;
1211         }
1212
1213         h->thread_context[0] = h;
1214
1215         // frame_start may not be called for the next thread (if it's decoding a bottom field)
1216         // so this has to be allocated here
1217         h->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
1218
1219         s->dsp.clear_blocks(h->mb);
1220         s->dsp.clear_blocks(h->mb+(24*16<<h->pixel_shift));
1221     }
1222
1223     //extradata/NAL handling
1224     h->is_avc          = h1->is_avc;
1225
1226     //SPS/PPS
1227     copy_parameter_set((void**)h->sps_buffers, (void**)h1->sps_buffers, MAX_SPS_COUNT, sizeof(SPS));
1228     h->sps             = h1->sps;
1229     copy_parameter_set((void**)h->pps_buffers, (void**)h1->pps_buffers, MAX_PPS_COUNT, sizeof(PPS));
1230     h->pps             = h1->pps;
1231
1232     //Dequantization matrices
1233     //FIXME these are big - can they be only copied when PPS changes?
1234     copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1235
1236     for(i=0; i<6; i++)
1237         h->dequant4_coeff[i] = h->dequant4_buffer[0] + (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
1238
1239     for(i=0; i<6; i++)
1240         h->dequant8_coeff[i] = h->dequant8_buffer[0] + (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
1241
1242     h->dequant_coeff_pps = h1->dequant_coeff_pps;
1243
1244     //POC timing
1245     copy_fields(h, h1, poc_lsb, redundant_pic_count);
1246
1247     //reference lists
1248     copy_fields(h, h1, ref_count, list_count);
1249     copy_fields(h, h1, ref_list,  intra_gb);
1250     copy_fields(h, h1, short_ref, cabac_init_idc);
1251
1252     copy_picture_range(h->short_ref,   h1->short_ref,   32, s, s1);
1253     copy_picture_range(h->long_ref,    h1->long_ref,    32, s, s1);
1254     copy_picture_range(h->delayed_pic, h1->delayed_pic, MAX_DELAYED_PIC_COUNT+2, s, s1);
1255
1256     h->last_slice_type = h1->last_slice_type;
1257
1258     if(!s->current_picture_ptr) return 0;
1259
1260     if(!s->dropable) {
1261         err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1262         h->prev_poc_msb     = h->poc_msb;
1263         h->prev_poc_lsb     = h->poc_lsb;
1264     }
1265     h->prev_frame_num_offset= h->frame_num_offset;
1266     h->prev_frame_num       = h->frame_num;
1267     h->outputed_poc         = h->next_outputed_poc;
1268
1269     return err;
1270 }
1271
1272 int ff_h264_frame_start(H264Context *h){
1273     MpegEncContext * const s = &h->s;
1274     int i;
1275     const int pixel_shift = h->pixel_shift;
1276     int thread_count = (s->avctx->active_thread_type & FF_THREAD_SLICE) ? s->avctx->thread_count : 1;
1277
1278     if(MPV_frame_start(s, s->avctx) < 0)
1279         return -1;
1280     ff_er_frame_start(s);
1281     /*
1282      * MPV_frame_start uses pict_type to derive key_frame.
1283      * This is incorrect for H.264; IDR markings must be used.
1284      * Zero here; IDR markings per slice in frame or fields are ORed in later.
1285      * See decode_nal_units().
1286      */
1287     s->current_picture_ptr->f.key_frame = 0;
1288     s->current_picture_ptr->mmco_reset= 0;
1289
1290     assert(s->linesize && s->uvlinesize);
1291
1292     for(i=0; i<16; i++){
1293         h->block_offset[i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
1294         h->block_offset[48+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
1295     }
1296     for(i=0; i<16; i++){
1297         h->block_offset[16+i]=
1298         h->block_offset[32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1299         h->block_offset[48+16+i]=
1300         h->block_offset[48+32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1301     }
1302
1303     /* can't be in alloc_tables because linesize isn't known there.
1304      * FIXME: redo bipred weight to not require extra buffer? */
1305     for(i = 0; i < thread_count; i++)
1306         if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
1307             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
1308
1309     /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
1310     memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
1311
1312 //    s->decode = (s->flags & CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.f.reference /*|| h->contains_intra*/ || 1;
1313
1314     // We mark the current picture as non-reference after allocating it, so
1315     // that if we break out due to an error it can be released automatically
1316     // in the next MPV_frame_start().
1317     // SVQ3 as well as most other codecs have only last/next/current and thus
1318     // get released even with set reference, besides SVQ3 and others do not
1319     // mark frames as reference later "naturally".
1320     if(s->codec_id != CODEC_ID_SVQ3)
1321         s->current_picture_ptr->f.reference = 0;
1322
1323     s->current_picture_ptr->field_poc[0]=
1324     s->current_picture_ptr->field_poc[1]= INT_MAX;
1325
1326     h->next_output_pic = NULL;
1327
1328     assert(s->current_picture_ptr->long_ref==0);
1329
1330     return 0;
1331 }
1332
1333 /**
1334   * Run setup operations that must be run after slice header decoding.
1335   * This includes finding the next displayed frame.
1336   *
1337   * @param h h264 master context
1338   * @param setup_finished enough NALs have been read that we can call
1339   * ff_thread_finish_setup()
1340   */
1341 static void decode_postinit(H264Context *h, int setup_finished){
1342     MpegEncContext * const s = &h->s;
1343     Picture *out = s->current_picture_ptr;
1344     Picture *cur = s->current_picture_ptr;
1345     int i, pics, out_of_order, out_idx;
1346
1347     s->current_picture_ptr->f.qscale_type = FF_QSCALE_TYPE_H264;
1348     s->current_picture_ptr->f.pict_type   = s->pict_type;
1349
1350     if (h->next_output_pic) return;
1351
1352     if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
1353         //FIXME: if we have two PAFF fields in one packet, we can't start the next thread here.
1354         //If we have one field per packet, we can. The check in decode_nal_units() is not good enough
1355         //to find this yet, so we assume the worst for now.
1356         //if (setup_finished)
1357         //    ff_thread_finish_setup(s->avctx);
1358         return;
1359     }
1360
1361     cur->f.interlaced_frame = 0;
1362     cur->f.repeat_pict      = 0;
1363
1364     /* Signal interlacing information externally. */
1365     /* Prioritize picture timing SEI information over used decoding process if it exists. */
1366
1367     if(h->sps.pic_struct_present_flag){
1368         switch (h->sei_pic_struct)
1369         {
1370         case SEI_PIC_STRUCT_FRAME:
1371             break;
1372         case SEI_PIC_STRUCT_TOP_FIELD:
1373         case SEI_PIC_STRUCT_BOTTOM_FIELD:
1374             cur->f.interlaced_frame = 1;
1375             break;
1376         case SEI_PIC_STRUCT_TOP_BOTTOM:
1377         case SEI_PIC_STRUCT_BOTTOM_TOP:
1378             if (FIELD_OR_MBAFF_PICTURE)
1379                 cur->f.interlaced_frame = 1;
1380             else
1381                 // try to flag soft telecine progressive
1382                 cur->f.interlaced_frame = h->prev_interlaced_frame;
1383             break;
1384         case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1385         case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
1386             // Signal the possibility of telecined film externally (pic_struct 5,6)
1387             // From these hints, let the applications decide if they apply deinterlacing.
1388             cur->f.repeat_pict = 1;
1389             break;
1390         case SEI_PIC_STRUCT_FRAME_DOUBLING:
1391             // Force progressive here, as doubling interlaced frame is a bad idea.
1392             cur->f.repeat_pict = 2;
1393             break;
1394         case SEI_PIC_STRUCT_FRAME_TRIPLING:
1395             cur->f.repeat_pict = 4;
1396             break;
1397         }
1398
1399         if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
1400             cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
1401     }else{
1402         /* Derive interlacing flag from used decoding process. */
1403         cur->f.interlaced_frame = FIELD_OR_MBAFF_PICTURE;
1404     }
1405     h->prev_interlaced_frame = cur->f.interlaced_frame;
1406
1407     if (cur->field_poc[0] != cur->field_poc[1]){
1408         /* Derive top_field_first from field pocs. */
1409         cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1];
1410     }else{
1411         if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) {
1412             /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
1413             if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
1414               || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
1415                 cur->f.top_field_first = 1;
1416             else
1417                 cur->f.top_field_first = 0;
1418         }else{
1419             /* Most likely progressive */
1420             cur->f.top_field_first = 0;
1421         }
1422     }
1423
1424     //FIXME do something with unavailable reference frames
1425
1426     /* Sort B-frames into display order */
1427
1428     if(h->sps.bitstream_restriction_flag
1429        && s->avctx->has_b_frames < h->sps.num_reorder_frames){
1430         s->avctx->has_b_frames = h->sps.num_reorder_frames;
1431         s->low_delay = 0;
1432     }
1433
1434     if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
1435        && !h->sps.bitstream_restriction_flag){
1436         s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
1437         s->low_delay= 0;
1438     }
1439
1440     pics = 0;
1441     while(h->delayed_pic[pics]) pics++;
1442
1443     assert(pics <= MAX_DELAYED_PIC_COUNT);
1444
1445     h->delayed_pic[pics++] = cur;
1446     if (cur->f.reference == 0)
1447         cur->f.reference = DELAYED_PIC_REF;
1448
1449     out = h->delayed_pic[0];
1450     out_idx = 0;
1451     for (i = 1; h->delayed_pic[i] && !h->delayed_pic[i]->f.key_frame && !h->delayed_pic[i]->mmco_reset; i++)
1452         if(h->delayed_pic[i]->poc < out->poc){
1453             out = h->delayed_pic[i];
1454             out_idx = i;
1455         }
1456     if (s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset))
1457         h->next_outputed_poc= INT_MIN;
1458     out_of_order = out->poc < h->next_outputed_poc;
1459
1460     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
1461         { }
1462     else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
1463        || (s->low_delay &&
1464         ((h->next_outputed_poc != INT_MIN && out->poc > h->next_outputed_poc + 2)
1465          || cur->f.pict_type == AV_PICTURE_TYPE_B)))
1466     {
1467         s->low_delay = 0;
1468         s->avctx->has_b_frames++;
1469     }
1470
1471     if(out_of_order || pics > s->avctx->has_b_frames){
1472         out->f.reference &= ~DELAYED_PIC_REF;
1473         out->owner2 = s; // for frame threading, the owner must be the second field's thread
1474                          // or else the first thread can release the picture and reuse it unsafely
1475         for(i=out_idx; h->delayed_pic[i]; i++)
1476             h->delayed_pic[i] = h->delayed_pic[i+1];
1477     }
1478     if(!out_of_order && pics > s->avctx->has_b_frames){
1479         h->next_output_pic = out;
1480         if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset)) {
1481             h->next_outputed_poc = INT_MIN;
1482         } else
1483             h->next_outputed_poc = out->poc;
1484     }else{
1485         av_log(s->avctx, AV_LOG_DEBUG, "no picture\n");
1486     }
1487
1488     if (setup_finished)
1489         ff_thread_finish_setup(s->avctx);
1490 }
1491
1492 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
1493                                               uint8_t *src_cb, uint8_t *src_cr,
1494                                               int linesize, int uvlinesize, int simple)
1495 {
1496     MpegEncContext * const s = &h->s;
1497     uint8_t *top_border;
1498     int top_idx = 1;
1499     const int pixel_shift = h->pixel_shift;
1500     int chroma444 = CHROMA444;
1501     int chroma422 = CHROMA422;
1502
1503     src_y  -=   linesize;
1504     src_cb -= uvlinesize;
1505     src_cr -= uvlinesize;
1506
1507     if(!simple && FRAME_MBAFF){
1508         if(s->mb_y&1){
1509             if(!MB_MBAFF){
1510                 top_border = h->top_borders[0][s->mb_x];
1511                 AV_COPY128(top_border, src_y + 15*linesize);
1512                 if (pixel_shift)
1513                     AV_COPY128(top_border+16, src_y+15*linesize+16);
1514                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1515                     if(chroma444){
1516                         if (pixel_shift){
1517                             AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
1518                             AV_COPY128(top_border+48, src_cb + 15*uvlinesize+16);
1519                             AV_COPY128(top_border+64, src_cr + 15*uvlinesize);
1520                             AV_COPY128(top_border+80, src_cr + 15*uvlinesize+16);
1521                         } else {
1522                             AV_COPY128(top_border+16, src_cb + 15*uvlinesize);
1523                             AV_COPY128(top_border+32, src_cr + 15*uvlinesize);
1524                         }
1525                     } else if(chroma422) {
1526                         if (pixel_shift) {
1527                             AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
1528                             AV_COPY128(top_border+48, src_cr + 15*uvlinesize);
1529                         } else {
1530                             AV_COPY64(top_border+16, src_cb +  15*uvlinesize);
1531                             AV_COPY64(top_border+24, src_cr +  15*uvlinesize);
1532                         }
1533                     } else {
1534                         if (pixel_shift) {
1535                             AV_COPY128(top_border+32, src_cb+7*uvlinesize);
1536                             AV_COPY128(top_border+48, src_cr+7*uvlinesize);
1537                         } else {
1538                             AV_COPY64(top_border+16, src_cb+7*uvlinesize);
1539                             AV_COPY64(top_border+24, src_cr+7*uvlinesize);
1540                         }
1541                     }
1542                 }
1543             }
1544         }else if(MB_MBAFF){
1545             top_idx = 0;
1546         }else
1547             return;
1548     }
1549
1550     top_border = h->top_borders[top_idx][s->mb_x];
1551     // There are two lines saved, the line above the the top macroblock of a pair,
1552     // and the line above the bottom macroblock
1553     AV_COPY128(top_border, src_y + 16*linesize);
1554     if (pixel_shift)
1555         AV_COPY128(top_border+16, src_y+16*linesize+16);
1556
1557     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1558         if(chroma444){
1559             if (pixel_shift){
1560                 AV_COPY128(top_border+32, src_cb + 16*linesize);
1561                 AV_COPY128(top_border+48, src_cb + 16*linesize+16);
1562                 AV_COPY128(top_border+64, src_cr + 16*linesize);
1563                 AV_COPY128(top_border+80, src_cr + 16*linesize+16);
1564             } else {
1565                 AV_COPY128(top_border+16, src_cb + 16*linesize);
1566                 AV_COPY128(top_border+32, src_cr + 16*linesize);
1567             }
1568         } else if(chroma422) {
1569             if (pixel_shift) {
1570                 AV_COPY128(top_border+32, src_cb+16*uvlinesize);
1571                 AV_COPY128(top_border+48, src_cr+16*uvlinesize);
1572             } else {
1573                 AV_COPY64(top_border+16, src_cb+16*uvlinesize);
1574                 AV_COPY64(top_border+24, src_cr+16*uvlinesize);
1575             }
1576         } else {
1577             if (pixel_shift) {
1578                 AV_COPY128(top_border+32, src_cb+8*uvlinesize);
1579                 AV_COPY128(top_border+48, src_cr+8*uvlinesize);
1580             } else {
1581                 AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1582                 AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1583             }
1584         }
1585     }
1586 }
1587
1588 static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
1589                                   uint8_t *src_cb, uint8_t *src_cr,
1590                                   int linesize, int uvlinesize,
1591                                   int xchg, int chroma444,
1592                                   int simple, int pixel_shift){
1593     MpegEncContext * const s = &h->s;
1594     int deblock_topleft;
1595     int deblock_top;
1596     int top_idx = 1;
1597     uint8_t *top_border_m1;
1598     uint8_t *top_border;
1599
1600     if(!simple && FRAME_MBAFF){
1601         if(s->mb_y&1){
1602             if(!MB_MBAFF)
1603                 return;
1604         }else{
1605             top_idx = MB_MBAFF ? 0 : 1;
1606         }
1607     }
1608
1609     if(h->deblocking_filter == 2) {
1610         deblock_topleft = h->slice_table[h->mb_xy - 1 - s->mb_stride] == h->slice_num;
1611         deblock_top     = h->top_type;
1612     } else {
1613         deblock_topleft = (s->mb_x > 0);
1614         deblock_top     = (s->mb_y > !!MB_FIELD);
1615     }
1616
1617     src_y  -=   linesize + 1 + pixel_shift;
1618     src_cb -= uvlinesize + 1 + pixel_shift;
1619     src_cr -= uvlinesize + 1 + pixel_shift;
1620
1621     top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1622     top_border    = h->top_borders[top_idx][s->mb_x];
1623
1624 #define XCHG(a,b,xchg)\
1625     if (pixel_shift) {\
1626         if (xchg) {\
1627             AV_SWAP64(b+0,a+0);\
1628             AV_SWAP64(b+8,a+8);\
1629         } else {\
1630             AV_COPY128(b,a); \
1631         }\
1632     } else \
1633 if (xchg) AV_SWAP64(b,a);\
1634 else      AV_COPY64(b,a);
1635
1636     if(deblock_top){
1637         if(deblock_topleft){
1638             XCHG(top_border_m1 + (8 << pixel_shift), src_y - (7 << pixel_shift), 1);
1639         }
1640         XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
1641         XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
1642         if(s->mb_x+1 < s->mb_width){
1643             XCHG(h->top_borders[top_idx][s->mb_x+1], src_y + (17 << pixel_shift), 1);
1644         }
1645     }
1646     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1647         if(chroma444){
1648             if(deblock_topleft){
1649                 XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1650                 XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1651             }
1652             XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
1653             XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
1654             XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
1655             XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
1656             if(s->mb_x+1 < s->mb_width){
1657                 XCHG(h->top_borders[top_idx][s->mb_x+1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
1658                 XCHG(h->top_borders[top_idx][s->mb_x+1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
1659             }
1660         } else {
1661             if(deblock_top){
1662                 if(deblock_topleft){
1663                     XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1664                     XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1665                 }
1666                 XCHG(top_border + (16 << pixel_shift), src_cb+1+pixel_shift, 1);
1667                 XCHG(top_border + (24 << pixel_shift), src_cr+1+pixel_shift, 1);
1668             }
1669         }
1670     }
1671 }
1672
1673 static av_always_inline int dctcoef_get(DCTELEM *mb, int high_bit_depth, int index) {
1674     if (high_bit_depth) {
1675         return AV_RN32A(((int32_t*)mb) + index);
1676     } else
1677         return AV_RN16A(mb + index);
1678 }
1679
1680 static av_always_inline void dctcoef_set(DCTELEM *mb, int high_bit_depth, int index, int value) {
1681     if (high_bit_depth) {
1682         AV_WN32A(((int32_t*)mb) + index, value);
1683     } else
1684         AV_WN16A(mb + index, value);
1685 }
1686
1687 static av_always_inline void hl_decode_mb_predict_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1688                                                        int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1689 {
1690     MpegEncContext * const s = &h->s;
1691     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1692     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1693     int i;
1694     int qscale = p == 0 ? s->qscale : h->chroma_qp[p-1];
1695     block_offset += 16*p;
1696     if(IS_INTRA4x4(mb_type)){
1697         if(simple || !s->encoding){
1698             if(IS_8x8DCT(mb_type)){
1699                 if(transform_bypass){
1700                     idct_dc_add =
1701                     idct_add    = s->dsp.add_pixels8;
1702                 }else{
1703                     idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1704                     idct_add    = h->h264dsp.h264_idct8_add;
1705                 }
1706                 for(i=0; i<16; i+=4){
1707                     uint8_t * const ptr= dest_y + block_offset[i];
1708                     const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1709                     if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1710                         h->hpc.pred8x8l_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1711                     }else{
1712                         const int nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1713                         h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1714                                                     (h->topright_samples_available<<i)&0x4000, linesize);
1715                         if(nnz){
1716                             if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1717                                 idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1718                             else
1719                                 idct_add   (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1720                         }
1721                     }
1722                 }
1723             }else{
1724                 if(transform_bypass){
1725                     idct_dc_add =
1726                     idct_add    = s->dsp.add_pixels4;
1727                 }else{
1728                     idct_dc_add = h->h264dsp.h264_idct_dc_add;
1729                     idct_add    = h->h264dsp.h264_idct_add;
1730                 }
1731                 for(i=0; i<16; i++){
1732                     uint8_t * const ptr= dest_y + block_offset[i];
1733                     const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1734
1735                     if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1736                         h->hpc.pred4x4_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1737                     }else{
1738                         uint8_t *topright;
1739                         int nnz, tr;
1740                         uint64_t tr_high;
1741                         if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1742                             const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1743                             assert(s->mb_y || linesize <= block_offset[i]);
1744                             if(!topright_avail){
1745                                 if (pixel_shift) {
1746                                     tr_high= ((uint16_t*)ptr)[3 - linesize/2]*0x0001000100010001ULL;
1747                                     topright= (uint8_t*) &tr_high;
1748                                 } else {
1749                                     tr= ptr[3 - linesize]*0x01010101u;
1750                                     topright= (uint8_t*) &tr;
1751                                 }
1752                             }else
1753                                 topright= ptr + (4 << pixel_shift) - linesize;
1754                         }else
1755                             topright= NULL;
1756
1757                         h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1758                         nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1759                         if(nnz){
1760                             if(is_h264){
1761                                 if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1762                                     idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1763                                 else
1764                                     idct_add   (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1765                             }else
1766                                 ff_svq3_add_idct_c(ptr, h->mb + i*16+p*256, linesize, qscale, 0);
1767                         }
1768                     }
1769                 }
1770             }
1771         }
1772     }else{
1773         h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1774         if(is_h264){
1775             if(h->non_zero_count_cache[ scan8[LUMA_DC_BLOCK_INDEX+p] ]){
1776                 if(!transform_bypass)
1777                     h->h264dsp.h264_luma_dc_dequant_idct(h->mb+(p*256 << pixel_shift), h->mb_luma_dc[p], h->dequant4_coeff[p][qscale][0]);
1778                 else{
1779                     static const uint8_t dc_mapping[16] = { 0*16, 1*16, 4*16, 5*16, 2*16, 3*16, 6*16, 7*16,
1780                                                             8*16, 9*16,12*16,13*16,10*16,11*16,14*16,15*16};
1781                     for(i = 0; i < 16; i++)
1782                         dctcoef_set(h->mb+(p*256 << pixel_shift), pixel_shift, dc_mapping[i], dctcoef_get(h->mb_luma_dc[p], pixel_shift, i));
1783                 }
1784             }
1785         }else
1786             ff_svq3_luma_dc_dequant_idct_c(h->mb+p*256, h->mb_luma_dc[p], qscale);
1787     }
1788 }
1789
1790 static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1791                                                     int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1792 {
1793     MpegEncContext * const s = &h->s;
1794     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1795     int i;
1796     block_offset += 16*p;
1797     if(!IS_INTRA4x4(mb_type)){
1798         if(is_h264){
1799             if(IS_INTRA16x16(mb_type)){
1800                 if(transform_bypass){
1801                     if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1802                         h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize);
1803                     }else{
1804                         for(i=0; i<16; i++){
1805                             if(h->non_zero_count_cache[ scan8[i+p*16] ] || dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1806                                 s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1807                         }
1808                     }
1809                 }else{
1810                     h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1811                 }
1812             }else if(h->cbp&15){
1813                 if(transform_bypass){
1814                     const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1815                     idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1816                     for(i=0; i<16; i+=di){
1817                         if(h->non_zero_count_cache[ scan8[i+p*16] ]){
1818                             idct_add(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1819                         }
1820                     }
1821                 }else{
1822                     if(IS_8x8DCT(mb_type)){
1823                         h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1824                     }else{
1825                         h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1826                     }
1827                 }
1828             }
1829         }else{
1830             for(i=0; i<16; i++){
1831                 if(h->non_zero_count_cache[ scan8[i+p*16] ] || h->mb[i*16+p*256]){ //FIXME benchmark weird rule, & below
1832                     uint8_t * const ptr= dest_y + block_offset[i];
1833                     ff_svq3_add_idct_c(ptr, h->mb + i*16 + p*256, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1834                 }
1835             }
1836         }
1837     }
1838 }
1839
1840 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, int pixel_shift)
1841 {
1842     MpegEncContext * const s = &h->s;
1843     const int mb_x= s->mb_x;
1844     const int mb_y= s->mb_y;
1845     const int mb_xy= h->mb_xy;
1846     const int mb_type = s->current_picture.f.mb_type[mb_xy];
1847     uint8_t  *dest_y, *dest_cb, *dest_cr;
1848     int linesize, uvlinesize /*dct_offset*/;
1849     int i, j;
1850     int *block_offset = &h->block_offset[0];
1851     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1852     /* is_h264 should always be true if SVQ3 is disabled. */
1853     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1854     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1855     const int block_h = 16 >> s->chroma_y_shift;
1856     const int chroma422 = CHROMA422;
1857
1858     dest_y  = s->current_picture.f.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize  ) * 16;
1859     dest_cb = s->current_picture.f.data[1] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
1860     dest_cr = s->current_picture.f.data[2] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
1861
1862     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
1863     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + (64 << pixel_shift), dest_cr - dest_cb, 2);
1864
1865     h->list_counts[mb_xy]= h->list_count;
1866
1867     if (!simple && MB_FIELD) {
1868         linesize   = h->mb_linesize   = s->linesize * 2;
1869         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1870         block_offset = &h->block_offset[48];
1871         if(mb_y&1){ //FIXME move out of this function?
1872             dest_y -= s->linesize*15;
1873             dest_cb-= s->uvlinesize * (block_h - 1);
1874             dest_cr-= s->uvlinesize * (block_h - 1);
1875         }
1876         if(FRAME_MBAFF) {
1877             int list;
1878             for(list=0; list<h->list_count; list++){
1879                 if(!USES_LIST(mb_type, list))
1880                     continue;
1881                 if(IS_16X16(mb_type)){
1882                     int8_t *ref = &h->ref_cache[list][scan8[0]];
1883                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1884                 }else{
1885                     for(i=0; i<16; i+=4){
1886                         int ref = h->ref_cache[list][scan8[i]];
1887                         if(ref >= 0)
1888                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1889                     }
1890                 }
1891             }
1892         }
1893     } else {
1894         linesize   = h->mb_linesize   = s->linesize;
1895         uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1896 //        dct_offset = s->linesize * 16;
1897     }
1898
1899     if (!simple && IS_INTRA_PCM(mb_type)) {
1900         if (pixel_shift) {
1901             const int bit_depth = h->sps.bit_depth_luma;
1902             int j;
1903             GetBitContext gb;
1904             init_get_bits(&gb, (uint8_t*)h->mb, 384*bit_depth);
1905
1906             for (i = 0; i < 16; i++) {
1907                 uint16_t *tmp_y  = (uint16_t*)(dest_y  + i*linesize);
1908                 for (j = 0; j < 16; j++)
1909                     tmp_y[j] = get_bits(&gb, bit_depth);
1910             }
1911             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1912                 if (!h->sps.chroma_format_idc) {
1913                     for (i = 0; i < block_h; i++) {
1914                         uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1915                         for (j = 0; j < 8; j++) {
1916                             tmp_cb[j] = 1 << (bit_depth - 1);
1917                         }
1918                     }
1919                     for (i = 0; i < block_h; i++) {
1920                         uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1921                         for (j = 0; j < 8; j++) {
1922                             tmp_cr[j] = 1 << (bit_depth - 1);
1923                         }
1924                     }
1925                 } else {
1926                     for (i = 0; i < block_h; i++) {
1927                         uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1928                         for (j = 0; j < 8; j++)
1929                             tmp_cb[j] = get_bits(&gb, bit_depth);
1930                     }
1931                     for (i = 0; i < block_h; i++) {
1932                         uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1933                         for (j = 0; j < 8; j++)
1934                             tmp_cr[j] = get_bits(&gb, bit_depth);
1935                     }
1936                 }
1937             }
1938         } else {
1939             for (i=0; i<16; i++) {
1940                 memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1941             }
1942             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1943                 if (!h->sps.chroma_format_idc) {
1944                     for (i = 0; i < block_h; i++) {
1945                         memset(dest_cb + i*uvlinesize, 128, 8);
1946                         memset(dest_cr + i*uvlinesize, 128, 8);
1947                     }
1948                 } else {
1949                     for (i = 0; i < block_h; i++) {
1950                         memcpy(dest_cb + i*uvlinesize, h->mb + 128 + i*4,  8);
1951                         memcpy(dest_cr + i*uvlinesize, h->mb + 160 + i*4,  8);
1952                     }
1953                 }
1954             }
1955         }
1956     } else {
1957         if(IS_INTRA(mb_type)){
1958             if(h->deblocking_filter)
1959                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, 0, simple, pixel_shift);
1960
1961             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1962                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1963                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1964             }
1965
1966             hl_decode_mb_predict_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
1967
1968             if(h->deblocking_filter)
1969                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, 0, simple, pixel_shift);
1970         }else if(is_h264){
1971             if (chroma422) {
1972                 hl_motion_422(h, dest_y, dest_cb, dest_cr,
1973                               s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1974                               s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1975                               h->h264dsp.weight_h264_pixels_tab,
1976                               h->h264dsp.biweight_h264_pixels_tab,
1977                               pixel_shift);
1978             } else {
1979                 hl_motion_420(h, dest_y, dest_cb, dest_cr,
1980                               s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1981                               s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1982                               h->h264dsp.weight_h264_pixels_tab,
1983                               h->h264dsp.biweight_h264_pixels_tab,
1984                               pixel_shift);
1985             }
1986         }
1987
1988         hl_decode_mb_idct_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
1989
1990         if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1991             uint8_t *dest[2] = {dest_cb, dest_cr};
1992             if(transform_bypass){
1993                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1994                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + (16*16*1 << pixel_shift), uvlinesize);
1995                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 32, h->mb + (16*16*2 << pixel_shift), uvlinesize);
1996                 }else{
1997                     idct_add = s->dsp.add_pixels4;
1998                     for(j=1; j<3; j++){
1999                         for(i=j*16; i<j*16+4; 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], h->mb + (i*16 << pixel_shift), uvlinesize);
2002                         }
2003                         if (chroma422) {
2004                             for(i=j*16+4; i<j*16+8; i++){
2005                                 if(h->non_zero_count_cache[ scan8[i+4] ] || dctcoef_get(h->mb, pixel_shift, i*16))
2006                                     idct_add   (dest[j-1] + block_offset[i+4], h->mb + (i*16 << pixel_shift), uvlinesize);
2007                             }
2008                         }
2009                     }
2010                 }
2011             }else{
2012                 if(is_h264){
2013                     int qp[2];
2014                     if (chroma422) {
2015                         qp[0] = h->chroma_qp[0] + 3;
2016                         qp[1] = h->chroma_qp[1] + 3;
2017                     } else {
2018                         qp[0] = h->chroma_qp[0];
2019                         qp[1] = h->chroma_qp[1];
2020                     }
2021                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+0] ])
2022                         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]);
2023                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+1] ])
2024                         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]);
2025                     h->h264dsp.h264_idct_add8(dest, block_offset,
2026                                               h->mb, uvlinesize,
2027                                               h->non_zero_count_cache);
2028                 }else{
2029                     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]);
2030                     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]);
2031                     for(j=1; j<3; j++){
2032                         for(i=j*16; i<j*16+4; i++){
2033                             if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2034                                 uint8_t * const ptr= dest[j-1] + block_offset[i];
2035                                 ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[0][s->qscale + 12] - 12, 2);
2036                             }
2037                         }
2038                     }
2039                 }
2040             }
2041         }
2042     }
2043     if(h->cbp || IS_INTRA(mb_type))
2044     {
2045         s->dsp.clear_blocks(h->mb);
2046         s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
2047     }
2048 }
2049
2050 static av_always_inline void hl_decode_mb_444_internal(H264Context *h, int simple, int pixel_shift){
2051     MpegEncContext * const s = &h->s;
2052     const int mb_x= s->mb_x;
2053     const int mb_y= s->mb_y;
2054     const int mb_xy= h->mb_xy;
2055     const int mb_type = s->current_picture.f.mb_type[mb_xy];
2056     uint8_t  *dest[3];
2057     int linesize;
2058     int i, j, p;
2059     int *block_offset = &h->block_offset[0];
2060     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
2061     const int plane_count = (simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) ? 3 : 1;
2062
2063     for (p = 0; p < plane_count; p++)
2064     {
2065         dest[p] = s->current_picture.f.data[p] + ((mb_x << pixel_shift) + mb_y * s->linesize) * 16;
2066         s->dsp.prefetch(dest[p] + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
2067     }
2068
2069     h->list_counts[mb_xy]= h->list_count;
2070
2071     if (!simple && MB_FIELD) {
2072         linesize   = h->mb_linesize = h->mb_uvlinesize = s->linesize * 2;
2073         block_offset = &h->block_offset[48];
2074         if(mb_y&1) //FIXME move out of this function?
2075             for (p = 0; p < 3; p++)
2076                 dest[p] -= s->linesize*15;
2077         if(FRAME_MBAFF) {
2078             int list;
2079             for(list=0; list<h->list_count; list++){
2080                 if(!USES_LIST(mb_type, list))
2081                     continue;
2082                 if(IS_16X16(mb_type)){
2083                     int8_t *ref = &h->ref_cache[list][scan8[0]];
2084                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
2085                 }else{
2086                     for(i=0; i<16; i+=4){
2087                         int ref = h->ref_cache[list][scan8[i]];
2088                         if(ref >= 0)
2089                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
2090                     }
2091                 }
2092             }
2093         }
2094     } else {
2095         linesize   = h->mb_linesize = h->mb_uvlinesize = s->linesize;
2096     }
2097
2098     if (!simple && IS_INTRA_PCM(mb_type)) {
2099         if (pixel_shift) {
2100             const int bit_depth = h->sps.bit_depth_luma;
2101             GetBitContext gb;
2102             init_get_bits(&gb, (uint8_t*)h->mb, 768*bit_depth);
2103
2104             for (p = 0; p < plane_count; p++) {
2105                 for (i = 0; i < 16; i++) {
2106                     uint16_t *tmp = (uint16_t*)(dest[p] + i*linesize);
2107                     for (j = 0; j < 16; j++)
2108                         tmp[j] = get_bits(&gb, bit_depth);
2109                 }
2110             }
2111         } else {
2112             for (p = 0; p < plane_count; p++) {
2113                 for (i = 0; i < 16; i++) {
2114                     memcpy(dest[p] + i*linesize, h->mb + p*128 + i*8, 16);
2115                 }
2116             }
2117         }
2118     } else {
2119         if(IS_INTRA(mb_type)){
2120             if(h->deblocking_filter)
2121                 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 1, 1, simple, pixel_shift);
2122
2123             for (p = 0; p < plane_count; p++)
2124                 hl_decode_mb_predict_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2125
2126             if(h->deblocking_filter)
2127                 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 0, 1, simple, pixel_shift);
2128         }else{
2129             hl_motion(h, dest[0], dest[1], dest[2],
2130                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2131                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2132                       h->h264dsp.weight_h264_pixels_tab,
2133                       h->h264dsp.biweight_h264_pixels_tab, pixel_shift, 3);
2134         }
2135
2136         for (p = 0; p < plane_count; p++)
2137             hl_decode_mb_idct_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2138     }
2139     if(h->cbp || IS_INTRA(mb_type))
2140     {
2141         s->dsp.clear_blocks(h->mb);
2142         s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
2143     }
2144 }
2145
2146 /**
2147  * Process a macroblock; this case avoids checks for expensive uncommon cases.
2148  */
2149 #define hl_decode_mb_simple(sh, bits) \
2150 static void hl_decode_mb_simple_ ## bits(H264Context *h){ \
2151     hl_decode_mb_internal(h, 1, sh); \
2152 }
2153 hl_decode_mb_simple(0, 8);
2154 hl_decode_mb_simple(1, 16);
2155
2156 /**
2157  * Process a macroblock; this handles edge cases, such as interlacing.
2158  */
2159 static void av_noinline hl_decode_mb_complex(H264Context *h){
2160     hl_decode_mb_internal(h, 0, h->pixel_shift);
2161 }
2162
2163 static void av_noinline hl_decode_mb_444_complex(H264Context *h){
2164     hl_decode_mb_444_internal(h, 0, h->pixel_shift);
2165 }
2166
2167 static void av_noinline hl_decode_mb_444_simple(H264Context *h){
2168     hl_decode_mb_444_internal(h, 1, 0);
2169 }
2170
2171 void ff_h264_hl_decode_mb(H264Context *h){
2172     MpegEncContext * const s = &h->s;
2173     const int mb_xy= h->mb_xy;
2174     const int mb_type = s->current_picture.f.mb_type[mb_xy];
2175     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
2176
2177     if (CHROMA444) {
2178         if(is_complex || h->pixel_shift)
2179             hl_decode_mb_444_complex(h);
2180         else
2181             hl_decode_mb_444_simple(h);
2182     } else if (is_complex) {
2183         hl_decode_mb_complex(h);
2184     } else if (h->pixel_shift) {
2185         hl_decode_mb_simple_16(h);
2186     } else
2187         hl_decode_mb_simple_8(h);
2188 }
2189
2190 static int pred_weight_table(H264Context *h){
2191     MpegEncContext * const s = &h->s;
2192     int list, i;
2193     int luma_def, chroma_def;
2194
2195     h->use_weight= 0;
2196     h->use_weight_chroma= 0;
2197     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2198     if(h->sps.chroma_format_idc)
2199         h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2200     luma_def = 1<<h->luma_log2_weight_denom;
2201     chroma_def = 1<<h->chroma_log2_weight_denom;
2202
2203     for(list=0; list<2; list++){
2204         h->luma_weight_flag[list]   = 0;
2205         h->chroma_weight_flag[list] = 0;
2206         for(i=0; i<h->ref_count[list]; i++){
2207             int luma_weight_flag, chroma_weight_flag;
2208
2209             luma_weight_flag= get_bits1(&s->gb);
2210             if(luma_weight_flag){
2211                 h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
2212                 h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
2213                 if(   h->luma_weight[i][list][0] != luma_def
2214                    || h->luma_weight[i][list][1] != 0) {
2215                     h->use_weight= 1;
2216                     h->luma_weight_flag[list]= 1;
2217                 }
2218             }else{
2219                 h->luma_weight[i][list][0]= luma_def;
2220                 h->luma_weight[i][list][1]= 0;
2221             }
2222
2223             if(h->sps.chroma_format_idc){
2224                 chroma_weight_flag= get_bits1(&s->gb);
2225                 if(chroma_weight_flag){
2226                     int j;
2227                     for(j=0; j<2; j++){
2228                         h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
2229                         h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
2230                         if(   h->chroma_weight[i][list][j][0] != chroma_def
2231                            || h->chroma_weight[i][list][j][1] != 0) {
2232                             h->use_weight_chroma= 1;
2233                             h->chroma_weight_flag[list]= 1;
2234                         }
2235                     }
2236                 }else{
2237                     int j;
2238                     for(j=0; j<2; j++){
2239                         h->chroma_weight[i][list][j][0]= chroma_def;
2240                         h->chroma_weight[i][list][j][1]= 0;
2241                     }
2242                 }
2243             }
2244         }
2245         if(h->slice_type_nos != AV_PICTURE_TYPE_B) break;
2246     }
2247     h->use_weight= h->use_weight || h->use_weight_chroma;
2248     return 0;
2249 }
2250
2251 /**
2252  * Initialize implicit_weight table.
2253  * @param field  0/1 initialize the weight for interlaced MBAFF
2254  *                -1 initializes the rest
2255  */
2256 static void implicit_weight_table(H264Context *h, int field){
2257     MpegEncContext * const s = &h->s;
2258     int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
2259
2260     for (i = 0; i < 2; i++) {
2261         h->luma_weight_flag[i]   = 0;
2262         h->chroma_weight_flag[i] = 0;
2263     }
2264
2265     if(field < 0){
2266         if (s->picture_structure == PICT_FRAME) {
2267             cur_poc = s->current_picture_ptr->poc;
2268         } else {
2269             cur_poc = s->current_picture_ptr->field_poc[s->picture_structure - 1];
2270         }
2271     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
2272        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
2273         h->use_weight= 0;
2274         h->use_weight_chroma= 0;
2275         return;
2276     }
2277         ref_start= 0;
2278         ref_count0= h->ref_count[0];
2279         ref_count1= h->ref_count[1];
2280     }else{
2281         cur_poc = s->current_picture_ptr->field_poc[field];
2282         ref_start= 16;
2283         ref_count0= 16+2*h->ref_count[0];
2284         ref_count1= 16+2*h->ref_count[1];
2285     }
2286
2287     h->use_weight= 2;
2288     h->use_weight_chroma= 2;
2289     h->luma_log2_weight_denom= 5;
2290     h->chroma_log2_weight_denom= 5;
2291
2292     for(ref0=ref_start; ref0 < ref_count0; ref0++){
2293         int poc0 = h->ref_list[0][ref0].poc;
2294         for(ref1=ref_start; ref1 < ref_count1; ref1++){
2295             int w = 32;
2296             if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
2297                 int poc1 = h->ref_list[1][ref1].poc;
2298                 int td = av_clip(poc1 - poc0, -128, 127);
2299                 if(td){
2300                     int tb = av_clip(cur_poc - poc0, -128, 127);
2301                     int tx = (16384 + (FFABS(td) >> 1)) / td;
2302                     int dist_scale_factor = (tb*tx + 32) >> 8;
2303                     if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
2304                         w = 64 - dist_scale_factor;
2305                 }
2306             }
2307             if(field<0){
2308                 h->implicit_weight[ref0][ref1][0]=
2309                 h->implicit_weight[ref0][ref1][1]= w;
2310             }else{
2311                 h->implicit_weight[ref0][ref1][field]=w;
2312             }
2313         }
2314     }
2315 }
2316
2317 /**
2318  * instantaneous decoder refresh.
2319  */
2320 static void idr(H264Context *h){
2321     ff_h264_remove_all_refs(h);
2322     h->prev_frame_num= 0;
2323     h->prev_frame_num_offset= 0;
2324     h->prev_poc_msb=
2325     h->prev_poc_lsb= 0;
2326 }
2327
2328 /* forget old pics after a seek */
2329 static void flush_dpb(AVCodecContext *avctx){
2330     H264Context *h= avctx->priv_data;
2331     int i;
2332     for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
2333         if(h->delayed_pic[i])
2334             h->delayed_pic[i]->f.reference = 0;
2335         h->delayed_pic[i]= NULL;
2336     }
2337     h->outputed_poc=h->next_outputed_poc= INT_MIN;
2338     h->prev_interlaced_frame = 1;
2339     idr(h);
2340     if(h->s.current_picture_ptr)
2341         h->s.current_picture_ptr->f.reference = 0;
2342     h->s.first_field= 0;
2343     ff_h264_reset_sei(h);
2344     ff_mpeg_flush(avctx);
2345 }
2346
2347 static int init_poc(H264Context *h){
2348     MpegEncContext * const s = &h->s;
2349     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2350     int field_poc[2];
2351     Picture *cur = s->current_picture_ptr;
2352
2353     h->frame_num_offset= h->prev_frame_num_offset;
2354     if(h->frame_num < h->prev_frame_num)
2355         h->frame_num_offset += max_frame_num;
2356
2357     if(h->sps.poc_type==0){
2358         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2359
2360         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2361             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2362         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2363             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2364         else
2365             h->poc_msb = h->prev_poc_msb;
2366 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2367         field_poc[0] =
2368         field_poc[1] = h->poc_msb + h->poc_lsb;
2369         if(s->picture_structure == PICT_FRAME)
2370             field_poc[1] += h->delta_poc_bottom;
2371     }else if(h->sps.poc_type==1){
2372         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2373         int i;
2374
2375         if(h->sps.poc_cycle_length != 0)
2376             abs_frame_num = h->frame_num_offset + h->frame_num;
2377         else
2378             abs_frame_num = 0;
2379
2380         if(h->nal_ref_idc==0 && abs_frame_num > 0)
2381             abs_frame_num--;
2382
2383         expected_delta_per_poc_cycle = 0;
2384         for(i=0; i < h->sps.poc_cycle_length; i++)
2385             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2386
2387         if(abs_frame_num > 0){
2388             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2389             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2390
2391             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2392             for(i = 0; i <= frame_num_in_poc_cycle; i++)
2393                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2394         } else
2395             expectedpoc = 0;
2396
2397         if(h->nal_ref_idc == 0)
2398             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2399
2400         field_poc[0] = expectedpoc + h->delta_poc[0];
2401         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2402
2403         if(s->picture_structure == PICT_FRAME)
2404             field_poc[1] += h->delta_poc[1];
2405     }else{
2406         int poc= 2*(h->frame_num_offset + h->frame_num);
2407
2408         if(!h->nal_ref_idc)
2409             poc--;
2410
2411         field_poc[0]= poc;
2412         field_poc[1]= poc;
2413     }
2414
2415     if(s->picture_structure != PICT_BOTTOM_FIELD)
2416         s->current_picture_ptr->field_poc[0]= field_poc[0];
2417     if(s->picture_structure != PICT_TOP_FIELD)
2418         s->current_picture_ptr->field_poc[1]= field_poc[1];
2419     cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
2420
2421     return 0;
2422 }
2423
2424
2425 /**
2426  * initialize scan tables
2427  */
2428 static void init_scan_tables(H264Context *h){
2429     int i;
2430     for(i=0; i<16; i++){
2431 #define T(x) (x>>2) | ((x<<2) & 0xF)
2432         h->zigzag_scan[i] = T(zigzag_scan[i]);
2433         h-> field_scan[i] = T( field_scan[i]);
2434 #undef T
2435     }
2436     for(i=0; i<64; i++){
2437 #define T(x) (x>>3) | ((x&7)<<3)
2438         h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
2439         h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
2440         h->field_scan8x8[i]        = T(field_scan8x8[i]);
2441         h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
2442 #undef T
2443     }
2444     if(h->sps.transform_bypass){ //FIXME same ugly
2445         h->zigzag_scan_q0          = zigzag_scan;
2446         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
2447         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
2448         h->field_scan_q0           = field_scan;
2449         h->field_scan8x8_q0        = field_scan8x8;
2450         h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
2451     }else{
2452         h->zigzag_scan_q0          = h->zigzag_scan;
2453         h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
2454         h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
2455         h->field_scan_q0           = h->field_scan;
2456         h->field_scan8x8_q0        = h->field_scan8x8;
2457         h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
2458     }
2459 }
2460
2461 static int field_end(H264Context *h, int in_setup){
2462     MpegEncContext * const s = &h->s;
2463     AVCodecContext * const avctx= s->avctx;
2464     int err = 0;
2465     s->mb_y= 0;
2466
2467     if (!in_setup && !s->dropable)
2468         ff_thread_report_progress((AVFrame*)s->current_picture_ptr, (16*s->mb_height >> FIELD_PICTURE) - 1,
2469                                  s->picture_structure==PICT_BOTTOM_FIELD);
2470
2471     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2472         ff_vdpau_h264_set_reference_frames(s);
2473
2474     if(in_setup || !(avctx->active_thread_type&FF_THREAD_FRAME)){
2475         if(!s->dropable) {
2476             err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
2477             h->prev_poc_msb= h->poc_msb;
2478             h->prev_poc_lsb= h->poc_lsb;
2479         }
2480         h->prev_frame_num_offset= h->frame_num_offset;
2481         h->prev_frame_num= h->frame_num;
2482         h->outputed_poc = h->next_outputed_poc;
2483     }
2484
2485     if (avctx->hwaccel) {
2486         if (avctx->hwaccel->end_frame(avctx) < 0)
2487             av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
2488     }
2489
2490     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2491         ff_vdpau_h264_picture_complete(s);
2492
2493     /*
2494      * FIXME: Error handling code does not seem to support interlaced
2495      * when slices span multiple rows
2496      * The ff_er_add_slice calls don't work right for bottom
2497      * fields; they cause massive erroneous error concealing
2498      * Error marking covers both fields (top and bottom).
2499      * This causes a mismatched s->error_count
2500      * and a bad error table. Further, the error count goes to
2501      * INT_MAX when called for bottom field, because mb_y is
2502      * past end by one (callers fault) and resync_mb_y != 0
2503      * causes problems for the first MB line, too.
2504      */
2505     if (!FIELD_PICTURE)
2506         ff_er_frame_end(s);
2507
2508     MPV_frame_end(s);
2509
2510     h->current_slice=0;
2511
2512     return err;
2513 }
2514
2515 /**
2516  * Replicate H264 "master" context to thread contexts.
2517  */
2518 static void clone_slice(H264Context *dst, H264Context *src)
2519 {
2520     memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
2521     dst->s.current_picture_ptr  = src->s.current_picture_ptr;
2522     dst->s.current_picture      = src->s.current_picture;
2523     dst->s.linesize             = src->s.linesize;
2524     dst->s.uvlinesize           = src->s.uvlinesize;
2525     dst->s.first_field          = src->s.first_field;
2526
2527     dst->prev_poc_msb           = src->prev_poc_msb;
2528     dst->prev_poc_lsb           = src->prev_poc_lsb;
2529     dst->prev_frame_num_offset  = src->prev_frame_num_offset;
2530     dst->prev_frame_num         = src->prev_frame_num;
2531     dst->short_ref_count        = src->short_ref_count;
2532
2533     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
2534     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
2535     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
2536     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
2537
2538     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
2539     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
2540 }
2541
2542 /**
2543  * computes profile from profile_idc and constraint_set?_flags
2544  *
2545  * @param sps SPS
2546  *
2547  * @return profile as defined by FF_PROFILE_H264_*
2548  */
2549 int ff_h264_get_profile(SPS *sps)
2550 {
2551     int profile = sps->profile_idc;
2552
2553     switch(sps->profile_idc) {
2554     case FF_PROFILE_H264_BASELINE:
2555         // constraint_set1_flag set to 1
2556         profile |= (sps->constraint_set_flags & 1<<1) ? FF_PROFILE_H264_CONSTRAINED : 0;
2557         break;
2558     case FF_PROFILE_H264_HIGH_10:
2559     case FF_PROFILE_H264_HIGH_422:
2560     case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
2561         // constraint_set3_flag set to 1
2562         profile |= (sps->constraint_set_flags & 1<<3) ? FF_PROFILE_H264_INTRA : 0;
2563         break;
2564     }
2565
2566     return profile;
2567 }
2568
2569 /**
2570  * decodes a slice header.
2571  * This will also call MPV_common_init() and frame_start() as needed.
2572  *
2573  * @param h h264context
2574  * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
2575  *
2576  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
2577  */
2578 static int decode_slice_header(H264Context *h, H264Context *h0){
2579     MpegEncContext * const s = &h->s;
2580     MpegEncContext * const s0 = &h0->s;
2581     unsigned int first_mb_in_slice;
2582     unsigned int pps_id;
2583     int num_ref_idx_active_override_flag;
2584     unsigned int slice_type, tmp, i, j;
2585     int default_ref_list_done = 0;
2586     int last_pic_structure;
2587
2588     s->dropable= h->nal_ref_idc == 0;
2589
2590     /* FIXME: 2tap qpel isn't implemented for high bit depth. */
2591     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc && !h->pixel_shift){
2592         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
2593         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
2594     }else{
2595         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
2596         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
2597     }
2598
2599     first_mb_in_slice= get_ue_golomb(&s->gb);
2600
2601     if(first_mb_in_slice == 0){ //FIXME better field boundary detection
2602         if(h0->current_slice && FIELD_PICTURE){
2603             field_end(h, 1);
2604         }
2605
2606         h0->current_slice = 0;
2607         if (!s0->first_field)
2608             s->current_picture_ptr= NULL;
2609     }
2610
2611     slice_type= get_ue_golomb_31(&s->gb);
2612     if(slice_type > 9){
2613         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);
2614         return -1;
2615     }
2616     if(slice_type > 4){
2617         slice_type -= 5;
2618         h->slice_type_fixed=1;
2619     }else
2620         h->slice_type_fixed=0;
2621
2622     slice_type= golomb_to_pict_type[ slice_type ];
2623     if (slice_type == AV_PICTURE_TYPE_I
2624         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
2625         default_ref_list_done = 1;
2626     }
2627     h->slice_type= slice_type;
2628     h->slice_type_nos= slice_type & 3;
2629
2630     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
2631
2632     pps_id= get_ue_golomb(&s->gb);
2633     if(pps_id>=MAX_PPS_COUNT){
2634         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
2635         return -1;
2636     }
2637     if(!h0->pps_buffers[pps_id]) {
2638         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
2639         return -1;
2640     }
2641     h->pps= *h0->pps_buffers[pps_id];
2642
2643     if(!h0->sps_buffers[h->pps.sps_id]) {
2644         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
2645         return -1;
2646     }
2647     h->sps = *h0->sps_buffers[h->pps.sps_id];
2648
2649     s->avctx->profile = ff_h264_get_profile(&h->sps);
2650     s->avctx->level   = h->sps.level_idc;
2651     s->avctx->refs    = h->sps.ref_frame_count;
2652
2653     if(h == h0 && h->dequant_coeff_pps != pps_id){
2654         h->dequant_coeff_pps = pps_id;
2655         init_dequant_tables(h);
2656     }
2657
2658     s->mb_width= h->sps.mb_width;
2659     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
2660
2661     h->b_stride=  s->mb_width*4;
2662
2663     s->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
2664
2665     s->width = 16*s->mb_width - (2>>CHROMA444)*FFMIN(h->sps.crop_right, (8<<CHROMA444)-1);
2666     if(h->sps.frame_mbs_only_flag)
2667         s->height= 16*s->mb_height - (1<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1);
2668     else
2669         s->height= 16*s->mb_height - (2<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1);
2670
2671     if (s->context_initialized
2672         && (   s->width != s->avctx->width || s->height != s->avctx->height
2673             || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
2674         if(h != h0) {
2675             av_log_missing_feature(s->avctx, "Width/height changing with threads is", 0);
2676             return -1;   // width / height changed during parallelized decoding
2677         }
2678         free_tables(h, 0);
2679         flush_dpb(s->avctx);
2680         MPV_common_end(s);
2681     }
2682     if (!s->context_initialized) {
2683         if (h != h0) {
2684             av_log(h->s.avctx, AV_LOG_ERROR, "Cannot (re-)initialize context during parallel decoding.\n");
2685             return -1;
2686         }
2687
2688         avcodec_set_dimensions(s->avctx, s->width, s->height);
2689         s->avctx->sample_aspect_ratio= h->sps.sar;
2690         av_assert0(s->avctx->sample_aspect_ratio.den);
2691
2692         if(h->sps.video_signal_type_present_flag){
2693             s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
2694             if(h->sps.colour_description_present_flag){
2695                 s->avctx->color_primaries = h->sps.color_primaries;
2696                 s->avctx->color_trc       = h->sps.color_trc;
2697                 s->avctx->colorspace      = h->sps.colorspace;
2698             }
2699         }
2700
2701         if(h->sps.timing_info_present_flag){
2702             int64_t den= h->sps.time_scale;
2703             if(h->x264_build < 44U)
2704                 den *= 2;
2705             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
2706                       h->sps.num_units_in_tick, den, 1<<30);
2707         }
2708
2709         switch (h->sps.bit_depth_luma) {
2710             case 9 :
2711                 if (CHROMA444)
2712                     s->avctx->pix_fmt = PIX_FMT_YUV444P9;
2713                 else if (CHROMA422)
2714                     s->avctx->pix_fmt = PIX_FMT_YUV422P9;
2715                 else
2716                     s->avctx->pix_fmt = PIX_FMT_YUV420P9;
2717                 break;
2718             case 10 :
2719                 if (CHROMA444)
2720                     s->avctx->pix_fmt = PIX_FMT_YUV444P10;
2721                 else if (CHROMA422)
2722                     s->avctx->pix_fmt = PIX_FMT_YUV422P10;
2723                 else
2724                     s->avctx->pix_fmt = PIX_FMT_YUV420P10;
2725                 break;
2726             default:
2727                 if (CHROMA444){
2728                     s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ444P : PIX_FMT_YUV444P;
2729                 } else if (CHROMA422) {
2730                     s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ422P : PIX_FMT_YUV422P;
2731                 }else{
2732                     s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
2733                                                              s->avctx->codec->pix_fmts ?
2734                                                              s->avctx->codec->pix_fmts :
2735                                                              s->avctx->color_range == AVCOL_RANGE_JPEG ?
2736                                                              hwaccel_pixfmt_list_h264_jpeg_420 :
2737                                                              ff_hwaccel_pixfmt_list_420);
2738                 }
2739         }
2740
2741         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
2742
2743         if (MPV_common_init(s) < 0) {
2744             av_log(h->s.avctx, AV_LOG_ERROR, "MPV_common_init() failed.\n");
2745             return -1;
2746         }
2747         s->first_field = 0;
2748         h->prev_interlaced_frame = 1;
2749
2750         init_scan_tables(h);
2751         if (ff_h264_alloc_tables(h) < 0) {
2752             av_log(h->s.avctx, AV_LOG_ERROR, "Could not allocate memory for h264\n");
2753             return AVERROR(ENOMEM);
2754         }
2755
2756         if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_SLICE)) {
2757             if (context_init(h) < 0) {
2758                 av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2759                 return -1;
2760             }
2761         } else {
2762             for(i = 1; i < s->avctx->thread_count; i++) {
2763                 H264Context *c;
2764                 c = h->thread_context[i] = av_malloc(sizeof(H264Context));
2765                 memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
2766                 memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
2767                 c->h264dsp = h->h264dsp;
2768                 c->sps = h->sps;
2769                 c->pps = h->pps;
2770                 c->pixel_shift = h->pixel_shift;
2771                 init_scan_tables(c);
2772                 clone_tables(c, h, i);
2773             }
2774
2775             for(i = 0; i < s->avctx->thread_count; i++)
2776                 if (context_init(h->thread_context[i]) < 0) {
2777                     av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2778                     return -1;
2779                 }
2780         }
2781     }
2782
2783     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
2784
2785     h->mb_mbaff = 0;
2786     h->mb_aff_frame = 0;
2787     last_pic_structure = s0->picture_structure;
2788     if(h->sps.frame_mbs_only_flag){
2789         s->picture_structure= PICT_FRAME;
2790     }else{
2791         if(get_bits1(&s->gb)) { //field_pic_flag
2792             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
2793         } else {
2794             s->picture_structure= PICT_FRAME;
2795             h->mb_aff_frame = h->sps.mb_aff;
2796         }
2797     }
2798     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
2799
2800     if(h0->current_slice == 0){
2801         // Shorten frame num gaps so we don't have to allocate reference frames just to throw them away
2802         if(h->frame_num != h->prev_frame_num) {
2803             int unwrap_prev_frame_num = h->prev_frame_num, max_frame_num = 1<<h->sps.log2_max_frame_num;
2804
2805             if (unwrap_prev_frame_num > h->frame_num) unwrap_prev_frame_num -= max_frame_num;
2806
2807             if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
2808                 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
2809                 if (unwrap_prev_frame_num < 0)
2810                     unwrap_prev_frame_num += max_frame_num;
2811
2812                 h->prev_frame_num = unwrap_prev_frame_num;
2813             }
2814         }
2815
2816         while(h->frame_num !=  h->prev_frame_num &&
2817               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
2818             Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
2819             av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
2820             if (ff_h264_frame_start(h) < 0)
2821                 return -1;
2822             h->prev_frame_num++;
2823             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
2824             s->current_picture_ptr->frame_num= h->prev_frame_num;
2825             ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 0);
2826             ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 1);
2827             ff_generate_sliding_window_mmcos(h);
2828             if (ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index) < 0 &&
2829                 (s->avctx->err_recognition & AV_EF_EXPLODE))
2830                 return AVERROR_INVALIDDATA;
2831             /* Error concealment: if a ref is missing, copy the previous ref in its place.
2832              * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
2833              * about there being no actual duplicates.
2834              * FIXME: this doesn't copy padding for out-of-frame motion vectors.  Given we're
2835              * concealing a lost frame, this probably isn't noticable by comparison, but it should
2836              * be fixed. */
2837             if (h->short_ref_count) {
2838                 if (prev) {
2839                     av_image_copy(h->short_ref[0]->f.data, h->short_ref[0]->f.linesize,
2840                                   (const uint8_t**)prev->f.data, prev->f.linesize,
2841                                   s->avctx->pix_fmt, s->mb_width*16, s->mb_height*16);
2842                     h->short_ref[0]->poc = prev->poc+2;
2843                 }
2844                 h->short_ref[0]->frame_num = h->prev_frame_num;
2845             }
2846         }
2847
2848         /* See if we have a decoded first field looking for a pair... */
2849         if (s0->first_field) {
2850             assert(s0->current_picture_ptr);
2851             assert(s0->current_picture_ptr->f.data[0]);
2852             assert(s0->current_picture_ptr->f.reference != DELAYED_PIC_REF);
2853
2854             /* figure out if we have a complementary field pair */
2855             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2856                 /*
2857                  * Previous field is unmatched. Don't display it, but let it
2858                  * remain for reference if marked as such.
2859                  */
2860                 s0->current_picture_ptr = NULL;
2861                 s0->first_field = FIELD_PICTURE;
2862
2863             } else {
2864                 if (h->nal_ref_idc &&
2865                         s0->current_picture_ptr->f.reference &&
2866                         s0->current_picture_ptr->frame_num != h->frame_num) {
2867                     /*
2868                      * This and previous field were reference, but had
2869                      * different frame_nums. Consider this field first in
2870                      * pair. Throw away previous field except for reference
2871                      * purposes.
2872                      */
2873                     s0->first_field = 1;
2874                     s0->current_picture_ptr = NULL;
2875
2876                 } else {
2877                     /* Second field in complementary pair */
2878                     s0->first_field = 0;
2879                 }
2880             }
2881
2882         } else {
2883             /* Frame or first field in a potentially complementary pair */
2884             assert(!s0->current_picture_ptr);
2885             s0->first_field = FIELD_PICTURE;
2886         }
2887
2888         if(!FIELD_PICTURE || s0->first_field) {
2889             if (ff_h264_frame_start(h) < 0) {
2890                 s0->first_field = 0;
2891                 return -1;
2892             }
2893         } else {
2894             ff_release_unused_pictures(s, 0);
2895         }
2896     }
2897     if(h != h0)
2898         clone_slice(h, h0);
2899
2900     s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
2901
2902     assert(s->mb_num == s->mb_width * s->mb_height);
2903     if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
2904        first_mb_in_slice                    >= s->mb_num){
2905         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
2906         return -1;
2907     }
2908     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
2909     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
2910     if (s->picture_structure == PICT_BOTTOM_FIELD)
2911         s->resync_mb_y = s->mb_y = s->mb_y + 1;
2912     assert(s->mb_y < s->mb_height);
2913
2914     if(s->picture_structure==PICT_FRAME){
2915         h->curr_pic_num=   h->frame_num;
2916         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
2917     }else{
2918         h->curr_pic_num= 2*h->frame_num + 1;
2919         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
2920     }
2921
2922     if(h->nal_unit_type == NAL_IDR_SLICE){
2923         get_ue_golomb(&s->gb); /* idr_pic_id */
2924     }
2925
2926     if(h->sps.poc_type==0){
2927         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2928
2929         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
2930             h->delta_poc_bottom= get_se_golomb(&s->gb);
2931         }
2932     }
2933
2934     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
2935         h->delta_poc[0]= get_se_golomb(&s->gb);
2936
2937         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2938             h->delta_poc[1]= get_se_golomb(&s->gb);
2939     }
2940
2941     init_poc(h);
2942
2943     if(h->pps.redundant_pic_cnt_present){
2944         h->redundant_pic_count= get_ue_golomb(&s->gb);
2945     }
2946
2947     //set defaults, might be overridden a few lines later
2948     h->ref_count[0]= h->pps.ref_count[0];
2949     h->ref_count[1]= h->pps.ref_count[1];
2950
2951     if(h->slice_type_nos != AV_PICTURE_TYPE_I){
2952         if(h->slice_type_nos == AV_PICTURE_TYPE_B){
2953             h->direct_spatial_mv_pred= get_bits1(&s->gb);
2954         }
2955         num_ref_idx_active_override_flag= get_bits1(&s->gb);
2956
2957         if(num_ref_idx_active_override_flag){
2958             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2959             if(h->slice_type_nos==AV_PICTURE_TYPE_B)
2960                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2961
2962             if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
2963                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
2964                 h->ref_count[0]= h->ref_count[1]= 1;
2965                 return -1;
2966             }
2967         }
2968         if(h->slice_type_nos == AV_PICTURE_TYPE_B)
2969             h->list_count= 2;
2970         else
2971             h->list_count= 1;
2972     }else
2973         h->list_count= 0;
2974
2975     if(!default_ref_list_done){
2976         ff_h264_fill_default_ref_list(h);
2977     }
2978
2979     if(h->slice_type_nos!=AV_PICTURE_TYPE_I && ff_h264_decode_ref_pic_list_reordering(h) < 0) {
2980         h->ref_count[1]= h->ref_count[0]= 0;
2981         return -1;
2982     }
2983
2984     if(h->slice_type_nos!=AV_PICTURE_TYPE_I){
2985         s->last_picture_ptr= &h->ref_list[0][0];
2986         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
2987     }
2988     if(h->slice_type_nos==AV_PICTURE_TYPE_B){
2989         s->next_picture_ptr= &h->ref_list[1][0];
2990         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
2991     }
2992
2993     if(   (h->pps.weighted_pred          && h->slice_type_nos == AV_PICTURE_TYPE_P )
2994        ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== AV_PICTURE_TYPE_B ) )
2995         pred_weight_table(h);
2996     else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
2997         implicit_weight_table(h, -1);
2998     }else {
2999         h->use_weight = 0;
3000         for (i = 0; i < 2; i++) {
3001             h->luma_weight_flag[i]   = 0;
3002             h->chroma_weight_flag[i] = 0;
3003         }
3004     }
3005
3006     if(h->nal_ref_idc && ff_h264_decode_ref_pic_marking(h0, &s->gb) < 0 &&
3007        (s->avctx->err_recognition & AV_EF_EXPLODE))
3008         return AVERROR_INVALIDDATA;
3009
3010     if(FRAME_MBAFF){
3011         ff_h264_fill_mbaff_ref_list(h);
3012
3013         if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
3014             implicit_weight_table(h, 0);
3015             implicit_weight_table(h, 1);
3016         }
3017     }
3018
3019     if(h->slice_type_nos==AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
3020         ff_h264_direct_dist_scale_factor(h);
3021     ff_h264_direct_ref_list_init(h);
3022
3023     if( h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac ){
3024         tmp = get_ue_golomb_31(&s->gb);
3025         if(tmp > 2){
3026             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3027             return -1;
3028         }
3029         h->cabac_init_idc= tmp;
3030     }
3031
3032     h->last_qscale_diff = 0;
3033     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
3034     if(tmp>51+6*(h->sps.bit_depth_luma-8)){
3035         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3036         return -1;
3037     }
3038     s->qscale= tmp;
3039     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3040     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3041     //FIXME qscale / qp ... stuff
3042     if(h->slice_type == AV_PICTURE_TYPE_SP){
3043         get_bits1(&s->gb); /* sp_for_switch_flag */
3044     }
3045     if(h->slice_type==AV_PICTURE_TYPE_SP || h->slice_type == AV_PICTURE_TYPE_SI){
3046         get_se_golomb(&s->gb); /* slice_qs_delta */
3047     }
3048
3049     h->deblocking_filter = 1;
3050     h->slice_alpha_c0_offset = 52;
3051     h->slice_beta_offset = 52;
3052     if( h->pps.deblocking_filter_parameters_present ) {
3053         tmp= get_ue_golomb_31(&s->gb);
3054         if(tmp > 2){
3055             av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
3056             return -1;
3057         }
3058         h->deblocking_filter= tmp;
3059         if(h->deblocking_filter < 2)
3060             h->deblocking_filter^= 1; // 1<->0
3061
3062         if( h->deblocking_filter ) {
3063             h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
3064             h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
3065             if(   h->slice_alpha_c0_offset > 104U
3066                || h->slice_beta_offset     > 104U){
3067                 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);
3068                 return -1;
3069             }
3070         }
3071     }
3072
3073     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
3074        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != AV_PICTURE_TYPE_I)
3075        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == AV_PICTURE_TYPE_B)
3076        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
3077         h->deblocking_filter= 0;
3078
3079     if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
3080         if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
3081             /* Cheat slightly for speed:
3082                Do not bother to deblock across slices. */
3083             h->deblocking_filter = 2;
3084         } else {
3085             h0->max_contexts = 1;
3086             if(!h0->single_decode_warning) {
3087                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
3088                 h0->single_decode_warning = 1;
3089             }
3090             if (h != h0) {
3091                 av_log(h->s.avctx, AV_LOG_ERROR, "Deblocking switched inside frame.\n");
3092                 return 1;
3093             }
3094         }
3095     }
3096     h->qp_thresh = 15 + 52 - FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset)
3097                  - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1])
3098                  + 6 * (h->sps.bit_depth_luma - 8);
3099
3100 #if 0 //FMO
3101     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3102         slice_group_change_cycle= get_bits(&s->gb, ?);
3103 #endif
3104
3105     h0->last_slice_type = slice_type;
3106     h->slice_num = ++h0->current_slice;
3107     if(h->slice_num >= MAX_SLICES){
3108         av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
3109     }
3110
3111     for(j=0; j<2; j++){
3112         int id_list[16];
3113         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
3114         for(i=0; i<16; i++){
3115             id_list[i]= 60;
3116             if (h->ref_list[j][i].f.data[0]) {
3117                 int k;
3118                 uint8_t *base = h->ref_list[j][i].f.base[0];
3119                 for(k=0; k<h->short_ref_count; k++)
3120                     if (h->short_ref[k]->f.base[0] == base) {
3121                         id_list[i]= k;
3122                         break;
3123                     }
3124                 for(k=0; k<h->long_ref_count; k++)
3125                     if (h->long_ref[k] && h->long_ref[k]->f.base[0] == base) {
3126                         id_list[i]= h->short_ref_count + k;
3127                         break;
3128                     }
3129             }
3130         }
3131
3132         ref2frm[0]=
3133         ref2frm[1]= -1;
3134         for(i=0; i<16; i++)
3135             ref2frm[i+2]= 4*id_list[i]
3136                           + (h->ref_list[j][i].f.reference & 3);
3137         ref2frm[18+0]=
3138         ref2frm[18+1]= -1;
3139         for(i=16; i<48; i++)
3140             ref2frm[i+4]= 4*id_list[(i-16)>>1]
3141                           + (h->ref_list[j][i].f.reference & 3);
3142     }
3143
3144     //FIXME: fix draw_edges+PAFF+frame threads
3145     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE || (!h->sps.frame_mbs_only_flag && s->avctx->active_thread_type)) ? 0 : 16;
3146     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
3147
3148     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3149         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",
3150                h->slice_num,
3151                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
3152                first_mb_in_slice,
3153                av_get_picture_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
3154                pps_id, h->frame_num,
3155                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3156                h->ref_count[0], h->ref_count[1],
3157                s->qscale,
3158                h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
3159                h->use_weight,
3160                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
3161                h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
3162                );
3163     }
3164
3165     return 0;
3166 }
3167
3168 int ff_h264_get_slice_type(const H264Context *h)
3169 {
3170     switch (h->slice_type) {
3171     case AV_PICTURE_TYPE_P:  return 0;
3172     case AV_PICTURE_TYPE_B:  return 1;
3173     case AV_PICTURE_TYPE_I:  return 2;
3174     case AV_PICTURE_TYPE_SP: return 3;
3175     case AV_PICTURE_TYPE_SI: return 4;
3176     default:         return -1;
3177     }
3178 }
3179
3180 static av_always_inline void fill_filter_caches_inter(H264Context *h, MpegEncContext * const s, int mb_type, int top_xy,
3181                                                       int left_xy[LEFT_MBS], int top_type, int left_type[LEFT_MBS], int mb_xy, int list)
3182 {
3183     int b_stride = h->b_stride;
3184     int16_t (*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
3185     int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
3186     if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
3187         if(USES_LIST(top_type, list)){
3188             const int b_xy= h->mb2b_xy[top_xy] + 3*b_stride;
3189             const int b8_xy= 4*top_xy + 2;
3190             int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3191             AV_COPY128(mv_dst - 1*8, s->current_picture.f.motion_val[list][b_xy + 0]);
3192             ref_cache[0 - 1*8]=
3193             ref_cache[1 - 1*8]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 0]];
3194             ref_cache[2 - 1*8]=
3195             ref_cache[3 - 1*8]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 1]];
3196         }else{
3197             AV_ZERO128(mv_dst - 1*8);
3198             AV_WN32A(&ref_cache[0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3199         }
3200
3201         if(!IS_INTERLACED(mb_type^left_type[LTOP])){
3202             if(USES_LIST(left_type[LTOP], list)){
3203                 const int b_xy= h->mb2b_xy[left_xy[LTOP]] + 3;
3204                 const int b8_xy= 4*left_xy[LTOP] + 1;
3205                 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[LTOP]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3206                 AV_COPY32(mv_dst - 1 +  0, s->current_picture.f.motion_val[list][b_xy + b_stride*0]);
3207                 AV_COPY32(mv_dst - 1 +  8, s->current_picture.f.motion_val[list][b_xy + b_stride*1]);
3208                 AV_COPY32(mv_dst - 1 + 16, s->current_picture.f.motion_val[list][b_xy + b_stride*2]);
3209                 AV_COPY32(mv_dst - 1 + 24, s->current_picture.f.motion_val[list][b_xy + b_stride*3]);
3210                 ref_cache[-1 +  0]=
3211                 ref_cache[-1 +  8]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2*0]];
3212                 ref_cache[-1 + 16]=
3213                 ref_cache[-1 + 24]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2*1]];
3214             }else{
3215                 AV_ZERO32(mv_dst - 1 + 0);
3216                 AV_ZERO32(mv_dst - 1 + 8);
3217                 AV_ZERO32(mv_dst - 1 +16);
3218                 AV_ZERO32(mv_dst - 1 +24);
3219                 ref_cache[-1 +  0]=
3220                 ref_cache[-1 +  8]=
3221                 ref_cache[-1 + 16]=
3222                 ref_cache[-1 + 24]= LIST_NOT_USED;
3223             }
3224         }
3225     }
3226
3227     if(!USES_LIST(mb_type, list)){
3228         fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0,0), 4);
3229         AV_WN32A(&ref_cache[0*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3230         AV_WN32A(&ref_cache[1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3231         AV_WN32A(&ref_cache[2*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3232         AV_WN32A(&ref_cache[3*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3233         return;
3234     }
3235
3236     {
3237         int8_t *ref = &s->current_picture.f.ref_index[list][4*mb_xy];
3238         int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3239         uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101;
3240         uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]],ref2frm[list][ref[3]])&0x00FF00FF)*0x0101;
3241         AV_WN32A(&ref_cache[0*8], ref01);
3242         AV_WN32A(&ref_cache[1*8], ref01);
3243         AV_WN32A(&ref_cache[2*8], ref23);
3244         AV_WN32A(&ref_cache[3*8], ref23);
3245     }
3246
3247     {
3248         int16_t (*mv_src)[2] = &s->current_picture.f.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
3249         AV_COPY128(mv_dst + 8*0, mv_src + 0*b_stride);
3250         AV_COPY128(mv_dst + 8*1, mv_src + 1*b_stride);
3251         AV_COPY128(mv_dst + 8*2, mv_src + 2*b_stride);
3252         AV_COPY128(mv_dst + 8*3, mv_src + 3*b_stride);
3253     }
3254 }
3255
3256 /**
3257  *
3258  * @return non zero if the loop filter can be skiped
3259  */
3260 static int fill_filter_caches(H264Context *h, int mb_type){
3261     MpegEncContext * const s = &h->s;
3262     const int mb_xy= h->mb_xy;
3263     int top_xy, left_xy[LEFT_MBS];
3264     int top_type, left_type[LEFT_MBS];
3265     uint8_t *nnz;
3266     uint8_t *nnz_cache;
3267
3268     top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
3269
3270     /* Wow, what a mess, why didn't they simplify the interlacing & intra
3271      * stuff, I can't imagine that these complex rules are worth it. */
3272
3273     left_xy[LBOT] = left_xy[LTOP] = mb_xy-1;
3274     if(FRAME_MBAFF){
3275         const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.f.mb_type[mb_xy - 1]);
3276         const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
3277         if(s->mb_y&1){
3278             if (left_mb_field_flag != curr_mb_field_flag) {
3279                 left_xy[LTOP] -= s->mb_stride;
3280             }
3281         }else{
3282             if(curr_mb_field_flag){
3283                 top_xy += s->mb_stride & (((s->current_picture.f.mb_type[top_xy] >> 7) & 1) - 1);
3284             }
3285             if (left_mb_field_flag != curr_mb_field_flag) {
3286                 left_xy[LBOT] += s->mb_stride;
3287             }
3288         }
3289     }
3290
3291     h->top_mb_xy = top_xy;
3292     h->left_mb_xy[LTOP] = left_xy[LTOP];
3293     h->left_mb_xy[LBOT] = left_xy[LBOT];
3294     {
3295         //for sufficiently low qp, filtering wouldn't do anything
3296         //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
3297         int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
3298         int qp = s->current_picture.f.qscale_table[mb_xy];
3299         if(qp <= qp_thresh
3300            && (left_xy[LTOP] < 0 || ((qp + s->current_picture.f.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh)
3301            && (top_xy        < 0 || ((qp + s->current_picture.f.qscale_table[top_xy       ] + 1) >> 1) <= qp_thresh)) {
3302             if(!FRAME_MBAFF)
3303                 return 1;
3304             if ((left_xy[LTOP] < 0            || ((qp + s->current_picture.f.qscale_table[left_xy[LBOT]        ] + 1) >> 1) <= qp_thresh) &&
3305                 (top_xy        < s->mb_stride || ((qp + s->current_picture.f.qscale_table[top_xy - s->mb_stride] + 1) >> 1) <= qp_thresh))
3306                 return 1;
3307         }
3308     }
3309
3310     top_type        = s->current_picture.f.mb_type[top_xy];
3311     left_type[LTOP] = s->current_picture.f.mb_type[left_xy[LTOP]];
3312     left_type[LBOT] = s->current_picture.f.mb_type[left_xy[LBOT]];
3313     if(h->deblocking_filter == 2){
3314         if(h->slice_table[top_xy       ] != h->slice_num) top_type= 0;
3315         if(h->slice_table[left_xy[LBOT]] != h->slice_num) left_type[LTOP]= left_type[LBOT]= 0;
3316     }else{
3317         if(h->slice_table[top_xy       ] == 0xFFFF) top_type= 0;
3318         if(h->slice_table[left_xy[LBOT]] == 0xFFFF) left_type[LTOP]= left_type[LBOT] =0;
3319     }
3320     h->top_type       = top_type;
3321     h->left_type[LTOP]= left_type[LTOP];
3322     h->left_type[LBOT]= left_type[LBOT];
3323
3324     if(IS_INTRA(mb_type))
3325         return 0;
3326
3327     fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy, top_type, left_type, mb_xy, 0);
3328     if(h->list_count == 2)
3329         fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy, top_type, left_type, mb_xy, 1);
3330
3331     nnz = h->non_zero_count[mb_xy];
3332     nnz_cache = h->non_zero_count_cache;
3333     AV_COPY32(&nnz_cache[4+8*1], &nnz[ 0]);
3334     AV_COPY32(&nnz_cache[4+8*2], &nnz[ 4]);
3335     AV_COPY32(&nnz_cache[4+8*3], &nnz[ 8]);
3336     AV_COPY32(&nnz_cache[4+8*4], &nnz[12]);
3337     h->cbp= h->cbp_table[mb_xy];
3338
3339     if(top_type){
3340         nnz = h->non_zero_count[top_xy];
3341         AV_COPY32(&nnz_cache[4+8*0], &nnz[3*4]);
3342     }
3343
3344     if(left_type[LTOP]){
3345         nnz = h->non_zero_count[left_xy[LTOP]];
3346         nnz_cache[3+8*1]= nnz[3+0*4];
3347         nnz_cache[3+8*2]= nnz[3+1*4];
3348         nnz_cache[3+8*3]= nnz[3+2*4];
3349         nnz_cache[3+8*4]= nnz[3+3*4];
3350     }
3351
3352     // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
3353     if(!CABAC && h->pps.transform_8x8_mode){
3354         if(IS_8x8DCT(top_type)){
3355             nnz_cache[4+8*0]=
3356             nnz_cache[5+8*0]= (h->cbp_table[top_xy] & 0x4000) >> 12;
3357             nnz_cache[6+8*0]=
3358             nnz_cache[7+8*0]= (h->cbp_table[top_xy] & 0x8000) >> 12;
3359         }
3360         if(IS_8x8DCT(left_type[LTOP])){
3361             nnz_cache[3+8*1]=
3362             nnz_cache[3+8*2]= (h->cbp_table[left_xy[LTOP]]&0x2000) >> 12; //FIXME check MBAFF
3363         }
3364         if(IS_8x8DCT(left_type[LBOT])){
3365             nnz_cache[3+8*3]=
3366             nnz_cache[3+8*4]= (h->cbp_table[left_xy[LBOT]]&0x8000) >> 12; //FIXME check MBAFF
3367         }
3368
3369         if(IS_8x8DCT(mb_type)){
3370             nnz_cache[scan8[0   ]]= nnz_cache[scan8[1   ]]=
3371             nnz_cache[scan8[2   ]]= nnz_cache[scan8[3   ]]= (h->cbp & 0x1000) >> 12;
3372
3373             nnz_cache[scan8[0+ 4]]= nnz_cache[scan8[1+ 4]]=
3374             nnz_cache[scan8[2+ 4]]= nnz_cache[scan8[3+ 4]]= (h->cbp & 0x2000) >> 12;
3375
3376             nnz_cache[scan8[0+ 8]]= nnz_cache[scan8[1+ 8]]=
3377             nnz_cache[scan8[2+ 8]]= nnz_cache[scan8[3+ 8]]= (h->cbp & 0x4000) >> 12;
3378
3379             nnz_cache[scan8[0+12]]= nnz_cache[scan8[1+12]]=
3380             nnz_cache[scan8[2+12]]= nnz_cache[scan8[3+12]]= (h->cbp & 0x8000) >> 12;
3381         }
3382     }
3383
3384     return 0;
3385 }
3386
3387 static void loop_filter(H264Context *h, int start_x, int end_x){
3388     MpegEncContext * const s = &h->s;
3389     uint8_t  *dest_y, *dest_cb, *dest_cr;
3390     int linesize, uvlinesize, mb_x, mb_y;
3391     const int end_mb_y= s->mb_y + FRAME_MBAFF;
3392     const int old_slice_type= h->slice_type;
3393     const int pixel_shift = h->pixel_shift;
3394     const int block_h = 16 >> s->chroma_y_shift;
3395
3396     if(h->deblocking_filter) {
3397         for(mb_x= start_x; mb_x<end_x; mb_x++){
3398             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
3399                 int mb_xy, mb_type;
3400                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
3401                 h->slice_num= h->slice_table[mb_xy];
3402                 mb_type = s->current_picture.f.mb_type[mb_xy];
3403                 h->list_count= h->list_counts[mb_xy];
3404
3405                 if(FRAME_MBAFF)
3406                     h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
3407
3408                 s->mb_x= mb_x;
3409                 s->mb_y= mb_y;
3410                 dest_y  = s->current_picture.f.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize  ) * 16;
3411                 dest_cb = s->current_picture.f.data[1] + (mb_x << pixel_shift) * (8 << CHROMA444) + mb_y * s->uvlinesize * block_h;
3412                 dest_cr = s->current_picture.f.data[2] + (mb_x << pixel_shift) * (8 << CHROMA444) + mb_y * s->uvlinesize * block_h;
3413                     //FIXME simplify above
3414
3415                 if (MB_FIELD) {
3416                     linesize   = h->mb_linesize   = s->linesize * 2;
3417                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
3418                     if(mb_y&1){ //FIXME move out of this function?
3419                         dest_y -= s->linesize*15;
3420                         dest_cb-= s->uvlinesize * (block_h - 1);
3421                         dest_cr-= s->uvlinesize * (block_h - 1);
3422                     }
3423                 } else {
3424                     linesize   = h->mb_linesize   = s->linesize;
3425                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
3426                 }
3427                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
3428                 if(fill_filter_caches(h, mb_type))
3429                     continue;
3430                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.f.qscale_table[mb_xy]);
3431                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.f.qscale_table[mb_xy]);
3432
3433                 if (FRAME_MBAFF) {
3434                     ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3435                 } else {
3436                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3437                 }
3438             }
3439         }
3440     }
3441     h->slice_type= old_slice_type;
3442     s->mb_x= end_x;
3443     s->mb_y= end_mb_y - FRAME_MBAFF;
3444     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3445     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3446 }
3447
3448 static void predict_field_decoding_flag(H264Context *h){
3449     MpegEncContext * const s = &h->s;
3450     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3451     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
3452                 ? s->current_picture.f.mb_type[mb_xy - 1]
3453                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
3454                 ? s->current_picture.f.mb_type[mb_xy - s->mb_stride]
3455                 : 0;
3456     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
3457 }
3458
3459 /**
3460  * Draw edges and report progress for the last MB row.
3461  */
3462 static void decode_finish_row(H264Context *h){
3463     MpegEncContext * const s = &h->s;
3464     int top = 16*(s->mb_y >> FIELD_PICTURE);
3465     int height = 16 << FRAME_MBAFF;
3466     int deblock_border = (16 + 4) << FRAME_MBAFF;
3467     int pic_height = 16*s->mb_height >> FIELD_PICTURE;
3468
3469     if (h->deblocking_filter) {
3470         if((top + height) >= pic_height)
3471             height += deblock_border;
3472
3473         top -= deblock_border;
3474     }
3475
3476     if (top >= pic_height || (top + height) < h->emu_edge_height)
3477         return;
3478
3479     height = FFMIN(height, pic_height - top);
3480     if (top < h->emu_edge_height) {
3481         height = top+height;
3482         top = 0;
3483     }
3484
3485     ff_draw_horiz_band(s, top, height);
3486
3487     if (s->dropable) return;
3488
3489     ff_thread_report_progress((AVFrame*)s->current_picture_ptr, top + height - 1,
3490                              s->picture_structure==PICT_BOTTOM_FIELD);
3491 }
3492
3493 static int decode_slice(struct AVCodecContext *avctx, void *arg){
3494     H264Context *h = *(void**)arg;
3495     MpegEncContext * const s = &h->s;
3496     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
3497     int lf_x_start = s->mb_x;
3498
3499     s->mb_skip_run= -1;
3500
3501     h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
3502                     (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
3503
3504     if( h->pps.cabac ) {
3505         /* realign */
3506         align_get_bits( &s->gb );
3507
3508         /* init cabac */
3509         ff_init_cabac_states( &h->cabac);
3510         ff_init_cabac_decoder( &h->cabac,
3511                                s->gb.buffer + get_bits_count(&s->gb)/8,
3512                                (get_bits_left(&s->gb) + 7)/8);
3513
3514         ff_h264_init_cabac_states(h);
3515
3516         for(;;){
3517 //START_TIMER
3518             int ret = ff_h264_decode_mb_cabac(h);
3519             int eos;
3520 //STOP_TIMER("decode_mb_cabac")
3521
3522             if(ret>=0) ff_h264_hl_decode_mb(h);
3523
3524             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
3525                 s->mb_y++;
3526
3527                 ret = ff_h264_decode_mb_cabac(h);
3528
3529                 if(ret>=0) ff_h264_hl_decode_mb(h);
3530                 s->mb_y--;
3531             }
3532             eos = get_cabac_terminate( &h->cabac );
3533
3534             if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
3535                 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);
3536                 if (s->mb_x >= lf_x_start) loop_filter(h, lf_x_start, s->mb_x + 1);
3537                 return 0;
3538             }
3539             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
3540                 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);
3541                 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);
3542                 return -1;
3543             }
3544
3545             if( ++s->mb_x >= s->mb_width ) {
3546                 loop_filter(h, lf_x_start, s->mb_x);
3547                 s->mb_x = lf_x_start = 0;
3548                 decode_finish_row(h);
3549                 ++s->mb_y;
3550                 if(FIELD_OR_MBAFF_PICTURE) {
3551                     ++s->mb_y;
3552                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
3553                         predict_field_decoding_flag(h);
3554                 }
3555             }
3556
3557             if( eos || s->mb_y >= s->mb_height ) {
3558                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3559                 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);
3560                 if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
3561                 return 0;
3562             }
3563         }
3564
3565     } else {
3566         for(;;){
3567             int ret = ff_h264_decode_mb_cavlc(h);
3568
3569             if(ret>=0) ff_h264_hl_decode_mb(h);
3570
3571             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
3572                 s->mb_y++;
3573                 ret = ff_h264_decode_mb_cavlc(h);
3574
3575                 if(ret>=0) ff_h264_hl_decode_mb(h);
3576                 s->mb_y--;
3577             }
3578
3579             if(ret<0){
3580                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3581                 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);
3582                 return -1;
3583             }
3584
3585             if(++s->mb_x >= s->mb_width){
3586                 loop_filter(h, lf_x_start, s->mb_x);
3587                 s->mb_x = lf_x_start = 0;
3588                 decode_finish_row(h);
3589                 ++s->mb_y;
3590                 if(FIELD_OR_MBAFF_PICTURE) {
3591                     ++s->mb_y;
3592                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
3593                         predict_field_decoding_flag(h);
3594                 }
3595                 if(s->mb_y >= s->mb_height){
3596                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3597
3598                     if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
3599                         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);
3600
3601                         return 0;
3602                     }else{
3603                         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);
3604
3605                         return -1;
3606                     }
3607                 }
3608             }
3609
3610             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
3611                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3612                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
3613                     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);
3614                     if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
3615
3616                     return 0;
3617                 }else{
3618                     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);
3619
3620                     return -1;
3621                 }
3622             }
3623         }
3624     }
3625 }
3626
3627 /**
3628  * Call decode_slice() for each context.
3629  *
3630  * @param h h264 master context
3631  * @param context_count number of contexts to execute
3632  */
3633 static int execute_decode_slices(H264Context *h, int context_count){
3634     MpegEncContext * const s = &h->s;
3635     AVCodecContext * const avctx= s->avctx;
3636     H264Context *hx;
3637     int i;
3638
3639     if (s->avctx->hwaccel || s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
3640         return 0;
3641     if(context_count == 1) {
3642         return decode_slice(avctx, &h);
3643     } else {
3644         for(i = 1; i < context_count; i++) {
3645             hx = h->thread_context[i];
3646             hx->s.error_recognition = avctx->error_recognition;
3647             hx->s.error_count = 0;
3648         }
3649
3650         avctx->execute(avctx, (void *)decode_slice,
3651                        h->thread_context, NULL, context_count, sizeof(void*));
3652
3653         /* pull back stuff from slices to master context */
3654         hx = h->thread_context[context_count - 1];
3655         s->mb_x = hx->s.mb_x;
3656         s->mb_y = hx->s.mb_y;
3657         s->dropable = hx->s.dropable;
3658         s->picture_structure = hx->s.picture_structure;
3659         for(i = 1; i < context_count; i++)
3660             h->s.error_count += h->thread_context[i]->s.error_count;
3661     }
3662
3663     return 0;
3664 }
3665
3666
3667 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
3668     MpegEncContext * const s = &h->s;
3669     AVCodecContext * const avctx= s->avctx;
3670     H264Context *hx; ///< thread context
3671     int buf_index;
3672     int context_count;
3673     int next_avc;
3674     int pass = !(avctx->active_thread_type & FF_THREAD_FRAME);
3675     int nals_needed=0; ///< number of NALs that need decoding before the next frame thread starts
3676     int nal_index;
3677
3678     h->max_contexts = (HAVE_THREADS && (s->avctx->active_thread_type&FF_THREAD_SLICE)) ? avctx->thread_count : 1;
3679     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
3680         h->current_slice = 0;
3681         if (!s->first_field)
3682             s->current_picture_ptr= NULL;
3683         ff_h264_reset_sei(h);
3684     }
3685
3686     for(;pass <= 1;pass++){
3687         buf_index = 0;
3688         context_count = 0;
3689         next_avc = h->is_avc ? 0 : buf_size;
3690         nal_index = 0;
3691     for(;;){
3692         int consumed;
3693         int dst_length;
3694         int bit_length;
3695         const uint8_t *ptr;
3696         int i, nalsize = 0;
3697         int err;
3698
3699         if(buf_index >= next_avc) {
3700             if(buf_index >= buf_size) break;
3701             nalsize = 0;
3702             for(i = 0; i < h->nal_length_size; i++)
3703                 nalsize = (nalsize << 8) | buf[buf_index++];
3704             if(nalsize <= 0 || nalsize > buf_size - buf_index){
3705                 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
3706                 break;
3707             }
3708             next_avc= buf_index + nalsize;
3709         } else {
3710             // start code prefix search
3711             for(; buf_index + 3 < next_avc; buf_index++){
3712                 // This should always succeed in the first iteration.
3713                 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
3714                     break;
3715             }
3716
3717             if(buf_index+3 >= buf_size) break;
3718
3719             buf_index+=3;
3720             if(buf_index >= next_avc) continue;
3721         }
3722
3723         hx = h->thread_context[context_count];
3724
3725         ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
3726         if (ptr==NULL || dst_length < 0){
3727             return -1;
3728         }
3729         i= buf_index + consumed;
3730         if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
3731            buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
3732             s->workaround_bugs |= FF_BUG_TRUNCATED;
3733
3734         if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
3735         while(ptr[dst_length - 1] == 0 && dst_length > 0)
3736             dst_length--;
3737         }
3738         bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
3739
3740         if(s->avctx->debug&FF_DEBUG_STARTCODE){
3741             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);
3742         }
3743
3744         if (h->is_avc && (nalsize != consumed) && nalsize){
3745             av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
3746         }
3747
3748         buf_index += consumed;
3749         nal_index++;
3750
3751         if(pass == 0) {
3752             // packets can sometimes contain multiple PPS/SPS
3753             // e.g. two PAFF field pictures in one packet, or a demuxer which splits NALs strangely
3754             // if so, when frame threading we can't start the next thread until we've read all of them
3755             switch (hx->nal_unit_type) {
3756                 case NAL_SPS:
3757                 case NAL_PPS:
3758                     nals_needed = nal_index;
3759                     break;
3760                 case NAL_IDR_SLICE:
3761                 case NAL_SLICE:
3762                     init_get_bits(&hx->s.gb, ptr, bit_length);
3763                     if (!get_ue_golomb(&hx->s.gb))
3764                         nals_needed = nal_index;
3765             }
3766             continue;
3767         }
3768
3769         //FIXME do not discard SEI id
3770         if(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0)
3771             continue;
3772
3773       again:
3774         err = 0;
3775         switch(hx->nal_unit_type){
3776         case NAL_IDR_SLICE:
3777             if (h->nal_unit_type != NAL_IDR_SLICE) {
3778                 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
3779                 return -1;
3780             }
3781             idr(h); //FIXME ensure we don't loose some frames if there is reordering
3782         case NAL_SLICE:
3783             init_get_bits(&hx->s.gb, ptr, bit_length);
3784             hx->intra_gb_ptr=
3785             hx->inter_gb_ptr= &hx->s.gb;
3786             hx->s.data_partitioning = 0;
3787
3788             if((err = decode_slice_header(hx, h)))
3789                break;
3790
3791             s->current_picture_ptr->f.key_frame |=
3792                     (hx->nal_unit_type == NAL_IDR_SLICE) ||
3793                     (h->sei_recovery_frame_cnt >= 0);
3794
3795             if (h->current_slice == 1) {
3796                 if(!(s->flags2 & CODEC_FLAG2_CHUNKS)) {
3797                     decode_postinit(h, nal_index >= nals_needed);
3798                 }
3799
3800                 if (s->avctx->hwaccel && s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
3801                     return -1;
3802                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
3803                     ff_vdpau_h264_picture_start(s);
3804             }
3805
3806             if(hx->redundant_pic_count==0
3807                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
3808                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
3809                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
3810                && avctx->skip_frame < AVDISCARD_ALL){
3811                 if(avctx->hwaccel) {
3812                     if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
3813                         return -1;
3814                 }else
3815                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
3816                     static const uint8_t start_code[] = {0x00, 0x00, 0x01};
3817                     ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
3818                     ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
3819                 }else
3820                     context_count++;
3821             }
3822             break;
3823         case NAL_DPA:
3824             init_get_bits(&hx->s.gb, ptr, bit_length);
3825             hx->intra_gb_ptr=
3826             hx->inter_gb_ptr= NULL;
3827
3828             if ((err = decode_slice_header(hx, h)) < 0)
3829                 break;
3830
3831             hx->s.data_partitioning = 1;
3832
3833             break;
3834         case NAL_DPB:
3835             init_get_bits(&hx->intra_gb, ptr, bit_length);
3836             hx->intra_gb_ptr= &hx->intra_gb;
3837             break;
3838         case NAL_DPC:
3839             init_get_bits(&hx->inter_gb, ptr, bit_length);
3840             hx->inter_gb_ptr= &hx->inter_gb;
3841
3842             if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
3843                && s->context_initialized
3844                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
3845                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
3846                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
3847                && avctx->skip_frame < AVDISCARD_ALL)
3848                 context_count++;
3849             break;
3850         case NAL_SEI:
3851             init_get_bits(&s->gb, ptr, bit_length);
3852             ff_h264_decode_sei(h);
3853             break;
3854         case NAL_SPS:
3855             init_get_bits(&s->gb, ptr, bit_length);
3856             ff_h264_decode_seq_parameter_set(h);
3857
3858             if (s->flags& CODEC_FLAG_LOW_DELAY ||
3859                 (h->sps.bitstream_restriction_flag && !h->sps.num_reorder_frames))
3860                 s->low_delay=1;
3861
3862             if(avctx->has_b_frames < 2)
3863                 avctx->has_b_frames= !s->low_delay;
3864
3865             if (avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
3866                 h->cur_chroma_format_idc != h->sps.chroma_format_idc) {
3867                 if (h->sps.bit_depth_luma >= 8 && h->sps.bit_depth_luma <= 10) {
3868                     avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
3869                     h->cur_chroma_format_idc = h->sps.chroma_format_idc;
3870                     h->pixel_shift = h->sps.bit_depth_luma > 8;
3871
3872                     ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma, h->sps.chroma_format_idc);
3873                     ff_h264_pred_init(&h->hpc, s->codec_id, h->sps.bit_depth_luma, h->sps.chroma_format_idc);
3874                     s->dsp.dct_bits = h->sps.bit_depth_luma > 8 ? 32 : 16;
3875                     dsputil_init(&s->dsp, s->avctx);
3876                 } else {
3877                     av_log(avctx, AV_LOG_ERROR, "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
3878                     return -1;
3879                 }
3880             }
3881             break;
3882         case NAL_PPS:
3883             init_get_bits(&s->gb, ptr, bit_length);
3884
3885             ff_h264_decode_picture_parameter_set(h, bit_length);
3886
3887             break;
3888         case NAL_AUD:
3889         case NAL_END_SEQUENCE:
3890         case NAL_END_STREAM:
3891         case NAL_FILLER_DATA:
3892         case NAL_SPS_EXT:
3893         case NAL_AUXILIARY_SLICE:
3894             break;
3895         default:
3896             av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
3897         }
3898
3899         if(context_count == h->max_contexts) {
3900             execute_decode_slices(h, context_count);
3901             context_count = 0;
3902         }
3903
3904         if (err < 0)
3905             av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
3906         else if(err == 1) {
3907             /* Slice could not be decoded in parallel mode, copy down
3908              * NAL unit stuff to context 0 and restart. Note that
3909              * rbsp_buffer is not transferred, but since we no longer
3910              * run in parallel mode this should not be an issue. */
3911             h->nal_unit_type = hx->nal_unit_type;
3912             h->nal_ref_idc   = hx->nal_ref_idc;
3913             hx = h;
3914             goto again;
3915         }
3916     }
3917     }
3918     if(context_count)
3919         execute_decode_slices(h, context_count);
3920     return buf_index;
3921 }
3922
3923 /**
3924  * returns the number of bytes consumed for building the current frame
3925  */
3926 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
3927         if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
3928         if(pos+10>buf_size) pos=buf_size; // oops ;)
3929
3930         return pos;
3931 }
3932
3933 static int decode_frame(AVCodecContext *avctx,
3934                              void *data, int *data_size,
3935                              AVPacket *avpkt)
3936 {
3937     const uint8_t *buf = avpkt->data;
3938     int buf_size = avpkt->size;
3939     H264Context *h = avctx->priv_data;
3940     MpegEncContext *s = &h->s;
3941     AVFrame *pict = data;
3942     int buf_index;
3943
3944     s->flags= avctx->flags;
3945     s->flags2= avctx->flags2;
3946
3947    /* end of stream, output what is still in the buffers */
3948  out:
3949     if (buf_size == 0) {
3950         Picture *out;
3951         int i, out_idx;
3952
3953         s->current_picture_ptr = NULL;
3954
3955 //FIXME factorize this with the output code below
3956         out = h->delayed_pic[0];
3957         out_idx = 0;
3958         for (i = 1; h->delayed_pic[i] && !h->delayed_pic[i]->f.key_frame && !h->delayed_pic[i]->mmco_reset; i++)
3959             if(h->delayed_pic[i]->poc < out->poc){
3960                 out = h->delayed_pic[i];
3961                 out_idx = i;
3962             }
3963
3964         for(i=out_idx; h->delayed_pic[i]; i++)
3965             h->delayed_pic[i] = h->delayed_pic[i+1];
3966
3967         if(out){
3968             *data_size = sizeof(AVFrame);
3969             *pict= *(AVFrame*)out;
3970         }
3971
3972         return 0;
3973     }
3974
3975     buf_index=decode_nal_units(h, buf, buf_size);
3976     if(buf_index < 0)
3977         return -1;
3978
3979     if (!s->current_picture_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
3980         buf_size = 0;
3981         goto out;
3982     }
3983
3984     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
3985         if (avctx->skip_frame >= AVDISCARD_NONREF)
3986             return 0;
3987         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
3988         return -1;
3989     }
3990
3991     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
3992
3993         if(s->flags2 & CODEC_FLAG2_CHUNKS) decode_postinit(h, 1);
3994
3995         field_end(h, 0);
3996
3997         if (!h->next_output_pic) {
3998             /* Wait for second field. */
3999             *data_size = 0;
4000
4001         } else {
4002             *data_size = sizeof(AVFrame);
4003             *pict = *(AVFrame*)h->next_output_pic;
4004         }
4005     }
4006
4007     assert(pict->data[0] || !*data_size);
4008     ff_print_debug_info(s, pict);
4009 //printf("out %d\n", (int)pict->data[0]);
4010
4011     return get_consumed_bytes(s, buf_index, buf_size);
4012 }
4013 #if 0
4014 static inline void fill_mb_avail(H264Context *h){
4015     MpegEncContext * const s = &h->s;
4016     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4017
4018     if(s->mb_y){
4019         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
4020         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
4021         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
4022     }else{
4023         h->mb_avail[0]=
4024         h->mb_avail[1]=
4025         h->mb_avail[2]= 0;
4026     }
4027     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
4028     h->mb_avail[4]= 1; //FIXME move out
4029     h->mb_avail[5]= 0; //FIXME move out
4030 }
4031 #endif
4032
4033 #ifdef TEST
4034 #undef printf
4035 #undef random
4036 #define COUNT 8000