flvdec: Do not call parse_keyframes_index with a NULL stream
[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     s->decode=1; //FIXME
1037
1038     dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
1039
1040     memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
1041     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
1042 }
1043
1044 int ff_h264_decode_extradata(H264Context *h)
1045 {
1046     AVCodecContext *avctx = h->s.avctx;
1047
1048     if(avctx->extradata[0] == 1){
1049         int i, cnt, nalsize;
1050         unsigned char *p = avctx->extradata;
1051
1052         h->is_avc = 1;
1053
1054         if(avctx->extradata_size < 7) {
1055             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
1056             return -1;
1057         }
1058         /* sps and pps in the avcC always have length coded with 2 bytes,
1059            so put a fake nal_length_size = 2 while parsing them */
1060         h->nal_length_size = 2;
1061         // Decode sps from avcC
1062         cnt = *(p+5) & 0x1f; // Number of sps
1063         p += 6;
1064         for (i = 0; i < cnt; i++) {
1065             nalsize = AV_RB16(p) + 2;
1066             if (p - avctx->extradata + nalsize > avctx->extradata_size)
1067                 return -1;
1068             if(decode_nal_units(h, p, nalsize) < 0) {
1069                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
1070                 return -1;
1071             }
1072             p += nalsize;
1073         }
1074         // Decode pps from avcC
1075         cnt = *(p++); // Number of pps
1076         for (i = 0; i < cnt; i++) {
1077             nalsize = AV_RB16(p) + 2;
1078             if (p - avctx->extradata + nalsize > avctx->extradata_size)
1079                 return -1;
1080             if (decode_nal_units(h, p, nalsize) < 0) {
1081                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
1082                 return -1;
1083             }
1084             p += nalsize;
1085         }
1086         // Now store right nal length size, that will be use to parse all other nals
1087         h->nal_length_size = (avctx->extradata[4] & 0x03) + 1;
1088     } else {
1089         h->is_avc = 0;
1090         if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
1091             return -1;
1092     }
1093     return 0;
1094 }
1095
1096 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
1097     H264Context *h= avctx->priv_data;
1098     MpegEncContext * const s = &h->s;
1099
1100     MPV_decode_defaults(s);
1101
1102     s->avctx = avctx;
1103     common_init(h);
1104
1105     s->out_format = FMT_H264;
1106     s->workaround_bugs= avctx->workaround_bugs;
1107
1108     // set defaults
1109 //    s->decode_mb= ff_h263_decode_mb;
1110     s->quarter_sample = 1;
1111     if(!avctx->has_b_frames)
1112     s->low_delay= 1;
1113
1114     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
1115
1116     ff_h264_decode_init_vlc();
1117
1118     h->pixel_shift = 0;
1119     h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
1120
1121     h->thread_context[0] = h;
1122     h->outputed_poc = h->next_outputed_poc = INT_MIN;
1123     h->prev_poc_msb= 1<<16;
1124     h->x264_build = -1;
1125     ff_h264_reset_sei(h);
1126     if(avctx->codec_id == CODEC_ID_H264){
1127         if(avctx->ticks_per_frame == 1){
1128             s->avctx->time_base.den *=2;
1129         }
1130         avctx->ticks_per_frame = 2;
1131     }
1132
1133     if(avctx->extradata_size > 0 && avctx->extradata &&
1134         ff_h264_decode_extradata(h))
1135         return -1;
1136
1137     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
1138         s->avctx->has_b_frames = h->sps.num_reorder_frames;
1139         s->low_delay = 0;
1140     }
1141
1142     return 0;
1143 }
1144
1145 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b)+(size))))
1146 static void copy_picture_range(Picture **to, Picture **from, int count, MpegEncContext *new_base, MpegEncContext *old_base)
1147 {
1148     int i;
1149
1150     for (i=0; i<count; i++){
1151         assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
1152                 IN_RANGE(from[i], old_base->picture, sizeof(Picture) * old_base->picture_count) ||
1153                 !from[i]));
1154         to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1155     }
1156 }
1157
1158 static void copy_parameter_set(void **to, void **from, int count, int size)
1159 {
1160     int i;
1161
1162     for (i=0; i<count; i++){
1163         if (to[i] && !from[i]) av_freep(&to[i]);
1164         else if (from[i] && !to[i]) to[i] = av_malloc(size);
1165
1166         if (from[i]) memcpy(to[i], from[i], size);
1167     }
1168 }
1169
1170 static int decode_init_thread_copy(AVCodecContext *avctx){
1171     H264Context *h= avctx->priv_data;
1172
1173     if (!avctx->is_copy) return 0;
1174     memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1175     memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1176
1177     return 0;
1178 }
1179
1180 #define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
1181 static int decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src){
1182     H264Context *h= dst->priv_data, *h1= src->priv_data;
1183     MpegEncContext * const s = &h->s, * const s1 = &h1->s;
1184     int inited = s->context_initialized, err;
1185     int i;
1186
1187     if(dst == src || !s1->context_initialized) return 0;
1188
1189     err = ff_mpeg_update_thread_context(dst, src);
1190     if(err) return err;
1191
1192     //FIXME handle width/height changing
1193     if(!inited){
1194         for(i = 0; i < MAX_SPS_COUNT; i++)
1195             av_freep(h->sps_buffers + i);
1196
1197         for(i = 0; i < MAX_PPS_COUNT; i++)
1198             av_freep(h->pps_buffers + i);
1199
1200         memcpy(&h->s + 1, &h1->s + 1, sizeof(H264Context) - sizeof(MpegEncContext)); //copy all fields after MpegEnc
1201         memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1202         memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1203         if (ff_h264_alloc_tables(h) < 0) {
1204             av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
1205             return AVERROR(ENOMEM);
1206         }
1207         context_init(h);
1208
1209         for(i=0; i<2; i++){
1210             h->rbsp_buffer[i] = NULL;
1211             h->rbsp_buffer_size[i] = 0;
1212         }
1213
1214         h->thread_context[0] = h;
1215
1216         // frame_start may not be called for the next thread (if it's decoding a bottom field)
1217         // so this has to be allocated here
1218         h->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
1219
1220         s->dsp.clear_blocks(h->mb);
1221         s->dsp.clear_blocks(h->mb+(24*16<<h->pixel_shift));
1222     }
1223
1224     //extradata/NAL handling
1225     h->is_avc          = h1->is_avc;
1226
1227     //SPS/PPS
1228     copy_parameter_set((void**)h->sps_buffers, (void**)h1->sps_buffers, MAX_SPS_COUNT, sizeof(SPS));
1229     h->sps             = h1->sps;
1230     copy_parameter_set((void**)h->pps_buffers, (void**)h1->pps_buffers, MAX_PPS_COUNT, sizeof(PPS));
1231     h->pps             = h1->pps;
1232
1233     //Dequantization matrices
1234     //FIXME these are big - can they be only copied when PPS changes?
1235     copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1236
1237     for(i=0; i<6; i++)
1238         h->dequant4_coeff[i] = h->dequant4_buffer[0] + (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
1239
1240     for(i=0; i<6; i++)
1241         h->dequant8_coeff[i] = h->dequant8_buffer[0] + (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
1242
1243     h->dequant_coeff_pps = h1->dequant_coeff_pps;
1244
1245     //POC timing
1246     copy_fields(h, h1, poc_lsb, redundant_pic_count);
1247
1248     //reference lists
1249     copy_fields(h, h1, ref_count, list_count);
1250     copy_fields(h, h1, ref_list,  intra_gb);
1251     copy_fields(h, h1, short_ref, cabac_init_idc);
1252
1253     copy_picture_range(h->short_ref,   h1->short_ref,   32, s, s1);
1254     copy_picture_range(h->long_ref,    h1->long_ref,    32, s, s1);
1255     copy_picture_range(h->delayed_pic, h1->delayed_pic, MAX_DELAYED_PIC_COUNT+2, s, s1);
1256
1257     h->last_slice_type = h1->last_slice_type;
1258
1259     if(!s->current_picture_ptr) return 0;
1260
1261     if(!s->dropable) {
1262         err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1263         h->prev_poc_msb     = h->poc_msb;
1264         h->prev_poc_lsb     = h->poc_lsb;
1265     }
1266     h->prev_frame_num_offset= h->frame_num_offset;
1267     h->prev_frame_num       = h->frame_num;
1268     h->outputed_poc         = h->next_outputed_poc;
1269
1270     return err;
1271 }
1272
1273 int ff_h264_frame_start(H264Context *h){
1274     MpegEncContext * const s = &h->s;
1275     int i;
1276     const int pixel_shift = h->pixel_shift;
1277     int thread_count = (s->avctx->active_thread_type & FF_THREAD_SLICE) ? s->avctx->thread_count : 1;
1278
1279     if(MPV_frame_start(s, s->avctx) < 0)
1280         return -1;
1281     ff_er_frame_start(s);
1282     /*
1283      * MPV_frame_start uses pict_type to derive key_frame.
1284      * This is incorrect for H.264; IDR markings must be used.
1285      * Zero here; IDR markings per slice in frame or fields are ORed in later.
1286      * See decode_nal_units().
1287      */
1288     s->current_picture_ptr->f.key_frame = 0;
1289     s->current_picture_ptr->mmco_reset= 0;
1290
1291     assert(s->linesize && s->uvlinesize);
1292
1293     for(i=0; i<16; i++){
1294         h->block_offset[i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
1295         h->block_offset[48+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
1296     }
1297     for(i=0; i<16; i++){
1298         h->block_offset[16+i]=
1299         h->block_offset[32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1300         h->block_offset[48+16+i]=
1301         h->block_offset[48+32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1302     }
1303
1304     /* can't be in alloc_tables because linesize isn't known there.
1305      * FIXME: redo bipred weight to not require extra buffer? */
1306     for(i = 0; i < thread_count; i++)
1307         if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
1308             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
1309
1310     /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
1311     memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
1312
1313 //    s->decode = (s->flags & CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.f.reference /*|| h->contains_intra*/ || 1;
1314
1315     // We mark the current picture as non-reference after allocating it, so
1316     // that if we break out due to an error it can be released automatically
1317     // in the next MPV_frame_start().
1318     // SVQ3 as well as most other codecs have only last/next/current and thus
1319     // get released even with set reference, besides SVQ3 and others do not
1320     // mark frames as reference later "naturally".
1321     if(s->codec_id != CODEC_ID_SVQ3)
1322         s->current_picture_ptr->f.reference = 0;
1323
1324     s->current_picture_ptr->field_poc[0]=
1325     s->current_picture_ptr->field_poc[1]= INT_MAX;
1326
1327     h->next_output_pic = NULL;
1328
1329     assert(s->current_picture_ptr->long_ref==0);
1330
1331     return 0;
1332 }
1333
1334 /**
1335   * Run setup operations that must be run after slice header decoding.
1336   * This includes finding the next displayed frame.
1337   *
1338   * @param h h264 master context
1339   * @param setup_finished enough NALs have been read that we can call
1340   * ff_thread_finish_setup()
1341   */
1342 static void decode_postinit(H264Context *h, int setup_finished){
1343     MpegEncContext * const s = &h->s;
1344     Picture *out = s->current_picture_ptr;
1345     Picture *cur = s->current_picture_ptr;
1346     int i, pics, out_of_order, out_idx;
1347
1348     s->current_picture_ptr->f.qscale_type = FF_QSCALE_TYPE_H264;
1349     s->current_picture_ptr->f.pict_type   = s->pict_type;
1350
1351     if (h->next_output_pic) return;
1352
1353     if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
1354         //FIXME: if we have two PAFF fields in one packet, we can't start the next thread here.
1355         //If we have one field per packet, we can. The check in decode_nal_units() is not good enough
1356         //to find this yet, so we assume the worst for now.
1357         //if (setup_finished)
1358         //    ff_thread_finish_setup(s->avctx);
1359         return;
1360     }
1361
1362     cur->f.interlaced_frame = 0;
1363     cur->f.repeat_pict      = 0;
1364
1365     /* Signal interlacing information externally. */
1366     /* Prioritize picture timing SEI information over used decoding process if it exists. */
1367
1368     if(h->sps.pic_struct_present_flag){
1369         switch (h->sei_pic_struct)
1370         {
1371         case SEI_PIC_STRUCT_FRAME:
1372             break;
1373         case SEI_PIC_STRUCT_TOP_FIELD:
1374         case SEI_PIC_STRUCT_BOTTOM_FIELD:
1375             cur->f.interlaced_frame = 1;
1376             break;
1377         case SEI_PIC_STRUCT_TOP_BOTTOM:
1378         case SEI_PIC_STRUCT_BOTTOM_TOP:
1379             if (FIELD_OR_MBAFF_PICTURE)
1380                 cur->f.interlaced_frame = 1;
1381             else
1382                 // try to flag soft telecine progressive
1383                 cur->f.interlaced_frame = h->prev_interlaced_frame;
1384             break;
1385         case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1386         case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
1387             // Signal the possibility of telecined film externally (pic_struct 5,6)
1388             // From these hints, let the applications decide if they apply deinterlacing.
1389             cur->f.repeat_pict = 1;
1390             break;
1391         case SEI_PIC_STRUCT_FRAME_DOUBLING:
1392             // Force progressive here, as doubling interlaced frame is a bad idea.
1393             cur->f.repeat_pict = 2;
1394             break;
1395         case SEI_PIC_STRUCT_FRAME_TRIPLING:
1396             cur->f.repeat_pict = 4;
1397             break;
1398         }
1399
1400         if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
1401             cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
1402     }else{
1403         /* Derive interlacing flag from used decoding process. */
1404         cur->f.interlaced_frame = FIELD_OR_MBAFF_PICTURE;
1405     }
1406     h->prev_interlaced_frame = cur->f.interlaced_frame;
1407
1408     if (cur->field_poc[0] != cur->field_poc[1]){
1409         /* Derive top_field_first from field pocs. */
1410         cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1];
1411     }else{
1412         if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) {
1413             /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
1414             if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
1415               || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
1416                 cur->f.top_field_first = 1;
1417             else
1418                 cur->f.top_field_first = 0;
1419         }else{
1420             /* Most likely progressive */
1421             cur->f.top_field_first = 0;
1422         }
1423     }
1424
1425     //FIXME do something with unavailable reference frames
1426
1427     /* Sort B-frames into display order */
1428
1429     if(h->sps.bitstream_restriction_flag
1430        && s->avctx->has_b_frames < h->sps.num_reorder_frames){
1431         s->avctx->has_b_frames = h->sps.num_reorder_frames;
1432         s->low_delay = 0;
1433     }
1434
1435     if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
1436        && !h->sps.bitstream_restriction_flag){
1437         s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
1438         s->low_delay= 0;
1439     }
1440
1441     pics = 0;
1442     while(h->delayed_pic[pics]) pics++;
1443
1444     assert(pics <= MAX_DELAYED_PIC_COUNT);
1445
1446     h->delayed_pic[pics++] = cur;
1447     if (cur->f.reference == 0)
1448         cur->f.reference = DELAYED_PIC_REF;
1449
1450     out = h->delayed_pic[0];
1451     out_idx = 0;
1452     for (i = 1; h->delayed_pic[i] && !h->delayed_pic[i]->f.key_frame && !h->delayed_pic[i]->mmco_reset; i++)
1453         if(h->delayed_pic[i]->poc < out->poc){
1454             out = h->delayed_pic[i];
1455             out_idx = i;
1456         }
1457     if (s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset))
1458         h->next_outputed_poc= INT_MIN;
1459     out_of_order = out->poc < h->next_outputed_poc;
1460
1461     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
1462         { }
1463     else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
1464        || (s->low_delay &&
1465         ((h->next_outputed_poc != INT_MIN && out->poc > h->next_outputed_poc + 2)
1466          || cur->f.pict_type == AV_PICTURE_TYPE_B)))
1467     {
1468         s->low_delay = 0;
1469         s->avctx->has_b_frames++;
1470     }
1471
1472     if(out_of_order || pics > s->avctx->has_b_frames){
1473         out->f.reference &= ~DELAYED_PIC_REF;
1474         out->owner2 = s; // for frame threading, the owner must be the second field's thread
1475                          // or else the first thread can release the picture and reuse it unsafely
1476         for(i=out_idx; h->delayed_pic[i]; i++)
1477             h->delayed_pic[i] = h->delayed_pic[i+1];
1478     }
1479     if(!out_of_order && pics > s->avctx->has_b_frames){
1480         h->next_output_pic = out;
1481         if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset)) {
1482             h->next_outputed_poc = INT_MIN;
1483         } else
1484             h->next_outputed_poc = out->poc;
1485     }else{
1486         av_log(s->avctx, AV_LOG_DEBUG, "no picture\n");
1487     }
1488
1489     if (setup_finished)
1490         ff_thread_finish_setup(s->avctx);
1491 }
1492
1493 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
1494                                               uint8_t *src_cb, uint8_t *src_cr,
1495                                               int linesize, int uvlinesize, int simple)
1496 {
1497     MpegEncContext * const s = &h->s;
1498     uint8_t *top_border;
1499     int top_idx = 1;
1500     const int pixel_shift = h->pixel_shift;
1501     int chroma444 = CHROMA444;
1502     int chroma422 = CHROMA422;
1503
1504     src_y  -=   linesize;
1505     src_cb -= uvlinesize;
1506     src_cr -= uvlinesize;
1507
1508     if(!simple && FRAME_MBAFF){
1509         if(s->mb_y&1){
1510             if(!MB_MBAFF){
1511                 top_border = h->top_borders[0][s->mb_x];
1512                 AV_COPY128(top_border, src_y + 15*linesize);
1513                 if (pixel_shift)
1514                     AV_COPY128(top_border+16, src_y+15*linesize+16);
1515                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1516                     if(chroma444){
1517                         if (pixel_shift){
1518                             AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
1519                             AV_COPY128(top_border+48, src_cb + 15*uvlinesize+16);
1520                             AV_COPY128(top_border+64, src_cr + 15*uvlinesize);
1521                             AV_COPY128(top_border+80, src_cr + 15*uvlinesize+16);
1522                         } else {
1523                             AV_COPY128(top_border+16, src_cb + 15*uvlinesize);
1524                             AV_COPY128(top_border+32, src_cr + 15*uvlinesize);
1525                         }
1526                     } else if(chroma422) {
1527                         if (pixel_shift) {
1528                             AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
1529                             AV_COPY128(top_border+48, src_cr + 15*uvlinesize);
1530                         } else {
1531                             AV_COPY64(top_border+16, src_cb +  15*uvlinesize);
1532                             AV_COPY64(top_border+24, src_cr +  15*uvlinesize);
1533                         }
1534                     } else {
1535                         if (pixel_shift) {
1536                             AV_COPY128(top_border+32, src_cb+7*uvlinesize);
1537                             AV_COPY128(top_border+48, src_cr+7*uvlinesize);
1538                         } else {
1539                             AV_COPY64(top_border+16, src_cb+7*uvlinesize);
1540                             AV_COPY64(top_border+24, src_cr+7*uvlinesize);
1541                         }
1542                     }
1543                 }
1544             }
1545         }else if(MB_MBAFF){
1546             top_idx = 0;
1547         }else
1548             return;
1549     }
1550
1551     top_border = h->top_borders[top_idx][s->mb_x];
1552     // There are two lines saved, the line above the the top macroblock of a pair,
1553     // and the line above the bottom macroblock
1554     AV_COPY128(top_border, src_y + 16*linesize);
1555     if (pixel_shift)
1556         AV_COPY128(top_border+16, src_y+16*linesize+16);
1557
1558     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1559         if(chroma444){
1560             if (pixel_shift){
1561                 AV_COPY128(top_border+32, src_cb + 16*linesize);
1562                 AV_COPY128(top_border+48, src_cb + 16*linesize+16);
1563                 AV_COPY128(top_border+64, src_cr + 16*linesize);
1564                 AV_COPY128(top_border+80, src_cr + 16*linesize+16);
1565             } else {
1566                 AV_COPY128(top_border+16, src_cb + 16*linesize);
1567                 AV_COPY128(top_border+32, src_cr + 16*linesize);
1568             }
1569         } else if(chroma422) {
1570             if (pixel_shift) {
1571                 AV_COPY128(top_border+32, src_cb+16*uvlinesize);
1572                 AV_COPY128(top_border+48, src_cr+16*uvlinesize);
1573             } else {
1574                 AV_COPY64(top_border+16, src_cb+16*uvlinesize);
1575                 AV_COPY64(top_border+24, src_cr+16*uvlinesize);
1576             }
1577         } else {
1578             if (pixel_shift) {
1579                 AV_COPY128(top_border+32, src_cb+8*uvlinesize);
1580                 AV_COPY128(top_border+48, src_cr+8*uvlinesize);
1581             } else {
1582                 AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1583                 AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1584             }
1585         }
1586     }
1587 }
1588
1589 static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
1590                                   uint8_t *src_cb, uint8_t *src_cr,
1591                                   int linesize, int uvlinesize,
1592                                   int xchg, int chroma444,
1593                                   int simple, int pixel_shift){
1594     MpegEncContext * const s = &h->s;
1595     int deblock_topleft;
1596     int deblock_top;
1597     int top_idx = 1;
1598     uint8_t *top_border_m1;
1599     uint8_t *top_border;
1600
1601     if(!simple && FRAME_MBAFF){
1602         if(s->mb_y&1){
1603             if(!MB_MBAFF)
1604                 return;
1605         }else{
1606             top_idx = MB_MBAFF ? 0 : 1;
1607         }
1608     }
1609
1610     if(h->deblocking_filter == 2) {
1611         deblock_topleft = h->slice_table[h->mb_xy - 1 - s->mb_stride] == h->slice_num;
1612         deblock_top     = h->top_type;
1613     } else {
1614         deblock_topleft = (s->mb_x > 0);
1615         deblock_top     = (s->mb_y > !!MB_FIELD);
1616     }
1617
1618     src_y  -=   linesize + 1 + pixel_shift;
1619     src_cb -= uvlinesize + 1 + pixel_shift;
1620     src_cr -= uvlinesize + 1 + pixel_shift;
1621
1622     top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1623     top_border    = h->top_borders[top_idx][s->mb_x];
1624
1625 #define XCHG(a,b,xchg)\
1626     if (pixel_shift) {\
1627         if (xchg) {\
1628             AV_SWAP64(b+0,a+0);\
1629             AV_SWAP64(b+8,a+8);\
1630         } else {\
1631             AV_COPY128(b,a); \
1632         }\
1633     } else \
1634 if (xchg) AV_SWAP64(b,a);\
1635 else      AV_COPY64(b,a);
1636
1637     if(deblock_top){
1638         if(deblock_topleft){
1639             XCHG(top_border_m1 + (8 << pixel_shift), src_y - (7 << pixel_shift), 1);
1640         }
1641         XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
1642         XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
1643         if(s->mb_x+1 < s->mb_width){
1644             XCHG(h->top_borders[top_idx][s->mb_x+1], src_y + (17 << pixel_shift), 1);
1645         }
1646     }
1647     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1648         if(chroma444){
1649             if(deblock_topleft){
1650                 XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1651                 XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1652             }
1653             XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
1654             XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
1655             XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
1656             XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
1657             if(s->mb_x+1 < s->mb_width){
1658                 XCHG(h->top_borders[top_idx][s->mb_x+1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
1659                 XCHG(h->top_borders[top_idx][s->mb_x+1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
1660             }
1661         } else {
1662             if(deblock_top){
1663                 if(deblock_topleft){
1664                     XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1665                     XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1666                 }
1667                 XCHG(top_border + (16 << pixel_shift), src_cb+1+pixel_shift, 1);
1668                 XCHG(top_border + (24 << pixel_shift), src_cr+1+pixel_shift, 1);
1669             }
1670         }
1671     }
1672 }
1673
1674 static av_always_inline int dctcoef_get(DCTELEM *mb, int high_bit_depth, int index) {
1675     if (high_bit_depth) {
1676         return AV_RN32A(((int32_t*)mb) + index);
1677     } else
1678         return AV_RN16A(mb + index);
1679 }
1680
1681 static av_always_inline void dctcoef_set(DCTELEM *mb, int high_bit_depth, int index, int value) {
1682     if (high_bit_depth) {
1683         AV_WN32A(((int32_t*)mb) + index, value);
1684     } else
1685         AV_WN16A(mb + index, value);
1686 }
1687
1688 static av_always_inline void hl_decode_mb_predict_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1689                                                        int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1690 {
1691     MpegEncContext * const s = &h->s;
1692     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1693     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1694     int i;
1695     int qscale = p == 0 ? s->qscale : h->chroma_qp[p-1];
1696     block_offset += 16*p;
1697     if(IS_INTRA4x4(mb_type)){
1698         if(simple || !s->encoding){
1699             if(IS_8x8DCT(mb_type)){
1700                 if(transform_bypass){
1701                     idct_dc_add =
1702                     idct_add    = s->dsp.add_pixels8;
1703                 }else{
1704                     idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1705                     idct_add    = h->h264dsp.h264_idct8_add;
1706                 }
1707                 for(i=0; i<16; i+=4){
1708                     uint8_t * const ptr= dest_y + block_offset[i];
1709                     const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1710                     if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1711                         h->hpc.pred8x8l_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1712                     }else{
1713                         const int nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1714                         h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1715                                                     (h->topright_samples_available<<i)&0x4000, linesize);
1716                         if(nnz){
1717                             if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1718                                 idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1719                             else
1720                                 idct_add   (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1721                         }
1722                     }
1723                 }
1724             }else{
1725                 if(transform_bypass){
1726                     idct_dc_add =
1727                     idct_add    = s->dsp.add_pixels4;
1728                 }else{
1729                     idct_dc_add = h->h264dsp.h264_idct_dc_add;
1730                     idct_add    = h->h264dsp.h264_idct_add;
1731                 }
1732                 for(i=0; i<16; i++){
1733                     uint8_t * const ptr= dest_y + block_offset[i];
1734                     const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1735
1736                     if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1737                         h->hpc.pred4x4_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1738                     }else{
1739                         uint8_t *topright;
1740                         int nnz, tr;
1741                         uint64_t tr_high;
1742                         if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1743                             const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1744                             assert(s->mb_y || linesize <= block_offset[i]);
1745                             if(!topright_avail){
1746                                 if (pixel_shift) {
1747                                     tr_high= ((uint16_t*)ptr)[3 - linesize/2]*0x0001000100010001ULL;
1748                                     topright= (uint8_t*) &tr_high;
1749                                 } else {
1750                                     tr= ptr[3 - linesize]*0x01010101u;
1751                                     topright= (uint8_t*) &tr;
1752                                 }
1753                             }else
1754                                 topright= ptr + (4 << pixel_shift) - linesize;
1755                         }else
1756                             topright= NULL;
1757
1758                         h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1759                         nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1760                         if(nnz){
1761                             if(is_h264){
1762                                 if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1763                                     idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1764                                 else
1765                                     idct_add   (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1766                             }else
1767                                 ff_svq3_add_idct_c(ptr, h->mb + i*16+p*256, linesize, qscale, 0);
1768                         }
1769                     }
1770                 }
1771             }
1772         }
1773     }else{
1774         h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1775         if(is_h264){
1776             if(h->non_zero_count_cache[ scan8[LUMA_DC_BLOCK_INDEX+p] ]){
1777                 if(!transform_bypass)
1778                     h->h264dsp.h264_luma_dc_dequant_idct(h->mb+(p*256 << pixel_shift), h->mb_luma_dc[p], h->dequant4_coeff[p][qscale][0]);
1779                 else{
1780                     static const uint8_t dc_mapping[16] = { 0*16, 1*16, 4*16, 5*16, 2*16, 3*16, 6*16, 7*16,
1781                                                             8*16, 9*16,12*16,13*16,10*16,11*16,14*16,15*16};
1782                     for(i = 0; i < 16; i++)
1783                         dctcoef_set(h->mb+p*256, pixel_shift, dc_mapping[i], dctcoef_get(h->mb_luma_dc[p], pixel_shift, i));
1784                 }
1785             }
1786         }else
1787             ff_svq3_luma_dc_dequant_idct_c(h->mb+p*256, h->mb_luma_dc[p], qscale);
1788     }
1789 }
1790
1791 static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1792                                                     int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1793 {
1794     MpegEncContext * const s = &h->s;
1795     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1796     int i;
1797     block_offset += 16*p;
1798     if(!IS_INTRA4x4(mb_type)){
1799         if(is_h264){
1800             if(IS_INTRA16x16(mb_type)){
1801                 if(transform_bypass){
1802                     if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1803                         h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize);
1804                     }else{
1805                         for(i=0; i<16; i++){
1806                             if(h->non_zero_count_cache[ scan8[i+p*16] ] || dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1807                                 s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1808                         }
1809                     }
1810                 }else{
1811                     h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1812                 }
1813             }else if(h->cbp&15){
1814                 if(transform_bypass){
1815                     const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1816                     idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1817                     for(i=0; i<16; i+=di){
1818                         if(h->non_zero_count_cache[ scan8[i+p*16] ]){
1819                             idct_add(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1820                         }
1821                     }
1822                 }else{
1823                     if(IS_8x8DCT(mb_type)){
1824                         h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1825                     }else{
1826                         h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1827                     }
1828                 }
1829             }
1830         }else{
1831             for(i=0; i<16; i++){
1832                 if(h->non_zero_count_cache[ scan8[i+p*16] ] || h->mb[i*16+p*256]){ //FIXME benchmark weird rule, & below
1833                     uint8_t * const ptr= dest_y + block_offset[i];
1834                     ff_svq3_add_idct_c(ptr, h->mb + i*16 + p*256, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1835                 }
1836             }
1837         }
1838     }
1839 }
1840
1841 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, int pixel_shift)
1842 {
1843     MpegEncContext * const s = &h->s;
1844     const int mb_x= s->mb_x;
1845     const int mb_y= s->mb_y;
1846     const int mb_xy= h->mb_xy;
1847     const int mb_type = s->current_picture.f.mb_type[mb_xy];
1848     uint8_t  *dest_y, *dest_cb, *dest_cr;
1849     int linesize, uvlinesize /*dct_offset*/;
1850     int i, j;
1851     int *block_offset = &h->block_offset[0];
1852     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1853     /* is_h264 should always be true if SVQ3 is disabled. */
1854     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1855     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1856     const int block_h = 16 >> s->chroma_y_shift;
1857     const int chroma422 = CHROMA422;
1858
1859     dest_y  = s->current_picture.f.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize  ) * 16;
1860     dest_cb = s->current_picture.f.data[1] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
1861     dest_cr = s->current_picture.f.data[2] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
1862
1863     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
1864     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + (64 << pixel_shift), dest_cr - dest_cb, 2);
1865
1866     h->list_counts[mb_xy]= h->list_count;
1867
1868     if (!simple && MB_FIELD) {
1869         linesize   = h->mb_linesize   = s->linesize * 2;
1870         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1871         block_offset = &h->block_offset[48];
1872         if(mb_y&1){ //FIXME move out of this function?
1873             dest_y -= s->linesize*15;
1874             dest_cb-= s->uvlinesize * (block_h - 1);
1875             dest_cr-= s->uvlinesize * (block_h - 1);
1876         }
1877         if(FRAME_MBAFF) {
1878             int list;
1879             for(list=0; list<h->list_count; list++){
1880                 if(!USES_LIST(mb_type, list))
1881                     continue;
1882                 if(IS_16X16(mb_type)){
1883                     int8_t *ref = &h->ref_cache[list][scan8[0]];
1884                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1885                 }else{
1886                     for(i=0; i<16; i+=4){
1887                         int ref = h->ref_cache[list][scan8[i]];
1888                         if(ref >= 0)
1889                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1890                     }
1891                 }
1892             }
1893         }
1894     } else {
1895         linesize   = h->mb_linesize   = s->linesize;
1896         uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1897 //        dct_offset = s->linesize * 16;
1898     }
1899
1900     if (!simple && IS_INTRA_PCM(mb_type)) {
1901         if (pixel_shift) {
1902             const int bit_depth = h->sps.bit_depth_luma;
1903             int j;
1904             GetBitContext gb;
1905             init_get_bits(&gb, (uint8_t*)h->mb, 384*bit_depth);
1906
1907             for (i = 0; i < 16; i++) {
1908                 uint16_t *tmp_y  = (uint16_t*)(dest_y  + i*linesize);
1909                 for (j = 0; j < 16; j++)
1910                     tmp_y[j] = get_bits(&gb, bit_depth);
1911             }
1912             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1913                 if (!h->sps.chroma_format_idc) {
1914                     for (i = 0; i < block_h; i++) {
1915                         uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1916                         for (j = 0; j < 8; j++) {
1917                             tmp_cb[j] = 1 << (bit_depth - 1);
1918                         }
1919                     }
1920                     for (i = 0; i < block_h; i++) {
1921                         uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1922                         for (j = 0; j < 8; j++) {
1923                             tmp_cr[j] = 1 << (bit_depth - 1);
1924                         }
1925                     }
1926                 } else {
1927                     for (i = 0; i < block_h; i++) {
1928                         uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1929                         for (j = 0; j < 8; j++)
1930                             tmp_cb[j] = get_bits(&gb, bit_depth);
1931                     }
1932                     for (i = 0; i < block_h; i++) {
1933                         uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1934                         for (j = 0; j < 8; j++)
1935                             tmp_cr[j] = get_bits(&gb, bit_depth);
1936                     }
1937                 }
1938             }
1939         } else {
1940             for (i=0; i<16; i++) {
1941                 memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1942             }
1943             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1944                 if (!h->sps.chroma_format_idc) {
1945                     for (i = 0; i < block_h; i++) {
1946                         memset(dest_cb + i*uvlinesize, 128, 8);
1947                         memset(dest_cr + i*uvlinesize, 128, 8);
1948                     }
1949                 } else {
1950                     for (i = 0; i < block_h; i++) {
1951                         memcpy(dest_cb + i*uvlinesize, h->mb + 128 + i*4,  8);
1952                         memcpy(dest_cr + i*uvlinesize, h->mb + 160 + i*4,  8);
1953                     }
1954                 }
1955             }
1956         }
1957     } else {
1958         if(IS_INTRA(mb_type)){
1959             if(h->deblocking_filter)
1960                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, 0, simple, pixel_shift);
1961
1962             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1963                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1964                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1965             }
1966
1967             hl_decode_mb_predict_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
1968
1969             if(h->deblocking_filter)
1970                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, 0, simple, pixel_shift);
1971         }else if(is_h264){
1972             if (chroma422) {
1973                 hl_motion_422(h, dest_y, dest_cb, dest_cr,
1974                               s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1975                               s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1976                               h->h264dsp.weight_h264_pixels_tab,
1977                               h->h264dsp.biweight_h264_pixels_tab,
1978                               pixel_shift);
1979             } else {
1980                 hl_motion_420(h, dest_y, dest_cb, dest_cr,
1981                               s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1982                               s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1983                               h->h264dsp.weight_h264_pixels_tab,
1984                               h->h264dsp.biweight_h264_pixels_tab,
1985                               pixel_shift);
1986             }
1987         }
1988
1989         hl_decode_mb_idct_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
1990
1991         if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1992             uint8_t *dest[2] = {dest_cb, dest_cr};
1993             if(transform_bypass){
1994                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1995                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + (16*16*1 << pixel_shift), uvlinesize);
1996                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 32, h->mb + (16*16*2 << pixel_shift), uvlinesize);
1997                 }else{
1998                     idct_add = s->dsp.add_pixels4;
1999                     for(j=1; j<3; j++){
2000                         for(i=j*16; i<j*16+4; i++){
2001                             if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
2002                                 idct_add   (dest[j-1] + block_offset[i], h->mb + (i*16 << pixel_shift), uvlinesize);
2003                         }
2004                         if (chroma422) {
2005                             for(i=j*16+4; i<j*16+8; i++){
2006                                 if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
2007                                     idct_add   (dest[j-1] + block_offset[i+4], h->mb + (i*16 << pixel_shift), uvlinesize);
2008                             }
2009                         }
2010                     }
2011                 }
2012             }else{
2013                 if(is_h264){
2014                     int qp[2];
2015                     if (chroma422) {
2016                         qp[0] = h->chroma_qp[0] + 3;
2017                         qp[1] = h->chroma_qp[1] + 3;
2018                     } else {
2019                         qp[0] = h->chroma_qp[0];
2020                         qp[1] = h->chroma_qp[1];
2021                     }
2022                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+0] ])
2023                         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]);
2024                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+1] ])
2025                         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]);
2026                     h->h264dsp.h264_idct_add8(dest, block_offset,
2027                                               h->mb, uvlinesize,
2028                                               h->non_zero_count_cache);
2029                 }else{
2030                     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]);
2031                     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]);
2032                     for(j=1; j<3; j++){
2033                         for(i=j*16; i<j*16+4; i++){
2034                             if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2035                                 uint8_t * const ptr= dest[j-1] + block_offset[i];
2036                                 ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[0][s->qscale + 12] - 12, 2);
2037                             }
2038                         }
2039                     }
2040                 }
2041             }
2042         }
2043     }
2044     if(h->cbp || IS_INTRA(mb_type))
2045     {
2046         s->dsp.clear_blocks(h->mb);
2047         s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
2048     }
2049 }
2050
2051 static av_always_inline void hl_decode_mb_444_internal(H264Context *h, int simple, int pixel_shift){
2052     MpegEncContext * const s = &h->s;
2053     const int mb_x= s->mb_x;
2054     const int mb_y= s->mb_y;
2055     const int mb_xy= h->mb_xy;
2056     const int mb_type = s->current_picture.f.mb_type[mb_xy];
2057     uint8_t  *dest[3];
2058     int linesize;
2059     int i, j, p;
2060     int *block_offset = &h->block_offset[0];
2061     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
2062     const int plane_count = (simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) ? 3 : 1;
2063
2064     for (p = 0; p < plane_count; p++)
2065     {
2066         dest[p] = s->current_picture.f.data[p] + ((mb_x << pixel_shift) + mb_y * s->linesize) * 16;
2067         s->dsp.prefetch(dest[p] + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
2068     }
2069
2070     h->list_counts[mb_xy]= h->list_count;
2071
2072     if (!simple && MB_FIELD) {
2073         linesize   = h->mb_linesize = h->mb_uvlinesize = s->linesize * 2;
2074         block_offset = &h->block_offset[48];
2075         if(mb_y&1) //FIXME move out of this function?
2076             for (p = 0; p < 3; p++)
2077                 dest[p] -= s->linesize*15;
2078         if(FRAME_MBAFF) {
2079             int list;
2080             for(list=0; list<h->list_count; list++){
2081                 if(!USES_LIST(mb_type, list))
2082                     continue;
2083                 if(IS_16X16(mb_type)){
2084                     int8_t *ref = &h->ref_cache[list][scan8[0]];
2085                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
2086                 }else{
2087                     for(i=0; i<16; i+=4){
2088                         int ref = h->ref_cache[list][scan8[i]];
2089                         if(ref >= 0)
2090                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
2091                     }
2092                 }
2093             }
2094         }
2095     } else {
2096         linesize   = h->mb_linesize = h->mb_uvlinesize = s->linesize;
2097     }
2098
2099     if (!simple && IS_INTRA_PCM(mb_type)) {
2100         if (pixel_shift) {
2101             const int bit_depth = h->sps.bit_depth_luma;
2102             GetBitContext gb;
2103             init_get_bits(&gb, (uint8_t*)h->mb, 768*bit_depth);
2104
2105             for (p = 0; p < plane_count; p++) {
2106                 for (i = 0; i < 16; i++) {
2107                     uint16_t *tmp = (uint16_t*)(dest[p] + i*linesize);
2108                     for (j = 0; j < 16; j++)
2109                         tmp[j] = get_bits(&gb, bit_depth);
2110                 }
2111             }
2112         } else {
2113             for (p = 0; p < plane_count; p++) {
2114                 for (i = 0; i < 16; i++) {
2115                     memcpy(dest[p] + i*linesize, h->mb + p*128 + i*8, 16);
2116                 }
2117             }
2118         }
2119     } else {
2120         if(IS_INTRA(mb_type)){
2121             if(h->deblocking_filter)
2122                 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 1, 1, simple, pixel_shift);
2123
2124             for (p = 0; p < plane_count; p++)
2125                 hl_decode_mb_predict_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2126
2127             if(h->deblocking_filter)
2128                 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 0, 1, simple, pixel_shift);
2129         }else{
2130             hl_motion(h, dest[0], dest[1], dest[2],
2131                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2132                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2133                       h->h264dsp.weight_h264_pixels_tab,
2134                       h->h264dsp.biweight_h264_pixels_tab, pixel_shift, 3);
2135         }
2136
2137         for (p = 0; p < plane_count; p++)
2138             hl_decode_mb_idct_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2139     }
2140     if(h->cbp || IS_INTRA(mb_type))
2141     {
2142         s->dsp.clear_blocks(h->mb);
2143         s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
2144     }
2145 }
2146
2147 /**
2148  * Process a macroblock; this case avoids checks for expensive uncommon cases.
2149  */
2150 #define hl_decode_mb_simple(sh, bits) \
2151 static void hl_decode_mb_simple_ ## bits(H264Context *h){ \
2152     hl_decode_mb_internal(h, 1, sh); \
2153 }
2154 hl_decode_mb_simple(0, 8);
2155 hl_decode_mb_simple(1, 16);
2156
2157 /**
2158  * Process a macroblock; this handles edge cases, such as interlacing.
2159  */
2160 static void av_noinline hl_decode_mb_complex(H264Context *h){
2161     hl_decode_mb_internal(h, 0, h->pixel_shift);
2162 }
2163
2164 static void av_noinline hl_decode_mb_444_complex(H264Context *h){
2165     hl_decode_mb_444_internal(h, 0, h->pixel_shift);
2166 }
2167
2168 static void av_noinline hl_decode_mb_444_simple(H264Context *h){
2169     hl_decode_mb_444_internal(h, 1, 0);
2170 }
2171
2172 void ff_h264_hl_decode_mb(H264Context *h){
2173     MpegEncContext * const s = &h->s;
2174     const int mb_xy= h->mb_xy;
2175     const int mb_type = s->current_picture.f.mb_type[mb_xy];
2176     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
2177
2178     if (CHROMA444) {
2179         if(is_complex || h->pixel_shift)
2180             hl_decode_mb_444_complex(h);
2181         else
2182             hl_decode_mb_444_simple(h);
2183     } else if (is_complex) {
2184         hl_decode_mb_complex(h);
2185     } else if (h->pixel_shift) {
2186         hl_decode_mb_simple_16(h);
2187     } else
2188         hl_decode_mb_simple_8(h);
2189 }
2190
2191 static int pred_weight_table(H264Context *h){
2192     MpegEncContext * const s = &h->s;
2193     int list, i;
2194     int luma_def, chroma_def;
2195
2196     h->use_weight= 0;
2197     h->use_weight_chroma= 0;
2198     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2199     if(h->sps.chroma_format_idc)
2200         h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2201     luma_def = 1<<h->luma_log2_weight_denom;
2202     chroma_def = 1<<h->chroma_log2_weight_denom;
2203
2204     for(list=0; list<2; list++){
2205         h->luma_weight_flag[list]   = 0;
2206         h->chroma_weight_flag[list] = 0;
2207         for(i=0; i<h->ref_count[list]; i++){
2208             int luma_weight_flag, chroma_weight_flag;
2209
2210             luma_weight_flag= get_bits1(&s->gb);
2211             if(luma_weight_flag){
2212                 h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
2213                 h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
2214                 if(   h->luma_weight[i][list][0] != luma_def
2215                    || h->luma_weight[i][list][1] != 0) {
2216                     h->use_weight= 1;
2217                     h->luma_weight_flag[list]= 1;
2218                 }
2219             }else{
2220                 h->luma_weight[i][list][0]= luma_def;
2221                 h->luma_weight[i][list][1]= 0;
2222             }
2223
2224             if(h->sps.chroma_format_idc){
2225                 chroma_weight_flag= get_bits1(&s->gb);
2226                 if(chroma_weight_flag){
2227                     int j;
2228                     for(j=0; j<2; j++){
2229                         h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
2230                         h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
2231                         if(   h->chroma_weight[i][list][j][0] != chroma_def
2232                            || h->chroma_weight[i][list][j][1] != 0) {
2233                             h->use_weight_chroma= 1;
2234                             h->chroma_weight_flag[list]= 1;
2235                         }
2236                     }
2237                 }else{
2238                     int j;
2239                     for(j=0; j<2; j++){
2240                         h->chroma_weight[i][list][j][0]= chroma_def;
2241                         h->chroma_weight[i][list][j][1]= 0;
2242                     }
2243                 }
2244             }
2245         }
2246         if(h->slice_type_nos != AV_PICTURE_TYPE_B) break;
2247     }
2248     h->use_weight= h->use_weight || h->use_weight_chroma;
2249     return 0;
2250 }
2251
2252 /**
2253  * Initialize implicit_weight table.
2254  * @param field  0/1 initialize the weight for interlaced MBAFF
2255  *                -1 initializes the rest
2256  */
2257 static void implicit_weight_table(H264Context *h, int field){
2258     MpegEncContext * const s = &h->s;
2259     int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
2260
2261     for (i = 0; i < 2; i++) {
2262         h->luma_weight_flag[i]   = 0;
2263         h->chroma_weight_flag[i] = 0;
2264     }
2265
2266     if(field < 0){
2267         if (s->picture_structure == PICT_FRAME) {
2268             cur_poc = s->current_picture_ptr->poc;
2269         } else {
2270             cur_poc = s->current_picture_ptr->field_poc[s->picture_structure - 1];
2271         }
2272     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
2273        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
2274         h->use_weight= 0;
2275         h->use_weight_chroma= 0;
2276         return;
2277     }
2278         ref_start= 0;
2279         ref_count0= h->ref_count[0];
2280         ref_count1= h->ref_count[1];
2281     }else{
2282         cur_poc = s->current_picture_ptr->field_poc[field];
2283         ref_start= 16;
2284         ref_count0= 16+2*h->ref_count[0];
2285         ref_count1= 16+2*h->ref_count[1];
2286     }
2287
2288     h->use_weight= 2;
2289     h->use_weight_chroma= 2;
2290     h->luma_log2_weight_denom= 5;
2291     h->chroma_log2_weight_denom= 5;
2292
2293     for(ref0=ref_start; ref0 < ref_count0; ref0++){
2294         int poc0 = h->ref_list[0][ref0].poc;
2295         for(ref1=ref_start; ref1 < ref_count1; ref1++){
2296             int w = 32;
2297             if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
2298                 int poc1 = h->ref_list[1][ref1].poc;
2299                 int td = av_clip(poc1 - poc0, -128, 127);
2300                 if(td){
2301                     int tb = av_clip(cur_poc - poc0, -128, 127);
2302                     int tx = (16384 + (FFABS(td) >> 1)) / td;
2303                     int dist_scale_factor = (tb*tx + 32) >> 8;
2304                     if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
2305                         w = 64 - dist_scale_factor;
2306                 }
2307             }
2308             if(field<0){
2309                 h->implicit_weight[ref0][ref1][0]=
2310                 h->implicit_weight[ref0][ref1][1]= w;
2311             }else{
2312                 h->implicit_weight[ref0][ref1][field]=w;
2313             }
2314         }
2315     }
2316 }
2317
2318 /**
2319  * instantaneous decoder refresh.
2320  */
2321 static void idr(H264Context *h){
2322     ff_h264_remove_all_refs(h);
2323     h->prev_frame_num= 0;
2324     h->prev_frame_num_offset= 0;
2325     h->prev_poc_msb=
2326     h->prev_poc_lsb= 0;
2327 }
2328
2329 /* forget old pics after a seek */
2330 static void flush_dpb(AVCodecContext *avctx){
2331     H264Context *h= avctx->priv_data;
2332     int i;
2333     for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
2334         if(h->delayed_pic[i])
2335             h->delayed_pic[i]->f.reference = 0;
2336         h->delayed_pic[i]= NULL;
2337     }
2338     h->outputed_poc=h->next_outputed_poc= INT_MIN;
2339     h->prev_interlaced_frame = 1;
2340     idr(h);
2341     if(h->s.current_picture_ptr)
2342         h->s.current_picture_ptr->f.reference = 0;
2343     h->s.first_field= 0;
2344     ff_h264_reset_sei(h);
2345     ff_mpeg_flush(avctx);
2346 }
2347
2348 static int init_poc(H264Context *h){
2349     MpegEncContext * const s = &h->s;
2350     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2351     int field_poc[2];
2352     Picture *cur = s->current_picture_ptr;
2353
2354     h->frame_num_offset= h->prev_frame_num_offset;
2355     if(h->frame_num < h->prev_frame_num)
2356         h->frame_num_offset += max_frame_num;
2357
2358     if(h->sps.poc_type==0){
2359         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2360
2361         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2362             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2363         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2364             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2365         else
2366             h->poc_msb = h->prev_poc_msb;
2367 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2368         field_poc[0] =
2369         field_poc[1] = h->poc_msb + h->poc_lsb;
2370         if(s->picture_structure == PICT_FRAME)
2371             field_poc[1] += h->delta_poc_bottom;
2372     }else if(h->sps.poc_type==1){
2373         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2374         int i;
2375
2376         if(h->sps.poc_cycle_length != 0)
2377             abs_frame_num = h->frame_num_offset + h->frame_num;
2378         else
2379             abs_frame_num = 0;
2380
2381         if(h->nal_ref_idc==0 && abs_frame_num > 0)
2382             abs_frame_num--;
2383
2384         expected_delta_per_poc_cycle = 0;
2385         for(i=0; i < h->sps.poc_cycle_length; i++)
2386             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2387
2388         if(abs_frame_num > 0){
2389             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2390             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2391
2392             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2393             for(i = 0; i <= frame_num_in_poc_cycle; i++)
2394                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2395         } else
2396             expectedpoc = 0;
2397
2398         if(h->nal_ref_idc == 0)
2399             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2400
2401         field_poc[0] = expectedpoc + h->delta_poc[0];
2402         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2403
2404         if(s->picture_structure == PICT_FRAME)
2405             field_poc[1] += h->delta_poc[1];
2406     }else{
2407         int poc= 2*(h->frame_num_offset + h->frame_num);
2408
2409         if(!h->nal_ref_idc)
2410             poc--;
2411
2412         field_poc[0]= poc;
2413         field_poc[1]= poc;
2414     }
2415
2416     if(s->picture_structure != PICT_BOTTOM_FIELD)
2417         s->current_picture_ptr->field_poc[0]= field_poc[0];
2418     if(s->picture_structure != PICT_TOP_FIELD)
2419         s->current_picture_ptr->field_poc[1]= field_poc[1];
2420     cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
2421
2422     return 0;
2423 }
2424
2425
2426 /**
2427  * initialize scan tables
2428  */
2429 static void init_scan_tables(H264Context *h){
2430     int i;
2431     for(i=0; i<16; i++){
2432 #define T(x) (x>>2) | ((x<<2) & 0xF)
2433         h->zigzag_scan[i] = T(zigzag_scan[i]);
2434         h-> field_scan[i] = T( field_scan[i]);
2435 #undef T
2436     }
2437     for(i=0; i<64; i++){
2438 #define T(x) (x>>3) | ((x&7)<<3)
2439         h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
2440         h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
2441         h->field_scan8x8[i]        = T(field_scan8x8[i]);
2442         h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
2443 #undef T
2444     }
2445     if(h->sps.transform_bypass){ //FIXME same ugly
2446         h->zigzag_scan_q0          = zigzag_scan;
2447         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
2448         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
2449         h->field_scan_q0           = field_scan;
2450         h->field_scan8x8_q0        = field_scan8x8;
2451         h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
2452     }else{
2453         h->zigzag_scan_q0          = h->zigzag_scan;
2454         h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
2455         h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
2456         h->field_scan_q0           = h->field_scan;
2457         h->field_scan8x8_q0        = h->field_scan8x8;
2458         h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
2459     }
2460 }
2461
2462 static int field_end(H264Context *h, int in_setup){
2463     MpegEncContext * const s = &h->s;
2464     AVCodecContext * const avctx= s->avctx;
2465     int err = 0;
2466     s->mb_y= 0;
2467
2468     if (!in_setup && !s->dropable)
2469         ff_thread_report_progress((AVFrame*)s->current_picture_ptr, (16*s->mb_height >> FIELD_PICTURE) - 1,
2470                                  s->picture_structure==PICT_BOTTOM_FIELD);
2471
2472     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2473         ff_vdpau_h264_set_reference_frames(s);
2474
2475     if(in_setup || !(avctx->active_thread_type&FF_THREAD_FRAME)){
2476         if(!s->dropable) {
2477             err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
2478             h->prev_poc_msb= h->poc_msb;
2479             h->prev_poc_lsb= h->poc_lsb;
2480         }
2481         h->prev_frame_num_offset= h->frame_num_offset;
2482         h->prev_frame_num= h->frame_num;
2483         h->outputed_poc = h->next_outputed_poc;
2484     }
2485
2486     if (avctx->hwaccel) {
2487         if (avctx->hwaccel->end_frame(avctx) < 0)
2488             av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
2489     }
2490
2491     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2492         ff_vdpau_h264_picture_complete(s);
2493
2494     /*
2495      * FIXME: Error handling code does not seem to support interlaced
2496      * when slices span multiple rows
2497      * The ff_er_add_slice calls don't work right for bottom
2498      * fields; they cause massive erroneous error concealing
2499      * Error marking covers both fields (top and bottom).
2500      * This causes a mismatched s->error_count
2501      * and a bad error table. Further, the error count goes to
2502      * INT_MAX when called for bottom field, because mb_y is
2503      * past end by one (callers fault) and resync_mb_y != 0
2504      * causes problems for the first MB line, too.
2505      */
2506     if (!FIELD_PICTURE)
2507         ff_er_frame_end(s);
2508
2509     MPV_frame_end(s);
2510
2511     h->current_slice=0;
2512
2513     return err;
2514 }
2515
2516 /**
2517  * Replicate H264 "master" context to thread contexts.
2518  */
2519 static void clone_slice(H264Context *dst, H264Context *src)
2520 {
2521     memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
2522     dst->s.current_picture_ptr  = src->s.current_picture_ptr;
2523     dst->s.current_picture      = src->s.current_picture;
2524     dst->s.linesize             = src->s.linesize;
2525     dst->s.uvlinesize           = src->s.uvlinesize;
2526     dst->s.first_field          = src->s.first_field;
2527
2528     dst->prev_poc_msb           = src->prev_poc_msb;
2529     dst->prev_poc_lsb           = src->prev_poc_lsb;
2530     dst->prev_frame_num_offset  = src->prev_frame_num_offset;
2531     dst->prev_frame_num         = src->prev_frame_num;
2532     dst->short_ref_count        = src->short_ref_count;
2533
2534     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
2535     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
2536     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
2537     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
2538
2539     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
2540     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
2541 }
2542
2543 /**
2544  * computes profile from profile_idc and constraint_set?_flags
2545  *
2546  * @param sps SPS
2547  *
2548  * @return profile as defined by FF_PROFILE_H264_*
2549  */
2550 int ff_h264_get_profile(SPS *sps)
2551 {
2552     int profile = sps->profile_idc;
2553
2554     switch(sps->profile_idc) {
2555     case FF_PROFILE_H264_BASELINE:
2556         // constraint_set1_flag set to 1
2557         profile |= (sps->constraint_set_flags & 1<<1) ? FF_PROFILE_H264_CONSTRAINED : 0;
2558         break;
2559     case FF_PROFILE_H264_HIGH_10:
2560     case FF_PROFILE_H264_HIGH_422:
2561     case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
2562         // constraint_set3_flag set to 1
2563         profile |= (sps->constraint_set_flags & 1<<3) ? FF_PROFILE_H264_INTRA : 0;
2564         break;
2565     }
2566
2567     return profile;
2568 }
2569
2570 /**
2571  * decodes a slice header.
2572  * This will also call MPV_common_init() and frame_start() as needed.
2573  *
2574  * @param h h264context
2575  * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
2576  *
2577  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
2578  */
2579 static int decode_slice_header(H264Context *h, H264Context *h0){
2580     MpegEncContext * const s = &h->s;
2581     MpegEncContext * const s0 = &h0->s;
2582     unsigned int first_mb_in_slice;
2583     unsigned int pps_id;
2584     int num_ref_idx_active_override_flag;
2585     unsigned int slice_type, tmp, i, j;
2586     int default_ref_list_done = 0;
2587     int last_pic_structure;
2588
2589     s->dropable= h->nal_ref_idc == 0;
2590
2591     /* FIXME: 2tap qpel isn't implemented for high bit depth. */
2592     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc && !h->pixel_shift){
2593         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
2594         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
2595     }else{
2596         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
2597         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
2598     }
2599
2600     first_mb_in_slice= get_ue_golomb(&s->gb);
2601
2602     if(first_mb_in_slice == 0){ //FIXME better field boundary detection
2603         if(h0->current_slice && FIELD_PICTURE){
2604             field_end(h, 1);
2605         }
2606
2607         h0->current_slice = 0;
2608         if (!s0->first_field)
2609             s->current_picture_ptr= NULL;
2610     }
2611
2612     slice_type= get_ue_golomb_31(&s->gb);
2613     if(slice_type > 9){
2614         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);
2615         return -1;
2616     }
2617     if(slice_type > 4){
2618         slice_type -= 5;
2619         h->slice_type_fixed=1;
2620     }else
2621         h->slice_type_fixed=0;
2622
2623     slice_type= golomb_to_pict_type[ slice_type ];
2624     if (slice_type == AV_PICTURE_TYPE_I
2625         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
2626         default_ref_list_done = 1;
2627     }
2628     h->slice_type= slice_type;
2629     h->slice_type_nos= slice_type & 3;
2630
2631     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
2632
2633     pps_id= get_ue_golomb(&s->gb);
2634     if(pps_id>=MAX_PPS_COUNT){
2635         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
2636         return -1;
2637     }
2638     if(!h0->pps_buffers[pps_id]) {
2639         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
2640         return -1;
2641     }
2642     h->pps= *h0->pps_buffers[pps_id];
2643
2644     if(!h0->sps_buffers[h->pps.sps_id]) {
2645         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
2646         return -1;
2647     }
2648     h->sps = *h0->sps_buffers[h->pps.sps_id];
2649
2650     s->avctx->profile = ff_h264_get_profile(&h->sps);
2651     s->avctx->level   = h->sps.level_idc;
2652     s->avctx->refs    = h->sps.ref_frame_count;
2653
2654     if(h == h0 && h->dequant_coeff_pps != pps_id){
2655         h->dequant_coeff_pps = pps_id;
2656         init_dequant_tables(h);
2657     }
2658
2659     s->mb_width= h->sps.mb_width;
2660     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
2661
2662     h->b_stride=  s->mb_width*4;
2663
2664     s->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
2665
2666     s->width = 16*s->mb_width - (2>>CHROMA444)*FFMIN(h->sps.crop_right, (8<<CHROMA444)-1);
2667     if(h->sps.frame_mbs_only_flag)
2668         s->height= 16*s->mb_height - (1<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1);
2669     else
2670         s->height= 16*s->mb_height - (2<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1);
2671
2672     if (s->context_initialized
2673         && (   s->width != s->avctx->width || s->height != s->avctx->height
2674             || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
2675         if(h != h0) {
2676             av_log_missing_feature(s->avctx, "Width/height changing with threads is", 0);
2677             return -1;   // width / height changed during parallelized decoding
2678         }
2679         free_tables(h, 0);
2680         flush_dpb(s->avctx);
2681         MPV_common_end(s);
2682     }
2683     if (!s->context_initialized) {
2684         if (h != h0) {
2685             av_log(h->s.avctx, AV_LOG_ERROR, "Cannot (re-)initialize context during parallel decoding.\n");
2686             return -1;
2687         }
2688
2689         avcodec_set_dimensions(s->avctx, s->width, s->height);
2690         s->avctx->sample_aspect_ratio= h->sps.sar;
2691         av_assert0(s->avctx->sample_aspect_ratio.den);
2692
2693         if(h->sps.video_signal_type_present_flag){
2694             s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
2695             if(h->sps.colour_description_present_flag){
2696                 s->avctx->color_primaries = h->sps.color_primaries;
2697                 s->avctx->color_trc       = h->sps.color_trc;
2698                 s->avctx->colorspace      = h->sps.colorspace;
2699             }
2700         }
2701
2702         if(h->sps.timing_info_present_flag){
2703             int64_t den= h->sps.time_scale;
2704             if(h->x264_build < 44U)
2705                 den *= 2;
2706             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
2707                       h->sps.num_units_in_tick, den, 1<<30);
2708         }
2709
2710         switch (h->sps.bit_depth_luma) {
2711             case 9 :
2712                 if (CHROMA444)
2713                     s->avctx->pix_fmt = PIX_FMT_YUV444P9;
2714                 else if (CHROMA422)
2715                     s->avctx->pix_fmt = PIX_FMT_YUV422P9;
2716                 else
2717                     s->avctx->pix_fmt = PIX_FMT_YUV420P9;
2718                 break;
2719             case 10 :
2720                 if (CHROMA444)
2721                     s->avctx->pix_fmt = PIX_FMT_YUV444P10;
2722                 else if (CHROMA422)
2723                     s->avctx->pix_fmt = PIX_FMT_YUV422P10;
2724                 else
2725                     s->avctx->pix_fmt = PIX_FMT_YUV420P10;
2726                 break;
2727             default:
2728                 if (CHROMA444){
2729                     s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ444P : PIX_FMT_YUV444P;
2730                 } else if (CHROMA422) {
2731                     s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ422P : PIX_FMT_YUV422P;
2732                 }else{
2733                     s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
2734                                                              s->avctx->codec->pix_fmts ?
2735                                                              s->avctx->codec->pix_fmts :
2736                                                              s->avctx->color_range == AVCOL_RANGE_JPEG ?
2737                                                              hwaccel_pixfmt_list_h264_jpeg_420 :
2738                                                              ff_hwaccel_pixfmt_list_420);
2739                 }
2740         }
2741
2742         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
2743
2744         if (MPV_common_init(s) < 0) {
2745             av_log(h->s.avctx, AV_LOG_ERROR, "MPV_common_init() failed.\n");
2746             return -1;
2747         }
2748         s->first_field = 0;
2749         h->prev_interlaced_frame = 1;
2750
2751         init_scan_tables(h);
2752         if (ff_h264_alloc_tables(h) < 0) {
2753             av_log(h->s.avctx, AV_LOG_ERROR, "Could not allocate memory for h264\n");
2754             return AVERROR(ENOMEM);
2755         }
2756
2757         if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_SLICE)) {
2758             if (context_init(h) < 0) {
2759                 av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2760                 return -1;
2761             }
2762         } else {
2763             for(i = 1; i < s->avctx->thread_count; i++) {
2764                 H264Context *c;
2765                 c = h->thread_context[i] = av_malloc(sizeof(H264Context));
2766                 memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
2767                 memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
2768                 c->h264dsp = h->h264dsp;
2769                 c->sps = h->sps;
2770                 c->pps = h->pps;
2771                 c->pixel_shift = h->pixel_shift;
2772                 init_scan_tables(c);
2773                 clone_tables(c, h, i);
2774             }
2775
2776             for(i = 0; i < s->avctx->thread_count; i++)
2777                 if (context_init(h->thread_context[i]) < 0) {
2778                     av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2779                     return -1;
2780                 }
2781         }
2782     }
2783
2784     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
2785
2786     h->mb_mbaff = 0;
2787     h->mb_aff_frame = 0;
2788     last_pic_structure = s0->picture_structure;
2789     if(h->sps.frame_mbs_only_flag){
2790         s->picture_structure= PICT_FRAME;
2791     }else{
2792         if(get_bits1(&s->gb)) { //field_pic_flag
2793             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
2794         } else {
2795             s->picture_structure= PICT_FRAME;
2796             h->mb_aff_frame = h->sps.mb_aff;
2797         }
2798     }
2799     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
2800
2801     if(h0->current_slice == 0){
2802         // Shorten frame num gaps so we don't have to allocate reference frames just to throw them away
2803         if(h->frame_num != h->prev_frame_num) {
2804             int unwrap_prev_frame_num = h->prev_frame_num, max_frame_num = 1<<h->sps.log2_max_frame_num;
2805
2806             if (unwrap_prev_frame_num > h->frame_num) unwrap_prev_frame_num -= max_frame_num;
2807
2808             if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
2809                 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
2810                 if (unwrap_prev_frame_num < 0)
2811                     unwrap_prev_frame_num += max_frame_num;
2812
2813                 h->prev_frame_num = unwrap_prev_frame_num;
2814             }
2815         }
2816
2817         while(h->frame_num !=  h->prev_frame_num &&
2818               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
2819             Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
2820             av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
2821             if (ff_h264_frame_start(h) < 0)
2822                 return -1;
2823             h->prev_frame_num++;
2824             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
2825             s->current_picture_ptr->frame_num= h->prev_frame_num;
2826             ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 0);
2827             ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 1);
2828             ff_generate_sliding_window_mmcos(h);
2829             if (ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index) < 0 &&
2830                 s->avctx->error_recognition >= FF_ER_EXPLODE)
2831                 return AVERROR_INVALIDDATA;
2832             /* Error concealment: if a ref is missing, copy the previous ref in its place.
2833              * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
2834              * about there being no actual duplicates.
2835              * FIXME: this doesn't copy padding for out-of-frame motion vectors.  Given we're
2836              * concealing a lost frame, this probably isn't noticable by comparison, but it should
2837              * be fixed. */
2838             if (h->short_ref_count) {
2839                 if (prev) {
2840                     av_image_copy(h->short_ref[0]->f.data, h->short_ref[0]->f.linesize,
2841                                   (const uint8_t**)prev->f.data, prev->f.linesize,
2842                                   s->avctx->pix_fmt, s->mb_width*16, s->mb_height*16);
2843                     h->short_ref[0]->poc = prev->poc+2;
2844                 }
2845                 h->short_ref[0]->frame_num = h->prev_frame_num;
2846             }
2847         }
2848
2849         /* See if we have a decoded first field looking for a pair... */
2850         if (s0->first_field) {
2851             assert(s0->current_picture_ptr);
2852             assert(s0->current_picture_ptr->f.data[0]);
2853             assert(s0->current_picture_ptr->f.reference != DELAYED_PIC_REF);
2854
2855             /* figure out if we have a complementary field pair */
2856             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2857                 /*
2858                  * Previous field is unmatched. Don't display it, but let it
2859                  * remain for reference if marked as such.
2860                  */
2861                 s0->current_picture_ptr = NULL;
2862                 s0->first_field = FIELD_PICTURE;
2863
2864             } else {
2865                 if (h->nal_ref_idc &&
2866                         s0->current_picture_ptr->f.reference &&
2867                         s0->current_picture_ptr->frame_num != h->frame_num) {
2868                     /*
2869                      * This and previous field were reference, but had
2870                      * different frame_nums. Consider this field first in
2871                      * pair. Throw away previous field except for reference
2872                      * purposes.
2873                      */
2874                     s0->first_field = 1;
2875                     s0->current_picture_ptr = NULL;
2876
2877                 } else {
2878                     /* Second field in complementary pair */
2879                     s0->first_field = 0;
2880                 }
2881             }
2882
2883         } else {
2884             /* Frame or first field in a potentially complementary pair */
2885             assert(!s0->current_picture_ptr);
2886             s0->first_field = FIELD_PICTURE;
2887         }
2888
2889         if(!FIELD_PICTURE || s0->first_field) {
2890             if (ff_h264_frame_start(h) < 0) {
2891                 s0->first_field = 0;
2892                 return -1;
2893             }
2894         } else {
2895             ff_release_unused_pictures(s, 0);
2896         }
2897     }
2898     if(h != h0)
2899         clone_slice(h, h0);
2900
2901     s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
2902
2903     assert(s->mb_num == s->mb_width * s->mb_height);
2904     if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
2905        first_mb_in_slice                    >= s->mb_num){
2906         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
2907         return -1;
2908     }
2909     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
2910     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
2911     if (s->picture_structure == PICT_BOTTOM_FIELD)
2912         s->resync_mb_y = s->mb_y = s->mb_y + 1;
2913     assert(s->mb_y < s->mb_height);
2914
2915     if(s->picture_structure==PICT_FRAME){
2916         h->curr_pic_num=   h->frame_num;
2917         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
2918     }else{
2919         h->curr_pic_num= 2*h->frame_num + 1;
2920         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
2921     }
2922
2923     if(h->nal_unit_type == NAL_IDR_SLICE){
2924         get_ue_golomb(&s->gb); /* idr_pic_id */
2925     }
2926
2927     if(h->sps.poc_type==0){
2928         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2929
2930         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
2931             h->delta_poc_bottom= get_se_golomb(&s->gb);
2932         }
2933     }
2934
2935     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
2936         h->delta_poc[0]= get_se_golomb(&s->gb);
2937
2938         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2939             h->delta_poc[1]= get_se_golomb(&s->gb);
2940     }
2941
2942     init_poc(h);
2943
2944     if(h->pps.redundant_pic_cnt_present){
2945         h->redundant_pic_count= get_ue_golomb(&s->gb);
2946     }
2947
2948     //set defaults, might be overridden a few lines later
2949     h->ref_count[0]= h->pps.ref_count[0];
2950     h->ref_count[1]= h->pps.ref_count[1];
2951
2952     if(h->slice_type_nos != AV_PICTURE_TYPE_I){
2953         if(h->slice_type_nos == AV_PICTURE_TYPE_B){
2954             h->direct_spatial_mv_pred= get_bits1(&s->gb);
2955         }
2956         num_ref_idx_active_override_flag= get_bits1(&s->gb);
2957
2958         if(num_ref_idx_active_override_flag){
2959             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2960             if(h->slice_type_nos==AV_PICTURE_TYPE_B)
2961                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2962
2963             if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
2964                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
2965                 h->ref_count[0]= h->ref_count[1]= 1;
2966                 return -1;
2967             }
2968         }
2969         if(h->slice_type_nos == AV_PICTURE_TYPE_B)
2970             h->list_count= 2;
2971         else
2972             h->list_count= 1;
2973     }else
2974         h->list_count= 0;
2975
2976     if(!default_ref_list_done){
2977         ff_h264_fill_default_ref_list(h);
2978     }
2979
2980     if(h->slice_type_nos!=AV_PICTURE_TYPE_I && ff_h264_decode_ref_pic_list_reordering(h) < 0) {
2981         h->ref_count[1]= h->ref_count[0]= 0;
2982         return -1;
2983     }
2984
2985     if(h->slice_type_nos!=AV_PICTURE_TYPE_I){
2986         s->last_picture_ptr= &h->ref_list[0][0];
2987         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
2988     }
2989     if(h->slice_type_nos==AV_PICTURE_TYPE_B){
2990         s->next_picture_ptr= &h->ref_list[1][0];
2991         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
2992     }
2993
2994     if(   (h->pps.weighted_pred          && h->slice_type_nos == AV_PICTURE_TYPE_P )
2995        ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== AV_PICTURE_TYPE_B ) )
2996         pred_weight_table(h);
2997     else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
2998         implicit_weight_table(h, -1);
2999     }else {
3000         h->use_weight = 0;
3001         for (i = 0; i < 2; i++) {
3002             h->luma_weight_flag[i]   = 0;
3003             h->chroma_weight_flag[i] = 0;
3004         }
3005     }
3006
3007     if(h->nal_ref_idc && ff_h264_decode_ref_pic_marking(h0, &s->gb) < 0 &&
3008        s->avctx->error_recognition >= FF_ER_EXPLODE)
3009         return AVERROR_INVALIDDATA;
3010
3011     if(FRAME_MBAFF){
3012         ff_h264_fill_mbaff_ref_list(h);
3013
3014         if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
3015             implicit_weight_table(h, 0);
3016             implicit_weight_table(h, 1);
3017         }
3018     }
3019
3020     if(h->slice_type_nos==AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
3021         ff_h264_direct_dist_scale_factor(h);
3022     ff_h264_direct_ref_list_init(h);
3023
3024     if( h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac ){
3025         tmp = get_ue_golomb_31(&s->gb);
3026         if(tmp > 2){
3027             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3028             return -1;
3029         }
3030         h->cabac_init_idc= tmp;
3031     }
3032
3033     h->last_qscale_diff = 0;
3034     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
3035     if(tmp>51+6*(h->sps.bit_depth_luma-8)){
3036         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3037         return -1;
3038     }
3039     s->qscale= tmp;
3040     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3041     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3042     //FIXME qscale / qp ... stuff
3043     if(h->slice_type == AV_PICTURE_TYPE_SP){
3044         get_bits1(&s->gb); /* sp_for_switch_flag */
3045     }
3046     if(h->slice_type==AV_PICTURE_TYPE_SP || h->slice_type == AV_PICTURE_TYPE_SI){
3047         get_se_golomb(&s->gb); /* slice_qs_delta */
3048     }
3049
3050     h->deblocking_filter = 1;
3051     h->slice_alpha_c0_offset = 52;
3052     h->slice_beta_offset = 52;
3053     if( h->pps.deblocking_filter_parameters_present ) {
3054         tmp= get_ue_golomb_31(&s->gb);
3055         if(tmp > 2){
3056             av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
3057             return -1;
3058         }
3059         h->deblocking_filter= tmp;
3060         if(h->deblocking_filter < 2)
3061             h->deblocking_filter^= 1; // 1<->0
3062
3063         if( h->deblocking_filter ) {
3064             h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
3065             h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
3066             if(   h->slice_alpha_c0_offset > 104U
3067                || h->slice_beta_offset     > 104U){
3068                 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);
3069                 return -1;
3070             }
3071         }
3072     }
3073
3074     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
3075        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != AV_PICTURE_TYPE_I)
3076        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == AV_PICTURE_TYPE_B)
3077        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
3078         h->deblocking_filter= 0;
3079
3080     if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
3081         if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
3082             /* Cheat slightly for speed:
3083                Do not bother to deblock across slices. */
3084             h->deblocking_filter = 2;
3085         } else {
3086             h0->max_contexts = 1;
3087             if(!h0->single_decode_warning) {
3088                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
3089                 h0->single_decode_warning = 1;
3090             }
3091             if (h != h0) {
3092                 av_log(h->s.avctx, AV_LOG_ERROR, "Deblocking switched inside frame.\n");
3093                 return 1;
3094             }
3095         }
3096     }
3097     h->qp_thresh = 15 + 52 - FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset)
3098                  - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1])
3099                  + 6 * (h->sps.bit_depth_luma - 8);
3100
3101 #if 0 //FMO
3102     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3103         slice_group_change_cycle= get_bits(&s->gb, ?);
3104 #endif
3105
3106     h0->last_slice_type = slice_type;
3107     h->slice_num = ++h0->current_slice;
3108     if(h->slice_num >= MAX_SLICES){
3109         av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
3110     }
3111
3112     for(j=0; j<2; j++){
3113         int id_list[16];
3114         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
3115         for(i=0; i<16; i++){
3116             id_list[i]= 60;
3117             if (h->ref_list[j][i].f.data[0]) {
3118                 int k;
3119                 uint8_t *base = h->ref_list[j][i].f.base[0];
3120                 for(k=0; k<h->short_ref_count; k++)
3121                     if (h->short_ref[k]->f.base[0] == base) {
3122                         id_list[i]= k;
3123                         break;
3124                     }
3125                 for(k=0; k<h->long_ref_count; k++)
3126                     if (h->long_ref[k] && h->long_ref[k]->f.base[0] == base) {
3127                         id_list[i]= h->short_ref_count + k;
3128                         break;
3129                     }
3130             }
3131         }
3132
3133         ref2frm[0]=
3134         ref2frm[1]= -1;
3135         for(i=0; i<16; i++)
3136             ref2frm[i+2]= 4*id_list[i]
3137                           + (h->ref_list[j][i].f.reference & 3);
3138         ref2frm[18+0]=
3139         ref2frm[18+1]= -1;
3140         for(i=16; i<48; i++)
3141             ref2frm[i+4]= 4*id_list[(i-16)>>1]
3142                           + (h->ref_list[j][i].f.reference & 3);
3143     }
3144
3145     //FIXME: fix draw_edges+PAFF+frame threads
3146     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE || (!h->sps.frame_mbs_only_flag && s->avctx->active_thread_type)) ? 0 : 16;
3147     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
3148
3149     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3150         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",
3151                h->slice_num,
3152                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
3153                first_mb_in_slice,
3154                av_get_picture_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
3155                pps_id, h->frame_num,
3156                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3157                h->ref_count[0], h->ref_count[1],
3158                s->qscale,
3159                h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
3160                h->use_weight,
3161                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
3162                h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
3163                );
3164     }
3165
3166     return 0;
3167 }
3168
3169 int ff_h264_get_slice_type(const H264Context *h)
3170 {
3171     switch (h->slice_type) {
3172     case AV_PICTURE_TYPE_P:  return 0;
3173     case AV_PICTURE_TYPE_B:  return 1;
3174     case AV_PICTURE_TYPE_I:  return 2;
3175     case AV_PICTURE_TYPE_SP: return 3;
3176     case AV_PICTURE_TYPE_SI: return 4;
3177     default:         return -1;
3178     }
3179 }
3180
3181 static av_always_inline void fill_filter_caches_inter(H264Context *h, MpegEncContext * const s, int mb_type, int top_xy,
3182                                                       int left_xy[LEFT_MBS], int top_type, int left_type[LEFT_MBS], int mb_xy, int list)
3183 {
3184     int b_stride = h->b_stride;
3185     int16_t (*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
3186     int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
3187     if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
3188         if(USES_LIST(top_type, list)){
3189             const int b_xy= h->mb2b_xy[top_xy] + 3*b_stride;
3190             const int b8_xy= 4*top_xy + 2;
3191             int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3192             AV_COPY128(mv_dst - 1*8, s->current_picture.f.motion_val[list][b_xy + 0]);
3193             ref_cache[0 - 1*8]=
3194             ref_cache[1 - 1*8]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 0]];
3195             ref_cache[2 - 1*8]=
3196             ref_cache[3 - 1*8]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 1]];
3197         }else{
3198             AV_ZERO128(mv_dst - 1*8);
3199             AV_WN32A(&ref_cache[0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3200         }
3201
3202         if(!IS_INTERLACED(mb_type^left_type[LTOP])){
3203             if(USES_LIST(left_type[LTOP], list)){
3204                 const int b_xy= h->mb2b_xy[left_xy[LTOP]] + 3;
3205                 const int b8_xy= 4*left_xy[LTOP] + 1;
3206                 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[LTOP]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3207                 AV_COPY32(mv_dst - 1 +  0, s->current_picture.f.motion_val[list][b_xy + b_stride*0]);
3208                 AV_COPY32(mv_dst - 1 +  8, s->current_picture.f.motion_val[list][b_xy + b_stride*1]);
3209                 AV_COPY32(mv_dst - 1 + 16, s->current_picture.f.motion_val[list][b_xy + b_stride*2]);
3210                 AV_COPY32(mv_dst - 1 + 24, s->current_picture.f.motion_val[list][b_xy + b_stride*3]);
3211                 ref_cache[-1 +  0]=
3212                 ref_cache[-1 +  8]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2*0]];
3213                 ref_cache[-1 + 16]=
3214                 ref_cache[-1 + 24]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2*1]];
3215             }else{
3216                 AV_ZERO32(mv_dst - 1 + 0);
3217                 AV_ZERO32(mv_dst - 1 + 8);
3218                 AV_ZERO32(mv_dst - 1 +16);
3219                 AV_ZERO32(mv_dst - 1 +24);
3220                 ref_cache[-1 +  0]=
3221                 ref_cache[-1 +  8]=
3222                 ref_cache[-1 + 16]=
3223                 ref_cache[-1 + 24]= LIST_NOT_USED;
3224             }
3225         }
3226     }
3227
3228     if(!USES_LIST(mb_type, list)){
3229         fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0,0), 4);
3230         AV_WN32A(&ref_cache[0*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3231         AV_WN32A(&ref_cache[1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3232         AV_WN32A(&ref_cache[2*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3233         AV_WN32A(&ref_cache[3*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3234         return;
3235     }
3236
3237     {
3238         int8_t *ref = &s->current_picture.f.ref_index[list][4*mb_xy];
3239         int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3240         uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101;
3241         uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]],ref2frm[list][ref[3]])&0x00FF00FF)*0x0101;
3242         AV_WN32A(&ref_cache[0*8], ref01);
3243         AV_WN32A(&ref_cache[1*8], ref01);
3244         AV_WN32A(&ref_cache[2*8], ref23);
3245         AV_WN32A(&ref_cache[3*8], ref23);
3246     }
3247
3248     {
3249         int16_t (*mv_src)[2] = &s->current_picture.f.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
3250         AV_COPY128(mv_dst + 8*0, mv_src + 0*b_stride);
3251         AV_COPY128(mv_dst + 8*1, mv_src + 1*b_stride);
3252         AV_COPY128(mv_dst + 8*2, mv_src + 2*b_stride);
3253         AV_COPY128(mv_dst + 8*3, mv_src + 3*b_stride);
3254     }
3255 }
3256
3257 /**
3258  *
3259  * @return non zero if the loop filter can be skiped
3260  */
3261 static int fill_filter_caches(H264Context *h, int mb_type){
3262     MpegEncContext * const s = &h->s;
3263     const int mb_xy= h->mb_xy;
3264     int top_xy, left_xy[LEFT_MBS];
3265     int top_type, left_type[LEFT_MBS];
3266     uint8_t *nnz;
3267     uint8_t *nnz_cache;
3268
3269     top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
3270
3271     /* Wow, what a mess, why didn't they simplify the interlacing & intra
3272      * stuff, I can't imagine that these complex rules are worth it. */
3273
3274     left_xy[LBOT] = left_xy[LTOP] = mb_xy-1;
3275     if(FRAME_MBAFF){
3276         const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.f.mb_type[mb_xy - 1]);
3277         const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
3278         if(s->mb_y&1){
3279             if (left_mb_field_flag != curr_mb_field_flag) {
3280                 left_xy[LTOP] -= s->mb_stride;
3281             }
3282         }else{
3283             if(curr_mb_field_flag){
3284                 top_xy += s->mb_stride & (((s->current_picture.f.mb_type[top_xy] >> 7) & 1) - 1);
3285             }
3286             if (left_mb_field_flag != curr_mb_field_flag) {
3287                 left_xy[LBOT] += s->mb_stride;
3288             }
3289         }
3290     }
3291
3292     h->top_mb_xy = top_xy;
3293     h->left_mb_xy[LTOP] = left_xy[LTOP];
3294     h->left_mb_xy[LBOT] = left_xy[LBOT];
3295     {
3296         //for sufficiently low qp, filtering wouldn't do anything
3297         //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
3298         int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
3299         int qp = s->current_picture.f.qscale_table[mb_xy];
3300         if(qp <= qp_thresh
3301            && (left_xy[LTOP] < 0 || ((qp + s->current_picture.f.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh)
3302            && (top_xy        < 0 || ((qp + s->current_picture.f.qscale_table[top_xy       ] + 1) >> 1) <= qp_thresh)) {
3303             if(!FRAME_MBAFF)
3304                 return 1;
3305             if ((left_xy[LTOP] < 0            || ((qp + s->current_picture.f.qscale_table[left_xy[LBOT]        ] + 1) >> 1) <= qp_thresh) &&
3306                 (top_xy        < s->mb_stride || ((qp + s->current_picture.f.qscale_table[top_xy - s->mb_stride] + 1) >> 1) <= qp_thresh))
3307                 return 1;
3308         }
3309     }
3310
3311     top_type        = s->current_picture.f.mb_type[top_xy];
3312     left_type[LTOP] = s->current_picture.f.mb_type[left_xy[LTOP]];
3313     left_type[LBOT] = s->current_picture.f.mb_type[left_xy[LBOT]];
3314     if(h->deblocking_filter == 2){
3315         if(h->slice_table[top_xy       ] != h->slice_num) top_type= 0;
3316         if(h->slice_table[left_xy[LBOT]] != h->slice_num) left_type[LTOP]= left_type[LBOT]= 0;
3317     }else{
3318         if(h->slice_table[top_xy       ] == 0xFFFF) top_type= 0;
3319         if(h->slice_table[left_xy[LBOT]] == 0xFFFF) left_type[LTOP]= left_type[LBOT] =0;
3320     }
3321     h->top_type       = top_type;
3322     h->left_type[LTOP]= left_type[LTOP];
3323     h->left_type[LBOT]= left_type[LBOT];
3324
3325     if(IS_INTRA(mb_type))
3326         return 0;
3327
3328     fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy, top_type, left_type, mb_xy, 0);
3329     if(h->list_count == 2)
3330         fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy, top_type, left_type, mb_xy, 1);
3331
3332     nnz = h->non_zero_count[mb_xy];
3333     nnz_cache = h->non_zero_count_cache;
3334     AV_COPY32(&nnz_cache[4+8*1], &nnz[ 0]);
3335     AV_COPY32(&nnz_cache[4+8*2], &nnz[ 4]);
3336     AV_COPY32(&nnz_cache[4+8*3], &nnz[ 8]);
3337     AV_COPY32(&nnz_cache[4+8*4], &nnz[12]);
3338     h->cbp= h->cbp_table[mb_xy];
3339
3340     if(top_type){
3341         nnz = h->non_zero_count[top_xy];
3342         AV_COPY32(&nnz_cache[4+8*0], &nnz[3*4]);
3343     }
3344
3345     if(left_type[LTOP]){
3346         nnz = h->non_zero_count[left_xy[LTOP]];
3347         nnz_cache[3+8*1]= nnz[3+0*4];
3348         nnz_cache[3+8*2]= nnz[3+1*4];
3349         nnz_cache[3+8*3]= nnz[3+2*4];
3350         nnz_cache[3+8*4]= nnz[3+3*4];
3351     }
3352
3353     // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
3354     if(!CABAC && h->pps.transform_8x8_mode){
3355         if(IS_8x8DCT(top_type)){
3356             nnz_cache[4+8*0]=
3357             nnz_cache[5+8*0]= (h->cbp_table[top_xy] & 0x4000) >> 12;
3358             nnz_cache[6+8*0]=
3359             nnz_cache[7+8*0]= (h->cbp_table[top_xy] & 0x8000) >> 12;
3360         }
3361         if(IS_8x8DCT(left_type[LTOP])){
3362             nnz_cache[3+8*1]=
3363             nnz_cache[3+8*2]= (h->cbp_table[left_xy[LTOP]]&0x2000) >> 12; //FIXME check MBAFF
3364         }
3365         if(IS_8x8DCT(left_type[LBOT])){
3366             nnz_cache[3+8*3]=
3367             nnz_cache[3+8*4]= (h->cbp_table[left_xy[LBOT]]&0x8000) >> 12; //FIXME check MBAFF
3368         }
3369
3370         if(IS_8x8DCT(mb_type)){
3371             nnz_cache[scan8[0   ]]= nnz_cache[scan8[1   ]]=
3372             nnz_cache[scan8[2   ]]= nnz_cache[scan8[3   ]]= (h->cbp & 0x1000) >> 12;
3373
3374             nnz_cache[scan8[0+ 4]]= nnz_cache[scan8[1+ 4]]=
3375             nnz_cache[scan8[2+ 4]]= nnz_cache[scan8[3+ 4]]= (h->cbp & 0x2000) >> 12;
3376
3377             nnz_cache[scan8[0+ 8]]= nnz_cache[scan8[1+ 8]]=
3378             nnz_cache[scan8[2+ 8]]= nnz_cache[scan8[3+ 8]]= (h->cbp & 0x4000) >> 12;
3379
3380             nnz_cache[scan8[0+12]]= nnz_cache[scan8[1+12]]=
3381             nnz_cache[scan8[2+12]]= nnz_cache[scan8[3+12]]= (h->cbp & 0x8000) >> 12;
3382         }
3383     }
3384
3385     return 0;
3386 }
3387
3388 static void loop_filter(H264Context *h, int start_x, int end_x){
3389     MpegEncContext * const s = &h->s;
3390     uint8_t  *dest_y, *dest_cb, *dest_cr;
3391     int linesize, uvlinesize, mb_x, mb_y;
3392     const int end_mb_y= s->mb_y + FRAME_MBAFF;
3393     const int old_slice_type= h->slice_type;
3394     const int pixel_shift = h->pixel_shift;
3395     const int block_h = 16 >> s->chroma_y_shift;
3396
3397     if(h->deblocking_filter) {
3398         for(mb_x= start_x; mb_x<end_x; mb_x++){
3399             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
3400                 int mb_xy, mb_type;
3401                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
3402                 h->slice_num= h->slice_table[mb_xy];
3403                 mb_type = s->current_picture.f.mb_type[mb_xy];
3404                 h->list_count= h->list_counts[mb_xy];
3405
3406                 if(FRAME_MBAFF)
3407                     h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
3408
3409                 s->mb_x= mb_x;
3410                 s->mb_y= mb_y;
3411                 dest_y  = s->current_picture.f.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize  ) * 16;
3412                 dest_cb = s->current_picture.f.data[1] + (mb_x << pixel_shift) * (8 << CHROMA444) + mb_y * s->uvlinesize * block_h;
3413                 dest_cr = s->current_picture.f.data[2] + (mb_x << pixel_shift) * (8 << CHROMA444) + mb_y * s->uvlinesize * block_h;
3414                     //FIXME simplify above
3415
3416                 if (MB_FIELD) {
3417                     linesize   = h->mb_linesize   = s->linesize * 2;
3418                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
3419                     if(mb_y&1){ //FIXME move out of this function?
3420                         dest_y -= s->linesize*15;
3421                         dest_cb-= s->uvlinesize * (block_h - 1);
3422                         dest_cr-= s->uvlinesize * (block_h - 1);
3423                     }
3424                 } else {
3425                     linesize   = h->mb_linesize   = s->linesize;
3426                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
3427                 }
3428                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
3429                 if(fill_filter_caches(h, mb_type))
3430                     continue;
3431                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.f.qscale_table[mb_xy]);
3432                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.f.qscale_table[mb_xy]);
3433
3434                 if (FRAME_MBAFF) {
3435                     ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3436                 } else {
3437                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3438                 }
3439             }
3440         }
3441     }
3442     h->slice_type= old_slice_type;
3443     s->mb_x= end_x;
3444     s->mb_y= end_mb_y - FRAME_MBAFF;
3445     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3446     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3447 }
3448
3449 static void predict_field_decoding_flag(H264Context *h){
3450     MpegEncContext * const s = &h->s;
3451     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3452     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
3453                 ? s->current_picture.f.mb_type[mb_xy - 1]
3454       &nbs