H.264: move avctx->{profile,level} init before AVCodecContext.get_format().
[ffmpeg.git] / libavcodec / h264.c
1 /*
2  * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * FFmpeg is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21
22 /**
23  * @file libavcodec/h264.c
24  * H.264 / AVC / MPEG4 part10 codec.
25  * @author Michael Niedermayer <michaelni@gmx.at>
26  */
27
28 #include "internal.h"
29 #include "dsputil.h"
30 #include "avcodec.h"
31 #include "mpegvideo.h"
32 #include "h264.h"
33 #include "h264data.h"
34 #include "h264_mvpred.h"
35 #include "h264_parser.h"
36 #include "golomb.h"
37 #include "mathops.h"
38 #include "rectangle.h"
39 #include "vdpau_internal.h"
40
41 #include "cabac.h"
42
43 //#undef NDEBUG
44 #include <assert.h>
45
46 static const uint8_t rem6[52]={
47 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,
48 };
49
50 static const uint8_t div6[52]={
51 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,
52 };
53
54 void ff_h264_write_back_intra_pred_mode(H264Context *h){
55     int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
56
57     AV_COPY32(mode, h->intra4x4_pred_mode_cache + 4 + 8*4);
58     mode[4]= h->intra4x4_pred_mode_cache[7+8*3];
59     mode[5]= h->intra4x4_pred_mode_cache[7+8*2];
60     mode[6]= h->intra4x4_pred_mode_cache[7+8*1];
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 #if 0
148     for(i=0; i<length; i++)
149         printf("%2X ", src[i]);
150 #endif
151
152 #if HAVE_FAST_UNALIGNED
153 # if HAVE_FAST_64BIT
154 #   define RS 7
155     for(i=0; i+1<length; i+=9){
156         if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
157 # else
158 #   define RS 3
159     for(i=0; i+1<length; i+=5){
160         if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
161 # endif
162             continue;
163         if(i>0 && !src[i]) i--;
164         while(src[i]) i++;
165 #else
166 #   define RS 0
167     for(i=0; i+1<length; i+=2){
168         if(src[i]) continue;
169         if(i>0 && src[i-1]==0) i--;
170 #endif
171         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
172             if(src[i+2]!=3){
173                 /* startcode, so we must be past the end */
174                 length=i;
175             }
176             break;
177         }
178         i-= RS;
179     }
180
181     if(i>=length-1){ //no escaped 0
182         *dst_length= length;
183         *consumed= length+1; //+1 for the header
184         return src;
185     }
186
187     bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
188     av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
189     dst= h->rbsp_buffer[bufidx];
190
191     if (dst == NULL){
192         return NULL;
193     }
194
195 //printf("decoding esc\n");
196     memcpy(dst, src, i);
197     si=di=i;
198     while(si+2<length){
199         //remove escapes (very rare 1:2^22)
200         if(src[si+2]>3){
201             dst[di++]= src[si++];
202             dst[di++]= src[si++];
203         }else if(src[si]==0 && src[si+1]==0){
204             if(src[si+2]==3){ //escape
205                 dst[di++]= 0;
206                 dst[di++]= 0;
207                 si+=3;
208                 continue;
209             }else //next start code
210                 goto nsc;
211         }
212
213         dst[di++]= src[si++];
214     }
215     while(si<length)
216         dst[di++]= src[si++];
217 nsc:
218
219     memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
220
221     *dst_length= di;
222     *consumed= si + 1;//+1 for the header
223 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
224     return dst;
225 }
226
227 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 /**
241  * IDCT transforms the 16 dc values and dequantizes them.
242  * @param qp quantization parameter
243  */
244 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
245 #define stride 16
246     int i;
247     int temp[16]; //FIXME check if this is a good idea
248     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
249     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
250
251 //memset(block, 64, 2*256);
252 //return;
253     for(i=0; i<4; i++){
254         const int offset= y_offset[i];
255         const int z0= block[offset+stride*0] + block[offset+stride*4];
256         const int z1= block[offset+stride*0] - block[offset+stride*4];
257         const int z2= block[offset+stride*1] - block[offset+stride*5];
258         const int z3= block[offset+stride*1] + block[offset+stride*5];
259
260         temp[4*i+0]= z0+z3;
261         temp[4*i+1]= z1+z2;
262         temp[4*i+2]= z1-z2;
263         temp[4*i+3]= z0-z3;
264     }
265
266     for(i=0; i<4; i++){
267         const int offset= x_offset[i];
268         const int z0= temp[4*0+i] + temp[4*2+i];
269         const int z1= temp[4*0+i] - temp[4*2+i];
270         const int z2= temp[4*1+i] - temp[4*3+i];
271         const int z3= temp[4*1+i] + temp[4*3+i];
272
273         block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
274         block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
275         block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
276         block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
277     }
278 }
279
280 #if 0
281 /**
282  * DCT transforms the 16 dc values.
283  * @param qp quantization parameter ??? FIXME
284  */
285 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
286 //    const int qmul= dequant_coeff[qp][0];
287     int i;
288     int temp[16]; //FIXME check if this is a good idea
289     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
290     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
291
292     for(i=0; i<4; i++){
293         const int offset= y_offset[i];
294         const int z0= block[offset+stride*0] + block[offset+stride*4];
295         const int z1= block[offset+stride*0] - block[offset+stride*4];
296         const int z2= block[offset+stride*1] - block[offset+stride*5];
297         const int z3= block[offset+stride*1] + block[offset+stride*5];
298
299         temp[4*i+0]= z0+z3;
300         temp[4*i+1]= z1+z2;
301         temp[4*i+2]= z1-z2;
302         temp[4*i+3]= z0-z3;
303     }
304
305     for(i=0; i<4; i++){
306         const int offset= x_offset[i];
307         const int z0= temp[4*0+i] + temp[4*2+i];
308         const int z1= temp[4*0+i] - temp[4*2+i];
309         const int z2= temp[4*1+i] - temp[4*3+i];
310         const int z3= temp[4*1+i] + temp[4*3+i];
311
312         block[stride*0 +offset]= (z0 + z3)>>1;
313         block[stride*2 +offset]= (z1 + z2)>>1;
314         block[stride*8 +offset]= (z1 - z2)>>1;
315         block[stride*10+offset]= (z0 - z3)>>1;
316     }
317 }
318 #endif
319
320 #undef xStride
321 #undef stride
322
323 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
324     const int stride= 16*2;
325     const int xStride= 16;
326     int a,b,c,d,e;
327
328     a= block[stride*0 + xStride*0];
329     b= block[stride*0 + xStride*1];
330     c= block[stride*1 + xStride*0];
331     d= block[stride*1 + xStride*1];
332
333     e= a-b;
334     a= a+b;
335     b= c-d;
336     c= c+d;
337
338     block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
339     block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
340     block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
341     block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
342 }
343
344 #if 0
345 static void chroma_dc_dct_c(DCTELEM *block){
346     const int stride= 16*2;
347     const int xStride= 16;
348     int a,b,c,d,e;
349
350     a= block[stride*0 + xStride*0];
351     b= block[stride*0 + xStride*1];
352     c= block[stride*1 + xStride*0];
353     d= block[stride*1 + xStride*1];
354
355     e= a-b;
356     a= a+b;
357     b= c-d;
358     c= c+d;
359
360     block[stride*0 + xStride*0]= (a+c);
361     block[stride*0 + xStride*1]= (e+b);
362     block[stride*1 + xStride*0]= (a-c);
363     block[stride*1 + xStride*1]= (e-b);
364 }
365 #endif
366
367 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
368                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
369                            int src_x_offset, int src_y_offset,
370                            qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
371     MpegEncContext * const s = &h->s;
372     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
373     int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
374     const int luma_xy= (mx&3) + ((my&3)<<2);
375     uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
376     uint8_t * src_cb, * src_cr;
377     int extra_width= h->emu_edge_width;
378     int extra_height= h->emu_edge_height;
379     int emu=0;
380     const int full_mx= mx>>2;
381     const int full_my= my>>2;
382     const int pic_width  = 16*s->mb_width;
383     const int pic_height = 16*s->mb_height >> MB_FIELD;
384
385     if(mx&7) extra_width -= 3;
386     if(my&7) extra_height -= 3;
387
388     if(   full_mx < 0-extra_width
389        || full_my < 0-extra_height
390        || full_mx + 16/*FIXME*/ > pic_width + extra_width
391        || full_my + 16/*FIXME*/ > pic_height + extra_height){
392         ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
393             src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
394         emu=1;
395     }
396
397     qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
398     if(!square){
399         qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
400     }
401
402     if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
403
404     if(MB_FIELD){
405         // chroma offset when predicting from a field of opposite parity
406         my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
407         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
408     }
409     src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
410     src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
411
412     if(emu){
413         ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
414             src_cb= s->edge_emu_buffer;
415     }
416     chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
417
418     if(emu){
419         ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
420             src_cr= s->edge_emu_buffer;
421     }
422     chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
423 }
424
425 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
426                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
427                            int x_offset, int y_offset,
428                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
429                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
430                            int list0, int list1){
431     MpegEncContext * const s = &h->s;
432     qpel_mc_func *qpix_op=  qpix_put;
433     h264_chroma_mc_func chroma_op= chroma_put;
434
435     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
436     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
437     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
438     x_offset += 8*s->mb_x;
439     y_offset += 8*(s->mb_y >> MB_FIELD);
440
441     if(list0){
442         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
443         mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
444                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
445                            qpix_op, chroma_op);
446
447         qpix_op=  qpix_avg;
448         chroma_op= chroma_avg;
449     }
450
451     if(list1){
452         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
453         mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
454                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
455                            qpix_op, chroma_op);
456     }
457 }
458
459 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
460                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
461                            int x_offset, int y_offset,
462                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
463                            h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
464                            h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
465                            int list0, int list1){
466     MpegEncContext * const s = &h->s;
467
468     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
469     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
470     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
471     x_offset += 8*s->mb_x;
472     y_offset += 8*(s->mb_y >> MB_FIELD);
473
474     if(list0 && list1){
475         /* don't optimize for luma-only case, since B-frames usually
476          * use implicit weights => chroma too. */
477         uint8_t *tmp_cb = s->obmc_scratchpad;
478         uint8_t *tmp_cr = s->obmc_scratchpad + 8;
479         uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
480         int refn0 = h->ref_cache[0][ scan8[n] ];
481         int refn1 = h->ref_cache[1][ scan8[n] ];
482
483         mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
484                     dest_y, dest_cb, dest_cr,
485                     x_offset, y_offset, qpix_put, chroma_put);
486         mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
487                     tmp_y, tmp_cb, tmp_cr,
488                     x_offset, y_offset, qpix_put, chroma_put);
489
490         if(h->use_weight == 2){
491             int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
492             int weight1 = 64 - weight0;
493             luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
494             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
495             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
496         }else{
497             luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
498                             h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
499                             h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
500             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
501                             h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
502                             h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
503             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
504                             h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
505                             h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
506         }
507     }else{
508         int list = list1 ? 1 : 0;
509         int refn = h->ref_cache[list][ scan8[n] ];
510         Picture *ref= &h->ref_list[list][refn];
511         mc_dir_part(h, ref, n, square, chroma_height, delta, list,
512                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
513                     qpix_put, chroma_put);
514
515         luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
516                        h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
517         if(h->use_weight_chroma){
518             chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
519                              h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
520             chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
521                              h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
522         }
523     }
524 }
525
526 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
527                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
528                            int x_offset, int y_offset,
529                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
530                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
531                            h264_weight_func *weight_op, h264_biweight_func *weight_avg,
532                            int list0, int list1){
533     if((h->use_weight==2 && list0 && list1
534         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
535        || h->use_weight==1)
536         mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
537                          x_offset, y_offset, qpix_put, chroma_put,
538                          weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
539     else
540         mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
541                     x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
542 }
543
544 static inline void prefetch_motion(H264Context *h, int list){
545     /* fetch pixels for estimated mv 4 macroblocks ahead
546      * optimized for 64byte cache lines */
547     MpegEncContext * const s = &h->s;
548     const int refn = h->ref_cache[list][scan8[0]];
549     if(refn >= 0){
550         const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
551         const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
552         uint8_t **src= h->ref_list[list][refn].data;
553         int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
554         s->dsp.prefetch(src[0]+off, s->linesize, 4);
555         off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
556         s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
557     }
558 }
559
560 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
561                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
562                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
563                       h264_weight_func *weight_op, h264_biweight_func *weight_avg){
564     MpegEncContext * const s = &h->s;
565     const int mb_xy= h->mb_xy;
566     const int mb_type= s->current_picture.mb_type[mb_xy];
567
568     assert(IS_INTER(mb_type));
569
570     prefetch_motion(h, 0);
571
572     if(IS_16X16(mb_type)){
573         mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
574                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
575                 weight_op, weight_avg,
576                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
577     }else if(IS_16X8(mb_type)){
578         mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
579                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
580                 &weight_op[1], &weight_avg[1],
581                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
582         mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
583                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
584                 &weight_op[1], &weight_avg[1],
585                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
586     }else if(IS_8X16(mb_type)){
587         mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
588                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
589                 &weight_op[2], &weight_avg[2],
590                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
591         mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
592                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
593                 &weight_op[2], &weight_avg[2],
594                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
595     }else{
596         int i;
597
598         assert(IS_8X8(mb_type));
599
600         for(i=0; i<4; i++){
601             const int sub_mb_type= h->sub_mb_type[i];
602             const int n= 4*i;
603             int x_offset= (i&1)<<2;
604             int y_offset= (i&2)<<1;
605
606             if(IS_SUB_8X8(sub_mb_type)){
607                 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
608                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
609                     &weight_op[3], &weight_avg[3],
610                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
611             }else if(IS_SUB_8X4(sub_mb_type)){
612                 mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
613                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
614                     &weight_op[4], &weight_avg[4],
615                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
616                 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
617                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
618                     &weight_op[4], &weight_avg[4],
619                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
620             }else if(IS_SUB_4X8(sub_mb_type)){
621                 mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
622                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
623                     &weight_op[5], &weight_avg[5],
624                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
625                 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
626                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
627                     &weight_op[5], &weight_avg[5],
628                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
629             }else{
630                 int j;
631                 assert(IS_SUB_4X4(sub_mb_type));
632                 for(j=0; j<4; j++){
633                     int sub_x_offset= x_offset + 2*(j&1);
634                     int sub_y_offset= y_offset +   (j&2);
635                     mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
636                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
637                         &weight_op[6], &weight_avg[6],
638                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
639                 }
640             }
641         }
642     }
643
644     prefetch_motion(h, 1);
645 }
646
647
648 static void free_tables(H264Context *h){
649     int i;
650     H264Context *hx;
651     av_freep(&h->intra4x4_pred_mode);
652     av_freep(&h->chroma_pred_mode_table);
653     av_freep(&h->cbp_table);
654     av_freep(&h->mvd_table[0]);
655     av_freep(&h->mvd_table[1]);
656     av_freep(&h->direct_table);
657     av_freep(&h->non_zero_count);
658     av_freep(&h->slice_table_base);
659     h->slice_table= NULL;
660     av_freep(&h->list_counts);
661
662     av_freep(&h->mb2b_xy);
663     av_freep(&h->mb2br_xy);
664
665     for(i = 0; i < MAX_THREADS; i++) {
666         hx = h->thread_context[i];
667         if(!hx) continue;
668         av_freep(&hx->top_borders[1]);
669         av_freep(&hx->top_borders[0]);
670         av_freep(&hx->s.obmc_scratchpad);
671         av_freep(&hx->rbsp_buffer[1]);
672         av_freep(&hx->rbsp_buffer[0]);
673         hx->rbsp_buffer_size[0] = 0;
674         hx->rbsp_buffer_size[1] = 0;
675         if (i) av_freep(&h->thread_context[i]);
676     }
677 }
678
679 static void init_dequant8_coeff_table(H264Context *h){
680     int i,q,x;
681     const int transpose = (h->h264dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
682     h->dequant8_coeff[0] = h->dequant8_buffer[0];
683     h->dequant8_coeff[1] = h->dequant8_buffer[1];
684
685     for(i=0; i<2; i++ ){
686         if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
687             h->dequant8_coeff[1] = h->dequant8_buffer[0];
688             break;
689         }
690
691         for(q=0; q<52; q++){
692             int shift = div6[q];
693             int idx = rem6[q];
694             for(x=0; x<64; x++)
695                 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
696                     ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
697                     h->pps.scaling_matrix8[i][x]) << shift;
698         }
699     }
700 }
701
702 static void init_dequant4_coeff_table(H264Context *h){
703     int i,j,q,x;
704     const int transpose = (h->h264dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
705     for(i=0; i<6; i++ ){
706         h->dequant4_coeff[i] = h->dequant4_buffer[i];
707         for(j=0; j<i; j++){
708             if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
709                 h->dequant4_coeff[i] = h->dequant4_buffer[j];
710                 break;
711             }
712         }
713         if(j<i)
714             continue;
715
716         for(q=0; q<52; q++){
717             int shift = div6[q] + 2;
718             int idx = rem6[q];
719             for(x=0; x<16; x++)
720                 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
721                     ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
722                     h->pps.scaling_matrix4[i][x]) << shift;
723         }
724     }
725 }
726
727 static void init_dequant_tables(H264Context *h){
728     int i,x;
729     init_dequant4_coeff_table(h);
730     if(h->pps.transform_8x8_mode)
731         init_dequant8_coeff_table(h);
732     if(h->sps.transform_bypass){
733         for(i=0; i<6; i++)
734             for(x=0; x<16; x++)
735                 h->dequant4_coeff[i][0][x] = 1<<6;
736         if(h->pps.transform_8x8_mode)
737             for(i=0; i<2; i++)
738                 for(x=0; x<64; x++)
739                     h->dequant8_coeff[i][0][x] = 1<<6;
740     }
741 }
742
743
744 int ff_h264_alloc_tables(H264Context *h){
745     MpegEncContext * const s = &h->s;
746     const int big_mb_num= s->mb_stride * (s->mb_height+1);
747     const int row_mb_num= 2*s->mb_stride*s->avctx->thread_count;
748     int x,y;
749
750     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8  * sizeof(uint8_t), fail)
751
752     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 32 * sizeof(uint8_t), fail)
753     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
754     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
755
756     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
757     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
758     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
759     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
760     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
761
762     memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
763     h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
764
765     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy  , big_mb_num * sizeof(uint32_t), fail);
766     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
767     for(y=0; y<s->mb_height; y++){
768         for(x=0; x<s->mb_width; x++){
769             const int mb_xy= x + y*s->mb_stride;
770             const int b_xy = 4*x + 4*y*h->b_stride;
771
772             h->mb2b_xy [mb_xy]= b_xy;
773             h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
774         }
775     }
776
777     s->obmc_scratchpad = NULL;
778
779     if(!h->dequant4_coeff[0])
780         init_dequant_tables(h);
781
782     return 0;
783 fail:
784     free_tables(h);
785     return -1;
786 }
787
788 /**
789  * Mimic alloc_tables(), but for every context thread.
790  */
791 static void clone_tables(H264Context *dst, H264Context *src, int i){
792     MpegEncContext * const s = &src->s;
793     dst->intra4x4_pred_mode       = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
794     dst->non_zero_count           = src->non_zero_count;
795     dst->slice_table              = src->slice_table;
796     dst->cbp_table                = src->cbp_table;
797     dst->mb2b_xy                  = src->mb2b_xy;
798     dst->mb2br_xy                 = src->mb2br_xy;
799     dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
800     dst->mvd_table[0]             = src->mvd_table[0] + i*8*2*s->mb_stride;
801     dst->mvd_table[1]             = src->mvd_table[1] + i*8*2*s->mb_stride;
802     dst->direct_table             = src->direct_table;
803     dst->list_counts              = src->list_counts;
804
805     dst->s.obmc_scratchpad = NULL;
806     ff_h264_pred_init(&dst->hpc, src->s.codec_id);
807 }
808
809 /**
810  * Init context
811  * Allocate buffers which are not shared amongst multiple threads.
812  */
813 static int context_init(H264Context *h){
814     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
815     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
816
817     h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
818     h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
819
820     return 0;
821 fail:
822     return -1; // free_tables will clean up for us
823 }
824
825 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
826
827 static av_cold void common_init(H264Context *h){
828     MpegEncContext * const s = &h->s;
829
830     s->width = s->avctx->width;
831     s->height = s->avctx->height;
832     s->codec_id= s->avctx->codec->id;
833
834     ff_h264dsp_init(&h->h264dsp);
835     ff_h264_pred_init(&h->hpc, s->codec_id);
836
837     h->dequant_coeff_pps= -1;
838     s->unrestricted_mv=1;
839     s->decode=1; //FIXME
840
841     dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
842
843     memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
844     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
845 }
846
847 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
848     H264Context *h= avctx->priv_data;
849     MpegEncContext * const s = &h->s;
850
851     MPV_decode_defaults(s);
852
853     s->avctx = avctx;
854     common_init(h);
855
856     s->out_format = FMT_H264;
857     s->workaround_bugs= avctx->workaround_bugs;
858
859     // set defaults
860 //    s->decode_mb= ff_h263_decode_mb;
861     s->quarter_sample = 1;
862     if(!avctx->has_b_frames)
863     s->low_delay= 1;
864
865     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
866
867     ff_h264_decode_init_vlc();
868
869     h->thread_context[0] = h;
870     h->outputed_poc = INT_MIN;
871     h->prev_poc_msb= 1<<16;
872     h->x264_build = -1;
873     ff_h264_reset_sei(h);
874     if(avctx->codec_id == CODEC_ID_H264){
875         if(avctx->ticks_per_frame == 1){
876             s->avctx->time_base.den *=2;
877         }
878         avctx->ticks_per_frame = 2;
879     }
880
881     if(avctx->extradata_size > 0 && avctx->extradata && *(char *)avctx->extradata == 1){
882         int i, cnt, nalsize;
883         unsigned char *p = avctx->extradata;
884
885         h->is_avc = 1;
886
887         if(avctx->extradata_size < 7) {
888             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
889             return -1;
890         }
891         /* sps and pps in the avcC always have length coded with 2 bytes,
892            so put a fake nal_length_size = 2 while parsing them */
893         h->nal_length_size = 2;
894         // Decode sps from avcC
895         cnt = *(p+5) & 0x1f; // Number of sps
896         p += 6;
897         for (i = 0; i < cnt; i++) {
898             nalsize = AV_RB16(p) + 2;
899             if(decode_nal_units(h, p, nalsize) < 0) {
900                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
901                 return -1;
902             }
903             p += nalsize;
904         }
905         // Decode pps from avcC
906         cnt = *(p++); // Number of pps
907         for (i = 0; i < cnt; i++) {
908             nalsize = AV_RB16(p) + 2;
909             if(decode_nal_units(h, p, nalsize)  != nalsize) {
910                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
911                 return -1;
912             }
913             p += nalsize;
914         }
915         // Now store right nal length size, that will be use to parse all other nals
916         h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
917     } else {
918         h->is_avc = 0;
919         if(decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) < 0)
920             return -1;
921     }
922     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
923         s->avctx->has_b_frames = h->sps.num_reorder_frames;
924         s->low_delay = 0;
925     }
926
927     return 0;
928 }
929
930 int ff_h264_frame_start(H264Context *h){
931     MpegEncContext * const s = &h->s;
932     int i;
933
934     if(MPV_frame_start(s, s->avctx) < 0)
935         return -1;
936     ff_er_frame_start(s);
937     /*
938      * MPV_frame_start uses pict_type to derive key_frame.
939      * This is incorrect for H.264; IDR markings must be used.
940      * Zero here; IDR markings per slice in frame or fields are ORed in later.
941      * See decode_nal_units().
942      */
943     s->current_picture_ptr->key_frame= 0;
944     s->current_picture_ptr->mmco_reset= 0;
945
946     assert(s->linesize && s->uvlinesize);
947
948     for(i=0; i<16; i++){
949         h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
950         h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
951     }
952     for(i=0; i<4; i++){
953         h->block_offset[16+i]=
954         h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
955         h->block_offset[24+16+i]=
956         h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
957     }
958
959     /* can't be in alloc_tables because linesize isn't known there.
960      * FIXME: redo bipred weight to not require extra buffer? */
961     for(i = 0; i < s->avctx->thread_count; i++)
962         if(!h->thread_context[i]->s.obmc_scratchpad)
963             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
964
965     /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
966     memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
967
968 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
969
970     // We mark the current picture as non-reference after allocating it, so
971     // that if we break out due to an error it can be released automatically
972     // in the next MPV_frame_start().
973     // SVQ3 as well as most other codecs have only last/next/current and thus
974     // get released even with set reference, besides SVQ3 and others do not
975     // mark frames as reference later "naturally".
976     if(s->codec_id != CODEC_ID_SVQ3)
977         s->current_picture_ptr->reference= 0;
978
979     s->current_picture_ptr->field_poc[0]=
980     s->current_picture_ptr->field_poc[1]= INT_MAX;
981     assert(s->current_picture_ptr->long_ref==0);
982
983     return 0;
984 }
985
986 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int simple){
987     MpegEncContext * const s = &h->s;
988     uint8_t *top_border;
989     int top_idx = 1;
990
991     src_y  -=   linesize;
992     src_cb -= uvlinesize;
993     src_cr -= uvlinesize;
994
995     if(!simple && FRAME_MBAFF){
996         if(s->mb_y&1){
997             if(!MB_MBAFF){
998                 top_border = h->top_borders[0][s->mb_x];
999                 AV_COPY128(top_border, src_y + 15*linesize);
1000                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1001                     AV_COPY64(top_border+16, src_cb+7*uvlinesize);
1002                     AV_COPY64(top_border+24, src_cr+7*uvlinesize);
1003                 }
1004             }
1005         }else if(MB_MBAFF){
1006             top_idx = 0;
1007         }else
1008             return;
1009     }
1010
1011     top_border = h->top_borders[top_idx][s->mb_x];
1012     // There are two lines saved, the line above the the top macroblock of a pair,
1013     // and the line above the bottom macroblock
1014     AV_COPY128(top_border, src_y + 16*linesize);
1015
1016     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1017         AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1018         AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1019     }
1020 }
1021
1022 static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg, int simple){
1023     MpegEncContext * const s = &h->s;
1024     int deblock_left;
1025     int deblock_top;
1026     int top_idx = 1;
1027     uint8_t *top_border_m1;
1028     uint8_t *top_border;
1029
1030     if(!simple && FRAME_MBAFF){
1031         if(s->mb_y&1){
1032             if(!MB_MBAFF)
1033                 return;
1034         }else{
1035             top_idx = MB_MBAFF ? 0 : 1;
1036         }
1037     }
1038
1039     if(h->deblocking_filter == 2) {
1040         deblock_left = h->left_type[0];
1041         deblock_top  = h->top_type;
1042     } else {
1043         deblock_left = (s->mb_x > 0);
1044         deblock_top =  (s->mb_y > !!MB_FIELD);
1045     }
1046
1047     src_y  -=   linesize + 1;
1048     src_cb -= uvlinesize + 1;
1049     src_cr -= uvlinesize + 1;
1050
1051     top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1052     top_border    = h->top_borders[top_idx][s->mb_x];
1053
1054 #define XCHG(a,b,xchg)\
1055 if (xchg) AV_SWAP64(b,a);\
1056 else      AV_COPY64(b,a);
1057
1058     if(deblock_top){
1059         if(deblock_left){
1060             XCHG(top_border_m1+8, src_y -7, 1);
1061         }
1062         XCHG(top_border+0, src_y +1, xchg);
1063         XCHG(top_border+8, src_y +9, 1);
1064         if(s->mb_x+1 < s->mb_width){
1065             XCHG(h->top_borders[top_idx][s->mb_x+1], src_y +17, 1);
1066         }
1067     }
1068
1069     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1070         if(deblock_top){
1071             if(deblock_left){
1072                 XCHG(top_border_m1+16, src_cb -7, 1);
1073                 XCHG(top_border_m1+24, src_cr -7, 1);
1074             }
1075             XCHG(top_border+16, src_cb+1, 1);
1076             XCHG(top_border+24, src_cr+1, 1);
1077         }
1078     }
1079 }
1080
1081 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
1082     MpegEncContext * const s = &h->s;
1083     const int mb_x= s->mb_x;
1084     const int mb_y= s->mb_y;
1085     const int mb_xy= h->mb_xy;
1086     const int mb_type= s->current_picture.mb_type[mb_xy];
1087     uint8_t  *dest_y, *dest_cb, *dest_cr;
1088     int linesize, uvlinesize /*dct_offset*/;
1089     int i;
1090     int *block_offset = &h->block_offset[0];
1091     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1092     /* is_h264 should always be true if SVQ3 is disabled. */
1093     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1094     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1095     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1096
1097     dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
1098     dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
1099     dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
1100
1101     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
1102     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
1103
1104     h->list_counts[mb_xy]= h->list_count;
1105
1106     if (!simple && MB_FIELD) {
1107         linesize   = h->mb_linesize   = s->linesize * 2;
1108         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1109         block_offset = &h->block_offset[24];
1110         if(mb_y&1){ //FIXME move out of this function?
1111             dest_y -= s->linesize*15;
1112             dest_cb-= s->uvlinesize*7;
1113             dest_cr-= s->uvlinesize*7;
1114         }
1115         if(FRAME_MBAFF) {
1116             int list;
1117             for(list=0; list<h->list_count; list++){
1118                 if(!USES_LIST(mb_type, list))
1119                     continue;
1120                 if(IS_16X16(mb_type)){
1121                     int8_t *ref = &h->ref_cache[list][scan8[0]];
1122                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1123                 }else{
1124                     for(i=0; i<16; i+=4){
1125                         int ref = h->ref_cache[list][scan8[i]];
1126                         if(ref >= 0)
1127                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1128                     }
1129                 }
1130             }
1131         }
1132     } else {
1133         linesize   = h->mb_linesize   = s->linesize;
1134         uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1135 //        dct_offset = s->linesize * 16;
1136     }
1137
1138     if (!simple && IS_INTRA_PCM(mb_type)) {
1139         for (i=0; i<16; i++) {
1140             memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1141         }
1142         for (i=0; i<8; i++) {
1143             memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
1144             memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
1145         }
1146     } else {
1147         if(IS_INTRA(mb_type)){
1148             if(h->deblocking_filter)
1149                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
1150
1151             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1152                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1153                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1154             }
1155
1156             if(IS_INTRA4x4(mb_type)){
1157                 if(simple || !s->encoding){
1158                     if(IS_8x8DCT(mb_type)){
1159                         if(transform_bypass){
1160                             idct_dc_add =
1161                             idct_add    = s->dsp.add_pixels8;
1162                         }else{
1163                             idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1164                             idct_add    = h->h264dsp.h264_idct8_add;
1165                         }
1166                         for(i=0; i<16; i+=4){
1167                             uint8_t * const ptr= dest_y + block_offset[i];
1168                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1169                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1170                                 h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
1171                             }else{
1172                                 const int nnz = h->non_zero_count_cache[ scan8[i] ];
1173                                 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1174                                                             (h->topright_samples_available<<i)&0x4000, linesize);
1175                                 if(nnz){
1176                                     if(nnz == 1 && h->mb[i*16])
1177                                         idct_dc_add(ptr, h->mb + i*16, linesize);
1178                                     else
1179                                         idct_add   (ptr, h->mb + i*16, linesize);
1180                                 }
1181                             }
1182                         }
1183                     }else{
1184                         if(transform_bypass){
1185                             idct_dc_add =
1186                             idct_add    = s->dsp.add_pixels4;
1187                         }else{
1188                             idct_dc_add = h->h264dsp.h264_idct_dc_add;
1189                             idct_add    = h->h264dsp.h264_idct_add;
1190                         }
1191                         for(i=0; i<16; i++){
1192                             uint8_t * const ptr= dest_y + block_offset[i];
1193                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1194
1195                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1196                                 h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
1197                             }else{
1198                                 uint8_t *topright;
1199                                 int nnz, tr;
1200                                 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1201                                     const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1202                                     assert(mb_y || linesize <= block_offset[i]);
1203                                     if(!topright_avail){
1204                                         tr= ptr[3 - linesize]*0x01010101;
1205                                         topright= (uint8_t*) &tr;
1206                                     }else
1207                                         topright= ptr + 4 - linesize;
1208                                 }else
1209                                     topright= NULL;
1210
1211                                 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1212                                 nnz = h->non_zero_count_cache[ scan8[i] ];
1213                                 if(nnz){
1214                                     if(is_h264){
1215                                         if(nnz == 1 && h->mb[i*16])
1216                                             idct_dc_add(ptr, h->mb + i*16, linesize);
1217                                         else
1218                                             idct_add   (ptr, h->mb + i*16, linesize);
1219                                     }else
1220                                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1221                                 }
1222                             }
1223                         }
1224                     }
1225                 }
1226             }else{
1227                 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1228                 if(is_h264){
1229                     if(!transform_bypass)
1230                         h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
1231                 }else
1232                     ff_svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
1233             }
1234             if(h->deblocking_filter)
1235                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
1236         }else if(is_h264){
1237             hl_motion(h, dest_y, dest_cb, dest_cr,
1238                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1239                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1240                       h->h264dsp.weight_h264_pixels_tab, h->h264dsp.biweight_h264_pixels_tab);
1241         }
1242
1243
1244         if(!IS_INTRA4x4(mb_type)){
1245             if(is_h264){
1246                 if(IS_INTRA16x16(mb_type)){
1247                     if(transform_bypass){
1248                         if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1249                             h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1250                         }else{
1251                             for(i=0; i<16; i++){
1252                                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1253                                     s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
1254                             }
1255                         }
1256                     }else{
1257                          h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1258                     }
1259                 }else if(h->cbp&15){
1260                     if(transform_bypass){
1261                         const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1262                         idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1263                         for(i=0; i<16; i+=di){
1264                             if(h->non_zero_count_cache[ scan8[i] ]){
1265                                 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
1266                             }
1267                         }
1268                     }else{
1269                         if(IS_8x8DCT(mb_type)){
1270                             h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1271                         }else{
1272                             h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1273                         }
1274                     }
1275                 }
1276             }else{
1277                 for(i=0; i<16; i++){
1278                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1279                         uint8_t * const ptr= dest_y + block_offset[i];
1280                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1281                     }
1282                 }
1283             }
1284         }
1285
1286         if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1287             uint8_t *dest[2] = {dest_cb, dest_cr};
1288             if(transform_bypass){
1289                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1290                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
1291                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
1292                 }else{
1293                     idct_add = s->dsp.add_pixels4;
1294                     for(i=16; i<16+8; i++){
1295                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1296                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1297                     }
1298                 }
1299             }else{
1300                 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp[0], h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
1301                 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp[1], h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
1302                 if(is_h264){
1303                     idct_add = h->h264dsp.h264_idct_add;
1304                     idct_dc_add = h->h264dsp.h264_idct_dc_add;
1305                     for(i=16; i<16+8; i++){
1306                         if(h->non_zero_count_cache[ scan8[i] ])
1307                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1308                         else if(h->mb[i*16])
1309                             idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1310                     }
1311                 }else{
1312                     for(i=16; i<16+8; i++){
1313                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1314                             uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1315                             ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
1316                         }
1317                     }
1318                 }
1319             }
1320         }
1321     }
1322     if(h->cbp || IS_INTRA(mb_type))
1323         s->dsp.clear_blocks(h->mb);
1324 }
1325
1326 /**
1327  * Process a macroblock; this case avoids checks for expensive uncommon cases.
1328  */
1329 static void hl_decode_mb_simple(H264Context *h){
1330     hl_decode_mb_internal(h, 1);
1331 }
1332
1333 /**
1334  * Process a macroblock; this handles edge cases, such as interlacing.
1335  */
1336 static void av_noinline hl_decode_mb_complex(H264Context *h){
1337     hl_decode_mb_internal(h, 0);
1338 }
1339
1340 void ff_h264_hl_decode_mb(H264Context *h){
1341     MpegEncContext * const s = &h->s;
1342     const int mb_xy= h->mb_xy;
1343     const int mb_type= s->current_picture.mb_type[mb_xy];
1344     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1345
1346     if (is_complex)
1347         hl_decode_mb_complex(h);
1348     else hl_decode_mb_simple(h);
1349 }
1350
1351 static int pred_weight_table(H264Context *h){
1352     MpegEncContext * const s = &h->s;
1353     int list, i;
1354     int luma_def, chroma_def;
1355
1356     h->use_weight= 0;
1357     h->use_weight_chroma= 0;
1358     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1359     h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1360     luma_def = 1<<h->luma_log2_weight_denom;
1361     chroma_def = 1<<h->chroma_log2_weight_denom;
1362
1363     for(list=0; list<2; list++){
1364         h->luma_weight_flag[list]   = 0;
1365         h->chroma_weight_flag[list] = 0;
1366         for(i=0; i<h->ref_count[list]; i++){
1367             int luma_weight_flag, chroma_weight_flag;
1368
1369             luma_weight_flag= get_bits1(&s->gb);
1370             if(luma_weight_flag){
1371                 h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
1372                 h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
1373                 if(   h->luma_weight[i][list][0] != luma_def
1374                    || h->luma_weight[i][list][1] != 0) {
1375                     h->use_weight= 1;
1376                     h->luma_weight_flag[list]= 1;
1377                 }
1378             }else{
1379                 h->luma_weight[i][list][0]= luma_def;
1380                 h->luma_weight[i][list][1]= 0;
1381             }
1382
1383             if(CHROMA){
1384                 chroma_weight_flag= get_bits1(&s->gb);
1385                 if(chroma_weight_flag){
1386                     int j;
1387                     for(j=0; j<2; j++){
1388                         h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
1389                         h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
1390                         if(   h->chroma_weight[i][list][j][0] != chroma_def
1391                            || h->chroma_weight[i][list][j][1] != 0) {
1392                             h->use_weight_chroma= 1;
1393                             h->chroma_weight_flag[list]= 1;
1394                         }
1395                     }
1396                 }else{
1397                     int j;
1398                     for(j=0; j<2; j++){
1399                         h->chroma_weight[i][list][j][0]= chroma_def;
1400                         h->chroma_weight[i][list][j][1]= 0;
1401                     }
1402                 }
1403             }
1404         }
1405         if(h->slice_type_nos != FF_B_TYPE) break;
1406     }
1407     h->use_weight= h->use_weight || h->use_weight_chroma;
1408     return 0;
1409 }
1410
1411 /**
1412  * Initialize implicit_weight table.
1413  * @param field, 0/1 initialize the weight for interlaced MBAFF
1414  *                -1 initializes the rest
1415  */
1416 static void implicit_weight_table(H264Context *h, int field){
1417     MpegEncContext * const s = &h->s;
1418     int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
1419
1420     for (i = 0; i < 2; i++) {
1421         h->luma_weight_flag[i]   = 0;
1422         h->chroma_weight_flag[i] = 0;
1423     }
1424
1425     if(field < 0){
1426         cur_poc = s->current_picture_ptr->poc;
1427     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
1428        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1429         h->use_weight= 0;
1430         h->use_weight_chroma= 0;
1431         return;
1432     }
1433         ref_start= 0;
1434         ref_count0= h->ref_count[0];
1435         ref_count1= h->ref_count[1];
1436     }else{
1437         cur_poc = s->current_picture_ptr->field_poc[field];
1438         ref_start= 16;
1439         ref_count0= 16+2*h->ref_count[0];
1440         ref_count1= 16+2*h->ref_count[1];
1441     }
1442
1443     h->use_weight= 2;
1444     h->use_weight_chroma= 2;
1445     h->luma_log2_weight_denom= 5;
1446     h->chroma_log2_weight_denom= 5;
1447
1448     for(ref0=ref_start; ref0 < ref_count0; ref0++){
1449         int poc0 = h->ref_list[0][ref0].poc;
1450         for(ref1=ref_start; ref1 < ref_count1; ref1++){
1451             int poc1 = h->ref_list[1][ref1].poc;
1452             int td = av_clip(poc1 - poc0, -128, 127);
1453             int w= 32;
1454             if(td){
1455                 int tb = av_clip(cur_poc - poc0, -128, 127);
1456                 int tx = (16384 + (FFABS(td) >> 1)) / td;
1457                 int dist_scale_factor = (tb*tx + 32) >> 8;
1458                 if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
1459                     w = 64 - dist_scale_factor;
1460             }
1461             if(field<0){
1462                 h->implicit_weight[ref0][ref1][0]=
1463                 h->implicit_weight[ref0][ref1][1]= w;
1464             }else{
1465                 h->implicit_weight[ref0][ref1][field]=w;
1466             }
1467         }
1468     }
1469 }
1470
1471 /**
1472  * instantaneous decoder refresh.
1473  */
1474 static void idr(H264Context *h){
1475     ff_h264_remove_all_refs(h);
1476     h->prev_frame_num= 0;
1477     h->prev_frame_num_offset= 0;
1478     h->prev_poc_msb=
1479     h->prev_poc_lsb= 0;
1480 }
1481
1482 /* forget old pics after a seek */
1483 static void flush_dpb(AVCodecContext *avctx){
1484     H264Context *h= avctx->priv_data;
1485     int i;
1486     for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1487         if(h->delayed_pic[i])
1488             h->delayed_pic[i]->reference= 0;
1489         h->delayed_pic[i]= NULL;
1490     }
1491     h->outputed_poc= INT_MIN;
1492     h->prev_interlaced_frame = 1;
1493     idr(h);
1494     if(h->s.current_picture_ptr)
1495         h->s.current_picture_ptr->reference= 0;
1496     h->s.first_field= 0;
1497     ff_h264_reset_sei(h);
1498     ff_mpeg_flush(avctx);
1499 }
1500
1501 static int init_poc(H264Context *h){
1502     MpegEncContext * const s = &h->s;
1503     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1504     int field_poc[2];
1505     Picture *cur = s->current_picture_ptr;
1506
1507     h->frame_num_offset= h->prev_frame_num_offset;
1508     if(h->frame_num < h->prev_frame_num)
1509         h->frame_num_offset += max_frame_num;
1510
1511     if(h->sps.poc_type==0){
1512         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1513
1514         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1515             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1516         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1517             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1518         else
1519             h->poc_msb = h->prev_poc_msb;
1520 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1521         field_poc[0] =
1522         field_poc[1] = h->poc_msb + h->poc_lsb;
1523         if(s->picture_structure == PICT_FRAME)
1524             field_poc[1] += h->delta_poc_bottom;
1525     }else if(h->sps.poc_type==1){
1526         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1527         int i;
1528
1529         if(h->sps.poc_cycle_length != 0)
1530             abs_frame_num = h->frame_num_offset + h->frame_num;
1531         else
1532             abs_frame_num = 0;
1533
1534         if(h->nal_ref_idc==0 && abs_frame_num > 0)
1535             abs_frame_num--;
1536
1537         expected_delta_per_poc_cycle = 0;
1538         for(i=0; i < h->sps.poc_cycle_length; i++)
1539             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1540
1541         if(abs_frame_num > 0){
1542             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1543             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1544
1545             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1546             for(i = 0; i <= frame_num_in_poc_cycle; i++)
1547                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1548         } else
1549             expectedpoc = 0;
1550
1551         if(h->nal_ref_idc == 0)
1552             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1553
1554         field_poc[0] = expectedpoc + h->delta_poc[0];
1555         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1556
1557         if(s->picture_structure == PICT_FRAME)
1558             field_poc[1] += h->delta_poc[1];
1559     }else{
1560         int poc= 2*(h->frame_num_offset + h->frame_num);
1561
1562         if(!h->nal_ref_idc)
1563             poc--;
1564
1565         field_poc[0]= poc;
1566         field_poc[1]= poc;
1567     }
1568
1569     if(s->picture_structure != PICT_BOTTOM_FIELD)
1570         s->current_picture_ptr->field_poc[0]= field_poc[0];
1571     if(s->picture_structure != PICT_TOP_FIELD)
1572         s->current_picture_ptr->field_poc[1]= field_poc[1];
1573     cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1574
1575     return 0;
1576 }
1577
1578
1579 /**
1580  * initialize scan tables
1581  */
1582 static void init_scan_tables(H264Context *h){
1583     int i;
1584     if(h->h264dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
1585         memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
1586         memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
1587     }else{
1588         for(i=0; i<16; i++){
1589 #define T(x) (x>>2) | ((x<<2) & 0xF)
1590             h->zigzag_scan[i] = T(zigzag_scan[i]);
1591             h-> field_scan[i] = T( field_scan[i]);
1592 #undef T
1593         }
1594     }
1595     if(h->h264dsp.h264_idct8_add == ff_h264_idct8_add_c){
1596         memcpy(h->zigzag_scan8x8,       ff_zigzag_direct,     64*sizeof(uint8_t));
1597         memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
1598         memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
1599         memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
1600     }else{
1601         for(i=0; i<64; i++){
1602 #define T(x) (x>>3) | ((x&7)<<3)
1603             h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
1604             h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1605             h->field_scan8x8[i]        = T(field_scan8x8[i]);
1606             h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
1607 #undef T
1608         }
1609     }
1610     if(h->sps.transform_bypass){ //FIXME same ugly
1611         h->zigzag_scan_q0          = zigzag_scan;
1612         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
1613         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1614         h->field_scan_q0           = field_scan;
1615         h->field_scan8x8_q0        = field_scan8x8;
1616         h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
1617     }else{
1618         h->zigzag_scan_q0          = h->zigzag_scan;
1619         h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
1620         h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1621         h->field_scan_q0           = h->field_scan;
1622         h->field_scan8x8_q0        = h->field_scan8x8;
1623         h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
1624     }
1625 }
1626
1627 static void field_end(H264Context *h){
1628     MpegEncContext * const s = &h->s;
1629     AVCodecContext * const avctx= s->avctx;
1630     s->mb_y= 0;
1631
1632     s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1633     s->current_picture_ptr->pict_type= s->pict_type;
1634
1635     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1636         ff_vdpau_h264_set_reference_frames(s);
1637
1638     if(!s->dropable) {
1639         ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1640         h->prev_poc_msb= h->poc_msb;
1641         h->prev_poc_lsb= h->poc_lsb;
1642     }
1643     h->prev_frame_num_offset= h->frame_num_offset;
1644     h->prev_frame_num= h->frame_num;
1645
1646     if (avctx->hwaccel) {
1647         if (avctx->hwaccel->end_frame(avctx) < 0)
1648             av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1649     }
1650
1651     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1652         ff_vdpau_h264_picture_complete(s);
1653
1654     /*
1655      * FIXME: Error handling code does not seem to support interlaced
1656      * when slices span multiple rows
1657      * The ff_er_add_slice calls don't work right for bottom
1658      * fields; they cause massive erroneous error concealing
1659      * Error marking covers both fields (top and bottom).
1660      * This causes a mismatched s->error_count
1661      * and a bad error table. Further, the error count goes to
1662      * INT_MAX when called for bottom field, because mb_y is
1663      * past end by one (callers fault) and resync_mb_y != 0
1664      * causes problems for the first MB line, too.
1665      */
1666     if (!FIELD_PICTURE)
1667         ff_er_frame_end(s);
1668
1669     MPV_frame_end(s);
1670
1671     h->current_slice=0;
1672 }
1673
1674 /**
1675  * Replicates H264 "master" context to thread contexts.
1676  */
1677 static void clone_slice(H264Context *dst, H264Context *src)
1678 {
1679     memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
1680     dst->s.current_picture_ptr  = src->s.current_picture_ptr;
1681     dst->s.current_picture      = src->s.current_picture;
1682     dst->s.linesize             = src->s.linesize;
1683     dst->s.uvlinesize           = src->s.uvlinesize;
1684     dst->s.first_field          = src->s.first_field;
1685
1686     dst->prev_poc_msb           = src->prev_poc_msb;
1687     dst->prev_poc_lsb           = src->prev_poc_lsb;
1688     dst->prev_frame_num_offset  = src->prev_frame_num_offset;
1689     dst->prev_frame_num         = src->prev_frame_num;
1690     dst->short_ref_count        = src->short_ref_count;
1691
1692     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
1693     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
1694     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1695     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
1696
1697     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
1698     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
1699 }
1700
1701 /**
1702  * decodes a slice header.
1703  * This will also call MPV_common_init() and frame_start() as needed.
1704  *
1705  * @param h h264context
1706  * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1707  *
1708  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1709  */
1710 static int decode_slice_header(H264Context *h, H264Context *h0){
1711     MpegEncContext * const s = &h->s;
1712     MpegEncContext * const s0 = &h0->s;
1713     unsigned int first_mb_in_slice;
1714     unsigned int pps_id;
1715     int num_ref_idx_active_override_flag;
1716     unsigned int slice_type, tmp, i, j;
1717     int default_ref_list_done = 0;
1718     int last_pic_structure;
1719
1720     s->dropable= h->nal_ref_idc == 0;
1721
1722     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1723         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1724         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1725     }else{
1726         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1727         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1728     }
1729
1730     first_mb_in_slice= get_ue_golomb(&s->gb);
1731
1732     if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1733         if(h0->current_slice && FIELD_PICTURE){
1734             field_end(h);
1735         }
1736
1737         h0->current_slice = 0;
1738         if (!s0->first_field)
1739             s->current_picture_ptr= NULL;
1740     }
1741
1742     slice_type= get_ue_golomb_31(&s->gb);
1743     if(slice_type > 9){
1744         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);
1745         return -1;
1746     }
1747     if(slice_type > 4){
1748         slice_type -= 5;
1749         h->slice_type_fixed=1;
1750     }else
1751         h->slice_type_fixed=0;
1752
1753     slice_type= golomb_to_pict_type[ slice_type ];
1754     if (slice_type == FF_I_TYPE
1755         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
1756         default_ref_list_done = 1;
1757     }
1758     h->slice_type= slice_type;
1759     h->slice_type_nos= slice_type & 3;
1760
1761     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
1762
1763     pps_id= get_ue_golomb(&s->gb);
1764     if(pps_id>=MAX_PPS_COUNT){
1765         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
1766         return -1;
1767     }
1768     if(!h0->pps_buffers[pps_id]) {
1769         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
1770         return -1;
1771     }
1772     h->pps= *h0->pps_buffers[pps_id];
1773
1774     if(!h0->sps_buffers[h->pps.sps_id]) {
1775         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
1776         return -1;
1777     }
1778     h->sps = *h0->sps_buffers[h->pps.sps_id];
1779
1780     s->avctx->profile = h->sps.profile_idc;
1781     s->avctx->level   = h->sps.level_idc;
1782     s->avctx->refs= h->sps.ref_frame_count;
1783
1784     if(h == h0 && h->dequant_coeff_pps != pps_id){
1785         h->dequant_coeff_pps = pps_id;
1786         init_dequant_tables(h);
1787     }
1788
1789     s->mb_width= h->sps.mb_width;
1790     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1791
1792     h->b_stride=  s->mb_width*4;
1793
1794     s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
1795     if(h->sps.frame_mbs_only_flag)
1796         s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
1797     else
1798         s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
1799
1800     if (s->context_initialized
1801         && (   s->width != s->avctx->width || s->height != s->avctx->height
1802             || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
1803         if(h != h0)
1804             return -1;   // width / height changed during parallelized decoding
1805         free_tables(h);
1806         flush_dpb(s->avctx);
1807         MPV_common_end(s);
1808     }
1809     if (!s->context_initialized) {
1810         if(h != h0)
1811             return -1;  // we cant (re-)initialize context during parallel decoding
1812
1813         avcodec_set_dimensions(s->avctx, s->width, s->height);
1814         s->avctx->sample_aspect_ratio= h->sps.sar;
1815         if(!s->avctx->sample_aspect_ratio.den)
1816             s->avctx->sample_aspect_ratio.den = 1;
1817
1818         if(h->sps.video_signal_type_present_flag){
1819             s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1820             if(h->sps.colour_description_present_flag){
1821                 s->avctx->color_primaries = h->sps.color_primaries;
1822                 s->avctx->color_trc       = h->sps.color_trc;
1823                 s->avctx->colorspace      = h->sps.colorspace;
1824             }
1825         }
1826
1827         if(h->sps.timing_info_present_flag){
1828             int64_t den= h->sps.time_scale;
1829             if(h->x264_build < 44U)
1830                 den *= 2;
1831             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
1832                       h->sps.num_units_in_tick, den, 1<<30);
1833         }
1834         s->avctx->pix_fmt = s->avctx->get_format(s->avctx, s->avctx->codec->pix_fmts);
1835         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1836
1837         if (MPV_common_init(s) < 0)
1838             return -1;
1839         s->first_field = 0;
1840         h->prev_interlaced_frame = 1;
1841
1842         init_scan_tables(h);
1843         ff_h264_alloc_tables(h);
1844
1845         for(i = 1; i < s->avctx->thread_count; i++) {
1846             H264Context *c;
1847             c = h->thread_context[i] = av_malloc(sizeof(H264Context));
1848             memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
1849             memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
1850             c->h264dsp = h->h264dsp;
1851             c->sps = h->sps;
1852             c->pps = h->pps;
1853             init_scan_tables(c);
1854             clone_tables(c, h, i);
1855         }
1856
1857         for(i = 0; i < s->avctx->thread_count; i++)
1858             if(context_init(h->thread_context[i]) < 0)
1859                 return -1;
1860     }
1861
1862     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1863
1864     h->mb_mbaff = 0;
1865     h->mb_aff_frame = 0;
1866     last_pic_structure = s0->picture_structure;
1867     if(h->sps.frame_mbs_only_flag){
1868         s->picture_structure= PICT_FRAME;
1869     }else{
1870         if(get_bits1(&s->gb)) { //field_pic_flag
1871             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
1872         } else {
1873             s->picture_structure= PICT_FRAME;
1874             h->mb_aff_frame = h->sps.mb_aff;
1875         }
1876     }
1877     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
1878
1879     if(h0->current_slice == 0){
1880         while(h->frame_num !=  h->prev_frame_num &&
1881               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
1882             av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
1883             if (ff_h264_frame_start(h) < 0)
1884                 return -1;
1885             h->prev_frame_num++;
1886             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
1887             s->current_picture_ptr->frame_num= h->prev_frame_num;
1888             ff_h264_execute_ref_pic_marking(h, NULL, 0);
1889         }
1890
1891         /* See if we have a decoded first field looking for a pair... */
1892         if (s0->first_field) {
1893             assert(s0->current_picture_ptr);
1894             assert(s0->current_picture_ptr->data[0]);
1895             assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
1896
1897             /* figure out if we have a complementary field pair */
1898             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
1899                 /*
1900                  * Previous field is unmatched. Don't display it, but let it
1901                  * remain for reference if marked as such.
1902                  */
1903                 s0->current_picture_ptr = NULL;
1904                 s0->first_field = FIELD_PICTURE;
1905
1906             } else {
1907                 if (h->nal_ref_idc &&
1908                         s0->current_picture_ptr->reference &&
1909                         s0->current_picture_ptr->frame_num != h->frame_num) {
1910                     /*
1911                      * This and previous field were reference, but had
1912                      * different frame_nums. Consider this field first in
1913                      * pair. Throw away previous field except for reference
1914                      * purposes.
1915                      */
1916                     s0->first_field = 1;
1917                     s0->current_picture_ptr = NULL;
1918
1919                 } else {
1920                     /* Second field in complementary pair */
1921                     s0->first_field = 0;
1922                 }
1923             }
1924
1925         } else {
1926             /* Frame or first field in a potentially complementary pair */
1927             assert(!s0->current_picture_ptr);
1928             s0->first_field = FIELD_PICTURE;
1929         }
1930
1931         if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
1932             s0->first_field = 0;
1933             return -1;
1934         }
1935     }
1936     if(h != h0)
1937         clone_slice(h, h0);
1938
1939     s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
1940
1941     assert(s->mb_num == s->mb_width * s->mb_height);
1942     if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
1943        first_mb_in_slice                    >= s->mb_num){
1944         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1945         return -1;
1946     }
1947     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
1948     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
1949     if (s->picture_structure == PICT_BOTTOM_FIELD)
1950         s->resync_mb_y = s->mb_y = s->mb_y + 1;
1951     assert(s->mb_y < s->mb_height);
1952
1953     if(s->picture_structure==PICT_FRAME){
1954         h->curr_pic_num=   h->frame_num;
1955         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
1956     }else{
1957         h->curr_pic_num= 2*h->frame_num + 1;
1958         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
1959     }
1960
1961     if(h->nal_unit_type == NAL_IDR_SLICE){
1962         get_ue_golomb(&s->gb); /* idr_pic_id */
1963     }
1964
1965     if(h->sps.poc_type==0){
1966         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
1967
1968         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
1969             h->delta_poc_bottom= get_se_golomb(&s->gb);
1970         }
1971     }
1972
1973     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
1974         h->delta_poc[0]= get_se_golomb(&s->gb);
1975
1976         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
1977             h->delta_poc[1]= get_se_golomb(&s->gb);
1978     }
1979
1980     init_poc(h);
1981
1982     if(h->pps.redundant_pic_cnt_present){
1983         h->redundant_pic_count= get_ue_golomb(&s->gb);
1984     }
1985
1986     //set defaults, might be overridden a few lines later
1987     h->ref_count[0]= h->pps.ref_count[0];
1988     h->ref_count[1]= h->pps.ref_count[1];
1989
1990     if(h->slice_type_nos != FF_I_TYPE){
1991         if(h->slice_type_nos == FF_B_TYPE){
1992             h->direct_spatial_mv_pred= get_bits1(&s->gb);
1993         }
1994         num_ref_idx_active_override_flag= get_bits1(&s->gb);
1995
1996         if(num_ref_idx_active_override_flag){
1997             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
1998             if(h->slice_type_nos==FF_B_TYPE)
1999                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2000
2001             if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
2002                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
2003                 h->ref_count[0]= h->ref_count[1]= 1;
2004                 return -1;
2005             }
2006         }
2007         if(h->slice_type_nos == FF_B_TYPE)
2008             h->list_count= 2;
2009         else
2010             h->list_count= 1;
2011     }else
2012         h->list_count= 0;
2013
2014     if(!default_ref_list_done){
2015         ff_h264_fill_default_ref_list(h);
2016     }
2017
2018     if(h->slice_type_nos!=FF_I_TYPE && ff_h264_decode_ref_pic_list_reordering(h) < 0)
2019         return -1;
2020
2021     if(h->slice_type_nos!=FF_I_TYPE){
2022         s->last_picture_ptr= &h->ref_list[0][0];
2023         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
2024     }
2025     if(h->slice_type_nos==FF_B_TYPE){
2026         s->next_picture_ptr= &h->ref_list[1][0];
2027         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
2028     }
2029
2030     if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
2031        ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
2032         pred_weight_table(h);
2033     else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE){
2034         implicit_weight_table(h, -1);
2035     }else {
2036         h->use_weight = 0;
2037         for (i = 0; i < 2; i++) {
2038             h->luma_weight_flag[i]   = 0;
2039             h->chroma_weight_flag[i] = 0;
2040         }
2041     }
2042
2043     if(h->nal_ref_idc)
2044         ff_h264_decode_ref_pic_marking(h0, &s->gb);
2045
2046     if(FRAME_MBAFF){
2047         ff_h264_fill_mbaff_ref_list(h);
2048
2049         if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE){
2050             implicit_weight_table(h, 0);
2051             implicit_weight_table(h, 1);
2052         }
2053     }
2054
2055     if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
2056         ff_h264_direct_dist_scale_factor(h);
2057     ff_h264_direct_ref_list_init(h);
2058
2059     if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
2060         tmp = get_ue_golomb_31(&s->gb);
2061         if(tmp > 2){
2062             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2063             return -1;
2064         }
2065         h->cabac_init_idc= tmp;
2066     }
2067
2068     h->last_qscale_diff = 0;
2069     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2070     if(tmp>51){
2071         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2072         return -1;
2073     }
2074     s->qscale= tmp;
2075     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2076     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2077     //FIXME qscale / qp ... stuff
2078     if(h->slice_type == FF_SP_TYPE){
2079         get_bits1(&s->gb); /* sp_for_switch_flag */
2080     }
2081     if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
2082         get_se_golomb(&s->gb); /* slice_qs_delta */
2083     }
2084
2085     h->deblocking_filter = 1;
2086     h->slice_alpha_c0_offset = 52;
2087     h->slice_beta_offset = 52;
2088     if( h->pps.deblocking_filter_parameters_present ) {
2089         tmp= get_ue_golomb_31(&s->gb);
2090         if(tmp > 2){
2091             av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2092             return -1;
2093         }
2094         h->deblocking_filter= tmp;
2095         if(h->deblocking_filter < 2)
2096             h->deblocking_filter^= 1; // 1<->0
2097
2098         if( h->deblocking_filter ) {
2099             h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2100             h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
2101             if(   h->slice_alpha_c0_offset > 104U
2102                || h->slice_beta_offset     > 104U){
2103                 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);
2104                 return -1;
2105             }
2106         }
2107     }
2108
2109     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
2110        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
2111        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
2112        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2113         h->deblocking_filter= 0;
2114
2115     if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2116         if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2117             /* Cheat slightly for speed:
2118                Do not bother to deblock across slices. */
2119             h->deblocking_filter = 2;
2120         } else {
2121             h0->max_contexts = 1;
2122             if(!h0->single_decode_warning) {
2123                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2124                 h0->single_decode_warning = 1;
2125             }
2126             if(h != h0)
2127                 return 1; // deblocking switched inside frame
2128         }
2129     }
2130     h->qp_thresh= 15 + 52 - FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1]);
2131
2132 #if 0 //FMO
2133     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2134         slice_group_change_cycle= get_bits(&s->gb, ?);
2135 #endif
2136
2137     h0->last_slice_type = slice_type;
2138     h->slice_num = ++h0->current_slice;
2139     if(h->slice_num >= MAX_SLICES){
2140         av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2141     }
2142
2143     for(j=0; j<2; j++){
2144         int id_list[16];
2145         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2146         for(i=0; i<16; i++){
2147             id_list[i]= 60;
2148             if(h->ref_list[j][i].data[0]){
2149                 int k;
2150                 uint8_t *base= h->ref_list[j][i].base[0];
2151                 for(k=0; k<h->short_ref_count; k++)
2152                     if(h->short_ref[k]->base[0] == base){
2153                         id_list[i]= k;
2154                         break;
2155                     }
2156                 for(k=0; k<h->long_ref_count; k++)
2157                     if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2158                         id_list[i]= h->short_ref_count + k;
2159                         break;
2160                     }
2161             }
2162         }
2163
2164         ref2frm[0]=
2165         ref2frm[1]= -1;
2166         for(i=0; i<16; i++)
2167             ref2frm[i+2]= 4*id_list[i]
2168                           +(h->ref_list[j][i].reference&3);
2169         ref2frm[18+0]=
2170         ref2frm[18+1]= -1;
2171         for(i=16; i<48; i++)
2172             ref2frm[i+4]= 4*id_list[(i-16)>>1]
2173                           +(h->ref_list[j][i].reference&3);
2174     }
2175
2176     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
2177     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2178
2179     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2180         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",
2181                h->slice_num,
2182                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2183                first_mb_in_slice,
2184                av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2185                pps_id, h->frame_num,
2186                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2187                h->ref_count[0], h->ref_count[1],
2188                s->qscale,
2189                h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
2190                h->use_weight,
2191                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2192                h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2193                );
2194     }
2195
2196     return 0;
2197 }
2198
2199 int ff_h264_get_slice_type(const H264Context *h)
2200 {
2201     switch (h->slice_type) {
2202     case FF_P_TYPE:  return 0;
2203     case FF_B_TYPE:  return 1;
2204     case FF_I_TYPE:  return 2;
2205     case FF_SP_TYPE: return 3;
2206     case FF_SI_TYPE: return 4;
2207     default:         return -1;
2208     }
2209 }
2210
2211 /**
2212  *
2213  * @return non zero if the loop filter can be skiped
2214  */
2215 static int fill_filter_caches(H264Context *h, int mb_type){
2216     MpegEncContext * const s = &h->s;
2217     const int mb_xy= h->mb_xy;
2218     int top_xy, left_xy[2];
2219     int top_type, left_type[2];
2220
2221     top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
2222
2223     //FIXME deblocking could skip the intra and nnz parts.
2224
2225     /* Wow, what a mess, why didn't they simplify the interlacing & intra
2226      * stuff, I can't imagine that these complex rules are worth it. */
2227
2228     left_xy[1] = left_xy[0] = mb_xy-1;
2229     if(FRAME_MBAFF){
2230         const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
2231         const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
2232         if(s->mb_y&1){
2233             if (left_mb_field_flag != curr_mb_field_flag) {
2234                 left_xy[0] -= s->mb_stride;
2235             }
2236         }else{
2237             if(curr_mb_field_flag){
2238                 top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
2239             }
2240             if (left_mb_field_flag != curr_mb_field_flag) {
2241                 left_xy[1] += s->mb_stride;
2242             }
2243         }
2244     }
2245
2246     h->top_mb_xy = top_xy;
2247     h->left_mb_xy[0] = left_xy[0];
2248     h->left_mb_xy[1] = left_xy[1];
2249     {
2250         //for sufficiently low qp, filtering wouldn't do anything
2251         //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
2252         int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
2253         int qp = s->current_picture.qscale_table[mb_xy];
2254         if(qp <= qp_thresh
2255            && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
2256            && (top_xy   < 0 || ((qp + s->current_picture.qscale_table[top_xy    ] + 1)>>1) <= qp_thresh)){
2257             if(!FRAME_MBAFF)
2258                 return 1;
2259             if(   (left_xy[0]< 0            || ((qp + s->current_picture.qscale_table[left_xy[1]             ] + 1)>>1) <= qp_thresh)
2260                && (top_xy    < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy    -s->mb_stride] + 1)>>1) <= qp_thresh))
2261                 return 1;
2262         }
2263     }
2264
2265     top_type     = s->current_picture.mb_type[top_xy]    ;
2266     left_type[0] = s->current_picture.mb_type[left_xy[0]];
2267     left_type[1] = s->current_picture.mb_type[left_xy[1]];
2268     if(h->deblocking_filter == 2){
2269         if(h->slice_table[top_xy     ] != h->slice_num) top_type= 0;
2270         if(h->slice_table[left_xy[0] ] != h->slice_num) left_type[0]= left_type[1]= 0;
2271     }else{
2272         if(h->slice_table[top_xy     ] == 0xFFFF) top_type= 0;
2273         if(h->slice_table[left_xy[0] ] == 0xFFFF) left_type[0]= left_type[1] =0;
2274     }
2275     h->top_type    = top_type    ;
2276     h->left_type[0]= left_type[0];
2277     h->left_type[1]= left_type[1];
2278
2279     if(IS_INTRA(mb_type))
2280         return 0;
2281
2282     AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
2283     AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
2284     AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]);
2285     AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]);
2286     AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
2287
2288     h->cbp= h->cbp_table[mb_xy];
2289
2290     {
2291         int list;
2292         for(list=0; list<h->list_count; list++){
2293             int8_t *ref;
2294             int y, b_stride;
2295             int16_t (*mv_dst)[2];
2296             int16_t (*mv_src)[2];
2297
2298             if(!USES_LIST(mb_type, list)){
2299                 fill_rectangle(  h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
2300                 AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2301                 AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2302                 AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2303                 AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2304                 continue;
2305             }
2306
2307             ref = &s->current_picture.ref_index[list][4*mb_xy];
2308             {
2309                 int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2310                 AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2311                 AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2312                 ref += 2;
2313                 AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2314                 AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2315             }
2316
2317             b_stride = h->b_stride;
2318             mv_dst   = &h->mv_cache[list][scan8[0]];
2319             mv_src   = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
2320             for(y=0; y<4; y++){
2321                 AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
2322             }
2323
2324         }
2325     }
2326
2327
2328 /*
2329 0 . T T. T T T T
2330 1 L . .L . . . .
2331 2 L . .L . . . .
2332 3 . T TL . . . .
2333 4 L . .L . . . .
2334 5 L . .. . . . .
2335 */
2336 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
2337     if(top_type){
2338         AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
2339     }
2340
2341     if(left_type[0]){
2342         h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
2343         h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
2344         h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
2345         h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
2346     }
2347
2348     // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
2349     if(!CABAC && h->pps.transform_8x8_mode){
2350         if(IS_8x8DCT(top_type)){
2351             h->non_zero_count_cache[4+8*0]=
2352             h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
2353             h->non_zero_count_cache[6+8*0]=
2354             h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
2355         }
2356         if(IS_8x8DCT(left_type[0])){
2357             h->non_zero_count_cache[3+8*1]=
2358             h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
2359         }
2360         if(IS_8x8DCT(left_type[1])){
2361             h->non_zero_count_cache[3+8*3]=
2362             h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
2363         }
2364
2365         if(IS_8x8DCT(mb_type)){
2366             h->non_zero_count_cache[scan8[0   ]]= h->non_zero_count_cache[scan8[1   ]]=
2367             h->non_zero_count_cache[scan8[2   ]]= h->non_zero_count_cache[scan8[3   ]]= h->cbp & 1;
2368
2369             h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
2370             h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
2371
2372             h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
2373             h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
2374
2375             h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
2376             h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
2377         }
2378     }
2379
2380     if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
2381         int list;
2382         for(list=0; list<h->list_count; list++){
2383             if(USES_LIST(top_type, list)){
2384                 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
2385                 const int b8_xy= 4*top_xy + 2;
2386                 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2387                 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
2388                 h->ref_cache[list][scan8[0] + 0 - 1*8]=
2389                 h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
2390                 h->ref_cache[list][scan8[0] + 2 - 1*8]=
2391                 h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
2392             }else{
2393                 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
2394                 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2395             }
2396
2397             if(!IS_INTERLACED(mb_type^left_type[0])){
2398                 if(USES_LIST(left_type[0], list)){
2399                     const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
2400                     const int b8_xy= 4*left_xy[0] + 1;
2401                     int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2402                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
2403                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
2404                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
2405                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
2406                     h->ref_cache[list][scan8[0] - 1 + 0 ]=
2407                     h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]];
2408                     h->ref_cache[list][scan8[0] - 1 +16 ]=
2409                     h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]];
2410                 }else{
2411                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
2412                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
2413                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
2414                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
2415                     h->ref_cache[list][scan8[0] - 1 + 0  ]=
2416                     h->ref_cache[list][scan8[0] - 1 + 8  ]=
2417                     h->ref_cache[list][scan8[0] - 1 + 16 ]=
2418                     h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
2419                 }
2420             }
2421         }
2422     }
2423
2424     return 0;
2425 }
2426
2427 static void loop_filter(H264Context *h){
2428     MpegEncContext * const s = &h->s;
2429     uint8_t  *dest_y, *dest_cb, *dest_cr;
2430     int linesize, uvlinesize, mb_x, mb_y;
2431     const int end_mb_y= s->mb_y + FRAME_MBAFF;
2432     const int old_slice_type= h->slice_type;
2433
2434     if(h->deblocking_filter) {
2435         for(mb_x= 0; mb_x<s->mb_width; mb_x++){
2436             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
2437                 int mb_xy, mb_type;
2438                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2439                 h->slice_num= h->slice_table[mb_xy];
2440                 mb_type= s->current_picture.mb_type[mb_xy];
2441                 h->list_count= h->list_counts[mb_xy];
2442
2443                 if(FRAME_MBAFF)
2444                     h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2445
2446                 s->mb_x= mb_x;
2447                 s->mb_y= mb_y;
2448                 dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2449                 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2450                 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2451                     //FIXME simplify above
2452
2453                 if (MB_FIELD) {
2454                     linesize   = h->mb_linesize   = s->linesize * 2;
2455                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2456                     if(mb_y&1){ //FIXME move out of this function?
2457                         dest_y -= s->linesize*15;
2458                         dest_cb-= s->uvlinesize*7;
2459                         dest_cr-= s->uvlinesize*7;
2460                     }
2461                 } else {
2462                     linesize   = h->mb_linesize   = s->linesize;
2463                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2464                 }
2465                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2466                 if(fill_filter_caches(h, mb_type))
2467                     continue;
2468                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2469                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2470
2471                 if (FRAME_MBAFF) {
2472                     ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2473                 } else {
2474                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2475                 }
2476             }
2477         }
2478     }
2479     h->slice_type= old_slice_type;
2480     s->mb_x= 0;
2481     s->mb_y= end_mb_y - FRAME_MBAFF;
2482     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2483     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2484 }
2485
2486 static void predict_field_decoding_flag(H264Context *h){
2487     MpegEncContext * const s = &h->s;
2488     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2489     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
2490                 ? s->current_picture.mb_type[mb_xy-1]
2491                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
2492                 ? s->current_picture.mb_type[mb_xy-s->mb_stride]
2493                 : 0;
2494     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2495 }
2496
2497 static int decode_slice(struct AVCodecContext *avctx, void *arg){
2498     H264Context *h = *(void**)arg;
2499     MpegEncContext * const s = &h->s;
2500     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2501
2502     s->mb_skip_run= -1;
2503
2504     h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
2505                     (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
2506
2507     if( h->pps.cabac ) {
2508         /* realign */
2509         align_get_bits( &s->gb );
2510
2511         /* init cabac */
2512         ff_init_cabac_states( &h->cabac);
2513         ff_init_cabac_decoder( &h->cabac,
2514                                s->gb.buffer + get_bits_count(&s->gb)/8,
2515                                (get_bits_left(&s->gb) + 7)/8);
2516
2517         ff_h264_init_cabac_states(h);
2518
2519         for(;;){
2520 //START_TIMER
2521             int ret = ff_h264_decode_mb_cabac(h);
2522             int eos;
2523 //STOP_TIMER("decode_mb_cabac")
2524
2525             if(ret>=0) ff_h264_hl_decode_mb(h);
2526
2527             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
2528                 s->mb_y++;
2529
2530                 ret = ff_h264_decode_mb_cabac(h);
2531
2532                 if(ret>=0) ff_h264_hl_decode_mb(h);
2533                 s->mb_y--;
2534             }
2535             eos = get_cabac_terminate( &h->cabac );
2536
2537             if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
2538                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2539                 return 0;
2540             }
2541             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2542                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d, bytestream (%td)\n", s->mb_x, s->mb_y, h->cabac.bytestream_end - h->cabac.bytestream);
2543                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
2544                 return -1;
2545             }
2546
2547             if( ++s->mb_x >= s->mb_width ) {
2548                 s->mb_x = 0;
2549                 loop_filter(h);
2550                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2551                 ++s->mb_y;
2552                 if(FIELD_OR_MBAFF_PICTURE) {
2553                     ++s->mb_y;
2554                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2555                         predict_field_decoding_flag(h);
2556                 }
2557             }
2558
2559             if( eos || s->mb_y >= s->mb_height ) {
2560                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2561                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2562                 return 0;
2563             }
2564         }
2565
2566     } else {
2567         for(;;){
2568             int ret = ff_h264_decode_mb_cavlc(h);
2569
2570             if(ret>=0) ff_h264_hl_decode_mb(h);
2571
2572             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
2573                 s->mb_y++;
2574                 ret = ff_h264_decode_mb_cavlc(h);
2575
2576                 if(ret>=0) ff_h264_hl_decode_mb(h);
2577                 s->mb_y--;
2578             }
2579
2580             if(ret<0){
2581                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2582                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
2583
2584                 return -1;
2585             }
2586
2587             if(++s->mb_x >= s->mb_width){
2588                 s->mb_x=0;
2589                 loop_filter(h);
2590                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2591                 ++s->mb_y;
2592                 if(FIELD_OR_MBAFF_PICTURE) {
2593                     ++s->mb_y;
2594                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2595                         predict_field_decoding_flag(h);
2596                 }
2597                 if(s->mb_y >= s->mb_height){
2598                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2599
2600                     if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
2601                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2602
2603                         return 0;
2604                     }else{
2605                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2606
2607                         return -1;
2608                     }
2609                 }
2610             }
2611
2612             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
2613                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2614                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2615                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2616
2617                     return 0;
2618                 }else{
2619                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
2620
2621                     return -1;
2622                 }
2623             }
2624         }
2625     }
2626
2627 #if 0
2628     for(;s->mb_y < s->mb_height; s->mb_y++){
2629         for(;s->mb_x < s->mb_width; s->mb_x++){
2630             int ret= decode_mb(h);
2631
2632             ff_h264_hl_decode_mb(h);
2633
2634             if(ret<0){
2635                 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2636                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
2637
2638                 return -1;
2639             }
2640
2641             if(++s->mb_x >= s->mb_width){
2642                 s->mb_x=0;
2643                 if(++s->mb_y >= s->mb_height){
2644                     if(get_bits_count(s->gb) == s->gb.size_in_bits){
2645                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2646
2647                         return 0;
2648                     }else{
2649                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2650
2651                         return -1;
2652                     }
2653                 }
2654             }
2655
2656             if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2657                 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2658                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2659
2660                     return 0;
2661                 }else{
2662                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
2663
2664                     return -1;
2665                 }
2666             }
2667         }
2668         s->mb_x=0;
2669         ff_draw_horiz_band(s, 16*s->mb_y, 16);
2670     }
2671 #endif
2672     return -1; //not reached
2673 }
2674
2675 /**
2676  * Call decode_slice() for each context.
2677  *
2678  * @param h h264 master context
2679  * @param context_count number of contexts to execute
2680  */
2681 static void execute_decode_slices(H264Context *h, int context_count){
2682     MpegEncContext * const s = &h->s;
2683     AVCodecContext * const avctx= s->avctx;
2684     H264Context *hx;
2685     int i;
2686
2687     if (s->avctx->hwaccel)
2688         return;
2689     if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2690         return;
2691     if(context_count == 1) {
2692         decode_slice(avctx, &h);
2693     } else {
2694         for(i = 1; i < context_count; i++) {
2695             hx = h->thread_context[i];
2696             hx->s.error_recognition = avctx->error_recognition;
2697             hx->s.error_count = 0;
2698         }
2699
2700         avctx->execute(avctx, (void *)decode_slice,
2701                        h->thread_context, NULL, context_count, sizeof(void*));
2702
2703         /* pull back stuff from slices to master context */
2704         hx = h->thread_context[context_count - 1];
2705         s->mb_x = hx->s.mb_x;
2706         s->mb_y = hx->s.mb_y;
2707         s->dropable = hx->s.dropable;
2708         s->picture_structure = hx->s.picture_structure;
2709         for(i = 1; i < context_count; i++)
2710             h->s.error_count += h->thread_context[i]->s.error_count;
2711     }
2712 }
2713
2714
2715 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
2716     MpegEncContext * const s = &h->s;
2717     AVCodecContext * const avctx= s->avctx;
2718     int buf_index=0;
2719     H264Context *hx; ///< thread context
2720     int context_count = 0;
2721     int next_avc= h->is_avc ? 0 : buf_size;
2722
2723     h->max_contexts = avctx->thread_count;
2724 #if 0
2725     int i;
2726     for(i=0; i<50; i++){
2727         av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
2728     }
2729 #endif
2730     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
2731         h->current_slice = 0;
2732         if (!s->first_field)
2733             s->current_picture_ptr= NULL;
2734         ff_h264_reset_sei(h);
2735     }
2736
2737     for(;;){
2738         int consumed;
2739         int dst_length;
2740         int bit_length;
2741         const uint8_t *ptr;
2742         int i, nalsize = 0;
2743         int err;
2744
2745         if(buf_index >= next_avc) {
2746             if(buf_index >= buf_size) break;
2747             nalsize = 0;
2748             for(i = 0; i < h->nal_length_size; i++)
2749                 nalsize = (nalsize << 8) | buf[buf_index++];
2750             if(nalsize <= 1 || nalsize > buf_size - buf_index){
2751                 if(nalsize == 1){
2752                     buf_index++;
2753                     continue;
2754                 }else{
2755                     av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2756                     break;
2757                 }
2758             }
2759             next_avc= buf_index + nalsize;
2760         } else {
2761             // start code prefix search
2762             for(; buf_index + 3 < next_avc; buf_index++){
2763                 // This should always succeed in the first iteration.
2764                 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2765                     break;
2766             }
2767
2768             if(buf_index+3 >= buf_size) break;
2769
2770             buf_index+=3;
2771             if(buf_index >= next_avc) continue;
2772         }
2773
2774         hx = h->thread_context[context_count];
2775
2776         ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
2777         if (ptr==NULL || dst_length < 0){
2778             return -1;
2779         }
2780         i= buf_index + consumed;
2781         if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
2782            buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
2783             s->workaround_bugs |= FF_BUG_TRUNCATED;
2784
2785         if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
2786         while(ptr[dst_length - 1] == 0 && dst_length > 0)
2787             dst_length--;
2788         }
2789         bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
2790
2791         if(s->avctx->debug&FF_DEBUG_STARTCODE){
2792             av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d/%d length %d\n", hx->nal_unit_type, buf_index, buf_size, dst_length);
2793         }
2794
2795         if (h->is_avc && (nalsize != consumed) && nalsize){
2796             av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
2797         }
2798
2799         buf_index += consumed;
2800
2801         if(  (s->hurry_up == 1 && h->nal_ref_idc  == 0) //FIXME do not discard SEI id
2802            ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0))
2803             continue;
2804
2805       again:
2806         err = 0;
2807         switch(hx->nal_unit_type){
2808         case NAL_IDR_SLICE:
2809             if (h->nal_unit_type != NAL_IDR_SLICE) {
2810                 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2811                 return -1;
2812             }
2813             idr(h); //FIXME ensure we don't loose some frames if there is reordering
2814         case NAL_SLICE:
2815             init_get_bits(&hx->s.gb, ptr, bit_length);
2816             hx->intra_gb_ptr=
2817             hx->inter_gb_ptr= &hx->s.gb;
2818             hx->s.data_partitioning = 0;
2819
2820             if((err = decode_slice_header(hx, h)))
2821                break;
2822
2823             if (h->current_slice == 1) {
2824                 if (s->avctx->hwaccel && s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
2825                     return -1;
2826                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2827                     ff_vdpau_h264_picture_start(s);
2828             }
2829
2830             s->current_picture_ptr->key_frame |=
2831                     (hx->nal_unit_type == NAL_IDR_SLICE) ||
2832                     (h->sei_recovery_frame_cnt >= 0);
2833             if(hx->redundant_pic_count==0 && hx->s.hurry_up < 5
2834                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2835                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2836                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2837                && avctx->skip_frame < AVDISCARD_ALL){
2838                 if(avctx->hwaccel) {
2839                     if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
2840                         return -1;
2841                 }else
2842                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
2843                     static const uint8_t start_code[] = {0x00, 0x00, 0x01};
2844                     ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
2845                     ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
2846                 }else
2847                     context_count++;
2848             }
2849             break;
2850         case NAL_DPA:
2851             init_get_bits(&hx->s.gb, ptr, bit_length);
2852             hx->intra_gb_ptr=
2853             hx->inter_gb_ptr= NULL;
2854
2855             if ((err = decode_slice_header(hx, h)) < 0)
2856                 break;
2857
2858             hx->s.data_partitioning = 1;
2859
2860             break;
2861         case NAL_DPB:
2862             init_get_bits(&hx->intra_gb, ptr, bit_length);
2863             hx->intra_gb_ptr= &hx->intra_gb;
2864             break;
2865         case NAL_DPC:
2866             init_get_bits(&hx->inter_gb, ptr, bit_length);
2867             hx->inter_gb_ptr= &hx->inter_gb;
2868
2869             if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
2870                && s->context_initialized
2871                && s->hurry_up < 5
2872                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2873                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2874                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2875                && avctx->skip_frame < AVDISCARD_ALL)
2876                 context_count++;
2877             break;
2878         case NAL_SEI:
2879             init_get_bits(&s->gb, ptr, bit_length);
2880             ff_h264_decode_sei(h);
2881             break;
2882         case NAL_SPS:
2883             init_get_bits(&s->gb, ptr, bit_length);
2884             ff_h264_decode_seq_parameter_set(h);
2885
2886             if(s->flags& CODEC_FLAG_LOW_DELAY)
2887                 s->low_delay=1;
2888
2889             if(avctx->has_b_frames < 2)
2890                 avctx->has_b_frames= !s->low_delay;
2891             break;
2892         case NAL_PPS:
2893             init_get_bits(&s->gb, ptr, bit_length);
2894
2895             ff_h264_decode_picture_parameter_set(h, bit_length);
2896
2897             break;
2898         case NAL_AUD:
2899         case NAL_END_SEQUENCE:
2900         case NAL_END_STREAM:
2901         case NAL_FILLER_DATA:
2902         case NAL_SPS_EXT:
2903         case NAL_AUXILIARY_SLICE:
2904             break;
2905         default:
2906             av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
2907         }
2908
2909         if(context_count == h->max_contexts) {
2910             execute_decode_slices(h, context_count);
2911             context_count = 0;
2912         }
2913
2914         if (err < 0)
2915             av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
2916         else if(err == 1) {
2917             /* Slice could not be decoded in parallel mode, copy down
2918              * NAL unit stuff to context 0 and restart. Note that
2919              * rbsp_buffer is not transferred, but since we no longer
2920              * run in parallel mode this should not be an issue. */
2921             h->nal_unit_type = hx->nal_unit_type;
2922             h->nal_ref_idc   = hx->nal_ref_idc;
2923             hx = h;
2924             goto again;
2925         }
2926     }
2927     if(context_count)
2928         execute_decode_slices(h, context_count);
2929     return buf_index;
2930 }
2931
2932 /**
2933  * returns the number of bytes consumed for building the current frame
2934  */
2935 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
2936         if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
2937         if(pos+10>buf_size) pos=buf_size; // oops ;)
2938
2939         return pos;
2940 }
2941
2942 static int decode_frame(AVCodecContext *avctx,
2943                              void *data, int *data_size,
2944                              AVPacket *avpkt)
2945 {
2946     const uint8_t *buf = avpkt->data;
2947     int buf_size = avpkt->size;
2948     H264Context *h = avctx->priv_data;
2949     MpegEncContext *s = &h->s;
2950     AVFrame *pict = data;
2951     int buf_index;
2952
2953     s->flags= avctx->flags;
2954     s->flags2= avctx->flags2;
2955
2956    /* end of stream, output what is still in the buffers */
2957     if (buf_size == 0) {
2958         Picture *out;
2959         int i, out_idx;
2960
2961 //FIXME factorize this with the output code below
2962         out = h->delayed_pic[0];
2963         out_idx = 0;
2964         for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2965             if(h->delayed_pic[i]->poc < out->poc){
2966                 out = h->delayed_pic[i];
2967                 out_idx = i;
2968             }
2969
2970         for(i=out_idx; h->delayed_pic[i]; i++)
2971             h->delayed_pic[i] = h->delayed_pic[i+1];
2972
2973         if(out){
2974             *data_size = sizeof(AVFrame);
2975             *pict= *(AVFrame*)out;
2976         }
2977
2978         return 0;
2979     }
2980
2981     buf_index=decode_nal_units(h, buf, buf_size);
2982     if(buf_index < 0)
2983         return -1;
2984
2985     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
2986         if (avctx->skip_frame >= AVDISCARD_NONREF || s->hurry_up) return 0;
2987         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
2988         return -1;
2989     }
2990
2991     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
2992         Picture *out = s->current_picture_ptr;
2993         Picture *cur = s->current_picture_ptr;
2994         int i, pics, out_of_order, out_idx;
2995
2996         field_end(h);
2997
2998         if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
2999             /* Wait for second field. */
3000             *data_size = 0;
3001
3002         } else {
3003             cur->interlaced_frame = 0;
3004             cur->repeat_pict = 0;
3005
3006             /* Signal interlacing information externally. */
3007             /* Prioritize picture timing SEI information over used decoding process if it exists. */
3008
3009             if(h->sps.pic_struct_present_flag){
3010                 switch (h->sei_pic_struct)
3011                 {
3012                 case SEI_PIC_STRUCT_FRAME:
3013                     break;
3014                 case SEI_PIC_STRUCT_TOP_FIELD:
3015                 case SEI_PIC_STRUCT_BOTTOM_FIELD:
3016                     cur->interlaced_frame = 1;
3017                     break;
3018                 case SEI_PIC_STRUCT_TOP_BOTTOM:
3019                 case SEI_PIC_STRUCT_BOTTOM_TOP:
3020                     if (FIELD_OR_MBAFF_PICTURE)
3021                         cur->interlaced_frame = 1;
3022                     else
3023                         // try to flag soft telecine progressive
3024                         cur->interlaced_frame = h->prev_interlaced_frame;
3025                     break;
3026                 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
3027                 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
3028                     // Signal the possibility of telecined film externally (pic_struct 5,6)
3029                     // From these hints, let the applications decide if they apply deinterlacing.
3030                     cur->repeat_pict = 1;
3031                     break;
3032                 case SEI_PIC_STRUCT_FRAME_DOUBLING:
3033                     // Force progressive here, as doubling interlaced frame is a bad idea.
3034                     cur->repeat_pict = 2;
3035                     break;
3036                 case SEI_PIC_STRUCT_FRAME_TRIPLING:
3037                     cur->repeat_pict = 4;
3038                     break;
3039                 }
3040
3041                 if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
3042                     cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
3043             }else{
3044                 /* Derive interlacing flag from used decoding process. */
3045                 cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
3046             }
3047             h->prev_interlaced_frame = cur->interlaced_frame;
3048
3049             if (cur->field_poc[0] != cur->field_poc[1]){
3050                 /* Derive top_field_first from field pocs. */
3051                 cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
3052             }else{
3053                 if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
3054                     /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
3055                     if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
3056                       || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
3057                         cur->top_field_first = 1;
3058                     else
3059                         cur->top_field_first = 0;
3060                 }else{
3061                     /* Most likely progressive */
3062                     cur->top_field_first = 0;
3063                 }
3064             }
3065
3066         //FIXME do something with unavailable reference frames
3067
3068             /* Sort B-frames into display order */
3069
3070             if(h->sps.bitstream_restriction_flag
3071                && s->avctx->has_b_frames < h->sps.num_reorder_frames){
3072                 s->avctx->has_b_frames = h->sps.num_reorder_frames;
3073                 s->low_delay = 0;
3074             }
3075
3076             if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
3077                && !h->sps.bitstream_restriction_flag){
3078                 s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
3079                 s->low_delay= 0;
3080             }
3081
3082             pics = 0;
3083             while(h->delayed_pic[pics]) pics++;
3084
3085             assert(pics <= MAX_DELAYED_PIC_COUNT);
3086
3087             h->delayed_pic[pics++] = cur;
3088             if(cur->reference == 0)
3089                 cur->reference = DELAYED_PIC_REF;
3090
3091             out = h->delayed_pic[0];
3092             out_idx = 0;
3093             for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
3094                 if(h->delayed_pic[i]->poc < out->poc){
3095                     out = h->delayed_pic[i];
3096                     out_idx = i;
3097                 }
3098             if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
3099                 h->outputed_poc= INT_MIN;
3100             out_of_order = out->poc < h->outputed_poc;
3101
3102             if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
3103                 { }
3104             else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
3105                || (s->low_delay &&
3106                 ((h->outputed_poc != INT_MIN && out->poc > h->outputed_poc + 2)
3107                  || cur->pict_type == FF_B_TYPE)))
3108             {
3109                 s->low_delay = 0;
3110                 s->avctx->has_b_frames++;
3111             }
3112
3113             if(out_of_order || pics > s->avctx->has_b_frames){
3114                 out->reference &= ~DELAYED_PIC_REF;
3115                 for(i=out_idx; h->delayed_pic[i]; i++)
3116                     h->delayed_pic[i] = h->delayed_pic[i+1];
3117             }
3118             if(!out_of_order && pics > s->avctx->has_b_frames){
3119                 *data_size = sizeof(AVFrame);
3120
3121                 if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
3122                     h->outputed_poc = INT_MIN;
3123                 } else
3124                     h->outputed_poc = out->poc;
3125                 *pict= *(AVFrame*)out;
3126             }else{
3127                 av_log(avctx, AV_LOG_DEBUG, "no picture\n");
3128             }
3129         }
3130     }
3131
3132     assert(pict->data[0] || !*data_size);
3133     ff_print_debug_info(s, pict);
3134 //printf("out %d\n", (int)pict->data[0]);
3135
3136     return get_consumed_bytes(s, buf_index, buf_size);
3137 }
3138 #if 0
3139 static inline void fill_mb_avail(H264Context *h){
3140     MpegEncContext * const s = &h->s;
3141     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3142
3143     if(s->mb_y){
3144         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
3145         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
3146         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
3147     }else{
3148         h->mb_avail[0]=
3149         h->mb_avail[1]=
3150         h->mb_avail[2]= 0;
3151     }
3152     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
3153     h->mb_avail[4]= 1; //FIXME move out
3154     h->mb_avail[5]= 0; //FIXME move out
3155 }
3156 #endif
3157
3158 #ifdef TEST
3159 #undef printf
3160 #undef random
3161 #define COUNT 8000
3162 #define SIZE (COUNT*40)
3163 int main(void){
3164     int i;
3165     uint8_t temp[SIZE];
3166     PutBitContext pb;
3167     GetBitContext gb;
3168 //    int int_temp[10000];
3169     DSPContext dsp;
3170     AVCodecContext avctx;
3171
3172     dsputil_init(&dsp, &avctx);
3173
3174     init_put_bits(&pb, temp, SIZE);
3175     printf("testing unsigned exp golomb\n");
3176     for(i=0; i<COUNT; i++){
3177         START_TIMER
3178         set_ue_golomb(&pb, i);
3179         STOP_TIMER("set_ue_golomb");
3180     }
3181     flush_put_bits(&pb);
3182
3183     init_get_bits(&gb, temp, 8*SIZE);
3184     for(i=0; i<COUNT; i++){
3185         int j, s;
3186
3187         s= show_bits(&gb, 24);
3188
3189         START_TIMER
3190         j= get_ue_golomb(&gb);
3191         if(j != i){
3192             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
3193 //            return -1;
3194         }
3195         STOP_TIMER("get_ue_golomb");
3196     }
3197
3198
3199     init_put_bits(&pb, temp, SIZE);
3200     printf("testing signed exp golomb\n");
3201     for(i=0; i<COUNT; i++){
3202         START_TIMER
3203         set_se_golomb(&pb, i - COUNT/2);
3204         STOP_TIMER("set_se_golomb");
3205     }
3206     flush_put_bits(&pb);
3207
3208     init_get_bits(&gb, temp, 8*SIZE);
3209     for(i=0; i<COUNT; i++){
3210         int j, s;
3211
3212         s= show_bits(&gb, 24);
3213
3214         START_TIMER
3215         j= get_se_golomb(&gb);
3216         if(j != i - COUNT/2){
3217             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
3218 //            return -1;
3219         }
3220         STOP_TIMER("get_se_golomb");
3221     }
3222
3223 #if 0
3224     printf("testing 4x4 (I)DCT\n");
3225
3226     DCTELEM block[16];
3227     uint8_t src[16], ref[16];
3228     uint64_t error= 0, max_error=0;
3229
3230     for(i=0; i<COUNT; i++){
3231         int j;
3232 //        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
3233         for(j=0; j<16; j++){
3234             ref[j]= random()%255;
3235             src[j]= random()%255;
3236         }
3237
3238         h264_diff_dct_c(block, src, ref, 4);
3239
3240         //normalize
3241         for(j=0; j<16; j++){
3242 //            printf("%d ", block[j]);
3243             block[j]= block[j]*4;
3244             if(j&1) block[j]= (block[j]*4 + 2)/5;
3245             if(j&4) block[j]= (block[j]*4 + 2)/5;
3246         }
3247 //        printf("\n");
3248
3249         h->h264dsp.h264_idct_add(ref, block, 4);
3250 /*        for(j=0; j<16; j++){
3251             printf("%d ", ref[j]);
3252         }
3253         printf("\n");*/
3254
3255         for(j=0; j<16; j++){
3256             int diff= FFABS(src[j] - ref[j]);
3257
3258             error+= diff*diff;
3259             max_error= FFMAX(max_error, diff);
3260         }
3261     }
3262     printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
3263     printf("testing quantizer\n");
3264     for(qp=0; qp<52; qp++){
3265         for(i=0; i<16; i++)
3266             src1_block[i]= src2_block[i]= random()%255;
3267
3268     }
3269     printf("Testing NAL layer\n");
3270
3271     uint8_t bitstream[COUNT];
3272     uint8_t nal[COUNT*2];
3273     H264Context h;
3274     memset(&h, 0, sizeof(H264Context));
3275
3276     for(i=0; i<COUNT; i++){
3277         int zeros= i;
3278         int nal_length;
3279         int consumed;
3280         int out_length;
3281         uint8_t *out;
3282         int j;
3283
3284         for(j=0; j<COUNT; j++){
3285             bitstream[j]= (random() % 255) + 1;
3286         }
3287
3288         for(j=0; j<zeros; j++){
3289             int pos= random() % COUNT;
3290             while(bitstream[pos] == 0){
3291                 pos++;
3292                 pos %= COUNT;
3293             }
3294             bitstream[pos]=0;
3295         }
3296
3297         START_TIMER
3298
3299         nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
3300         if(nal_length<0){
3301             printf("encoding failed\n");
3302             return -1;
3303         }
3304
3305         out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
3306
3307         STOP_TIMER("NAL")
3308
3309         if(out_length != COUNT){
3310             printf("incorrect length %d %d\n", out_length, COUNT);
3311             return -1;
3312         }
3313
3314         if(consumed != nal_length){
3315             printf("incorrect consumed length %d %d\n", nal_length, consumed);
3316             return -1;
3317         }
3318
3319         if(memcmp(bitstream, out, COUNT)){
3320             printf("mismatch\n");
3321             return -1;
3322         }
3323     }
3324 #endif
3325
3326     printf("Testing RBSP\n");
3327
3328
3329     return 0;
3330 }
3331 #endif /* TEST */
3332
3333
3334 av_cold void ff_h264_free_context(H264Context *h)
3335 {
3336     int i;
3337
3338     free_tables(h); //FIXME cleanup init stuff perhaps
3339
3340     for(i = 0; i < MAX_SPS_COUNT; i++)
3341         av_freep(h->sps_buffers + i);
3342
3343     for(i = 0; i < MAX_PPS_COUNT; i++)
3344         av_freep(h->pps_buffers + i);
3345 }
3346
3347 av_cold int ff_h264_decode_end(AVCodecContext *avctx)
3348 {
3349     H264Context *h = avctx->priv_data;
3350     MpegEncContext *s = &h->s;
3351
3352     ff_h264_free_context(h);
3353
3354     MPV_common_end(s);
3355
3356 //    memset(h, 0, sizeof(H264Context));
3357
3358     return 0;
3359 }
3360
3361
3362 AVCodec h264_decoder = {
3363     "h264",
3364     AVMEDIA_TYPE_VIDEO,
3365     CODEC_ID_H264,
3366     sizeof(H264Context),
3367     ff_h264_decode_init,
3368     NULL,
3369     ff_h264_decode_end,
3370     decode_frame,
3371     /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3372     .flush= flush_dpb,
3373     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
3374     .pix_fmts= ff_hwaccel_pixfmt_list_420,
3375 };
3376
3377 #if CONFIG_H264_VDPAU_DECODER
3378 AVCodec h264_vdpau_decoder = {
3379     "h264_vdpau",
3380     AVMEDIA_TYPE_VIDEO,
3381     CODEC_ID_H264,
3382     sizeof(H264Context),
3383     ff_h264_decode_init,
3384     NULL,
3385     ff_h264_decode_end,
3386     decode_frame,
3387     CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3388     .flush= flush_dpb,
3389     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
3390     .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
3391 };
3392 #endif