Factorize ff_h264_decode_extradata().
[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
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 int ff_h264_decode_extradata(H264Context *h)
848 {
849     AVCodecContext *avctx = h->s.avctx;
850
851     if(*(char *)avctx->extradata == 1){
852         int i, cnt, nalsize;
853         unsigned char *p = avctx->extradata;
854
855         h->is_avc = 1;
856
857         if(avctx->extradata_size < 7) {
858             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
859             return -1;
860         }
861         /* sps and pps in the avcC always have length coded with 2 bytes,
862            so put a fake nal_length_size = 2 while parsing them */
863         h->nal_length_size = 2;
864         // Decode sps from avcC
865         cnt = *(p+5) & 0x1f; // Number of sps
866         p += 6;
867         for (i = 0; i < cnt; i++) {
868             nalsize = AV_RB16(p) + 2;
869             if(decode_nal_units(h, p, nalsize) < 0) {
870                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
871                 return -1;
872             }
873             p += nalsize;
874         }
875         // Decode pps from avcC
876         cnt = *(p++); // Number of pps
877         for (i = 0; i < cnt; i++) {
878             nalsize = AV_RB16(p) + 2;
879             if(decode_nal_units(h, p, nalsize)  != nalsize) {
880                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
881                 return -1;
882             }
883             p += nalsize;
884         }
885         // Now store right nal length size, that will be use to parse all other nals
886         h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
887     } else {
888         h->is_avc = 0;
889         if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
890             return -1;
891     }
892     return 0;
893 }
894
895 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
896     H264Context *h= avctx->priv_data;
897     MpegEncContext * const s = &h->s;
898
899     MPV_decode_defaults(s);
900
901     s->avctx = avctx;
902     common_init(h);
903
904     s->out_format = FMT_H264;
905     s->workaround_bugs= avctx->workaround_bugs;
906
907     // set defaults
908 //    s->decode_mb= ff_h263_decode_mb;
909     s->quarter_sample = 1;
910     if(!avctx->has_b_frames)
911     s->low_delay= 1;
912
913     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
914
915     ff_h264_decode_init_vlc();
916
917     h->thread_context[0] = h;
918     h->outputed_poc = INT_MIN;
919     h->prev_poc_msb= 1<<16;
920     h->x264_build = -1;
921     ff_h264_reset_sei(h);
922     if(avctx->codec_id == CODEC_ID_H264){
923         if(avctx->ticks_per_frame == 1){
924             s->avctx->time_base.den *=2;
925         }
926         avctx->ticks_per_frame = 2;
927     }
928
929     if(avctx->extradata_size > 0 && avctx->extradata &&
930         ff_h264_decode_extradata(h))
931         return -1;
932
933     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
934         s->avctx->has_b_frames = h->sps.num_reorder_frames;
935         s->low_delay = 0;
936     }
937
938     return 0;
939 }
940
941 int ff_h264_frame_start(H264Context *h){
942     MpegEncContext * const s = &h->s;
943     int i;
944
945     if(MPV_frame_start(s, s->avctx) < 0)
946         return -1;
947     ff_er_frame_start(s);
948     /*
949      * MPV_frame_start uses pict_type to derive key_frame.
950      * This is incorrect for H.264; IDR markings must be used.
951      * Zero here; IDR markings per slice in frame or fields are ORed in later.
952      * See decode_nal_units().
953      */
954     s->current_picture_ptr->key_frame= 0;
955     s->current_picture_ptr->mmco_reset= 0;
956
957     assert(s->linesize && s->uvlinesize);
958
959     for(i=0; i<16; i++){
960         h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
961         h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
962     }
963     for(i=0; i<4; i++){
964         h->block_offset[16+i]=
965         h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
966         h->block_offset[24+16+i]=
967         h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
968     }
969
970     /* can't be in alloc_tables because linesize isn't known there.
971      * FIXME: redo bipred weight to not require extra buffer? */
972     for(i = 0; i < s->avctx->thread_count; i++)
973         if(!h->thread_context[i]->s.obmc_scratchpad)
974             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
975
976     /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
977     memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
978
979 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
980
981     // We mark the current picture as non-reference after allocating it, so
982     // that if we break out due to an error it can be released automatically
983     // in the next MPV_frame_start().
984     // SVQ3 as well as most other codecs have only last/next/current and thus
985     // get released even with set reference, besides SVQ3 and others do not
986     // mark frames as reference later "naturally".
987     if(s->codec_id != CODEC_ID_SVQ3)
988         s->current_picture_ptr->reference= 0;
989
990     s->current_picture_ptr->field_poc[0]=
991     s->current_picture_ptr->field_poc[1]= INT_MAX;
992     assert(s->current_picture_ptr->long_ref==0);
993
994     return 0;
995 }
996
997 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){
998     MpegEncContext * const s = &h->s;
999     uint8_t *top_border;
1000     int top_idx = 1;
1001
1002     src_y  -=   linesize;
1003     src_cb -= uvlinesize;
1004     src_cr -= uvlinesize;
1005
1006     if(!simple && FRAME_MBAFF){
1007         if(s->mb_y&1){
1008             if(!MB_MBAFF){
1009                 top_border = h->top_borders[0][s->mb_x];
1010                 AV_COPY128(top_border, src_y + 15*linesize);
1011                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1012                     AV_COPY64(top_border+16, src_cb+7*uvlinesize);
1013                     AV_COPY64(top_border+24, src_cr+7*uvlinesize);
1014                 }
1015             }
1016         }else if(MB_MBAFF){
1017             top_idx = 0;
1018         }else
1019             return;
1020     }
1021
1022     top_border = h->top_borders[top_idx][s->mb_x];
1023     // There are two lines saved, the line above the the top macroblock of a pair,
1024     // and the line above the bottom macroblock
1025     AV_COPY128(top_border, src_y + 16*linesize);
1026
1027     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1028         AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1029         AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1030     }
1031 }
1032
1033 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){
1034     MpegEncContext * const s = &h->s;
1035     int deblock_left;
1036     int deblock_top;
1037     int top_idx = 1;
1038     uint8_t *top_border_m1;
1039     uint8_t *top_border;
1040
1041     if(!simple && FRAME_MBAFF){
1042         if(s->mb_y&1){
1043             if(!MB_MBAFF)
1044                 return;
1045         }else{
1046             top_idx = MB_MBAFF ? 0 : 1;
1047         }
1048     }
1049
1050     if(h->deblocking_filter == 2) {
1051         deblock_left = h->left_type[0];
1052         deblock_top  = h->top_type;
1053     } else {
1054         deblock_left = (s->mb_x > 0);
1055         deblock_top =  (s->mb_y > !!MB_FIELD);
1056     }
1057
1058     src_y  -=   linesize + 1;
1059     src_cb -= uvlinesize + 1;
1060     src_cr -= uvlinesize + 1;
1061
1062     top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1063     top_border    = h->top_borders[top_idx][s->mb_x];
1064
1065 #define XCHG(a,b,xchg)\
1066 if (xchg) AV_SWAP64(b,a);\
1067 else      AV_COPY64(b,a);
1068
1069     if(deblock_top){
1070         if(deblock_left){
1071             XCHG(top_border_m1+8, src_y -7, 1);
1072         }
1073         XCHG(top_border+0, src_y +1, xchg);
1074         XCHG(top_border+8, src_y +9, 1);
1075         if(s->mb_x+1 < s->mb_width){
1076             XCHG(h->top_borders[top_idx][s->mb_x+1], src_y +17, 1);
1077         }
1078     }
1079
1080     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1081         if(deblock_top){
1082             if(deblock_left){
1083                 XCHG(top_border_m1+16, src_cb -7, 1);
1084                 XCHG(top_border_m1+24, src_cr -7, 1);
1085             }
1086             XCHG(top_border+16, src_cb+1, 1);
1087             XCHG(top_border+24, src_cr+1, 1);
1088         }
1089     }
1090 }
1091
1092 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
1093     MpegEncContext * const s = &h->s;
1094     const int mb_x= s->mb_x;
1095     const int mb_y= s->mb_y;
1096     const int mb_xy= h->mb_xy;
1097     const int mb_type= s->current_picture.mb_type[mb_xy];
1098     uint8_t  *dest_y, *dest_cb, *dest_cr;
1099     int linesize, uvlinesize /*dct_offset*/;
1100     int i;
1101     int *block_offset = &h->block_offset[0];
1102     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1103     /* is_h264 should always be true if SVQ3 is disabled. */
1104     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1105     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1106     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1107
1108     dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
1109     dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
1110     dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
1111
1112     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
1113     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
1114
1115     h->list_counts[mb_xy]= h->list_count;
1116
1117     if (!simple && MB_FIELD) {
1118         linesize   = h->mb_linesize   = s->linesize * 2;
1119         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1120         block_offset = &h->block_offset[24];
1121         if(mb_y&1){ //FIXME move out of this function?
1122             dest_y -= s->linesize*15;
1123             dest_cb-= s->uvlinesize*7;
1124             dest_cr-= s->uvlinesize*7;
1125         }
1126         if(FRAME_MBAFF) {
1127             int list;
1128             for(list=0; list<h->list_count; list++){
1129                 if(!USES_LIST(mb_type, list))
1130                     continue;
1131                 if(IS_16X16(mb_type)){
1132                     int8_t *ref = &h->ref_cache[list][scan8[0]];
1133                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1134                 }else{
1135                     for(i=0; i<16; i+=4){
1136                         int ref = h->ref_cache[list][scan8[i]];
1137                         if(ref >= 0)
1138                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1139                     }
1140                 }
1141             }
1142         }
1143     } else {
1144         linesize   = h->mb_linesize   = s->linesize;
1145         uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1146 //        dct_offset = s->linesize * 16;
1147     }
1148
1149     if (!simple && IS_INTRA_PCM(mb_type)) {
1150         for (i=0; i<16; i++) {
1151             memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1152         }
1153         for (i=0; i<8; i++) {
1154             memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
1155             memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
1156         }
1157     } else {
1158         if(IS_INTRA(mb_type)){
1159             if(h->deblocking_filter)
1160                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
1161
1162             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1163                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1164                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1165             }
1166
1167             if(IS_INTRA4x4(mb_type)){
1168                 if(simple || !s->encoding){
1169                     if(IS_8x8DCT(mb_type)){
1170                         if(transform_bypass){
1171                             idct_dc_add =
1172                             idct_add    = s->dsp.add_pixels8;
1173                         }else{
1174                             idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1175                             idct_add    = h->h264dsp.h264_idct8_add;
1176                         }
1177                         for(i=0; i<16; i+=4){
1178                             uint8_t * const ptr= dest_y + block_offset[i];
1179                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1180                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1181                                 h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
1182                             }else{
1183                                 const int nnz = h->non_zero_count_cache[ scan8[i] ];
1184                                 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1185                                                             (h->topright_samples_available<<i)&0x4000, linesize);
1186                                 if(nnz){
1187                                     if(nnz == 1 && h->mb[i*16])
1188                                         idct_dc_add(ptr, h->mb + i*16, linesize);
1189                                     else
1190                                         idct_add   (ptr, h->mb + i*16, linesize);
1191                                 }
1192                             }
1193                         }
1194                     }else{
1195                         if(transform_bypass){
1196                             idct_dc_add =
1197                             idct_add    = s->dsp.add_pixels4;
1198                         }else{
1199                             idct_dc_add = h->h264dsp.h264_idct_dc_add;
1200                             idct_add    = h->h264dsp.h264_idct_add;
1201                         }
1202                         for(i=0; i<16; i++){
1203                             uint8_t * const ptr= dest_y + block_offset[i];
1204                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1205
1206                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1207                                 h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
1208                             }else{
1209                                 uint8_t *topright;
1210                                 int nnz, tr;
1211                                 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1212                                     const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1213                                     assert(mb_y || linesize <= block_offset[i]);
1214                                     if(!topright_avail){
1215                                         tr= ptr[3 - linesize]*0x01010101;
1216                                         topright= (uint8_t*) &tr;
1217                                     }else
1218                                         topright= ptr + 4 - linesize;
1219                                 }else
1220                                     topright= NULL;
1221
1222                                 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1223                                 nnz = h->non_zero_count_cache[ scan8[i] ];
1224                                 if(nnz){
1225                                     if(is_h264){
1226                                         if(nnz == 1 && h->mb[i*16])
1227                                             idct_dc_add(ptr, h->mb + i*16, linesize);
1228                                         else
1229                                             idct_add   (ptr, h->mb + i*16, linesize);
1230                                     }else
1231                                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1232                                 }
1233                             }
1234                         }
1235                     }
1236                 }
1237             }else{
1238                 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1239                 if(is_h264){
1240                     if(!transform_bypass)
1241                         h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
1242                 }else
1243                     ff_svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
1244             }
1245             if(h->deblocking_filter)
1246                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
1247         }else if(is_h264){
1248             hl_motion(h, dest_y, dest_cb, dest_cr,
1249                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1250                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1251                       h->h264dsp.weight_h264_pixels_tab, h->h264dsp.biweight_h264_pixels_tab);
1252         }
1253
1254
1255         if(!IS_INTRA4x4(mb_type)){
1256             if(is_h264){
1257                 if(IS_INTRA16x16(mb_type)){
1258                     if(transform_bypass){
1259                         if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1260                             h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1261                         }else{
1262                             for(i=0; i<16; i++){
1263                                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1264                                     s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
1265                             }
1266                         }
1267                     }else{
1268                          h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1269                     }
1270                 }else if(h->cbp&15){
1271                     if(transform_bypass){
1272                         const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1273                         idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1274                         for(i=0; i<16; i+=di){
1275                             if(h->non_zero_count_cache[ scan8[i] ]){
1276                                 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
1277                             }
1278                         }
1279                     }else{
1280                         if(IS_8x8DCT(mb_type)){
1281                             h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1282                         }else{
1283                             h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1284                         }
1285                     }
1286                 }
1287             }else{
1288                 for(i=0; i<16; i++){
1289                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1290                         uint8_t * const ptr= dest_y + block_offset[i];
1291                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1292                     }
1293                 }
1294             }
1295         }
1296
1297         if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1298             uint8_t *dest[2] = {dest_cb, dest_cr};
1299             if(transform_bypass){
1300                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1301                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
1302                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
1303                 }else{
1304                     idct_add = s->dsp.add_pixels4;
1305                     for(i=16; i<16+8; i++){
1306                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1307                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1308                     }
1309                 }
1310             }else{
1311                 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]);
1312                 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]);
1313                 if(is_h264){
1314                     idct_add = h->h264dsp.h264_idct_add;
1315                     idct_dc_add = h->h264dsp.h264_idct_dc_add;
1316                     for(i=16; i<16+8; i++){
1317                         if(h->non_zero_count_cache[ scan8[i] ])
1318                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1319                         else if(h->mb[i*16])
1320                             idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1321                     }
1322                 }else{
1323                     for(i=16; i<16+8; i++){
1324                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1325                             uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1326                             ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
1327                         }
1328                     }
1329                 }
1330             }
1331         }
1332     }
1333     if(h->cbp || IS_INTRA(mb_type))
1334         s->dsp.clear_blocks(h->mb);
1335 }
1336
1337 /**
1338  * Process a macroblock; this case avoids checks for expensive uncommon cases.
1339  */
1340 static void hl_decode_mb_simple(H264Context *h){
1341     hl_decode_mb_internal(h, 1);
1342 }
1343
1344 /**
1345  * Process a macroblock; this handles edge cases, such as interlacing.
1346  */
1347 static void av_noinline hl_decode_mb_complex(H264Context *h){
1348     hl_decode_mb_internal(h, 0);
1349 }
1350
1351 void ff_h264_hl_decode_mb(H264Context *h){
1352     MpegEncContext * const s = &h->s;
1353     const int mb_xy= h->mb_xy;
1354     const int mb_type= s->current_picture.mb_type[mb_xy];
1355     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1356
1357     if (is_complex)
1358         hl_decode_mb_complex(h);
1359     else hl_decode_mb_simple(h);
1360 }
1361
1362 static int pred_weight_table(H264Context *h){
1363     MpegEncContext * const s = &h->s;
1364     int list, i;
1365     int luma_def, chroma_def;
1366
1367     h->use_weight= 0;
1368     h->use_weight_chroma= 0;
1369     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1370     h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1371     luma_def = 1<<h->luma_log2_weight_denom;
1372     chroma_def = 1<<h->chroma_log2_weight_denom;
1373
1374     for(list=0; list<2; list++){
1375         h->luma_weight_flag[list]   = 0;
1376         h->chroma_weight_flag[list] = 0;
1377         for(i=0; i<h->ref_count[list]; i++){
1378             int luma_weight_flag, chroma_weight_flag;
1379
1380             luma_weight_flag= get_bits1(&s->gb);
1381             if(luma_weight_flag){
1382                 h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
1383                 h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
1384                 if(   h->luma_weight[i][list][0] != luma_def
1385                    || h->luma_weight[i][list][1] != 0) {
1386                     h->use_weight= 1;
1387                     h->luma_weight_flag[list]= 1;
1388                 }
1389             }else{
1390                 h->luma_weight[i][list][0]= luma_def;
1391                 h->luma_weight[i][list][1]= 0;
1392             }
1393
1394             if(CHROMA){
1395                 chroma_weight_flag= get_bits1(&s->gb);
1396                 if(chroma_weight_flag){
1397                     int j;
1398                     for(j=0; j<2; j++){
1399                         h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
1400                         h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
1401                         if(   h->chroma_weight[i][list][j][0] != chroma_def
1402                            || h->chroma_weight[i][list][j][1] != 0) {
1403                             h->use_weight_chroma= 1;
1404                             h->chroma_weight_flag[list]= 1;
1405                         }
1406                     }
1407                 }else{
1408                     int j;
1409                     for(j=0; j<2; j++){
1410                         h->chroma_weight[i][list][j][0]= chroma_def;
1411                         h->chroma_weight[i][list][j][1]= 0;
1412                     }
1413                 }
1414             }
1415         }
1416         if(h->slice_type_nos != FF_B_TYPE) break;
1417     }
1418     h->use_weight= h->use_weight || h->use_weight_chroma;
1419     return 0;
1420 }
1421
1422 /**
1423  * Initialize implicit_weight table.
1424  * @param field, 0/1 initialize the weight for interlaced MBAFF
1425  *                -1 initializes the rest
1426  */
1427 static void implicit_weight_table(H264Context *h, int field){
1428     MpegEncContext * const s = &h->s;
1429     int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
1430
1431     for (i = 0; i < 2; i++) {
1432         h->luma_weight_flag[i]   = 0;
1433         h->chroma_weight_flag[i] = 0;
1434     }
1435
1436     if(field < 0){
1437         cur_poc = s->current_picture_ptr->poc;
1438     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
1439        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1440         h->use_weight= 0;
1441         h->use_weight_chroma= 0;
1442         return;
1443     }
1444         ref_start= 0;
1445         ref_count0= h->ref_count[0];
1446         ref_count1= h->ref_count[1];
1447     }else{
1448         cur_poc = s->current_picture_ptr->field_poc[field];
1449         ref_start= 16;
1450         ref_count0= 16+2*h->ref_count[0];
1451         ref_count1= 16+2*h->ref_count[1];
1452     }
1453
1454     h->use_weight= 2;
1455     h->use_weight_chroma= 2;
1456     h->luma_log2_weight_denom= 5;
1457     h->chroma_log2_weight_denom= 5;
1458
1459     for(ref0=ref_start; ref0 < ref_count0; ref0++){
1460         int poc0 = h->ref_list[0][ref0].poc;
1461         for(ref1=ref_start; ref1 < ref_count1; ref1++){
1462             int poc1 = h->ref_list[1][ref1].poc;
1463             int td = av_clip(poc1 - poc0, -128, 127);
1464             int w= 32;
1465             if(td){
1466                 int tb = av_clip(cur_poc - poc0, -128, 127);
1467                 int tx = (16384 + (FFABS(td) >> 1)) / td;
1468                 int dist_scale_factor = (tb*tx + 32) >> 8;
1469                 if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
1470                     w = 64 - dist_scale_factor;
1471             }
1472             if(field<0){
1473                 h->implicit_weight[ref0][ref1][0]=
1474                 h->implicit_weight[ref0][ref1][1]= w;
1475             }else{
1476                 h->implicit_weight[ref0][ref1][field]=w;
1477             }
1478         }
1479     }
1480 }
1481
1482 /**
1483  * instantaneous decoder refresh.
1484  */
1485 static void idr(H264Context *h){
1486     ff_h264_remove_all_refs(h);
1487     h->prev_frame_num= 0;
1488     h->prev_frame_num_offset= 0;
1489     h->prev_poc_msb=
1490     h->prev_poc_lsb= 0;
1491 }
1492
1493 /* forget old pics after a seek */
1494 static void flush_dpb(AVCodecContext *avctx){
1495     H264Context *h= avctx->priv_data;
1496     int i;
1497     for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1498         if(h->delayed_pic[i])
1499             h->delayed_pic[i]->reference= 0;
1500         h->delayed_pic[i]= NULL;
1501     }
1502     h->outputed_poc= INT_MIN;
1503     h->prev_interlaced_frame = 1;
1504     idr(h);
1505     if(h->s.current_picture_ptr)
1506         h->s.current_picture_ptr->reference= 0;
1507     h->s.first_field= 0;
1508     ff_h264_reset_sei(h);
1509     ff_mpeg_flush(avctx);
1510 }
1511
1512 static int init_poc(H264Context *h){
1513     MpegEncContext * const s = &h->s;
1514     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1515     int field_poc[2];
1516     Picture *cur = s->current_picture_ptr;
1517
1518     h->frame_num_offset= h->prev_frame_num_offset;
1519     if(h->frame_num < h->prev_frame_num)
1520         h->frame_num_offset += max_frame_num;
1521
1522     if(h->sps.poc_type==0){
1523         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1524
1525         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1526             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1527         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1528             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1529         else
1530             h->poc_msb = h->prev_poc_msb;
1531 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1532         field_poc[0] =
1533         field_poc[1] = h->poc_msb + h->poc_lsb;
1534         if(s->picture_structure == PICT_FRAME)
1535             field_poc[1] += h->delta_poc_bottom;
1536     }else if(h->sps.poc_type==1){
1537         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1538         int i;
1539
1540         if(h->sps.poc_cycle_length != 0)
1541             abs_frame_num = h->frame_num_offset + h->frame_num;
1542         else
1543             abs_frame_num = 0;
1544
1545         if(h->nal_ref_idc==0 && abs_frame_num > 0)
1546             abs_frame_num--;
1547
1548         expected_delta_per_poc_cycle = 0;
1549         for(i=0; i < h->sps.poc_cycle_length; i++)
1550             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1551
1552         if(abs_frame_num > 0){
1553             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1554             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1555
1556             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1557             for(i = 0; i <= frame_num_in_poc_cycle; i++)
1558                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1559         } else
1560             expectedpoc = 0;
1561
1562         if(h->nal_ref_idc == 0)
1563             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1564
1565         field_poc[0] = expectedpoc + h->delta_poc[0];
1566         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1567
1568         if(s->picture_structure == PICT_FRAME)
1569             field_poc[1] += h->delta_poc[1];
1570     }else{
1571         int poc= 2*(h->frame_num_offset + h->frame_num);
1572
1573         if(!h->nal_ref_idc)
1574             poc--;
1575
1576         field_poc[0]= poc;
1577         field_poc[1]= poc;
1578     }
1579
1580     if(s->picture_structure != PICT_BOTTOM_FIELD)
1581         s->current_picture_ptr->field_poc[0]= field_poc[0];
1582     if(s->picture_structure != PICT_TOP_FIELD)
1583         s->current_picture_ptr->field_poc[1]= field_poc[1];
1584     cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1585
1586     return 0;
1587 }
1588
1589
1590 /**
1591  * initialize scan tables
1592  */
1593 static void init_scan_tables(H264Context *h){
1594     int i;
1595     if(h->h264dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
1596         memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
1597         memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
1598     }else{
1599         for(i=0; i<16; i++){
1600 #define T(x) (x>>2) | ((x<<2) & 0xF)
1601             h->zigzag_scan[i] = T(zigzag_scan[i]);
1602             h-> field_scan[i] = T( field_scan[i]);
1603 #undef T
1604         }
1605     }
1606     if(h->h264dsp.h264_idct8_add == ff_h264_idct8_add_c){
1607         memcpy(h->zigzag_scan8x8,       ff_zigzag_direct,     64*sizeof(uint8_t));
1608         memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
1609         memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
1610         memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
1611     }else{
1612         for(i=0; i<64; i++){
1613 #define T(x) (x>>3) | ((x&7)<<3)
1614             h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
1615             h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1616             h->field_scan8x8[i]        = T(field_scan8x8[i]);
1617             h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
1618 #undef T
1619         }
1620     }
1621     if(h->sps.transform_bypass){ //FIXME same ugly
1622         h->zigzag_scan_q0          = zigzag_scan;
1623         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
1624         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1625         h->field_scan_q0           = field_scan;
1626         h->field_scan8x8_q0        = field_scan8x8;
1627         h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
1628     }else{
1629         h->zigzag_scan_q0          = h->zigzag_scan;
1630         h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
1631         h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1632         h->field_scan_q0           = h->field_scan;
1633         h->field_scan8x8_q0        = h->field_scan8x8;
1634         h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
1635     }
1636 }
1637
1638 static void field_end(H264Context *h){
1639     MpegEncContext * const s = &h->s;
1640     AVCodecContext * const avctx= s->avctx;
1641     s->mb_y= 0;
1642
1643     s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1644     s->current_picture_ptr->pict_type= s->pict_type;
1645
1646     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1647         ff_vdpau_h264_set_reference_frames(s);
1648
1649     if(!s->dropable) {
1650         ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1651         h->prev_poc_msb= h->poc_msb;
1652         h->prev_poc_lsb= h->poc_lsb;
1653     }
1654     h->prev_frame_num_offset= h->frame_num_offset;
1655     h->prev_frame_num= h->frame_num;
1656
1657     if (avctx->hwaccel) {
1658         if (avctx->hwaccel->end_frame(avctx) < 0)
1659             av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1660     }
1661
1662     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1663         ff_vdpau_h264_picture_complete(s);
1664
1665     /*
1666      * FIXME: Error handling code does not seem to support interlaced
1667      * when slices span multiple rows
1668      * The ff_er_add_slice calls don't work right for bottom
1669      * fields; they cause massive erroneous error concealing
1670      * Error marking covers both fields (top and bottom).
1671      * This causes a mismatched s->error_count
1672      * and a bad error table. Further, the error count goes to
1673      * INT_MAX when called for bottom field, because mb_y is
1674      * past end by one (callers fault) and resync_mb_y != 0
1675      * causes problems for the first MB line, too.
1676      */
1677     if (!FIELD_PICTURE)
1678         ff_er_frame_end(s);
1679
1680     MPV_frame_end(s);
1681
1682     h->current_slice=0;
1683 }
1684
1685 /**
1686  * Replicates H264 "master" context to thread contexts.
1687  */
1688 static void clone_slice(H264Context *dst, H264Context *src)
1689 {
1690     memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
1691     dst->s.current_picture_ptr  = src->s.current_picture_ptr;
1692     dst->s.current_picture      = src->s.current_picture;
1693     dst->s.linesize             = src->s.linesize;
1694     dst->s.uvlinesize           = src->s.uvlinesize;
1695     dst->s.first_field          = src->s.first_field;
1696
1697     dst->prev_poc_msb           = src->prev_poc_msb;
1698     dst->prev_poc_lsb           = src->prev_poc_lsb;
1699     dst->prev_frame_num_offset  = src->prev_frame_num_offset;
1700     dst->prev_frame_num         = src->prev_frame_num;
1701     dst->short_ref_count        = src->short_ref_count;
1702
1703     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
1704     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
1705     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1706     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
1707
1708     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
1709     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
1710 }
1711
1712 /**
1713  * decodes a slice header.
1714  * This will also call MPV_common_init() and frame_start() as needed.
1715  *
1716  * @param h h264context
1717  * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1718  *
1719  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1720  */
1721 static int decode_slice_header(H264Context *h, H264Context *h0){
1722     MpegEncContext * const s = &h->s;
1723     MpegEncContext * const s0 = &h0->s;
1724     unsigned int first_mb_in_slice;
1725     unsigned int pps_id;
1726     int num_ref_idx_active_override_flag;
1727     unsigned int slice_type, tmp, i, j;
1728     int default_ref_list_done = 0;
1729     int last_pic_structure;
1730
1731     s->dropable= h->nal_ref_idc == 0;
1732
1733     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1734         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1735         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1736     }else{
1737         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1738         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1739     }
1740
1741     first_mb_in_slice= get_ue_golomb(&s->gb);
1742
1743     if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1744         if(h0->current_slice && FIELD_PICTURE){
1745             field_end(h);
1746         }
1747
1748         h0->current_slice = 0;
1749         if (!s0->first_field)
1750             s->current_picture_ptr= NULL;
1751     }
1752
1753     slice_type= get_ue_golomb_31(&s->gb);
1754     if(slice_type > 9){
1755         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);
1756         return -1;
1757     }
1758     if(slice_type > 4){
1759         slice_type -= 5;
1760         h->slice_type_fixed=1;
1761     }else
1762         h->slice_type_fixed=0;
1763
1764     slice_type= golomb_to_pict_type[ slice_type ];
1765     if (slice_type == FF_I_TYPE
1766         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
1767         default_ref_list_done = 1;
1768     }
1769     h->slice_type= slice_type;
1770     h->slice_type_nos= slice_type & 3;
1771
1772     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
1773
1774     pps_id= get_ue_golomb(&s->gb);
1775     if(pps_id>=MAX_PPS_COUNT){
1776         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
1777         return -1;
1778     }
1779     if(!h0->pps_buffers[pps_id]) {
1780         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
1781         return -1;
1782     }
1783     h->pps= *h0->pps_buffers[pps_id];
1784
1785     if(!h0->sps_buffers[h->pps.sps_id]) {
1786         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
1787         return -1;
1788     }
1789     h->sps = *h0->sps_buffers[h->pps.sps_id];
1790
1791     s->avctx->profile = h->sps.profile_idc;
1792     s->avctx->level   = h->sps.level_idc;
1793     s->avctx->refs    = h->sps.ref_frame_count;
1794
1795     if(h == h0 && h->dequant_coeff_pps != pps_id){
1796         h->dequant_coeff_pps = pps_id;
1797         init_dequant_tables(h);
1798     }
1799
1800     s->mb_width= h->sps.mb_width;
1801     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1802
1803     h->b_stride=  s->mb_width*4;
1804
1805     s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
1806     if(h->sps.frame_mbs_only_flag)
1807         s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
1808     else
1809         s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
1810
1811     if (s->context_initialized
1812         && (   s->width != s->avctx->width || s->height != s->avctx->height
1813             || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
1814         if(h != h0)
1815             return -1;   // width / height changed during parallelized decoding
1816         free_tables(h);
1817         flush_dpb(s->avctx);
1818         MPV_common_end(s);
1819     }
1820     if (!s->context_initialized) {
1821         if(h != h0)
1822             return -1;  // we cant (re-)initialize context during parallel decoding
1823
1824         avcodec_set_dimensions(s->avctx, s->width, s->height);
1825         s->avctx->sample_aspect_ratio= h->sps.sar;
1826         if(!s->avctx->sample_aspect_ratio.den)
1827             s->avctx->sample_aspect_ratio.den = 1;
1828
1829         if(h->sps.video_signal_type_present_flag){
1830             s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1831             if(h->sps.colour_description_present_flag){
1832                 s->avctx->color_primaries = h->sps.color_primaries;
1833                 s->avctx->color_trc       = h->sps.color_trc;
1834                 s->avctx->colorspace      = h->sps.colorspace;
1835             }
1836         }
1837
1838         if(h->sps.timing_info_present_flag){
1839             int64_t den= h->sps.time_scale;
1840             if(h->x264_build < 44U)
1841                 den *= 2;
1842             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
1843                       h->sps.num_units_in_tick, den, 1<<30);
1844         }
1845         s->avctx->pix_fmt = s->avctx->get_format(s->avctx, s->avctx->codec->pix_fmts);
1846         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1847
1848         if (MPV_common_init(s) < 0)
1849             return -1;
1850         s->first_field = 0;
1851         h->prev_interlaced_frame = 1;
1852
1853         init_scan_tables(h);
1854         ff_h264_alloc_tables(h);
1855
1856         for(i = 1; i < s->avctx->thread_count; i++) {
1857             H264Context *c;
1858             c = h->thread_context[i] = av_malloc(sizeof(H264Context));
1859             memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
1860             memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
1861             c->h264dsp = h->h264dsp;
1862             c->sps = h->sps;
1863             c->pps = h->pps;
1864             init_scan_tables(c);
1865             clone_tables(c, h, i);
1866         }
1867
1868         for(i = 0; i < s->avctx->thread_count; i++)
1869             if(context_init(h->thread_context[i]) < 0)
1870                 return -1;
1871     }
1872
1873     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1874
1875     h->mb_mbaff = 0;
1876     h->mb_aff_frame = 0;
1877     last_pic_structure = s0->picture_structure;
1878     if(h->sps.frame_mbs_only_flag){
1879         s->picture_structure= PICT_FRAME;
1880     }else{
1881         if(get_bits1(&s->gb)) { //field_pic_flag
1882             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
1883         } else {
1884             s->picture_structure= PICT_FRAME;
1885             h->mb_aff_frame = h->sps.mb_aff;
1886         }
1887     }
1888     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
1889
1890     if(h0->current_slice == 0){
1891         while(h->frame_num !=  h->prev_frame_num &&
1892               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
1893             av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
1894             if (ff_h264_frame_start(h) < 0)
1895                 return -1;
1896             h->prev_frame_num++;
1897             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
1898             s->current_picture_ptr->frame_num= h->prev_frame_num;
1899             ff_h264_execute_ref_pic_marking(h, NULL, 0);
1900         }
1901
1902         /* See if we have a decoded first field looking for a pair... */
1903         if (s0->first_field) {
1904             assert(s0->current_picture_ptr);
1905             assert(s0->current_picture_ptr->data[0]);
1906             assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
1907
1908             /* figure out if we have a complementary field pair */
1909             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
1910                 /*
1911                  * Previous field is unmatched. Don't display it, but let it
1912                  * remain for reference if marked as such.
1913                  */
1914                 s0->current_picture_ptr = NULL;
1915                 s0->first_field = FIELD_PICTURE;
1916
1917             } else {
1918                 if (h->nal_ref_idc &&
1919                         s0->current_picture_ptr->reference &&
1920                         s0->current_picture_ptr->frame_num != h->frame_num) {
1921                     /*
1922                      * This and previous field were reference, but had
1923                      * different frame_nums. Consider this field first in
1924                      * pair. Throw away previous field except for reference
1925                      * purposes.
1926                      */
1927                     s0->first_field = 1;
1928                     s0->current_picture_ptr = NULL;
1929
1930                 } else {
1931                     /* Second field in complementary pair */
1932                     s0->first_field = 0;
1933                 }
1934             }
1935
1936         } else {
1937             /* Frame or first field in a potentially complementary pair */
1938             assert(!s0->current_picture_ptr);
1939             s0->first_field = FIELD_PICTURE;
1940         }
1941
1942         if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
1943             s0->first_field = 0;
1944             return -1;
1945         }
1946     }
1947     if(h != h0)
1948         clone_slice(h, h0);
1949
1950     s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
1951
1952     assert(s->mb_num == s->mb_width * s->mb_height);
1953     if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
1954        first_mb_in_slice                    >= s->mb_num){
1955         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1956         return -1;
1957     }
1958     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
1959     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
1960     if (s->picture_structure == PICT_BOTTOM_FIELD)
1961         s->resync_mb_y = s->mb_y = s->mb_y + 1;
1962     assert(s->mb_y < s->mb_height);
1963
1964     if(s->picture_structure==PICT_FRAME){
1965         h->curr_pic_num=   h->frame_num;
1966         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
1967     }else{
1968         h->curr_pic_num= 2*h->frame_num + 1;
1969         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
1970     }
1971
1972     if(h->nal_unit_type == NAL_IDR_SLICE){
1973         get_ue_golomb(&s->gb); /* idr_pic_id */
1974     }
1975
1976     if(h->sps.poc_type==0){
1977         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
1978
1979         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
1980             h->delta_poc_bottom= get_se_golomb(&s->gb);
1981         }
1982     }
1983
1984     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
1985         h->delta_poc[0]= get_se_golomb(&s->gb);
1986
1987         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
1988             h->delta_poc[1]= get_se_golomb(&s->gb);
1989     }
1990
1991     init_poc(h);
1992
1993     if(h->pps.redundant_pic_cnt_present){
1994         h->redundant_pic_count= get_ue_golomb(&s->gb);
1995     }
1996
1997     //set defaults, might be overridden a few lines later
1998     h->ref_count[0]= h->pps.ref_count[0];
1999     h->ref_count[1]= h->pps.ref_count[1];
2000
2001     if(h->slice_type_nos != FF_I_TYPE){
2002         if(h->slice_type_nos == FF_B_TYPE){
2003             h->direct_spatial_mv_pred= get_bits1(&s->gb);
2004         }
2005         num_ref_idx_active_override_flag= get_bits1(&s->gb);
2006
2007         if(num_ref_idx_active_override_flag){
2008             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2009             if(h->slice_type_nos==FF_B_TYPE)
2010                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2011
2012             if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
2013                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
2014                 h->ref_count[0]= h->ref_count[1]= 1;
2015                 return -1;
2016             }
2017         }
2018         if(h->slice_type_nos == FF_B_TYPE)
2019             h->list_count= 2;
2020         else
2021             h->list_count= 1;
2022     }else
2023         h->list_count= 0;
2024
2025     if(!default_ref_list_done){
2026         ff_h264_fill_default_ref_list(h);
2027     }
2028
2029     if(h->slice_type_nos!=FF_I_TYPE && ff_h264_decode_ref_pic_list_reordering(h) < 0)
2030         return -1;
2031
2032     if(h->slice_type_nos!=FF_I_TYPE){
2033         s->last_picture_ptr= &h->ref_list[0][0];
2034         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
2035     }
2036     if(h->slice_type_nos==FF_B_TYPE){
2037         s->next_picture_ptr= &h->ref_list[1][0];
2038         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
2039     }
2040
2041     if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
2042        ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
2043         pred_weight_table(h);
2044     else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE){
2045         implicit_weight_table(h, -1);
2046     }else {
2047         h->use_weight = 0;
2048         for (i = 0; i < 2; i++) {
2049             h->luma_weight_flag[i]   = 0;
2050             h->chroma_weight_flag[i] = 0;
2051         }
2052     }
2053
2054     if(h->nal_ref_idc)
2055         ff_h264_decode_ref_pic_marking(h0, &s->gb);
2056
2057     if(FRAME_MBAFF){
2058         ff_h264_fill_mbaff_ref_list(h);
2059
2060         if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE){
2061             implicit_weight_table(h, 0);
2062             implicit_weight_table(h, 1);
2063         }
2064     }
2065
2066     if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
2067         ff_h264_direct_dist_scale_factor(h);
2068     ff_h264_direct_ref_list_init(h);
2069
2070     if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
2071         tmp = get_ue_golomb_31(&s->gb);
2072         if(tmp > 2){
2073             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2074             return -1;
2075         }
2076         h->cabac_init_idc= tmp;
2077     }
2078
2079     h->last_qscale_diff = 0;
2080     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2081     if(tmp>51){
2082         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2083         return -1;
2084     }
2085     s->qscale= tmp;
2086     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2087     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2088     //FIXME qscale / qp ... stuff
2089     if(h->slice_type == FF_SP_TYPE){
2090         get_bits1(&s->gb); /* sp_for_switch_flag */
2091     }
2092     if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
2093         get_se_golomb(&s->gb); /* slice_qs_delta */
2094     }
2095
2096     h->deblocking_filter = 1;
2097     h->slice_alpha_c0_offset = 52;
2098     h->slice_beta_offset = 52;
2099     if( h->pps.deblocking_filter_parameters_present ) {
2100         tmp= get_ue_golomb_31(&s->gb);
2101         if(tmp > 2){
2102             av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2103             return -1;
2104         }
2105         h->deblocking_filter= tmp;
2106         if(h->deblocking_filter < 2)
2107             h->deblocking_filter^= 1; // 1<->0
2108
2109         if( h->deblocking_filter ) {
2110             h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2111             h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
2112             if(   h->slice_alpha_c0_offset > 104U
2113                || h->slice_beta_offset     > 104U){
2114                 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);
2115                 return -1;
2116             }
2117         }
2118     }
2119
2120     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
2121        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
2122        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
2123        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2124         h->deblocking_filter= 0;
2125
2126     if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2127         if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2128             /* Cheat slightly for speed:
2129                Do not bother to deblock across slices. */
2130             h->deblocking_filter = 2;
2131         } else {
2132             h0->max_contexts = 1;
2133             if(!h0->single_decode_warning) {
2134                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2135                 h0->single_decode_warning = 1;
2136             }
2137             if(h != h0)
2138                 return 1; // deblocking switched inside frame
2139         }
2140     }
2141     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]);
2142
2143 #if 0 //FMO
2144     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2145         slice_group_change_cycle= get_bits(&s->gb, ?);
2146 #endif
2147
2148     h0->last_slice_type = slice_type;
2149     h->slice_num = ++h0->current_slice;
2150     if(h->slice_num >= MAX_SLICES){
2151         av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2152     }
2153
2154     for(j=0; j<2; j++){
2155         int id_list[16];
2156         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2157         for(i=0; i<16; i++){
2158             id_list[i]= 60;
2159             if(h->ref_list[j][i].data[0]){
2160                 int k;
2161                 uint8_t *base= h->ref_list[j][i].base[0];
2162                 for(k=0; k<h->short_ref_count; k++)
2163                     if(h->short_ref[k]->base[0] == base){
2164                         id_list[i]= k;
2165                         break;
2166                     }
2167                 for(k=0; k<h->long_ref_count; k++)
2168                     if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2169                         id_list[i]= h->short_ref_count + k;
2170                         break;
2171                     }
2172             }
2173         }
2174
2175         ref2frm[0]=
2176         ref2frm[1]= -1;
2177         for(i=0; i<16; i++)
2178             ref2frm[i+2]= 4*id_list[i]
2179                           +(h->ref_list[j][i].reference&3);
2180         ref2frm[18+0]=
2181         ref2frm[18+1]= -1;
2182         for(i=16; i<48; i++)
2183             ref2frm[i+4]= 4*id_list[(i-16)>>1]
2184                           +(h->ref_list[j][i].reference&3);
2185     }
2186
2187     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
2188     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2189
2190     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2191         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",
2192                h->slice_num,
2193                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2194                first_mb_in_slice,
2195                av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2196                pps_id, h->frame_num,
2197                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2198                h->ref_count[0], h->ref_count[1],
2199                s->qscale,
2200                h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
2201                h->use_weight,
2202                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2203                h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2204                );
2205     }
2206
2207     return 0;
2208 }
2209
2210 int ff_h264_get_slice_type(const H264Context *h)
2211 {
2212     switch (h->slice_type) {
2213     case FF_P_TYPE:  return 0;
2214     case FF_B_TYPE:  return 1;
2215     case FF_I_TYPE:  return 2;
2216     case FF_SP_TYPE: return 3;
2217     case FF_SI_TYPE: return 4;
2218     default:         return -1;
2219     }
2220 }
2221
2222 /**
2223  *
2224  * @return non zero if the loop filter can be skiped
2225  */
2226 static int fill_filter_caches(H264Context *h, int mb_type){
2227     MpegEncContext * const s = &h->s;
2228     const int mb_xy= h->mb_xy;
2229     int top_xy, left_xy[2];
2230     int top_type, left_type[2];
2231
2232     top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
2233
2234     //FIXME deblocking could skip the intra and nnz parts.
2235
2236     /* Wow, what a mess, why didn't they simplify the interlacing & intra
2237      * stuff, I can't imagine that these complex rules are worth it. */
2238
2239     left_xy[1] = left_xy[0] = mb_xy-1;
2240     if(FRAME_MBAFF){
2241         const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
2242         const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
2243         if(s->mb_y&1){
2244             if (left_mb_field_flag != curr_mb_field_flag) {
2245                 left_xy[0] -= s->mb_stride;
2246             }
2247         }else{
2248             if(curr_mb_field_flag){
2249                 top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
2250             }
2251             if (left_mb_field_flag != curr_mb_field_flag) {
2252                 left_xy[1] += s->mb_stride;
2253             }
2254         }
2255     }
2256
2257     h->top_mb_xy = top_xy;
2258     h->left_mb_xy[0] = left_xy[0];
2259     h->left_mb_xy[1] = left_xy[1];
2260     {
2261         //for sufficiently low qp, filtering wouldn't do anything
2262         //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
2263         int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
2264         int qp = s->current_picture.qscale_table[mb_xy];
2265         if(qp <= qp_thresh
2266            && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
2267            && (top_xy   < 0 || ((qp + s->current_picture.qscale_table[top_xy    ] + 1)>>1) <= qp_thresh)){
2268             if(!FRAME_MBAFF)
2269                 return 1;
2270             if(   (left_xy[0]< 0            || ((qp + s->current_picture.qscale_table[left_xy[1]             ] + 1)>>1) <= qp_thresh)
2271                && (top_xy    < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy    -s->mb_stride] + 1)>>1) <= qp_thresh))
2272                 return 1;
2273         }
2274     }
2275
2276     top_type     = s->current_picture.mb_type[top_xy]    ;
2277     left_type[0] = s->current_picture.mb_type[left_xy[0]];
2278     left_type[1] = s->current_picture.mb_type[left_xy[1]];
2279     if(h->deblocking_filter == 2){
2280         if(h->slice_table[top_xy     ] != h->slice_num) top_type= 0;
2281         if(h->slice_table[left_xy[0] ] != h->slice_num) left_type[0]= left_type[1]= 0;
2282     }else{
2283         if(h->slice_table[top_xy     ] == 0xFFFF) top_type= 0;
2284         if(h->slice_table[left_xy[0] ] == 0xFFFF) left_type[0]= left_type[1] =0;
2285     }
2286     h->top_type    = top_type    ;
2287     h->left_type[0]= left_type[0];
2288     h->left_type[1]= left_type[1];
2289
2290     if(IS_INTRA(mb_type))
2291         return 0;
2292
2293     AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
2294     AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
2295     AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]);
2296     AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]);
2297     AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
2298
2299     h->cbp= h->cbp_table[mb_xy];
2300
2301     {
2302         int list;
2303         for(list=0; list<h->list_count; list++){
2304             int8_t *ref;
2305             int y, b_stride;
2306             int16_t (*mv_dst)[2];
2307             int16_t (*mv_src)[2];
2308
2309             if(!USES_LIST(mb_type, list)){
2310                 fill_rectangle(  h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
2311                 AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2312                 AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2313                 AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2314                 AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2315                 continue;
2316             }
2317
2318             ref = &s->current_picture.ref_index[list][4*mb_xy];
2319             {
2320                 int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2321                 AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2322                 AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2323                 ref += 2;
2324                 AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2325                 AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2326             }
2327
2328             b_stride = h->b_stride;
2329             mv_dst   = &h->mv_cache[list][scan8[0]];
2330             mv_src   = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
2331             for(y=0; y<4; y++){
2332                 AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
2333             }
2334
2335         }
2336     }
2337
2338
2339 /*
2340 0 . T T. T T T T
2341 1 L . .L . . . .
2342 2 L . .L . . . .
2343 3 . T TL . . . .
2344 4 L . .L . . . .
2345 5 L . .. . . . .
2346 */
2347 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
2348     if(top_type){
2349         AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
2350     }
2351
2352     if(left_type[0]){
2353         h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
2354         h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
2355         h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
2356         h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
2357     }
2358
2359     // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
2360     if(!CABAC && h->pps.transform_8x8_mode){
2361         if(IS_8x8DCT(top_type)){
2362             h->non_zero_count_cache[4+8*0]=
2363             h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
2364             h->non_zero_count_cache[6+8*0]=
2365             h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
2366         }
2367         if(IS_8x8DCT(left_type[0])){
2368             h->non_zero_count_cache[3+8*1]=
2369             h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
2370         }
2371         if(IS_8x8DCT(left_type[1])){
2372             h->non_zero_count_cache[3+8*3]=
2373             h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
2374         }
2375
2376         if(IS_8x8DCT(mb_type)){
2377             h->non_zero_count_cache[scan8[0   ]]= h->non_zero_count_cache[scan8[1   ]]=
2378             h->non_zero_count_cache[scan8[2   ]]= h->non_zero_count_cache[scan8[3   ]]= h->cbp & 1;
2379
2380             h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
2381             h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
2382
2383             h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
2384             h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
2385
2386             h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
2387             h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
2388         }
2389     }
2390
2391     if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
2392         int list;
2393         for(list=0; list<h->list_count; list++){
2394             if(USES_LIST(top_type, list)){
2395                 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
2396                 const int b8_xy= 4*top_xy + 2;
2397                 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2398                 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
2399                 h->ref_cache[list][scan8[0] + 0 - 1*8]=
2400                 h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
2401                 h->ref_cache[list][scan8[0] + 2 - 1*8]=
2402                 h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
2403             }else{
2404                 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
2405                 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2406             }
2407
2408             if(!IS_INTERLACED(mb_type^left_type[0])){
2409                 if(USES_LIST(left_type[0], list)){
2410                     const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
2411                     const int b8_xy= 4*left_xy[0] + 1;
2412                     int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2413                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
2414                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
2415                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
2416                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
2417                     h->ref_cache[list][scan8[0] - 1 + 0 ]=
2418                     h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]];
2419                     h->ref_cache[list][scan8[0] - 1 +16 ]=
2420                     h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]];
2421                 }else{
2422                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
2423                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
2424                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
2425                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
2426                     h->ref_cache[list][scan8[0] - 1 + 0  ]=
2427                     h->ref_cache[list][scan8[0] - 1 + 8  ]=
2428                     h->ref_cache[list][scan8[0] - 1 + 16 ]=
2429                     h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
2430                 }
2431             }
2432         }
2433     }
2434
2435     return 0;
2436 }
2437
2438 static void loop_filter(H264Context *h){
2439     MpegEncContext * const s = &h->s;
2440     uint8_t  *dest_y, *dest_cb, *dest_cr;
2441     int linesize, uvlinesize, mb_x, mb_y;
2442     const int end_mb_y= s->mb_y + FRAME_MBAFF;
2443     const int old_slice_type= h->slice_type;
2444
2445     if(h->deblocking_filter) {
2446         for(mb_x= 0; mb_x<s->mb_width; mb_x++){
2447             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
2448                 int mb_xy, mb_type;
2449                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2450                 h->slice_num= h->slice_table[mb_xy];
2451                 mb_type= s->current_picture.mb_type[mb_xy];
2452                 h->list_count= h->list_counts[mb_xy];
2453
2454                 if(FRAME_MBAFF)
2455                     h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2456
2457                 s->mb_x= mb_x;
2458                 s->mb_y= mb_y;
2459                 dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2460                 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2461                 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2462                     //FIXME simplify above
2463
2464                 if (MB_FIELD) {
2465                     linesize   = h->mb_linesize   = s->linesize * 2;
2466                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2467                     if(mb_y&1){ //FIXME move out of this function?
2468                         dest_y -= s->linesize*15;
2469                         dest_cb-= s->uvlinesize*7;
2470                         dest_cr-= s->uvlinesize*7;
2471                     }
2472                 } else {
2473                     linesize   = h->mb_linesize   = s->linesize;
2474                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2475                 }
2476                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2477                 if(fill_filter_caches(h, mb_type))
2478                     continue;
2479                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2480                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2481
2482                 if (FRAME_MBAFF) {
2483                     ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2484                 } else {
2485                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2486                 }
2487             }
2488         }
2489     }
2490     h->slice_type= old_slice_type;
2491     s->mb_x= 0;
2492     s->mb_y= end_mb_y - FRAME_MBAFF;
2493     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2494     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2495 }
2496
2497 static void predict_field_decoding_flag(H264Context *h){
2498     MpegEncContext * const s = &h->s;
2499     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2500     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
2501                 ? s->current_picture.mb_type[mb_xy-1]
2502                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
2503                 ? s->current_picture.mb_type[mb_xy-s->mb_stride]
2504                 : 0;
2505     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2506 }
2507
2508 static int decode_slice(struct AVCodecContext *avctx, void *arg){
2509     H264Context *h = *(void**)arg;
2510     MpegEncContext * const s = &h->s;
2511     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2512
2513     s->mb_skip_run= -1;
2514
2515     h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
2516                     (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
2517
2518     if( h->pps.cabac ) {
2519         /* realign */
2520         align_get_bits( &s->gb );
2521
2522         /* init cabac */
2523         ff_init_cabac_states( &h->cabac);
2524         ff_init_cabac_decoder( &h->cabac,
2525                                s->gb.buffer + get_bits_count(&s->gb)/8,
2526                                (get_bits_left(&s->gb) + 7)/8);
2527
2528         ff_h264_init_cabac_states(h);
2529
2530         for(;;){
2531 //START_TIMER
2532             int ret = ff_h264_decode_mb_cabac(h);
2533             int eos;
2534 //STOP_TIMER("decode_mb_cabac")
2535
2536             if(ret>=0) ff_h264_hl_decode_mb(h);
2537
2538             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
2539                 s->mb_y++;
2540
2541                 ret = ff_h264_decode_mb_cabac(h);
2542
2543                 if(ret>=0) ff_h264_hl_decode_mb(h);
2544                 s->mb_y--;
2545             }
2546             eos = get_cabac_terminate( &h->cabac );
2547
2548             if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
2549                 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);
2550                 return 0;
2551             }
2552             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2553                 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);
2554                 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);
2555                 return -1;
2556             }
2557
2558             if( ++s->mb_x >= s->mb_width ) {
2559                 s->mb_x = 0;
2560                 loop_filter(h);
2561                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2562                 ++s->mb_y;
2563                 if(FIELD_OR_MBAFF_PICTURE) {
2564                     ++s->mb_y;
2565                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2566                         predict_field_decoding_flag(h);
2567                 }
2568             }
2569
2570             if( eos || s->mb_y >= s->mb_height ) {
2571                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2572                 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);
2573                 return 0;
2574             }
2575         }
2576
2577     } else {
2578         for(;;){
2579             int ret = ff_h264_decode_mb_cavlc(h);
2580
2581             if(ret>=0) ff_h264_hl_decode_mb(h);
2582
2583             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
2584                 s->mb_y++;
2585                 ret = ff_h264_decode_mb_cavlc(h);
2586
2587                 if(ret>=0) ff_h264_hl_decode_mb(h);
2588                 s->mb_y--;
2589             }
2590
2591             if(ret<0){
2592                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2593                 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);
2594
2595                 return -1;
2596             }
2597
2598             if(++s->mb_x >= s->mb_width){
2599                 s->mb_x=0;
2600                 loop_filter(h);
2601                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2602                 ++s->mb_y;
2603                 if(FIELD_OR_MBAFF_PICTURE) {
2604                     ++s->mb_y;
2605                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2606                         predict_field_decoding_flag(h);
2607                 }
2608                 if(s->mb_y >= s->mb_height){
2609                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2610
2611                     if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
2612                         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);
2613
2614                         return 0;
2615                     }else{
2616                         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);
2617
2618                         return -1;
2619                     }
2620                 }
2621             }
2622
2623             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
2624                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2625                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2626                     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);
2627
2628                     return 0;
2629                 }else{
2630                     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);
2631
2632                     return -1;
2633                 }
2634             }
2635         }
2636     }
2637
2638 #if 0
2639     for(;s->mb_y < s->mb_height; s->mb_y++){
2640         for(;s->mb_x < s->mb_width; s->mb_x++){
2641             int ret= decode_mb(h);
2642
2643             ff_h264_hl_decode_mb(h);
2644
2645             if(ret<0){
2646                 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2647                 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);
2648
2649                 return -1;
2650             }
2651
2652             if(++s->mb_x >= s->mb_width){
2653                 s->mb_x=0;
2654                 if(++s->mb_y >= s->mb_height){
2655                     if(get_bits_count(s->gb) == s->gb.size_in_bits){
2656                         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);
2657
2658                         return 0;
2659                     }else{
2660                         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);
2661
2662                         return -1;
2663                     }
2664                 }
2665             }
2666
2667             if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2668                 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2669                     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);
2670
2671                     return 0;
2672                 }else{
2673                     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);
2674
2675                     return -1;
2676                 }
2677             }
2678         }
2679         s->mb_x=0;
2680         ff_draw_horiz_band(s, 16*s->mb_y, 16);
2681     }
2682 #endif
2683     return -1; //not reached
2684 }
2685
2686 /**
2687  * Call decode_slice() for each context.
2688  *
2689  * @param h h264 master context
2690  * @param context_count number of contexts to execute
2691  */
2692 static void execute_decode_slices(H264Context *h, int context_count){
2693     MpegEncContext * const s = &h->s;
2694     AVCodecContext * const avctx= s->avctx;
2695     H264Context *hx;
2696     int i;
2697
2698     if (s->avctx->hwaccel)
2699         return;
2700     if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2701         return;
2702     if(context_count == 1) {
2703         decode_slice(avctx, &h);
2704     } else {
2705         for(i = 1; i < context_count; i++) {
2706             hx = h->thread_context[i];
2707             hx->s.error_recognition = avctx->error_recognition;
2708             hx->s.error_count = 0;
2709         }
2710
2711         avctx->execute(avctx, (void *)decode_slice,
2712                        h->thread_context, NULL, context_count, sizeof(void*));
2713
2714         /* pull back stuff from slices to master context */
2715         hx = h->thread_context[context_count - 1];
2716         s->mb_x = hx->s.mb_x;
2717         s->mb_y = hx->s.mb_y;
2718         s->dropable = hx->s.dropable;
2719         s->picture_structure = hx->s.picture_structure;
2720         for(i = 1; i < context_count; i++)
2721             h->s.error_count += h->thread_context[i]->s.error_count;
2722     }
2723 }
2724
2725
2726 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
2727     MpegEncContext * const s = &h->s;
2728     AVCodecContext * const avctx= s->avctx;
2729     int buf_index=0;
2730     H264Context *hx; ///< thread context
2731     int context_count = 0;
2732     int next_avc= h->is_avc ? 0 : buf_size;
2733
2734     h->max_contexts = avctx->thread_count;
2735 #if 0
2736     int i;
2737     for(i=0; i<50; i++){
2738         av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
2739     }
2740 #endif
2741     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
2742         h->current_slice = 0;
2743         if (!s->first_field)
2744             s->current_picture_ptr= NULL;
2745         ff_h264_reset_sei(h);
2746     }
2747
2748     for(;;){
2749         int consumed;
2750         int dst_length;
2751         int bit_length;
2752         const uint8_t *ptr;
2753         int i, nalsize = 0;
2754         int err;
2755
2756         if(buf_index >= next_avc) {
2757             if(buf_index >= buf_size) break;
2758             nalsize = 0;
2759             for(i = 0; i < h->nal_length_size; i++)
2760                 nalsize = (nalsize << 8) | buf[buf_index++];
2761             if(nalsize <= 1 || nalsize > buf_size - buf_index){
2762                 if(nalsize == 1){
2763                     buf_index++;
2764                     continue;
2765                 }else{
2766                     av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2767                     break;
2768                 }
2769             }
2770             next_avc= buf_index + nalsize;
2771         } else {
2772             // start code prefix search
2773             for(; buf_index + 3 < next_avc; buf_index++){
2774                 // This should always succeed in the first iteration.
2775                 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2776                     break;
2777             }
2778
2779             if(buf_index+3 >= buf_size) break;
2780
2781             buf_index+=3;
2782             if(buf_index >= next_avc) continue;
2783         }
2784
2785         hx = h->thread_context[context_count];
2786
2787         ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
2788         if (ptr==NULL || dst_length < 0){
2789             return -1;
2790         }
2791         i= buf_index + consumed;
2792         if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
2793            buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
2794             s->workaround_bugs |= FF_BUG_TRUNCATED;
2795
2796         if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
2797         while(ptr[dst_length - 1] == 0 && dst_length > 0)
2798             dst_length--;
2799         }
2800         bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
2801
2802         if(s->avctx->debug&FF_DEBUG_STARTCODE){
2803             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);
2804         }
2805
2806         if (h->is_avc && (nalsize != consumed) && nalsize){
2807             av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
2808         }
2809
2810         buf_index += consumed;
2811
2812         if(  (s->hurry_up == 1 && h->nal_ref_idc  == 0) //FIXME do not discard SEI id
2813            ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0))
2814             continue;
2815
2816       again:
2817         err = 0;
2818         switch(hx->nal_unit_type){
2819         case NAL_IDR_SLICE:
2820             if (h->nal_unit_type != NAL_IDR_SLICE) {
2821                 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2822                 return -1;
2823             }
2824             idr(h); //FIXME ensure we don't loose some frames if there is reordering
2825         case NAL_SLICE:
2826             init_get_bits(&hx->s.gb, ptr, bit_length);
2827             hx->intra_gb_ptr=
2828             hx->inter_gb_ptr= &hx->s.gb;
2829             hx->s.data_partitioning = 0;
2830
2831             if((err = decode_slice_header(hx, h)))
2832                break;
2833
2834             if (h->current_slice == 1) {
2835                 if (s->avctx->hwaccel && s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
2836                     return -1;
2837                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2838                     ff_vdpau_h264_picture_start(s);
2839             }
2840
2841             s->current_picture_ptr->key_frame |=
2842                     (hx->nal_unit_type == NAL_IDR_SLICE) ||
2843                     (h->sei_recovery_frame_cnt >= 0);
2844             if(hx->redundant_pic_count==0 && hx->s.hurry_up < 5
2845                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2846                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2847                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2848                && avctx->skip_frame < AVDISCARD_ALL){
2849                 if(avctx->hwaccel) {
2850                     if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
2851                         return -1;
2852                 }else
2853                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
2854                     static const uint8_t start_code[] = {0x00, 0x00, 0x01};
2855                     ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
2856                     ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
2857                 }else
2858                     context_count++;
2859             }
2860             break;
2861         case NAL_DPA:
2862             init_get_bits(&hx->s.gb, ptr, bit_length);
2863             hx->intra_gb_ptr=
2864             hx->inter_gb_ptr= NULL;
2865
2866             if ((err = decode_slice_header(hx, h)) < 0)
2867                 break;
2868
2869             hx->s.data_partitioning = 1;
2870
2871             break;
2872         case NAL_DPB:
2873             init_get_bits(&hx->intra_gb, ptr, bit_length);
2874             hx->intra_gb_ptr= &hx->intra_gb;
2875             break;
2876         case NAL_DPC:
2877             init_get_bits(&hx->inter_gb, ptr, bit_length);
2878             hx->inter_gb_ptr= &hx->inter_gb;
2879
2880             if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
2881                && s->context_initialized
2882                && s->hurry_up < 5
2883                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2884                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2885                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2886                && avctx->skip_frame < AVDISCARD_ALL)
2887                 context_count++;
2888             break;
2889         case NAL_SEI:
2890             init_get_bits(&s->gb, ptr, bit_length);
2891             ff_h264_decode_sei(h);
2892             break;
2893         case NAL_SPS:
2894             init_get_bits(&s->gb, ptr, bit_length);
2895             ff_h264_decode_seq_parameter_set(h);
2896
2897             if(s->flags& CODEC_FLAG_LOW_DELAY)
2898                 s->low_delay=1;
2899
2900             if(avctx->has_b_frames < 2)
2901                 avctx->has_b_frames= !s->low_delay;
2902             break;
2903         case NAL_PPS:
2904             init_get_bits(&s->gb, ptr, bit_length);
2905
2906             ff_h264_decode_picture_parameter_set(h, bit_length);
2907
2908             break;
2909         case NAL_AUD:
2910         case NAL_END_SEQUENCE:
2911         case NAL_END_STREAM:
2912         case NAL_FILLER_DATA:
2913         case NAL_SPS_EXT:
2914         case NAL_AUXILIARY_SLICE:
2915             break;
2916         default:
2917             av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
2918         }
2919
2920         if(context_count == h->max_contexts) {
2921             execute_decode_slices(h, context_count);
2922             context_count = 0;
2923         }
2924
2925         if (err < 0)
2926             av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
2927         else if(err == 1) {
2928             /* Slice could not be decoded in parallel mode, copy down
2929              * NAL unit stuff to context 0 and restart. Note that
2930              * rbsp_buffer is not transferred, but since we no longer
2931              * run in parallel mode this should not be an issue. */
2932             h->nal_unit_type = hx->nal_unit_type;
2933             h->nal_ref_idc   = hx->nal_ref_idc;
2934             hx = h;
2935             goto again;
2936         }
2937     }
2938     if(context_count)
2939         execute_decode_slices(h, context_count);
2940     return buf_index;
2941 }
2942
2943 /**
2944  * returns the number of bytes consumed for building the current frame
2945  */
2946 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
2947         if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
2948         if(pos+10>buf_size) pos=buf_size; // oops ;)
2949
2950         return pos;
2951 }
2952
2953 static int decode_frame(AVCodecContext *avctx,
2954                              void *data, int *data_size,
2955                              AVPacket *avpkt)
2956 {
2957     const uint8_t *buf = avpkt->data;
2958     int buf_size = avpkt->size;
2959     H264Context *h = avctx->priv_data;
2960     MpegEncContext *s = &h->s;
2961     AVFrame *pict = data;
2962     int buf_index;
2963
2964     s->flags= avctx->flags;
2965     s->flags2= avctx->flags2;
2966
2967    /* end of stream, output what is still in the buffers */
2968     if (buf_size == 0) {
2969         Picture *out;
2970         int i, out_idx;
2971
2972 //FIXME factorize this with the output code below
2973         out = h->delayed_pic[0];
2974         out_idx = 0;
2975         for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2976             if(h->delayed_pic[i]->poc < out->poc){
2977                 out = h->delayed_pic[i];
2978                 out_idx = i;
2979             }
2980
2981         for(i=out_idx; h->delayed_pic[i]; i++)
2982             h->delayed_pic[i] = h->delayed_pic[i+1];
2983
2984         if(out){
2985             *data_size = sizeof(AVFrame);
2986             *pict= *(AVFrame*)out;
2987         }
2988
2989         return 0;
2990     }
2991
2992     buf_index=decode_nal_units(h, buf, buf_size);
2993     if(buf_index < 0)
2994         return -1;
2995
2996     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
2997         if (avctx->skip_frame >= AVDISCARD_NONREF || s->hurry_up) return 0;
2998         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
2999         return -1;
3000     }
3001
3002     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
3003         Picture *out = s->current_picture_ptr;
3004         Picture *cur = s->current_picture_ptr;
3005         int i, pics, out_of_order, out_idx;
3006
3007         field_end(h);
3008
3009         if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
3010             /* Wait for second field. */
3011             *data_size = 0;
3012
3013         } else {
3014             cur->interlaced_frame = 0;
3015             cur->repeat_pict = 0;
3016
3017             /* Signal interlacing information externally. */
3018             /* Prioritize picture timing SEI information over used decoding process if it exists. */
3019
3020             if(h->sps.pic_struct_present_flag){
3021                 switch (h->sei_pic_struct)
3022                 {
3023                 case SEI_PIC_STRUCT_FRAME:
3024                     break;
3025                 case SEI_PIC_STRUCT_TOP_FIELD:
3026                 case SEI_PIC_STRUCT_BOTTOM_FIELD:
3027                     cur->interlaced_frame = 1;
3028                     break;
3029                 case SEI_PIC_STRUCT_TOP_BOTTOM:
3030                 case SEI_PIC_STRUCT_BOTTOM_TOP:
3031                     if (FIELD_OR_MBAFF_PICTURE)
3032                         cur->interlaced_frame = 1;
3033                     else
3034                         // try to flag soft telecine progressive
3035                         cur->interlaced_frame = h->prev_interlaced_frame;
3036                     break;
3037                 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
3038                 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
3039                     // Signal the possibility of telecined film externally (pic_struct 5,6)
3040                     // From these hints, let the applications decide if they apply deinterlacing.
3041                     cur->repeat_pict = 1;
3042                     break;
3043                 case SEI_PIC_STRUCT_FRAME_DOUBLING:
3044                     // Force progressive here, as doubling interlaced frame is a bad idea.
3045                     cur->repeat_pict = 2;
3046                     break;
3047                 case SEI_PIC_STRUCT_FRAME_TRIPLING:
3048                     cur->repeat_pict = 4;
3049                     break;
3050                 }
3051
3052                 if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
3053                     cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
3054             }else{
3055                 /* Derive interlacing flag from used decoding process. */
3056                 cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
3057             }
3058             h->prev_interlaced_frame = cur->interlaced_frame;
3059
3060             if (cur->field_poc[0] != cur->field_poc[1]){
3061                 /* Derive top_field_first from field pocs. */
3062                 cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
3063             }else{
3064                 if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
3065                     /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
3066                     if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
3067                       || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
3068                         cur->top_field_first = 1;
3069                     else
3070                         cur->top_field_first = 0;
3071                 }else{
3072                     /* Most likely progressive */
3073                     cur->top_field_first = 0;
3074                 }
3075             }
3076
3077         //FIXME do something with unavailable reference frames
3078
3079             /* Sort B-frames into display order */
3080
3081             if(h->sps.bitstream_restriction_flag
3082                && s->avctx->has_b_frames < h->sps.num_reorder_frames){
3083                 s->avctx->has_b_frames = h->sps.num_reorder_frames;
3084                 s->low_delay = 0;
3085             }
3086
3087             if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
3088                && !h->sps.bitstream_restriction_flag){
3089                 s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
3090                 s->low_delay= 0;
3091             }
3092
3093             pics = 0;
3094             while(h->delayed_pic[pics]) pics++;
3095
3096             assert(pics <= MAX_DELAYED_PIC_COUNT);
3097
3098             h->delayed_pic[pics++] = cur;
3099             if(cur->reference == 0)
3100                 cur->reference = DELAYED_PIC_REF;
3101
3102             out = h->delayed_pic[0];
3103             out_idx = 0;
3104             for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
3105                 if(h->delayed_pic[i]->poc < out->poc){
3106                     out = h->delayed_pic[i];
3107                     out_idx = i;
3108                 }
3109             if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
3110                 h->outputed_poc= INT_MIN;
3111             out_of_order = out->poc < h->outputed_poc;
3112
3113             if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
3114                 { }
3115             else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
3116                || (s->low_delay &&
3117                 ((h->outputed_poc != INT_MIN && out->poc > h->outputed_poc + 2)
3118                  || cur->pict_type == FF_B_TYPE)))
3119             {
3120                 s->low_delay = 0;
3121                 s->avctx->has_b_frames++;
3122             }
3123
3124             if(out_of_order || pics > s->avctx->has_b_frames){
3125                 out->reference &= ~DELAYED_PIC_REF;
3126                 for(i=out_idx; h->delayed_pic[i]; i++)
3127                     h->delayed_pic[i] = h->delayed_pic[i+1];
3128             }
3129             if(!out_of_order && pics > s->avctx->has_b_frames){
3130                 *data_size = sizeof(AVFrame);
3131
3132                 if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
3133                     h->outputed_poc = INT_MIN;
3134                 } else
3135                     h->outputed_poc = out->poc;
3136                 *pict= *(AVFrame*)out;
3137             }else{
3138                 av_log(avctx, AV_LOG_DEBUG, "no picture\n");
3139             }
3140         }
3141     }
3142
3143     assert(pict->data[0] || !*data_size);
3144     ff_print_debug_info(s, pict);
3145 //printf("out %d\n", (int)pict->data[0]);
3146
3147     return get_consumed_bytes(s, buf_index, buf_size);
3148 }
3149 #if 0
3150 static inline void fill_mb_avail(H264Context *h){
3151     MpegEncContext * const s = &h->s;
3152     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3153
3154     if(s->mb_y){
3155         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
3156         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
3157         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
3158     }else{
3159         h->mb_avail[0]=
3160         h->mb_avail[1]=
3161         h->mb_avail[2]= 0;
3162     }
3163     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
3164     h->mb_avail[4]= 1; //FIXME move out
3165     h->mb_avail[5]= 0; //FIXME move out
3166 }
3167 #endif
3168
3169 #ifdef TEST
3170 #undef printf
3171 #undef random
3172 #define COUNT 8000
3173 #define SIZE (COUNT*40)
3174 int main(void){
3175     int i;
3176     uint8_t temp[SIZE];
3177     PutBitContext pb;
3178     GetBitContext gb;
3179 //    int int_temp[10000];
3180     DSPContext dsp;
3181     AVCodecContext avctx;
3182
3183     dsputil_init(&dsp, &avctx);
3184
3185     init_put_bits(&pb, temp, SIZE);
3186     printf("testing unsigned exp golomb\n");
3187     for(i=0; i<COUNT; i++){
3188         START_TIMER
3189         set_ue_golomb(&pb, i);
3190         STOP_TIMER("set_ue_golomb");
3191     }
3192     flush_put_bits(&pb);
3193
3194     init_get_bits(&gb, temp, 8*SIZE);
3195     for(i=0; i<COUNT; i++){
3196         int j, s;
3197
3198         s= show_bits(&gb, 24);
3199
3200         START_TIMER
3201         j= get_ue_golomb(&gb);
3202         if(j != i){
3203             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
3204 //            return -1;
3205         }
3206         STOP_TIMER("get_ue_golomb");
3207     }
3208
3209
3210     init_put_bits(&pb, temp, SIZE);
3211     printf("testing signed exp golomb\n");
3212     for(i=0; i<COUNT; i++){
3213         START_TIMER
3214         set_se_golomb(&pb, i - COUNT/2);
3215         STOP_TIMER("set_se_golomb");
3216     }
3217     flush_put_bits(&pb);
3218
3219     init_get_bits(&gb, temp, 8*SIZE);
3220     for(i=0; i<COUNT; i++){
3221         int j, s;
3222
3223         s= show_bits(&gb, 24);
3224
3225         START_TIMER
3226         j= get_se_golomb(&gb);
3227         if(j != i - COUNT/2){
3228             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
3229 //            return -1;
3230         }
3231         STOP_TIMER("get_se_golomb");
3232     }
3233
3234 #if 0
3235     printf("testing 4x4 (I)DCT\n");
3236
3237     DCTELEM block[16];
3238     uint8_t src[16], ref[16];
3239     uint64_t error= 0, max_error=0;
3240
3241     for(i=0; i<COUNT; i++){
3242         int j;
3243 //        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
3244         for(j=0; j<16; j++){
3245             ref[j]= random()%255;
3246             src[j]= random()%255;
3247         }
3248
3249         h264_diff_dct_c(block, src, ref, 4);
3250
3251         //normalize
3252         for(j=0; j<16; j++){
3253 //            printf("%d ", block[j]);
3254             block[j]= block[j]*4;
3255             if(j&1) block[j]= (block[j]*4 + 2)/5;
3256             if(j&4) block[j]= (block[j]*4 + 2)/5;
3257         }
3258 //        printf("\n");
3259
3260         h->h264dsp.h264_idct_add(ref, block, 4);
3261 /*        for(j=0; j<16; j++){
3262             printf("%d ", ref[j]);
3263         }
3264         printf("\n");*/
3265
3266         for(j=0; j<16; j++){
3267             int diff= FFABS(src[j] - ref[j]);
3268
3269             error+= diff*diff;
3270             max_error= FFMAX(max_error, diff);
3271         }
3272     }
3273     printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
3274     printf("testing quantizer\n");
3275     for(qp=0; qp<52; qp++){
3276         for(i=0; i<16; i++)
3277             src1_block[i]= src2_block[i]= random()%255;
3278
3279     }
3280     printf("Testing NAL layer\n");
3281
3282     uint8_t bitstream[COUNT];
3283     uint8_t nal[COUNT*2];
3284     H264Context h;
3285     memset(&h, 0, sizeof(H264Context));
3286
3287     for(i=0; i<COUNT; i++){
3288         int zeros= i;
3289         int nal_length;
3290         int consumed;
3291         int out_length;
3292         uint8_t *out;
3293         int j;
3294
3295         for(j=0; j<COUNT; j++){
3296             bitstream[j]= (random() % 255) + 1;
3297         }
3298
3299         for(j=0; j<zeros; j++){
3300             int pos= random() % COUNT;
3301             while(bitstream[pos] == 0){
3302                 pos++;
3303                 pos %= COUNT;
3304             }
3305             bitstream[pos]=0;
3306         }
3307
3308         START_TIMER
3309
3310         nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
3311         if(nal_length<0){
3312             printf("encoding failed\n");
3313             return -1;
3314         }
3315
3316         out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
3317
3318         STOP_TIMER("NAL")
3319
3320         if(out_length != COUNT){
3321             printf("incorrect length %d %d\n", out_length, COUNT);
3322             return -1;
3323         }
3324
3325         if(consumed != nal_length){
3326             printf("incorrect consumed length %d %d\n", nal_length, consumed);
3327             return -1;
3328         }
3329
3330         if(memcmp(bitstream, out, COUNT)){
3331             printf("mismatch\n");
3332             return -1;
3333         }
3334     }
3335 #endif
3336
3337     printf("Testing RBSP\n");
3338
3339
3340     return 0;
3341 }
3342 #endif /* TEST */
3343
3344
3345 av_cold void ff_h264_free_context(H264Context *h)
3346 {
3347     int i;
3348
3349     free_tables(h); //FIXME cleanup init stuff perhaps
3350
3351     for(i = 0; i < MAX_SPS_COUNT; i++)
3352         av_freep(h->sps_buffers + i);
3353
3354     for(i = 0; i < MAX_PPS_COUNT; i++)
3355         av_freep(h->pps_buffers + i);
3356 }
3357
3358 av_cold int ff_h264_decode_end(AVCodecContext *avctx)
3359 {
3360     H264Context *h = avctx->priv_data;
3361     MpegEncContext *s = &h->s;
3362
3363     ff_h264_free_context(h);
3364
3365     MPV_common_end(s);
3366
3367 //    memset(h, 0, sizeof(H264Context));
3368
3369     return 0;
3370 }
3371
3372
3373 AVCodec h264_decoder = {
3374     "h264",
3375     AVMEDIA_TYPE_VIDEO,
3376     CODEC_ID_H264,
3377     sizeof(H264Context),
3378     ff_h264_decode_init,
3379     NULL,
3380     ff_h264_decode_end,
3381     decode_frame,
3382     /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3383     .flush= flush_dpb,
3384     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
3385     .pix_fmts= ff_hwaccel_pixfmt_list_420,
3386 };
3387
3388 #if CONFIG_H264_VDPAU_DECODER
3389 AVCodec h264_vdpau_decoder = {
3390     "h264_vdpau",
3391     AVMEDIA_TYPE_VIDEO,
3392     CODEC_ID_H264,
3393     sizeof(H264Context),
3394     ff_h264_decode_init,
3395     NULL,
3396     ff_h264_decode_end,
3397     decode_frame,
3398     CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3399     .flush= flush_dpb,
3400     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
3401     .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
3402 };
3403 #endif