H.264: split luma dc idct out and implement MMX/SSE2 versions
[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 "libavcore/imgutils.h"
29 #include "internal.h"
30 #include "dsputil.h"
31 #include "avcodec.h"
32 #include "mpegvideo.h"
33 #include "h264.h"
34 #include "h264data.h"
35 #include "h264_mvpred.h"
36 #include "h264_parser.h"
37 #include "golomb.h"
38 #include "mathops.h"
39 #include "rectangle.h"
40 #include "vdpau_internal.h"
41 #include "libavutil/avassert.h"
42
43 #include "cabac.h"
44
45 //#undef NDEBUG
46 #include <assert.h>
47
48 static const uint8_t rem6[52]={
49 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
50 };
51
52 static const uint8_t div6[52]={
53 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8,
54 };
55
56 static const enum PixelFormat hwaccel_pixfmt_list_h264_jpeg_420[] = {
57     PIX_FMT_DXVA2_VLD,
58     PIX_FMT_VAAPI_VLD,
59     PIX_FMT_YUVJ420P,
60     PIX_FMT_NONE
61 };
62
63 void ff_h264_write_back_intra_pred_mode(H264Context *h){
64     int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
65
66     AV_COPY32(mode, h->intra4x4_pred_mode_cache + 4 + 8*4);
67     mode[4]= h->intra4x4_pred_mode_cache[7+8*3];
68     mode[5]= h->intra4x4_pred_mode_cache[7+8*2];
69     mode[6]= h->intra4x4_pred_mode_cache[7+8*1];
70 }
71
72 /**
73  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
74  */
75 int ff_h264_check_intra4x4_pred_mode(H264Context *h){
76     MpegEncContext * const s = &h->s;
77     static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
78     static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
79     int i;
80
81     if(!(h->top_samples_available&0x8000)){
82         for(i=0; i<4; i++){
83             int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
84             if(status<0){
85                 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);
86                 return -1;
87             } else if(status){
88                 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
89             }
90         }
91     }
92
93     if((h->left_samples_available&0x8888)!=0x8888){
94         static const int mask[4]={0x8000,0x2000,0x80,0x20};
95         for(i=0; i<4; i++){
96             if(!(h->left_samples_available&mask[i])){
97                 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
98                 if(status<0){
99                     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);
100                     return -1;
101                 } else if(status){
102                     h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
103                 }
104             }
105         }
106     }
107
108     return 0;
109 } //FIXME cleanup like ff_h264_check_intra_pred_mode
110
111 /**
112  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
113  */
114 int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
115     MpegEncContext * const s = &h->s;
116     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
117     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
118
119     if(mode > 6U) {
120         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);
121         return -1;
122     }
123
124     if(!(h->top_samples_available&0x8000)){
125         mode= top[ mode ];
126         if(mode<0){
127             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);
128             return -1;
129         }
130     }
131
132     if((h->left_samples_available&0x8080) != 0x8080){
133         mode= left[ mode ];
134         if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
135             mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
136         }
137         if(mode<0){
138             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);
139             return -1;
140         }
141     }
142
143     return mode;
144 }
145
146 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
147     int i, si, di;
148     uint8_t *dst;
149     int bufidx;
150
151 //    src[0]&0x80;                //forbidden bit
152     h->nal_ref_idc= src[0]>>5;
153     h->nal_unit_type= src[0]&0x1F;
154
155     src++; length--;
156 #if 0
157     for(i=0; i<length; i++)
158         printf("%2X ", src[i]);
159 #endif
160
161 #if HAVE_FAST_UNALIGNED
162 # if HAVE_FAST_64BIT
163 #   define RS 7
164     for(i=0; i+1<length; i+=9){
165         if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
166 # else
167 #   define RS 3
168     for(i=0; i+1<length; i+=5){
169         if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
170 # endif
171             continue;
172         if(i>0 && !src[i]) i--;
173         while(src[i]) i++;
174 #else
175 #   define RS 0
176     for(i=0; i+1<length; i+=2){
177         if(src[i]) continue;
178         if(i>0 && src[i-1]==0) i--;
179 #endif
180         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
181             if(src[i+2]!=3){
182                 /* startcode, so we must be past the end */
183                 length=i;
184             }
185             break;
186         }
187         i-= RS;
188     }
189
190     if(i>=length-1){ //no escaped 0
191         *dst_length= length;
192         *consumed= length+1; //+1 for the header
193         return src;
194     }
195
196     bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
197     av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
198     dst= h->rbsp_buffer[bufidx];
199
200     if (dst == NULL){
201         return NULL;
202     }
203
204 //printf("decoding esc\n");
205     memcpy(dst, src, i);
206     si=di=i;
207     while(si+2<length){
208         //remove escapes (very rare 1:2^22)
209         if(src[si+2]>3){
210             dst[di++]= src[si++];
211             dst[di++]= src[si++];
212         }else if(src[si]==0 && src[si+1]==0){
213             if(src[si+2]==3){ //escape
214                 dst[di++]= 0;
215                 dst[di++]= 0;
216                 si+=3;
217                 continue;
218             }else //next start code
219                 goto nsc;
220         }
221
222         dst[di++]= src[si++];
223     }
224     while(si<length)
225         dst[di++]= src[si++];
226 nsc:
227
228     memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
229
230     *dst_length= di;
231     *consumed= si + 1;//+1 for the header
232 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
233     return dst;
234 }
235
236 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
237     int v= *src;
238     int r;
239
240     tprintf(h->s.avctx, "rbsp trailing %X\n", v);
241
242     for(r=1; r<9; r++){
243         if(v&1) return r;
244         v>>=1;
245     }
246     return 0;
247 }
248
249 #if 0
250 /**
251  * DCT transforms the 16 dc values.
252  * @param qp quantization parameter ??? FIXME
253  */
254 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
255 //    const int qmul= dequant_coeff[qp][0];
256     int i;
257     int temp[16]; //FIXME check if this is a good idea
258     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
259     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
260
261     for(i=0; i<4; i++){
262         const int offset= y_offset[i];
263         const int z0= block[offset+stride*0] + block[offset+stride*4];
264         const int z1= block[offset+stride*0] - block[offset+stride*4];
265         const int z2= block[offset+stride*1] - block[offset+stride*5];
266         const int z3= block[offset+stride*1] + block[offset+stride*5];
267
268         temp[4*i+0]= z0+z3;
269         temp[4*i+1]= z1+z2;
270         temp[4*i+2]= z1-z2;
271         temp[4*i+3]= z0-z3;
272     }
273
274     for(i=0; i<4; i++){
275         const int offset= x_offset[i];
276         const int z0= temp[4*0+i] + temp[4*2+i];
277         const int z1= temp[4*0+i] - temp[4*2+i];
278         const int z2= temp[4*1+i] - temp[4*3+i];
279         const int z3= temp[4*1+i] + temp[4*3+i];
280
281         block[stride*0 +offset]= (z0 + z3)>>1;
282         block[stride*2 +offset]= (z1 + z2)>>1;
283         block[stride*8 +offset]= (z1 - z2)>>1;
284         block[stride*10+offset]= (z0 - z3)>>1;
285     }
286 }
287 #endif
288
289 #undef xStride
290 #undef stride
291
292 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
293     const int stride= 16*2;
294     const int xStride= 16;
295     int a,b,c,d,e;
296
297     a= block[stride*0 + xStride*0];
298     b= block[stride*0 + xStride*1];
299     c= block[stride*1 + xStride*0];
300     d= block[stride*1 + xStride*1];
301
302     e= a-b;
303     a= a+b;
304     b= c-d;
305     c= c+d;
306
307     block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
308     block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
309     block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
310     block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
311 }
312
313 #if 0
314 static void chroma_dc_dct_c(DCTELEM *block){
315     const int stride= 16*2;
316     const int xStride= 16;
317     int a,b,c,d,e;
318
319     a= block[stride*0 + xStride*0];
320     b= block[stride*0 + xStride*1];
321     c= block[stride*1 + xStride*0];
322     d= block[stride*1 + xStride*1];
323
324     e= a-b;
325     a= a+b;
326     b= c-d;
327     c= c+d;
328
329     block[stride*0 + xStride*0]= (a+c);
330     block[stride*0 + xStride*1]= (e+b);
331     block[stride*1 + xStride*0]= (a-c);
332     block[stride*1 + xStride*1]= (e-b);
333 }
334 #endif
335
336 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
337                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
338                            int src_x_offset, int src_y_offset,
339                            qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
340     MpegEncContext * const s = &h->s;
341     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
342     int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
343     const int luma_xy= (mx&3) + ((my&3)<<2);
344     uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
345     uint8_t * src_cb, * src_cr;
346     int extra_width= h->emu_edge_width;
347     int extra_height= h->emu_edge_height;
348     int emu=0;
349     const int full_mx= mx>>2;
350     const int full_my= my>>2;
351     const int pic_width  = 16*s->mb_width;
352     const int pic_height = 16*s->mb_height >> MB_FIELD;
353
354     if(mx&7) extra_width -= 3;
355     if(my&7) extra_height -= 3;
356
357     if(   full_mx < 0-extra_width
358        || full_my < 0-extra_height
359        || full_mx + 16/*FIXME*/ > pic_width + extra_width
360        || full_my + 16/*FIXME*/ > pic_height + extra_height){
361         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);
362             src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
363         emu=1;
364     }
365
366     qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
367     if(!square){
368         qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
369     }
370
371     if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
372
373     if(MB_FIELD){
374         // chroma offset when predicting from a field of opposite parity
375         my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
376         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
377     }
378     src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
379     src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
380
381     if(emu){
382         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);
383             src_cb= s->edge_emu_buffer;
384     }
385     chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
386
387     if(emu){
388         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);
389             src_cr= s->edge_emu_buffer;
390     }
391     chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
392 }
393
394 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
395                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
396                            int x_offset, int y_offset,
397                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
398                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
399                            int list0, int list1){
400     MpegEncContext * const s = &h->s;
401     qpel_mc_func *qpix_op=  qpix_put;
402     h264_chroma_mc_func chroma_op= chroma_put;
403
404     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
405     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
406     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
407     x_offset += 8*s->mb_x;
408     y_offset += 8*(s->mb_y >> MB_FIELD);
409
410     if(list0){
411         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
412         mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
413                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
414                            qpix_op, chroma_op);
415
416         qpix_op=  qpix_avg;
417         chroma_op= chroma_avg;
418     }
419
420     if(list1){
421         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
422         mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
423                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
424                            qpix_op, chroma_op);
425     }
426 }
427
428 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
429                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
430                            int x_offset, int y_offset,
431                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
432                            h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
433                            h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
434                            int list0, int list1){
435     MpegEncContext * const s = &h->s;
436
437     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
438     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
439     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
440     x_offset += 8*s->mb_x;
441     y_offset += 8*(s->mb_y >> MB_FIELD);
442
443     if(list0 && list1){
444         /* don't optimize for luma-only case, since B-frames usually
445          * use implicit weights => chroma too. */
446         uint8_t *tmp_cb = s->obmc_scratchpad;
447         uint8_t *tmp_cr = s->obmc_scratchpad + 8;
448         uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
449         int refn0 = h->ref_cache[0][ scan8[n] ];
450         int refn1 = h->ref_cache[1][ scan8[n] ];
451
452         mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
453                     dest_y, dest_cb, dest_cr,
454                     x_offset, y_offset, qpix_put, chroma_put);
455         mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
456                     tmp_y, tmp_cb, tmp_cr,
457                     x_offset, y_offset, qpix_put, chroma_put);
458
459         if(h->use_weight == 2){
460             int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
461             int weight1 = 64 - weight0;
462             luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
463             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
464             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
465         }else{
466             luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
467                             h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
468                             h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
469             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
470                             h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
471                             h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
472             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
473                             h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
474                             h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
475         }
476     }else{
477         int list = list1 ? 1 : 0;
478         int refn = h->ref_cache[list][ scan8[n] ];
479         Picture *ref= &h->ref_list[list][refn];
480         mc_dir_part(h, ref, n, square, chroma_height, delta, list,
481                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
482                     qpix_put, chroma_put);
483
484         luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
485                        h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
486         if(h->use_weight_chroma){
487             chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
488                              h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
489             chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
490                              h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
491         }
492     }
493 }
494
495 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
496                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
497                            int x_offset, int y_offset,
498                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
499                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
500                            h264_weight_func *weight_op, h264_biweight_func *weight_avg,
501                            int list0, int list1){
502     if((h->use_weight==2 && list0 && list1
503         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
504        || h->use_weight==1)
505         mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
506                          x_offset, y_offset, qpix_put, chroma_put,
507                          weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
508     else
509         mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
510                     x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
511 }
512
513 static inline void prefetch_motion(H264Context *h, int list){
514     /* fetch pixels for estimated mv 4 macroblocks ahead
515      * optimized for 64byte cache lines */
516     MpegEncContext * const s = &h->s;
517     const int refn = h->ref_cache[list][scan8[0]];
518     if(refn >= 0){
519         const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
520         const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
521         uint8_t **src= h->ref_list[list][refn].data;
522         int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
523         s->dsp.prefetch(src[0]+off, s->linesize, 4);
524         off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
525         s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
526     }
527 }
528
529 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
530                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
531                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
532                       h264_weight_func *weight_op, h264_biweight_func *weight_avg){
533     MpegEncContext * const s = &h->s;
534     const int mb_xy= h->mb_xy;
535     const int mb_type= s->current_picture.mb_type[mb_xy];
536
537     assert(IS_INTER(mb_type));
538
539     prefetch_motion(h, 0);
540
541     if(IS_16X16(mb_type)){
542         mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
543                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
544                 weight_op, weight_avg,
545                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
546     }else if(IS_16X8(mb_type)){
547         mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
548                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
549                 &weight_op[1], &weight_avg[1],
550                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
551         mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
552                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
553                 &weight_op[1], &weight_avg[1],
554                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
555     }else if(IS_8X16(mb_type)){
556         mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
557                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
558                 &weight_op[2], &weight_avg[2],
559                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
560         mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
561                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
562                 &weight_op[2], &weight_avg[2],
563                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
564     }else{
565         int i;
566
567         assert(IS_8X8(mb_type));
568
569         for(i=0; i<4; i++){
570             const int sub_mb_type= h->sub_mb_type[i];
571             const int n= 4*i;
572             int x_offset= (i&1)<<2;
573             int y_offset= (i&2)<<1;
574
575             if(IS_SUB_8X8(sub_mb_type)){
576                 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
577                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
578                     &weight_op[3], &weight_avg[3],
579                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
580             }else if(IS_SUB_8X4(sub_mb_type)){
581                 mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
582                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
583                     &weight_op[4], &weight_avg[4],
584                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
585                 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
586                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
587                     &weight_op[4], &weight_avg[4],
588                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
589             }else if(IS_SUB_4X8(sub_mb_type)){
590                 mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
591                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
592                     &weight_op[5], &weight_avg[5],
593                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
594                 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
595                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
596                     &weight_op[5], &weight_avg[5],
597                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
598             }else{
599                 int j;
600                 assert(IS_SUB_4X4(sub_mb_type));
601                 for(j=0; j<4; j++){
602                     int sub_x_offset= x_offset + 2*(j&1);
603                     int sub_y_offset= y_offset +   (j&2);
604                     mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
605                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
606                         &weight_op[6], &weight_avg[6],
607                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
608                 }
609             }
610         }
611     }
612
613     prefetch_motion(h, 1);
614 }
615
616
617 static void free_tables(H264Context *h){
618     int i;
619     H264Context *hx;
620     av_freep(&h->intra4x4_pred_mode);
621     av_freep(&h->chroma_pred_mode_table);
622     av_freep(&h->cbp_table);
623     av_freep(&h->mvd_table[0]);
624     av_freep(&h->mvd_table[1]);
625     av_freep(&h->direct_table);
626     av_freep(&h->non_zero_count);
627     av_freep(&h->slice_table_base);
628     h->slice_table= NULL;
629     av_freep(&h->list_counts);
630
631     av_freep(&h->mb2b_xy);
632     av_freep(&h->mb2br_xy);
633
634     for(i = 0; i < MAX_THREADS; i++) {
635         hx = h->thread_context[i];
636         if(!hx) continue;
637         av_freep(&hx->top_borders[1]);
638         av_freep(&hx->top_borders[0]);
639         av_freep(&hx->s.obmc_scratchpad);
640         av_freep(&hx->rbsp_buffer[1]);
641         av_freep(&hx->rbsp_buffer[0]);
642         hx->rbsp_buffer_size[0] = 0;
643         hx->rbsp_buffer_size[1] = 0;
644         if (i) av_freep(&h->thread_context[i]);
645     }
646 }
647
648 static void init_dequant8_coeff_table(H264Context *h){
649     int i,q,x;
650     h->dequant8_coeff[0] = h->dequant8_buffer[0];
651     h->dequant8_coeff[1] = h->dequant8_buffer[1];
652
653     for(i=0; i<2; i++ ){
654         if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
655             h->dequant8_coeff[1] = h->dequant8_buffer[0];
656             break;
657         }
658
659         for(q=0; q<52; q++){
660             int shift = div6[q];
661             int idx = rem6[q];
662             for(x=0; x<64; x++)
663                 h->dequant8_coeff[i][q][(x>>3)|((x&7)<<3)] =
664                     ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
665                     h->pps.scaling_matrix8[i][x]) << shift;
666         }
667     }
668 }
669
670 static void init_dequant4_coeff_table(H264Context *h){
671     int i,j,q,x;
672     for(i=0; i<6; i++ ){
673         h->dequant4_coeff[i] = h->dequant4_buffer[i];
674         for(j=0; j<i; j++){
675             if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
676                 h->dequant4_coeff[i] = h->dequant4_buffer[j];
677                 break;
678             }
679         }
680         if(j<i)
681             continue;
682
683         for(q=0; q<52; q++){
684             int shift = div6[q] + 2;
685             int idx = rem6[q];
686             for(x=0; x<16; x++)
687                 h->dequant4_coeff[i][q][(x>>2)|((x<<2)&0xF)] =
688                     ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
689                     h->pps.scaling_matrix4[i][x]) << shift;
690         }
691     }
692 }
693
694 static void init_dequant_tables(H264Context *h){
695     int i,x;
696     init_dequant4_coeff_table(h);
697     if(h->pps.transform_8x8_mode)
698         init_dequant8_coeff_table(h);
699     if(h->sps.transform_bypass){
700         for(i=0; i<6; i++)
701             for(x=0; x<16; x++)
702                 h->dequant4_coeff[i][0][x] = 1<<6;
703         if(h->pps.transform_8x8_mode)
704             for(i=0; i<2; i++)
705                 for(x=0; x<64; x++)
706                     h->dequant8_coeff[i][0][x] = 1<<6;
707     }
708 }
709
710
711 int ff_h264_alloc_tables(H264Context *h){
712     MpegEncContext * const s = &h->s;
713     const int big_mb_num= s->mb_stride * (s->mb_height+1);
714     const int row_mb_num= 2*s->mb_stride*s->avctx->thread_count;
715     int x,y;
716
717     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8  * sizeof(uint8_t), fail)
718
719     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 32 * sizeof(uint8_t), fail)
720     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
721     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
722
723     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
724     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
725     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
726     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
727     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
728
729     memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
730     h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
731
732     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy  , big_mb_num * sizeof(uint32_t), fail);
733     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
734     for(y=0; y<s->mb_height; y++){
735         for(x=0; x<s->mb_width; x++){
736             const int mb_xy= x + y*s->mb_stride;
737             const int b_xy = 4*x + 4*y*h->b_stride;
738
739             h->mb2b_xy [mb_xy]= b_xy;
740             h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
741         }
742     }
743
744     s->obmc_scratchpad = NULL;
745
746     if(!h->dequant4_coeff[0])
747         init_dequant_tables(h);
748
749     return 0;
750 fail:
751     free_tables(h);
752     return -1;
753 }
754
755 /**
756  * Mimic alloc_tables(), but for every context thread.
757  */
758 static void clone_tables(H264Context *dst, H264Context *src, int i){
759     MpegEncContext * const s = &src->s;
760     dst->intra4x4_pred_mode       = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
761     dst->non_zero_count           = src->non_zero_count;
762     dst->slice_table              = src->slice_table;
763     dst->cbp_table                = src->cbp_table;
764     dst->mb2b_xy                  = src->mb2b_xy;
765     dst->mb2br_xy                 = src->mb2br_xy;
766     dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
767     dst->mvd_table[0]             = src->mvd_table[0] + i*8*2*s->mb_stride;
768     dst->mvd_table[1]             = src->mvd_table[1] + i*8*2*s->mb_stride;
769     dst->direct_table             = src->direct_table;
770     dst->list_counts              = src->list_counts;
771
772     dst->s.obmc_scratchpad = NULL;
773     ff_h264_pred_init(&dst->hpc, src->s.codec_id);
774 }
775
776 /**
777  * Init context
778  * Allocate buffers which are not shared amongst multiple threads.
779  */
780 static int context_init(H264Context *h){
781     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
782     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
783
784     h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
785     h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
786
787     return 0;
788 fail:
789     return -1; // free_tables will clean up for us
790 }
791
792 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
793
794 static av_cold void common_init(H264Context *h){
795     MpegEncContext * const s = &h->s;
796
797     s->width = s->avctx->width;
798     s->height = s->avctx->height;
799     s->codec_id= s->avctx->codec->id;
800
801     ff_h264dsp_init(&h->h264dsp);
802     ff_h264_pred_init(&h->hpc, s->codec_id);
803
804     h->dequant_coeff_pps= -1;
805     s->unrestricted_mv=1;
806     s->decode=1; //FIXME
807
808     dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
809
810     memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
811     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
812 }
813
814 int ff_h264_decode_extradata(H264Context *h)
815 {
816     AVCodecContext *avctx = h->s.avctx;
817
818     if(*(char *)avctx->extradata == 1){
819         int i, cnt, nalsize;
820         unsigned char *p = avctx->extradata;
821
822         h->is_avc = 1;
823
824         if(avctx->extradata_size < 7) {
825             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
826             return -1;
827         }
828         /* sps and pps in the avcC always have length coded with 2 bytes,
829            so put a fake nal_length_size = 2 while parsing them */
830         h->nal_length_size = 2;
831         // Decode sps from avcC
832         cnt = *(p+5) & 0x1f; // Number of sps
833         p += 6;
834         for (i = 0; i < cnt; i++) {
835             nalsize = AV_RB16(p) + 2;
836             if(decode_nal_units(h, p, nalsize) < 0) {
837                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
838                 return -1;
839             }
840             p += nalsize;
841         }
842         // Decode pps from avcC
843         cnt = *(p++); // Number of pps
844         for (i = 0; i < cnt; i++) {
845             nalsize = AV_RB16(p) + 2;
846             if(decode_nal_units(h, p, nalsize)  != nalsize) {
847                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
848                 return -1;
849             }
850             p += nalsize;
851         }
852         // Now store right nal length size, that will be use to parse all other nals
853         h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
854     } else {
855         h->is_avc = 0;
856         if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
857             return -1;
858     }
859     return 0;
860 }
861
862 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
863     H264Context *h= avctx->priv_data;
864     MpegEncContext * const s = &h->s;
865
866     MPV_decode_defaults(s);
867
868     s->avctx = avctx;
869     common_init(h);
870
871     s->out_format = FMT_H264;
872     s->workaround_bugs= avctx->workaround_bugs;
873
874     // set defaults
875 //    s->decode_mb= ff_h263_decode_mb;
876     s->quarter_sample = 1;
877     if(!avctx->has_b_frames)
878     s->low_delay= 1;
879
880     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
881
882     ff_h264_decode_init_vlc();
883
884     h->thread_context[0] = h;
885     h->outputed_poc = INT_MIN;
886     h->prev_poc_msb= 1<<16;
887     h->x264_build = -1;
888     ff_h264_reset_sei(h);
889     if(avctx->codec_id == CODEC_ID_H264){
890         if(avctx->ticks_per_frame == 1){
891             s->avctx->time_base.den *=2;
892         }
893         avctx->ticks_per_frame = 2;
894     }
895
896     if(avctx->extradata_size > 0 && avctx->extradata &&
897         ff_h264_decode_extradata(h))
898         return -1;
899
900     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
901         s->avctx->has_b_frames = h->sps.num_reorder_frames;
902         s->low_delay = 0;
903     }
904
905     return 0;
906 }
907
908 int ff_h264_frame_start(H264Context *h){
909     MpegEncContext * const s = &h->s;
910     int i;
911
912     if(MPV_frame_start(s, s->avctx) < 0)
913         return -1;
914     ff_er_frame_start(s);
915     /*
916      * MPV_frame_start uses pict_type to derive key_frame.
917      * This is incorrect for H.264; IDR markings must be used.
918      * Zero here; IDR markings per slice in frame or fields are ORed in later.
919      * See decode_nal_units().
920      */
921     s->current_picture_ptr->key_frame= 0;
922     s->current_picture_ptr->mmco_reset= 0;
923
924     assert(s->linesize && s->uvlinesize);
925
926     for(i=0; i<16; i++){
927         h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
928         h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
929     }
930     for(i=0; i<4; i++){
931         h->block_offset[16+i]=
932         h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
933         h->block_offset[24+16+i]=
934         h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
935     }
936
937     /* can't be in alloc_tables because linesize isn't known there.
938      * FIXME: redo bipred weight to not require extra buffer? */
939     for(i = 0; i < s->avctx->thread_count; i++)
940         if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
941             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
942
943     /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
944     memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
945
946 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
947
948     // We mark the current picture as non-reference after allocating it, so
949     // that if we break out due to an error it can be released automatically
950     // in the next MPV_frame_start().
951     // SVQ3 as well as most other codecs have only last/next/current and thus
952     // get released even with set reference, besides SVQ3 and others do not
953     // mark frames as reference later "naturally".
954     if(s->codec_id != CODEC_ID_SVQ3)
955         s->current_picture_ptr->reference= 0;
956
957     s->current_picture_ptr->field_poc[0]=
958     s->current_picture_ptr->field_poc[1]= INT_MAX;
959     assert(s->current_picture_ptr->long_ref==0);
960
961     return 0;
962 }
963
964 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){
965     MpegEncContext * const s = &h->s;
966     uint8_t *top_border;
967     int top_idx = 1;
968
969     src_y  -=   linesize;
970     src_cb -= uvlinesize;
971     src_cr -= uvlinesize;
972
973     if(!simple && FRAME_MBAFF){
974         if(s->mb_y&1){
975             if(!MB_MBAFF){
976                 top_border = h->top_borders[0][s->mb_x];
977                 AV_COPY128(top_border, src_y + 15*linesize);
978                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
979                     AV_COPY64(top_border+16, src_cb+7*uvlinesize);
980                     AV_COPY64(top_border+24, src_cr+7*uvlinesize);
981                 }
982             }
983         }else if(MB_MBAFF){
984             top_idx = 0;
985         }else
986             return;
987     }
988
989     top_border = h->top_borders[top_idx][s->mb_x];
990     // There are two lines saved, the line above the the top macroblock of a pair,
991     // and the line above the bottom macroblock
992     AV_COPY128(top_border, src_y + 16*linesize);
993
994     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
995         AV_COPY64(top_border+16, src_cb+8*uvlinesize);
996         AV_COPY64(top_border+24, src_cr+8*uvlinesize);
997     }
998 }
999
1000 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){
1001     MpegEncContext * const s = &h->s;
1002     int deblock_left;
1003     int deblock_top;
1004     int top_idx = 1;
1005     uint8_t *top_border_m1;
1006     uint8_t *top_border;
1007
1008     if(!simple && FRAME_MBAFF){
1009         if(s->mb_y&1){
1010             if(!MB_MBAFF)
1011                 return;
1012         }else{
1013             top_idx = MB_MBAFF ? 0 : 1;
1014         }
1015     }
1016
1017     if(h->deblocking_filter == 2) {
1018         deblock_left = h->left_type[0];
1019         deblock_top  = h->top_type;
1020     } else {
1021         deblock_left = (s->mb_x > 0);
1022         deblock_top =  (s->mb_y > !!MB_FIELD);
1023     }
1024
1025     src_y  -=   linesize + 1;
1026     src_cb -= uvlinesize + 1;
1027     src_cr -= uvlinesize + 1;
1028
1029     top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1030     top_border    = h->top_borders[top_idx][s->mb_x];
1031
1032 #define XCHG(a,b,xchg)\
1033 if (xchg) AV_SWAP64(b,a);\
1034 else      AV_COPY64(b,a);
1035
1036     if(deblock_top){
1037         if(deblock_left){
1038             XCHG(top_border_m1+8, src_y -7, 1);
1039         }
1040         XCHG(top_border+0, src_y +1, xchg);
1041         XCHG(top_border+8, src_y +9, 1);
1042         if(s->mb_x+1 < s->mb_width){
1043             XCHG(h->top_borders[top_idx][s->mb_x+1], src_y +17, 1);
1044         }
1045     }
1046
1047     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1048         if(deblock_top){
1049             if(deblock_left){
1050                 XCHG(top_border_m1+16, src_cb -7, 1);
1051                 XCHG(top_border_m1+24, src_cr -7, 1);
1052             }
1053             XCHG(top_border+16, src_cb+1, 1);
1054             XCHG(top_border+24, src_cr+1, 1);
1055         }
1056     }
1057 }
1058
1059 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
1060     MpegEncContext * const s = &h->s;
1061     const int mb_x= s->mb_x;
1062     const int mb_y= s->mb_y;
1063     const int mb_xy= h->mb_xy;
1064     const int mb_type= s->current_picture.mb_type[mb_xy];
1065     uint8_t  *dest_y, *dest_cb, *dest_cr;
1066     int linesize, uvlinesize /*dct_offset*/;
1067     int i;
1068     int *block_offset = &h->block_offset[0];
1069     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1070     /* is_h264 should always be true if SVQ3 is disabled. */
1071     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1072     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1073     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1074
1075     dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
1076     dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
1077     dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
1078
1079     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
1080     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
1081
1082     h->list_counts[mb_xy]= h->list_count;
1083
1084     if (!simple && MB_FIELD) {
1085         linesize   = h->mb_linesize   = s->linesize * 2;
1086         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1087         block_offset = &h->block_offset[24];
1088         if(mb_y&1){ //FIXME move out of this function?
1089             dest_y -= s->linesize*15;
1090             dest_cb-= s->uvlinesize*7;
1091             dest_cr-= s->uvlinesize*7;
1092         }
1093         if(FRAME_MBAFF) {
1094             int list;
1095             for(list=0; list<h->list_count; list++){
1096                 if(!USES_LIST(mb_type, list))
1097                     continue;
1098                 if(IS_16X16(mb_type)){
1099                     int8_t *ref = &h->ref_cache[list][scan8[0]];
1100                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1101                 }else{
1102                     for(i=0; i<16; i+=4){
1103                         int ref = h->ref_cache[list][scan8[i]];
1104                         if(ref >= 0)
1105                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1106                     }
1107                 }
1108             }
1109         }
1110     } else {
1111         linesize   = h->mb_linesize   = s->linesize;
1112         uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1113 //        dct_offset = s->linesize * 16;
1114     }
1115
1116     if (!simple && IS_INTRA_PCM(mb_type)) {
1117         for (i=0; i<16; i++) {
1118             memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1119         }
1120         for (i=0; i<8; i++) {
1121             memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
1122             memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
1123         }
1124     } else {
1125         if(IS_INTRA(mb_type)){
1126             if(h->deblocking_filter)
1127                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
1128
1129             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1130                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1131                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1132             }
1133
1134             if(IS_INTRA4x4(mb_type)){
1135                 if(simple || !s->encoding){
1136                     if(IS_8x8DCT(mb_type)){
1137                         if(transform_bypass){
1138                             idct_dc_add =
1139                             idct_add    = s->dsp.add_pixels8;
1140                         }else{
1141                             idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1142                             idct_add    = h->h264dsp.h264_idct8_add;
1143                         }
1144                         for(i=0; i<16; i+=4){
1145                             uint8_t * const ptr= dest_y + block_offset[i];
1146                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1147                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1148                                 h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
1149                             }else{
1150                                 const int nnz = h->non_zero_count_cache[ scan8[i] ];
1151                                 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1152                                                             (h->topright_samples_available<<i)&0x4000, linesize);
1153                                 if(nnz){
1154                                     if(nnz == 1 && h->mb[i*16])
1155                                         idct_dc_add(ptr, h->mb + i*16, linesize);
1156                                     else
1157                                         idct_add   (ptr, h->mb + i*16, linesize);
1158                                 }
1159                             }
1160                         }
1161                     }else{
1162                         if(transform_bypass){
1163                             idct_dc_add =
1164                             idct_add    = s->dsp.add_pixels4;
1165                         }else{
1166                             idct_dc_add = h->h264dsp.h264_idct_dc_add;
1167                             idct_add    = h->h264dsp.h264_idct_add;
1168                         }
1169                         for(i=0; i<16; i++){
1170                             uint8_t * const ptr= dest_y + block_offset[i];
1171                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1172
1173                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1174                                 h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
1175                             }else{
1176                                 uint8_t *topright;
1177                                 int nnz, tr;
1178                                 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1179                                     const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1180                                     assert(mb_y || linesize <= block_offset[i]);
1181                                     if(!topright_avail){
1182                                         tr= ptr[3 - linesize]*0x01010101;
1183                                         topright= (uint8_t*) &tr;
1184                                     }else
1185                                         topright= ptr + 4 - linesize;
1186                                 }else
1187                                     topright= NULL;
1188
1189                                 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1190                                 nnz = h->non_zero_count_cache[ scan8[i] ];
1191                                 if(nnz){
1192                                     if(is_h264){
1193                                         if(nnz == 1 && h->mb[i*16])
1194                                             idct_dc_add(ptr, h->mb + i*16, linesize);
1195                                         else
1196                                             idct_add   (ptr, h->mb + i*16, linesize);
1197                                     }else
1198                                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1199                                 }
1200                             }
1201                         }
1202                     }
1203                 }
1204             }else{
1205                 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1206                 if(is_h264){
1207                     if(!transform_bypass)
1208                         h->h264dsp.h264_luma_dc_dequant_idct(h->mb, h->mb_luma_dc, h->dequant4_coeff[0][s->qscale][0]);
1209                     else{
1210                         static const uint8_t dc_mapping[16] = { 0*16, 1*16, 4*16, 5*16, 2*16, 3*16, 6*16, 7*16,
1211                                                                 8*16, 9*16,12*16,13*16,10*16,11*16,14*16,15*16};
1212                         for(i = 0; i < 16; i++)
1213                             h->mb[dc_mapping[i]] = h->mb_luma_dc[i];
1214                     }
1215                 }else
1216                     ff_svq3_luma_dc_dequant_idct_c(h->mb, h->mb_luma_dc, s->qscale);
1217             }
1218             if(h->deblocking_filter)
1219                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
1220         }else if(is_h264){
1221             hl_motion(h, dest_y, dest_cb, dest_cr,
1222                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1223                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1224                       h->h264dsp.weight_h264_pixels_tab, h->h264dsp.biweight_h264_pixels_tab);
1225         }
1226
1227
1228         if(!IS_INTRA4x4(mb_type)){
1229             if(is_h264){
1230                 if(IS_INTRA16x16(mb_type)){
1231                     if(transform_bypass){
1232                         if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1233                             h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1234                         }else{
1235                             for(i=0; i<16; i++){
1236                                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1237                                     s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
1238                             }
1239                         }
1240                     }else{
1241                          h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1242                     }
1243                 }else if(h->cbp&15){
1244                     if(transform_bypass){
1245                         const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1246                         idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1247                         for(i=0; i<16; i+=di){
1248                             if(h->non_zero_count_cache[ scan8[i] ]){
1249                                 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
1250                             }
1251                         }
1252                     }else{
1253                         if(IS_8x8DCT(mb_type)){
1254                             h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1255                         }else{
1256                             h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1257                         }
1258                     }
1259                 }
1260             }else{
1261                 for(i=0; i<16; i++){
1262                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1263                         uint8_t * const ptr= dest_y + block_offset[i];
1264                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1265                     }
1266                 }
1267             }
1268         }
1269
1270         if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1271             uint8_t *dest[2] = {dest_cb, dest_cr};
1272             if(transform_bypass){
1273                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1274                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
1275                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
1276                 }else{
1277                     idct_add = s->dsp.add_pixels4;
1278                     for(i=16; i<16+8; i++){
1279                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1280                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1281                     }
1282                 }
1283             }else{
1284                 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]);
1285                 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]);
1286                 if(is_h264){
1287                     h->h264dsp.h264_idct_add8(dest, block_offset,
1288                                               h->mb, uvlinesize,
1289                                               h->non_zero_count_cache);
1290                 }else{
1291                     for(i=16; i<16+8; i++){
1292                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1293                             uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1294                             ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
1295                         }
1296                     }
1297                 }
1298             }
1299         }
1300     }
1301     if(h->cbp || IS_INTRA(mb_type))
1302         s->dsp.clear_blocks(h->mb);
1303 }
1304
1305 /**
1306  * Process a macroblock; this case avoids checks for expensive uncommon cases.
1307  */
1308 static void hl_decode_mb_simple(H264Context *h){
1309     hl_decode_mb_internal(h, 1);
1310 }
1311
1312 /**
1313  * Process a macroblock; this handles edge cases, such as interlacing.
1314  */
1315 static void av_noinline hl_decode_mb_complex(H264Context *h){
1316     hl_decode_mb_internal(h, 0);
1317 }
1318
1319 void ff_h264_hl_decode_mb(H264Context *h){
1320     MpegEncContext * const s = &h->s;
1321     const int mb_xy= h->mb_xy;
1322     const int mb_type= s->current_picture.mb_type[mb_xy];
1323     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1324
1325     if (is_complex)
1326         hl_decode_mb_complex(h);
1327     else hl_decode_mb_simple(h);
1328 }
1329
1330 static int pred_weight_table(H264Context *h){
1331     MpegEncContext * const s = &h->s;
1332     int list, i;
1333     int luma_def, chroma_def;
1334
1335     h->use_weight= 0;
1336     h->use_weight_chroma= 0;
1337     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1338     if(CHROMA)
1339         h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1340     luma_def = 1<<h->luma_log2_weight_denom;
1341     chroma_def = 1<<h->chroma_log2_weight_denom;
1342
1343     for(list=0; list<2; list++){
1344         h->luma_weight_flag[list]   = 0;
1345         h->chroma_weight_flag[list] = 0;
1346         for(i=0; i<h->ref_count[list]; i++){
1347             int luma_weight_flag, chroma_weight_flag;
1348
1349             luma_weight_flag= get_bits1(&s->gb);
1350             if(luma_weight_flag){
1351                 h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
1352                 h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
1353                 if(   h->luma_weight[i][list][0] != luma_def
1354                    || h->luma_weight[i][list][1] != 0) {
1355                     h->use_weight= 1;
1356                     h->luma_weight_flag[list]= 1;
1357                 }
1358             }else{
1359                 h->luma_weight[i][list][0]= luma_def;
1360                 h->luma_weight[i][list][1]= 0;
1361             }
1362
1363             if(CHROMA){
1364                 chroma_weight_flag= get_bits1(&s->gb);
1365                 if(chroma_weight_flag){
1366                     int j;
1367                     for(j=0; j<2; j++){
1368                         h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
1369                         h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
1370                         if(   h->chroma_weight[i][list][j][0] != chroma_def
1371                            || h->chroma_weight[i][list][j][1] != 0) {
1372                             h->use_weight_chroma= 1;
1373                             h->chroma_weight_flag[list]= 1;
1374                         }
1375                     }
1376                 }else{
1377                     int j;
1378                     for(j=0; j<2; j++){
1379                         h->chroma_weight[i][list][j][0]= chroma_def;
1380                         h->chroma_weight[i][list][j][1]= 0;
1381                     }
1382                 }
1383             }
1384         }
1385         if(h->slice_type_nos != FF_B_TYPE) break;
1386     }
1387     h->use_weight= h->use_weight || h->use_weight_chroma;
1388     return 0;
1389 }
1390
1391 /**
1392  * Initialize implicit_weight table.
1393  * @param field  0/1 initialize the weight for interlaced MBAFF
1394  *                -1 initializes the rest
1395  */
1396 static void implicit_weight_table(H264Context *h, int field){
1397     MpegEncContext * const s = &h->s;
1398     int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
1399
1400     for (i = 0; i < 2; i++) {
1401         h->luma_weight_flag[i]   = 0;
1402         h->chroma_weight_flag[i] = 0;
1403     }
1404
1405     if(field < 0){
1406         cur_poc = s->current_picture_ptr->poc;
1407     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
1408        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1409         h->use_weight= 0;
1410         h->use_weight_chroma= 0;
1411         return;
1412     }
1413         ref_start= 0;
1414         ref_count0= h->ref_count[0];
1415         ref_count1= h->ref_count[1];
1416     }else{
1417         cur_poc = s->current_picture_ptr->field_poc[field];
1418         ref_start= 16;
1419         ref_count0= 16+2*h->ref_count[0];
1420         ref_count1= 16+2*h->ref_count[1];
1421     }
1422
1423     h->use_weight= 2;
1424     h->use_weight_chroma= 2;
1425     h->luma_log2_weight_denom= 5;
1426     h->chroma_log2_weight_denom= 5;
1427
1428     for(ref0=ref_start; ref0 < ref_count0; ref0++){
1429         int poc0 = h->ref_list[0][ref0].poc;
1430         for(ref1=ref_start; ref1 < ref_count1; ref1++){
1431             int poc1 = h->ref_list[1][ref1].poc;
1432             int td = av_clip(poc1 - poc0, -128, 127);
1433             int w= 32;
1434             if(td){
1435                 int tb = av_clip(cur_poc - poc0, -128, 127);
1436                 int tx = (16384 + (FFABS(td) >> 1)) / td;
1437                 int dist_scale_factor = (tb*tx + 32) >> 8;
1438                 if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
1439                     w = 64 - dist_scale_factor;
1440             }
1441             if(field<0){
1442                 h->implicit_weight[ref0][ref1][0]=
1443                 h->implicit_weight[ref0][ref1][1]= w;
1444             }else{
1445                 h->implicit_weight[ref0][ref1][field]=w;
1446             }
1447         }
1448     }
1449 }
1450
1451 /**
1452  * instantaneous decoder refresh.
1453  */
1454 static void idr(H264Context *h){
1455     ff_h264_remove_all_refs(h);
1456     h->prev_frame_num= 0;
1457     h->prev_frame_num_offset= 0;
1458     h->prev_poc_msb=
1459     h->prev_poc_lsb= 0;
1460 }
1461
1462 /* forget old pics after a seek */
1463 static void flush_dpb(AVCodecContext *avctx){
1464     H264Context *h= avctx->priv_data;
1465     int i;
1466     for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1467         if(h->delayed_pic[i])
1468             h->delayed_pic[i]->reference= 0;
1469         h->delayed_pic[i]= NULL;
1470     }
1471     h->outputed_poc= INT_MIN;
1472     h->prev_interlaced_frame = 1;
1473     idr(h);
1474     if(h->s.current_picture_ptr)
1475         h->s.current_picture_ptr->reference= 0;
1476     h->s.first_field= 0;
1477     ff_h264_reset_sei(h);
1478     ff_mpeg_flush(avctx);
1479 }
1480
1481 static int init_poc(H264Context *h){
1482     MpegEncContext * const s = &h->s;
1483     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1484     int field_poc[2];
1485     Picture *cur = s->current_picture_ptr;
1486
1487     h->frame_num_offset= h->prev_frame_num_offset;
1488     if(h->frame_num < h->prev_frame_num)
1489         h->frame_num_offset += max_frame_num;
1490
1491     if(h->sps.poc_type==0){
1492         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1493
1494         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1495             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1496         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1497             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1498         else
1499             h->poc_msb = h->prev_poc_msb;
1500 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1501         field_poc[0] =
1502         field_poc[1] = h->poc_msb + h->poc_lsb;
1503         if(s->picture_structure == PICT_FRAME)
1504             field_poc[1] += h->delta_poc_bottom;
1505     }else if(h->sps.poc_type==1){
1506         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1507         int i;
1508
1509         if(h->sps.poc_cycle_length != 0)
1510             abs_frame_num = h->frame_num_offset + h->frame_num;
1511         else
1512             abs_frame_num = 0;
1513
1514         if(h->nal_ref_idc==0 && abs_frame_num > 0)
1515             abs_frame_num--;
1516
1517         expected_delta_per_poc_cycle = 0;
1518         for(i=0; i < h->sps.poc_cycle_length; i++)
1519             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1520
1521         if(abs_frame_num > 0){
1522             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1523             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1524
1525             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1526             for(i = 0; i <= frame_num_in_poc_cycle; i++)
1527                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1528         } else
1529             expectedpoc = 0;
1530
1531         if(h->nal_ref_idc == 0)
1532             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1533
1534         field_poc[0] = expectedpoc + h->delta_poc[0];
1535         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1536
1537         if(s->picture_structure == PICT_FRAME)
1538             field_poc[1] += h->delta_poc[1];
1539     }else{
1540         int poc= 2*(h->frame_num_offset + h->frame_num);
1541
1542         if(!h->nal_ref_idc)
1543             poc--;
1544
1545         field_poc[0]= poc;
1546         field_poc[1]= poc;
1547     }
1548
1549     if(s->picture_structure != PICT_BOTTOM_FIELD)
1550         s->current_picture_ptr->field_poc[0]= field_poc[0];
1551     if(s->picture_structure != PICT_TOP_FIELD)
1552         s->current_picture_ptr->field_poc[1]= field_poc[1];
1553     cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1554
1555     return 0;
1556 }
1557
1558
1559 /**
1560  * initialize scan tables
1561  */
1562 static void init_scan_tables(H264Context *h){
1563     int i;
1564     for(i=0; i<16; i++){
1565 #define T(x) (x>>2) | ((x<<2) & 0xF)
1566         h->zigzag_scan[i] = T(zigzag_scan[i]);
1567         h-> field_scan[i] = T( field_scan[i]);
1568 #undef T
1569     }
1570     for(i=0; i<64; i++){
1571 #define T(x) (x>>3) | ((x&7)<<3)
1572         h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
1573         h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1574         h->field_scan8x8[i]        = T(field_scan8x8[i]);
1575         h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
1576 #undef T
1577     }
1578     if(h->sps.transform_bypass){ //FIXME same ugly
1579         h->zigzag_scan_q0          = zigzag_scan;
1580         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
1581         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1582         h->field_scan_q0           = field_scan;
1583         h->field_scan8x8_q0        = field_scan8x8;
1584         h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
1585     }else{
1586         h->zigzag_scan_q0          = h->zigzag_scan;
1587         h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
1588         h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1589         h->field_scan_q0           = h->field_scan;
1590         h->field_scan8x8_q0        = h->field_scan8x8;
1591         h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
1592     }
1593 }
1594
1595 static void field_end(H264Context *h){
1596     MpegEncContext * const s = &h->s;
1597     AVCodecContext * const avctx= s->avctx;
1598     s->mb_y= 0;
1599
1600     s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1601     s->current_picture_ptr->pict_type= s->pict_type;
1602
1603     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1604         ff_vdpau_h264_set_reference_frames(s);
1605
1606     if(!s->dropable) {
1607         ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1608         h->prev_poc_msb= h->poc_msb;
1609         h->prev_poc_lsb= h->poc_lsb;
1610     }
1611     h->prev_frame_num_offset= h->frame_num_offset;
1612     h->prev_frame_num= h->frame_num;
1613
1614     if (avctx->hwaccel) {
1615         if (avctx->hwaccel->end_frame(avctx) < 0)
1616             av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1617     }
1618
1619     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1620         ff_vdpau_h264_picture_complete(s);
1621
1622     /*
1623      * FIXME: Error handling code does not seem to support interlaced
1624      * when slices span multiple rows
1625      * The ff_er_add_slice calls don't work right for bottom
1626      * fields; they cause massive erroneous error concealing
1627      * Error marking covers both fields (top and bottom).
1628      * This causes a mismatched s->error_count
1629      * and a bad error table. Further, the error count goes to
1630      * INT_MAX when called for bottom field, because mb_y is
1631      * past end by one (callers fault) and resync_mb_y != 0
1632      * causes problems for the first MB line, too.
1633      */
1634     if (!FIELD_PICTURE)
1635         ff_er_frame_end(s);
1636
1637     MPV_frame_end(s);
1638
1639     h->current_slice=0;
1640 }
1641
1642 /**
1643  * Replicate H264 "master" context to thread contexts.
1644  */
1645 static void clone_slice(H264Context *dst, H264Context *src)
1646 {
1647     memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
1648     dst->s.current_picture_ptr  = src->s.current_picture_ptr;
1649     dst->s.current_picture      = src->s.current_picture;
1650     dst->s.linesize             = src->s.linesize;
1651     dst->s.uvlinesize           = src->s.uvlinesize;
1652     dst->s.first_field          = src->s.first_field;
1653
1654     dst->prev_poc_msb           = src->prev_poc_msb;
1655     dst->prev_poc_lsb           = src->prev_poc_lsb;
1656     dst->prev_frame_num_offset  = src->prev_frame_num_offset;
1657     dst->prev_frame_num         = src->prev_frame_num;
1658     dst->short_ref_count        = src->short_ref_count;
1659
1660     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
1661     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
1662     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1663     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
1664
1665     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
1666     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
1667 }
1668
1669 /**
1670  * decodes a slice header.
1671  * This will also call MPV_common_init() and frame_start() as needed.
1672  *
1673  * @param h h264context
1674  * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1675  *
1676  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1677  */
1678 static int decode_slice_header(H264Context *h, H264Context *h0){
1679     MpegEncContext * const s = &h->s;
1680     MpegEncContext * const s0 = &h0->s;
1681     unsigned int first_mb_in_slice;
1682     unsigned int pps_id;
1683     int num_ref_idx_active_override_flag;
1684     unsigned int slice_type, tmp, i, j;
1685     int default_ref_list_done = 0;
1686     int last_pic_structure;
1687
1688     s->dropable= h->nal_ref_idc == 0;
1689
1690     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1691         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1692         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1693     }else{
1694         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1695         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1696     }
1697
1698     first_mb_in_slice= get_ue_golomb(&s->gb);
1699
1700     if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1701         if(h0->current_slice && FIELD_PICTURE){
1702             field_end(h);
1703         }
1704
1705         h0->current_slice = 0;
1706         if (!s0->first_field)
1707             s->current_picture_ptr= NULL;
1708     }
1709
1710     slice_type= get_ue_golomb_31(&s->gb);
1711     if(slice_type > 9){
1712         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);
1713         return -1;
1714     }
1715     if(slice_type > 4){
1716         slice_type -= 5;
1717         h->slice_type_fixed=1;
1718     }else
1719         h->slice_type_fixed=0;
1720
1721     slice_type= golomb_to_pict_type[ slice_type ];
1722     if (slice_type == FF_I_TYPE
1723         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
1724         default_ref_list_done = 1;
1725     }
1726     h->slice_type= slice_type;
1727     h->slice_type_nos= slice_type & 3;
1728
1729     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
1730
1731     pps_id= get_ue_golomb(&s->gb);
1732     if(pps_id>=MAX_PPS_COUNT){
1733         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
1734         return -1;
1735     }
1736     if(!h0->pps_buffers[pps_id]) {
1737         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
1738         return -1;
1739     }
1740     h->pps= *h0->pps_buffers[pps_id];
1741
1742     if(!h0->sps_buffers[h->pps.sps_id]) {
1743         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
1744         return -1;
1745     }
1746     h->sps = *h0->sps_buffers[h->pps.sps_id];
1747
1748     s->avctx->profile = h->sps.profile_idc;
1749     s->avctx->level   = h->sps.level_idc;
1750     s->avctx->refs    = h->sps.ref_frame_count;
1751
1752     if(h == h0 && h->dequant_coeff_pps != pps_id){
1753         h->dequant_coeff_pps = pps_id;
1754         init_dequant_tables(h);
1755     }
1756
1757     s->mb_width= h->sps.mb_width;
1758     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1759
1760     h->b_stride=  s->mb_width*4;
1761
1762     s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
1763     if(h->sps.frame_mbs_only_flag)
1764         s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
1765     else
1766         s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 7);
1767
1768     if (s->context_initialized
1769         && (   s->width != s->avctx->width || s->height != s->avctx->height
1770             || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
1771         if(h != h0)
1772             return -1;   // width / height changed during parallelized decoding
1773         free_tables(h);
1774         flush_dpb(s->avctx);
1775         MPV_common_end(s);
1776     }
1777     if (!s->context_initialized) {
1778         if(h != h0)
1779             return -1;  // we cant (re-)initialize context during parallel decoding
1780
1781         avcodec_set_dimensions(s->avctx, s->width, s->height);
1782         s->avctx->sample_aspect_ratio= h->sps.sar;
1783         av_assert0(s->avctx->sample_aspect_ratio.den);
1784
1785         if(h->sps.video_signal_type_present_flag){
1786             s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1787             if(h->sps.colour_description_present_flag){
1788                 s->avctx->color_primaries = h->sps.color_primaries;
1789                 s->avctx->color_trc       = h->sps.color_trc;
1790                 s->avctx->colorspace      = h->sps.colorspace;
1791             }
1792         }
1793
1794         if(h->sps.timing_info_present_flag){
1795             int64_t den= h->sps.time_scale;
1796             if(h->x264_build < 44U)
1797                 den *= 2;
1798             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
1799                       h->sps.num_units_in_tick, den, 1<<30);
1800         }
1801         s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
1802                                                  s->avctx->codec->pix_fmts ?
1803                                                  s->avctx->codec->pix_fmts :
1804                                                  s->avctx->color_range == AVCOL_RANGE_JPEG ?
1805                                                  hwaccel_pixfmt_list_h264_jpeg_420 :
1806                                                  ff_hwaccel_pixfmt_list_420);
1807         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1808
1809         if (MPV_common_init(s) < 0)
1810             return -1;
1811         s->first_field = 0;
1812         h->prev_interlaced_frame = 1;
1813
1814         init_scan_tables(h);
1815         ff_h264_alloc_tables(h);
1816
1817         for(i = 1; i < s->avctx->thread_count; i++) {
1818             H264Context *c;
1819             c = h->thread_context[i] = av_malloc(sizeof(H264Context));
1820             memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
1821             memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
1822             c->h264dsp = h->h264dsp;
1823             c->sps = h->sps;
1824             c->pps = h->pps;
1825             init_scan_tables(c);
1826             clone_tables(c, h, i);
1827         }
1828
1829         for(i = 0; i < s->avctx->thread_count; i++)
1830             if(context_init(h->thread_context[i]) < 0)
1831                 return -1;
1832     }
1833
1834     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1835
1836     h->mb_mbaff = 0;
1837     h->mb_aff_frame = 0;
1838     last_pic_structure = s0->picture_structure;
1839     if(h->sps.frame_mbs_only_flag){
1840         s->picture_structure= PICT_FRAME;
1841     }else{
1842         if(get_bits1(&s->gb)) { //field_pic_flag
1843             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
1844         } else {
1845             s->picture_structure= PICT_FRAME;
1846             h->mb_aff_frame = h->sps.mb_aff;
1847         }
1848     }
1849     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
1850
1851     if(h0->current_slice == 0){
1852         while(h->frame_num !=  h->prev_frame_num &&
1853               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
1854             Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1855             av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
1856             if (ff_h264_frame_start(h) < 0)
1857                 return -1;
1858             h->prev_frame_num++;
1859             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
1860             s->current_picture_ptr->frame_num= h->prev_frame_num;
1861             ff_generate_sliding_window_mmcos(h);
1862             ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1863             /* Error concealment: if a ref is missing, copy the previous ref in its place.
1864              * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
1865              * about there being no actual duplicates.
1866              * FIXME: this doesn't copy padding for out-of-frame motion vectors.  Given we're
1867              * concealing a lost frame, this probably isn't noticable by comparison, but it should
1868              * be fixed. */
1869             if (h->short_ref_count) {
1870                 if (prev) {
1871                     av_image_copy(h->short_ref[0]->data, h->short_ref[0]->linesize,
1872                                   (const uint8_t**)prev->data, prev->linesize,
1873                                   s->avctx->pix_fmt, s->mb_width*16, s->mb_height*16);
1874                     h->short_ref[0]->poc = prev->poc+2;
1875                 }
1876                 h->short_ref[0]->frame_num = h->prev_frame_num;
1877             }
1878         }
1879
1880         /* See if we have a decoded first field looking for a pair... */
1881         if (s0->first_field) {
1882             assert(s0->current_picture_ptr);
1883             assert(s0->current_picture_ptr->data[0]);
1884             assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
1885
1886             /* figure out if we have a complementary field pair */
1887             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
1888                 /*
1889                  * Previous field is unmatched. Don't display it, but let it
1890                  * remain for reference if marked as such.
1891                  */
1892                 s0->current_picture_ptr = NULL;
1893                 s0->first_field = FIELD_PICTURE;
1894
1895             } else {
1896                 if (h->nal_ref_idc &&
1897                         s0->current_picture_ptr->reference &&
1898                         s0->current_picture_ptr->frame_num != h->frame_num) {
1899                     /*
1900                      * This and previous field were reference, but had
1901                      * different frame_nums. Consider this field first in
1902                      * pair. Throw away previous field except for reference
1903                      * purposes.
1904                      */
1905                     s0->first_field = 1;
1906                     s0->current_picture_ptr = NULL;
1907
1908                 } else {
1909                     /* Second field in complementary pair */
1910                     s0->first_field = 0;
1911                 }
1912             }
1913
1914         } else {
1915             /* Frame or first field in a potentially complementary pair */
1916             assert(!s0->current_picture_ptr);
1917             s0->first_field = FIELD_PICTURE;
1918         }
1919
1920         if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
1921             s0->first_field = 0;
1922             return -1;
1923         }
1924     }
1925     if(h != h0)
1926         clone_slice(h, h0);
1927
1928     s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
1929
1930     assert(s->mb_num == s->mb_width * s->mb_height);
1931     if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
1932        first_mb_in_slice                    >= s->mb_num){
1933         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1934         return -1;
1935     }
1936     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
1937     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
1938     if (s->picture_structure == PICT_BOTTOM_FIELD)
1939         s->resync_mb_y = s->mb_y = s->mb_y + 1;
1940     assert(s->mb_y < s->mb_height);
1941
1942     if(s->picture_structure==PICT_FRAME){
1943         h->curr_pic_num=   h->frame_num;
1944         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
1945     }else{
1946         h->curr_pic_num= 2*h->frame_num + 1;
1947         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
1948     }
1949
1950     if(h->nal_unit_type == NAL_IDR_SLICE){
1951         get_ue_golomb(&s->gb); /* idr_pic_id */
1952     }
1953
1954     if(h->sps.poc_type==0){
1955         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
1956
1957         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
1958             h->delta_poc_bottom= get_se_golomb(&s->gb);
1959         }
1960     }
1961
1962     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
1963         h->delta_poc[0]= get_se_golomb(&s->gb);
1964
1965         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
1966             h->delta_poc[1]= get_se_golomb(&s->gb);
1967     }
1968
1969     init_poc(h);
1970
1971     if(h->pps.redundant_pic_cnt_present){
1972         h->redundant_pic_count= get_ue_golomb(&s->gb);
1973     }
1974
1975     //set defaults, might be overridden a few lines later
1976     h->ref_count[0]= h->pps.ref_count[0];
1977     h->ref_count[1]= h->pps.ref_count[1];
1978
1979     if(h->slice_type_nos != FF_I_TYPE){
1980         if(h->slice_type_nos == FF_B_TYPE){
1981             h->direct_spatial_mv_pred= get_bits1(&s->gb);
1982         }
1983         num_ref_idx_active_override_flag= get_bits1(&s->gb);
1984
1985         if(num_ref_idx_active_override_flag){
1986             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
1987             if(h->slice_type_nos==FF_B_TYPE)
1988                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
1989
1990             if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
1991                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
1992                 h->ref_count[0]= h->ref_count[1]= 1;
1993                 return -1;
1994             }
1995         }
1996         if(h->slice_type_nos == FF_B_TYPE)
1997             h->list_count= 2;
1998         else
1999             h->list_count= 1;
2000     }else
2001         h->list_count= 0;
2002
2003     if(!default_ref_list_done){
2004         ff_h264_fill_default_ref_list(h);
2005     }
2006
2007     if(h->slice_type_nos!=FF_I_TYPE && ff_h264_decode_ref_pic_list_reordering(h) < 0)
2008         return -1;
2009
2010     if(h->slice_type_nos!=FF_I_TYPE){
2011         s->last_picture_ptr= &h->ref_list[0][0];
2012         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
2013     }
2014     if(h->slice_type_nos==FF_B_TYPE){
2015         s->next_picture_ptr= &h->ref_list[1][0];
2016         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
2017     }
2018
2019     if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
2020        ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
2021         pred_weight_table(h);
2022     else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE){
2023         implicit_weight_table(h, -1);
2024     }else {
2025         h->use_weight = 0;
2026         for (i = 0; i < 2; i++) {
2027             h->luma_weight_flag[i]   = 0;
2028             h->chroma_weight_flag[i] = 0;
2029         }
2030     }
2031
2032     if(h->nal_ref_idc)
2033         ff_h264_decode_ref_pic_marking(h0, &s->gb);
2034
2035     if(FRAME_MBAFF){
2036         ff_h264_fill_mbaff_ref_list(h);
2037
2038         if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE){
2039             implicit_weight_table(h, 0);
2040             implicit_weight_table(h, 1);
2041         }
2042     }
2043
2044     if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
2045         ff_h264_direct_dist_scale_factor(h);
2046     ff_h264_direct_ref_list_init(h);
2047
2048     if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
2049         tmp = get_ue_golomb_31(&s->gb);
2050         if(tmp > 2){
2051             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2052             return -1;
2053         }
2054         h->cabac_init_idc= tmp;
2055     }
2056
2057     h->last_qscale_diff = 0;
2058     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2059     if(tmp>51){
2060         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2061         return -1;
2062     }
2063     s->qscale= tmp;
2064     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2065     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2066     //FIXME qscale / qp ... stuff
2067     if(h->slice_type == FF_SP_TYPE){
2068         get_bits1(&s->gb); /* sp_for_switch_flag */
2069     }
2070     if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
2071         get_se_golomb(&s->gb); /* slice_qs_delta */
2072     }
2073
2074     h->deblocking_filter = 1;
2075     h->slice_alpha_c0_offset = 52;
2076     h->slice_beta_offset = 52;
2077     if( h->pps.deblocking_filter_parameters_present ) {
2078         tmp= get_ue_golomb_31(&s->gb);
2079         if(tmp > 2){
2080             av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2081             return -1;
2082         }
2083         h->deblocking_filter= tmp;
2084         if(h->deblocking_filter < 2)
2085             h->deblocking_filter^= 1; // 1<->0
2086
2087         if( h->deblocking_filter ) {
2088             h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2089             h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
2090             if(   h->slice_alpha_c0_offset > 104U
2091                || h->slice_beta_offset     > 104U){
2092                 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);
2093                 return -1;
2094             }
2095         }
2096     }
2097
2098     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
2099        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
2100        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
2101        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2102         h->deblocking_filter= 0;
2103
2104     if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2105         if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2106             /* Cheat slightly for speed:
2107                Do not bother to deblock across slices. */
2108             h->deblocking_filter = 2;
2109         } else {
2110             h0->max_contexts = 1;
2111             if(!h0->single_decode_warning) {
2112                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2113                 h0->single_decode_warning = 1;
2114             }
2115             if(h != h0)
2116                 return 1; // deblocking switched inside frame
2117         }
2118     }
2119     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]);
2120
2121 #if 0 //FMO
2122     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2123         slice_group_change_cycle= get_bits(&s->gb, ?);
2124 #endif
2125
2126     h0->last_slice_type = slice_type;
2127     h->slice_num = ++h0->current_slice;
2128     if(h->slice_num >= MAX_SLICES){
2129         av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2130     }
2131
2132     for(j=0; j<2; j++){
2133         int id_list[16];
2134         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2135         for(i=0; i<16; i++){
2136             id_list[i]= 60;
2137             if(h->ref_list[j][i].data[0]){
2138                 int k;
2139                 uint8_t *base= h->ref_list[j][i].base[0];
2140                 for(k=0; k<h->short_ref_count; k++)
2141                     if(h->short_ref[k]->base[0] == base){
2142                         id_list[i]= k;
2143                         break;
2144                     }
2145                 for(k=0; k<h->long_ref_count; k++)
2146                     if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2147                         id_list[i]= h->short_ref_count + k;
2148                         break;
2149                     }
2150             }
2151         }
2152
2153         ref2frm[0]=
2154         ref2frm[1]= -1;
2155         for(i=0; i<16; i++)
2156             ref2frm[i+2]= 4*id_list[i]
2157                           +(h->ref_list[j][i].reference&3);
2158         ref2frm[18+0]=
2159         ref2frm[18+1]= -1;
2160         for(i=16; i<48; i++)
2161             ref2frm[i+4]= 4*id_list[(i-16)>>1]
2162                           +(h->ref_list[j][i].reference&3);
2163     }
2164
2165     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
2166     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2167
2168     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2169         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",
2170                h->slice_num,
2171                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2172                first_mb_in_slice,
2173                av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2174                pps_id, h->frame_num,
2175                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2176                h->ref_count[0], h->ref_count[1],
2177                s->qscale,
2178                h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
2179                h->use_weight,
2180                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2181                h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2182                );
2183     }
2184
2185     return 0;
2186 }
2187
2188 int ff_h264_get_slice_type(const H264Context *h)
2189 {
2190     switch (h->slice_type) {
2191     case FF_P_TYPE:  return 0;
2192     case FF_B_TYPE:  return 1;
2193     case FF_I_TYPE:  return 2;
2194     case FF_SP_TYPE: return 3;
2195     case FF_SI_TYPE: return 4;
2196     default:         return -1;
2197     }
2198 }
2199
2200 /**
2201  *
2202  * @return non zero if the loop filter can be skiped
2203  */
2204 static int fill_filter_caches(H264Context *h, int mb_type){
2205     MpegEncContext * const s = &h->s;
2206     const int mb_xy= h->mb_xy;
2207     int top_xy, left_xy[2];
2208     int top_type, left_type[2];
2209
2210     top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
2211
2212     //FIXME deblocking could skip the intra and nnz parts.
2213
2214     /* Wow, what a mess, why didn't they simplify the interlacing & intra
2215      * stuff, I can't imagine that these complex rules are worth it. */
2216
2217     left_xy[1] = left_xy[0] = mb_xy-1;
2218     if(FRAME_MBAFF){
2219         const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
2220         const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
2221         if(s->mb_y&1){
2222             if (left_mb_field_flag != curr_mb_field_flag) {
2223                 left_xy[0] -= s->mb_stride;
2224             }
2225         }else{
2226             if(curr_mb_field_flag){
2227                 top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
2228             }
2229             if (left_mb_field_flag != curr_mb_field_flag) {
2230                 left_xy[1] += s->mb_stride;
2231             }
2232         }
2233     }
2234
2235     h->top_mb_xy = top_xy;
2236     h->left_mb_xy[0] = left_xy[0];
2237     h->left_mb_xy[1] = left_xy[1];
2238     {
2239         //for sufficiently low qp, filtering wouldn't do anything
2240         //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
2241         int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
2242         int qp = s->current_picture.qscale_table[mb_xy];
2243         if(qp <= qp_thresh
2244            && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
2245            && (top_xy   < 0 || ((qp + s->current_picture.qscale_table[top_xy    ] + 1)>>1) <= qp_thresh)){
2246             if(!FRAME_MBAFF)
2247                 return 1;
2248             if(   (left_xy[0]< 0            || ((qp + s->current_picture.qscale_table[left_xy[1]             ] + 1)>>1) <= qp_thresh)
2249                && (top_xy    < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy    -s->mb_stride] + 1)>>1) <= qp_thresh))
2250                 return 1;
2251         }
2252     }
2253
2254     top_type     = s->current_picture.mb_type[top_xy]    ;
2255     left_type[0] = s->current_picture.mb_type[left_xy[0]];
2256     left_type[1] = s->current_picture.mb_type[left_xy[1]];
2257     if(h->deblocking_filter == 2){
2258         if(h->slice_table[top_xy     ] != h->slice_num) top_type= 0;
2259         if(h->slice_table[left_xy[0] ] != h->slice_num) left_type[0]= left_type[1]= 0;
2260     }else{
2261         if(h->slice_table[top_xy     ] == 0xFFFF) top_type= 0;
2262         if(h->slice_table[left_xy[0] ] == 0xFFFF) left_type[0]= left_type[1] =0;
2263     }
2264     h->top_type    = top_type    ;
2265     h->left_type[0]= left_type[0];
2266     h->left_type[1]= left_type[1];
2267
2268     if(IS_INTRA(mb_type))
2269         return 0;
2270
2271     AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
2272     AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
2273     AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]);
2274     AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]);
2275     AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
2276
2277     h->cbp= h->cbp_table[mb_xy];
2278
2279     {
2280         int list;
2281         for(list=0; list<h->list_count; list++){
2282             int8_t *ref;
2283             int y, b_stride;
2284             int16_t (*mv_dst)[2];
2285             int16_t (*mv_src)[2];
2286
2287             if(!USES_LIST(mb_type, list)){
2288                 fill_rectangle(  h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
2289                 AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2290                 AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2291                 AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2292                 AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2293                 continue;
2294             }
2295
2296             ref = &s->current_picture.ref_index[list][4*mb_xy];
2297             {
2298                 int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2299                 AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2300                 AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2301                 ref += 2;
2302                 AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2303                 AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2304             }
2305
2306             b_stride = h->b_stride;
2307             mv_dst   = &h->mv_cache[list][scan8[0]];
2308             mv_src   = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
2309             for(y=0; y<4; y++){
2310                 AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
2311             }
2312
2313         }
2314     }
2315
2316
2317 /*
2318 0 . T T. T T T T
2319 1 L . .L . . . .
2320 2 L . .L . . . .
2321 3 . T TL . . . .
2322 4 L . .L . . . .
2323 5 L . .. . . . .
2324 */
2325 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
2326     if(top_type){
2327         AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
2328     }
2329
2330     if(left_type[0]){
2331         h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
2332         h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
2333         h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
2334         h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
2335     }
2336
2337     // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
2338     if(!CABAC && h->pps.transform_8x8_mode){
2339         if(IS_8x8DCT(top_type)){
2340             h->non_zero_count_cache[4+8*0]=
2341             h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
2342             h->non_zero_count_cache[6+8*0]=
2343             h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
2344         }
2345         if(IS_8x8DCT(left_type[0])){
2346             h->non_zero_count_cache[3+8*1]=
2347             h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
2348         }
2349         if(IS_8x8DCT(left_type[1])){
2350             h->non_zero_count_cache[3+8*3]=
2351             h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
2352         }
2353
2354         if(IS_8x8DCT(mb_type)){
2355             h->non_zero_count_cache[scan8[0   ]]= h->non_zero_count_cache[scan8[1   ]]=
2356             h->non_zero_count_cache[scan8[2   ]]= h->non_zero_count_cache[scan8[3   ]]= h->cbp & 1;
2357
2358             h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
2359             h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
2360
2361             h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
2362             h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
2363
2364             h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
2365             h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
2366         }
2367     }
2368
2369     if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
2370         int list;
2371         for(list=0; list<h->list_count; list++){
2372             if(USES_LIST(top_type, list)){
2373                 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
2374                 const int b8_xy= 4*top_xy + 2;
2375                 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2376                 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
2377                 h->ref_cache[list][scan8[0] + 0 - 1*8]=
2378                 h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
2379                 h->ref_cache[list][scan8[0] + 2 - 1*8]=
2380                 h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
2381             }else{
2382                 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
2383                 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2384             }
2385
2386             if(!IS_INTERLACED(mb_type^left_type[0])){
2387                 if(USES_LIST(left_type[0], list)){
2388                     const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
2389                     const int b8_xy= 4*left_xy[0] + 1;
2390                     int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2391                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
2392                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
2393                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
2394                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
2395                     h->ref_cache[list][scan8[0] - 1 + 0 ]=
2396                     h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]];
2397                     h->ref_cache[list][scan8[0] - 1 +16 ]=
2398                     h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]];
2399                 }else{
2400                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
2401                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
2402                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
2403                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
2404                     h->ref_cache[list][scan8[0] - 1 + 0  ]=
2405                     h->ref_cache[list][scan8[0] - 1 + 8  ]=
2406                     h->ref_cache[list][scan8[0] - 1 + 16 ]=
2407                     h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
2408                 }
2409             }
2410         }
2411     }
2412
2413     return 0;
2414 }
2415
2416 static void loop_filter(H264Context *h){
2417     MpegEncContext * const s = &h->s;
2418     uint8_t  *dest_y, *dest_cb, *dest_cr;
2419     int linesize, uvlinesize, mb_x, mb_y;
2420     const int end_mb_y= s->mb_y + FRAME_MBAFF;
2421     const int old_slice_type= h->slice_type;
2422
2423     if(h->deblocking_filter) {
2424         for(mb_x= 0; mb_x<s->mb_width; mb_x++){
2425             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
2426                 int mb_xy, mb_type;
2427                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2428                 h->slice_num= h->slice_table[mb_xy];
2429                 mb_type= s->current_picture.mb_type[mb_xy];
2430                 h->list_count= h->list_counts[mb_xy];
2431
2432                 if(FRAME_MBAFF)
2433                     h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2434
2435                 s->mb_x= mb_x;
2436                 s->mb_y= mb_y;
2437                 dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2438                 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2439                 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2440                     //FIXME simplify above
2441
2442                 if (MB_FIELD) {
2443                     linesize   = h->mb_linesize   = s->linesize * 2;
2444                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2445                     if(mb_y&1){ //FIXME move out of this function?
2446                         dest_y -= s->linesize*15;
2447                         dest_cb-= s->uvlinesize*7;
2448                         dest_cr-= s->uvlinesize*7;
2449                     }
2450                 } else {
2451                     linesize   = h->mb_linesize   = s->linesize;
2452                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2453                 }
2454                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2455                 if(fill_filter_caches(h, mb_type))
2456                     continue;
2457                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2458                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2459
2460                 if (FRAME_MBAFF) {
2461                     ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2462                 } else {
2463                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2464                 }
2465             }
2466         }
2467     }
2468     h->slice_type= old_slice_type;
2469     s->mb_x= 0;
2470     s->mb_y= end_mb_y - FRAME_MBAFF;
2471     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2472     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2473 }
2474
2475 static void predict_field_decoding_flag(H264Context *h){
2476     MpegEncContext * const s = &h->s;
2477     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2478     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
2479                 ? s->current_picture.mb_type[mb_xy-1]
2480                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
2481                 ? s->current_picture.mb_type[mb_xy-s->mb_stride]
2482                 : 0;
2483     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2484 }
2485
2486 static int decode_slice(struct AVCodecContext *avctx, void *arg){
2487     H264Context *h = *(void**)arg;
2488     MpegEncContext * const s = &h->s;
2489     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2490
2491     s->mb_skip_run= -1;
2492
2493     h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
2494                     (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
2495
2496     if( h->pps.cabac ) {
2497         /* realign */
2498         align_get_bits( &s->gb );
2499
2500         /* init cabac */
2501         ff_init_cabac_states( &h->cabac);
2502         ff_init_cabac_decoder( &h->cabac,
2503                                s->gb.buffer + get_bits_count(&s->gb)/8,
2504                                (get_bits_left(&s->gb) + 7)/8);
2505
2506         ff_h264_init_cabac_states(h);
2507
2508         for(;;){
2509 //START_TIMER
2510             int ret = ff_h264_decode_mb_cabac(h);
2511             int eos;
2512 //STOP_TIMER("decode_mb_cabac")
2513
2514             if(ret>=0) ff_h264_hl_decode_mb(h);
2515
2516             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
2517                 s->mb_y++;
2518
2519                 ret = ff_h264_decode_mb_cabac(h);
2520
2521                 if(ret>=0) ff_h264_hl_decode_mb(h);
2522                 s->mb_y--;
2523             }
2524             eos = get_cabac_terminate( &h->cabac );
2525
2526             if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
2527                 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);
2528                 return 0;
2529             }
2530             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2531                 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);
2532                 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);
2533                 return -1;
2534             }
2535
2536             if( ++s->mb_x >= s->mb_width ) {
2537                 s->mb_x = 0;
2538                 loop_filter(h);
2539                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2540                 ++s->mb_y;
2541                 if(FIELD_OR_MBAFF_PICTURE) {
2542                     ++s->mb_y;
2543                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2544                         predict_field_decoding_flag(h);
2545                 }
2546             }
2547
2548             if( eos || s->mb_y >= s->mb_height ) {
2549                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2550                 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);
2551                 return 0;
2552             }
2553         }
2554
2555     } else {
2556         for(;;){
2557             int ret = ff_h264_decode_mb_cavlc(h);
2558
2559             if(ret>=0) ff_h264_hl_decode_mb(h);
2560
2561             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
2562                 s->mb_y++;
2563                 ret = ff_h264_decode_mb_cavlc(h);
2564
2565                 if(ret>=0) ff_h264_hl_decode_mb(h);
2566                 s->mb_y--;
2567             }
2568
2569             if(ret<0){
2570                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2571                 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);
2572
2573                 return -1;
2574             }
2575
2576             if(++s->mb_x >= s->mb_width){
2577                 s->mb_x=0;
2578                 loop_filter(h);
2579                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2580                 ++s->mb_y;
2581                 if(FIELD_OR_MBAFF_PICTURE) {
2582                     ++s->mb_y;
2583                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2584                         predict_field_decoding_flag(h);
2585                 }
2586                 if(s->mb_y >= s->mb_height){
2587                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2588
2589                     if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
2590                         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);
2591
2592                         return 0;
2593                     }else{
2594                         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);
2595
2596                         return -1;
2597                     }
2598                 }
2599             }
2600
2601             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
2602                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2603                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2604                     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);
2605
2606                     return 0;
2607                 }else{
2608                     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);
2609
2610                     return -1;
2611                 }
2612             }
2613         }
2614     }
2615
2616 #if 0
2617     for(;s->mb_y < s->mb_height; s->mb_y++){
2618         for(;s->mb_x < s->mb_width; s->mb_x++){
2619             int ret= decode_mb(h);
2620
2621             ff_h264_hl_decode_mb(h);
2622
2623             if(ret<0){
2624                 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2625                 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);
2626
2627                 return -1;
2628             }
2629
2630             if(++s->mb_x >= s->mb_width){
2631                 s->mb_x=0;
2632                 if(++s->mb_y >= s->mb_height){
2633                     if(get_bits_count(s->gb) == s->gb.size_in_bits){
2634                         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);
2635
2636                         return 0;
2637                     }else{
2638                         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);
2639
2640                         return -1;
2641                     }
2642                 }
2643             }
2644
2645             if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2646                 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2647                     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);
2648
2649                     return 0;
2650                 }else{
2651                     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);
2652
2653                     return -1;
2654                 }
2655             }
2656         }
2657         s->mb_x=0;
2658         ff_draw_horiz_band(s, 16*s->mb_y, 16);
2659     }
2660 #endif
2661     return -1; //not reached
2662 }
2663
2664 /**
2665  * Call decode_slice() for each context.
2666  *
2667  * @param h h264 master context
2668  * @param context_count number of contexts to execute
2669  */
2670 static void execute_decode_slices(H264Context *h, int context_count){
2671     MpegEncContext * const s = &h->s;
2672     AVCodecContext * const avctx= s->avctx;
2673     H264Context *hx;
2674     int i;
2675
2676     if (s->avctx->hwaccel)
2677         return;
2678     if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2679         return;
2680     if(context_count == 1) {
2681         decode_slice(avctx, &h);
2682     } else {
2683         for(i = 1; i < context_count; i++) {
2684             hx = h->thread_context[i];
2685             hx->s.error_recognition = avctx->error_recognition;
2686             hx->s.error_count = 0;
2687         }
2688
2689         avctx->execute(avctx, (void *)decode_slice,
2690                        h->thread_context, NULL, context_count, sizeof(void*));
2691
2692         /* pull back stuff from slices to master context */
2693         hx = h->thread_context[context_count - 1];
2694         s->mb_x = hx->s.mb_x;
2695         s->mb_y = hx->s.mb_y;
2696         s->dropable = hx->s.dropable;
2697         s->picture_structure = hx->s.picture_structure;
2698         for(i = 1; i < context_count; i++)
2699             h->s.error_count += h->thread_context[i]->s.error_count;
2700     }
2701 }
2702
2703
2704 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
2705     MpegEncContext * const s = &h->s;
2706     AVCodecContext * const avctx= s->avctx;
2707     int buf_index=0;
2708     H264Context *hx; ///< thread context
2709     int context_count = 0;
2710     int next_avc= h->is_avc ? 0 : buf_size;
2711
2712     h->max_contexts = avctx->thread_count;
2713 #if 0
2714     int i;
2715     for(i=0; i<50; i++){
2716         av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
2717     }
2718 #endif
2719     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
2720         h->current_slice = 0;
2721         if (!s->first_field)
2722             s->current_picture_ptr= NULL;
2723         ff_h264_reset_sei(h);
2724     }
2725
2726     for(;;){
2727         int consumed;
2728         int dst_length;
2729         int bit_length;
2730         const uint8_t *ptr;
2731         int i, nalsize = 0;
2732         int err;
2733
2734         if(buf_index >= next_avc) {
2735             if(buf_index >= buf_size) break;
2736             nalsize = 0;
2737             for(i = 0; i < h->nal_length_size; i++)
2738                 nalsize = (nalsize << 8) | buf[buf_index++];
2739             if(nalsize <= 0 || nalsize > buf_size - buf_index){
2740                 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2741                 break;
2742             }
2743             next_avc= buf_index + nalsize;
2744         } else {
2745             // start code prefix search
2746             for(; buf_index + 3 < next_avc; buf_index++){
2747                 // This should always succeed in the first iteration.
2748                 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2749                     break;
2750             }
2751
2752             if(buf_index+3 >= buf_size) break;
2753
2754             buf_index+=3;
2755             if(buf_index >= next_avc) continue;
2756         }
2757
2758         hx = h->thread_context[context_count];
2759
2760         ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
2761         if (ptr==NULL || dst_length < 0){
2762             return -1;
2763         }
2764         i= buf_index + consumed;
2765         if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
2766            buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
2767             s->workaround_bugs |= FF_BUG_TRUNCATED;
2768
2769         if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
2770         while(ptr[dst_length - 1] == 0 && dst_length > 0)
2771             dst_length--;
2772         }
2773         bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
2774
2775         if(s->avctx->debug&FF_DEBUG_STARTCODE){
2776             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);
2777         }
2778
2779         if (h->is_avc && (nalsize != consumed) && nalsize){
2780             av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
2781         }
2782
2783         buf_index += consumed;
2784
2785         if(  (s->hurry_up == 1 && h->nal_ref_idc  == 0) //FIXME do not discard SEI id
2786            ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0))
2787             continue;
2788
2789       again:
2790         err = 0;
2791         switch(hx->nal_unit_type){
2792         case NAL_IDR_SLICE:
2793             if (h->nal_unit_type != NAL_IDR_SLICE) {
2794                 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2795                 return -1;
2796             }
2797             idr(h); //FIXME ensure we don't loose some frames if there is reordering
2798         case NAL_SLICE:
2799             init_get_bits(&hx->s.gb, ptr, bit_length);
2800             hx->intra_gb_ptr=
2801             hx->inter_gb_ptr= &hx->s.gb;
2802             hx->s.data_partitioning = 0;
2803
2804             if((err = decode_slice_header(hx, h)))
2805                break;
2806
2807             if (h->current_slice == 1) {
2808                 if (s->avctx->hwaccel && s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
2809                     return -1;
2810                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2811                     ff_vdpau_h264_picture_start(s);
2812             }
2813
2814             s->current_picture_ptr->key_frame |=
2815                     (hx->nal_unit_type == NAL_IDR_SLICE) ||
2816                     (h->sei_recovery_frame_cnt >= 0);
2817             if(hx->redundant_pic_count==0 && hx->s.hurry_up < 5
2818                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2819                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2820                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2821                && avctx->skip_frame < AVDISCARD_ALL){
2822                 if(avctx->hwaccel) {
2823                     if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
2824                         return -1;
2825                 }else
2826                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
2827                     static const uint8_t start_code[] = {0x00, 0x00, 0x01};
2828                     ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
2829                     ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
2830                 }else
2831                     context_count++;
2832             }
2833             break;
2834         case NAL_DPA:
2835             init_get_bits(&hx->s.gb, ptr, bit_length);
2836             hx->intra_gb_ptr=
2837             hx->inter_gb_ptr= NULL;
2838
2839             if ((err = decode_slice_header(hx, h)) < 0)
2840                 break;
2841
2842             hx->s.data_partitioning = 1;
2843
2844             break;
2845         case NAL_DPB:
2846             init_get_bits(&hx->intra_gb, ptr, bit_length);
2847             hx->intra_gb_ptr= &hx->intra_gb;
2848             break;
2849         case NAL_DPC:
2850             init_get_bits(&hx->inter_gb, ptr, bit_length);
2851             hx->inter_gb_ptr= &hx->inter_gb;
2852
2853             if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
2854                && s->context_initialized
2855                && s->hurry_up < 5
2856                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2857                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2858                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2859                && avctx->skip_frame < AVDISCARD_ALL)
2860                 context_count++;
2861             break;
2862         case NAL_SEI:
2863             init_get_bits(&s->gb, ptr, bit_length);
2864             ff_h264_decode_sei(h);
2865             break;
2866         case NAL_SPS:
2867             init_get_bits(&s->gb, ptr, bit_length);
2868             ff_h264_decode_seq_parameter_set(h);
2869
2870             if(s->flags& CODEC_FLAG_LOW_DELAY)
2871                 s->low_delay=1;
2872
2873             if(avctx->has_b_frames < 2)
2874                 avctx->has_b_frames= !s->low_delay;
2875             break;
2876         case NAL_PPS:
2877             init_get_bits(&s->gb, ptr, bit_length);
2878
2879             ff_h264_decode_picture_parameter_set(h, bit_length);
2880
2881             break;
2882         case NAL_AUD:
2883         case NAL_END_SEQUENCE:
2884         case NAL_END_STREAM:
2885         case NAL_FILLER_DATA:
2886         case NAL_SPS_EXT:
2887         case NAL_AUXILIARY_SLICE:
2888             break;
2889         default:
2890             av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
2891         }
2892
2893         if(context_count == h->max_contexts) {
2894             execute_decode_slices(h, context_count);
2895             context_count = 0;
2896         }
2897
2898         if (err < 0)
2899             av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
2900         else if(err == 1) {
2901             /* Slice could not be decoded in parallel mode, copy down
2902              * NAL unit stuff to context 0 and restart. Note that
2903              * rbsp_buffer is not transferred, but since we no longer
2904              * run in parallel mode this should not be an issue. */
2905             h->nal_unit_type = hx->nal_unit_type;
2906             h->nal_ref_idc   = hx->nal_ref_idc;
2907             hx = h;
2908             goto again;
2909         }
2910     }
2911     if(context_count)
2912         execute_decode_slices(h, context_count);
2913     return buf_index;
2914 }
2915
2916 /**
2917  * returns the number of bytes consumed for building the current frame
2918  */
2919 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
2920         if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
2921         if(pos+10>buf_size) pos=buf_size; // oops ;)
2922
2923         return pos;
2924 }
2925
2926 static int decode_frame(AVCodecContext *avctx,
2927                              void *data, int *data_size,
2928                              AVPacket *avpkt)
2929 {
2930     const uint8_t *buf = avpkt->data;
2931     int buf_size = avpkt->size;
2932     H264Context *h = avctx->priv_data;
2933     MpegEncContext *s = &h->s;
2934     AVFrame *pict = data;
2935     int buf_index;
2936
2937     s->flags= avctx->flags;
2938     s->flags2= avctx->flags2;
2939
2940    /* end of stream, output what is still in the buffers */
2941  out:
2942     if (buf_size == 0) {
2943         Picture *out;
2944         int i, out_idx;
2945
2946 //FIXME factorize this with the output code below
2947         out = h->delayed_pic[0];
2948         out_idx = 0;
2949         for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2950             if(h->delayed_pic[i]->poc < out->poc){
2951                 out = h->delayed_pic[i];
2952                 out_idx = i;
2953             }
2954
2955         for(i=out_idx; h->delayed_pic[i]; i++)
2956             h->delayed_pic[i] = h->delayed_pic[i+1];
2957
2958         if(out){
2959             *data_size = sizeof(AVFrame);
2960             *pict= *(AVFrame*)out;
2961         }
2962
2963         return 0;
2964     }
2965
2966     buf_index=decode_nal_units(h, buf, buf_size);
2967     if(buf_index < 0)
2968         return -1;
2969
2970     if (!s->current_picture_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
2971         buf_size = 0;
2972         goto out;
2973     }
2974
2975     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
2976         if (avctx->skip_frame >= AVDISCARD_NONREF || s->hurry_up) return 0;
2977         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
2978         return -1;
2979     }
2980
2981     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
2982         Picture *out = s->current_picture_ptr;
2983         Picture *cur = s->current_picture_ptr;
2984         int i, pics, out_of_order, out_idx;
2985
2986         field_end(h);
2987
2988         if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
2989             /* Wait for second field. */
2990             *data_size = 0;
2991
2992         } else {
2993             cur->interlaced_frame = 0;
2994             cur->repeat_pict = 0;
2995
2996             /* Signal interlacing information externally. */
2997             /* Prioritize picture timing SEI information over used decoding process if it exists. */
2998
2999             if(h->sps.pic_struct_present_flag){
3000                 switch (h->sei_pic_struct)
3001                 {
3002                 case SEI_PIC_STRUCT_FRAME:
3003                     break;
3004                 case SEI_PIC_STRUCT_TOP_FIELD:
3005                 case SEI_PIC_STRUCT_BOTTOM_FIELD:
3006                     cur->interlaced_frame = 1;
3007                     break;
3008                 case SEI_PIC_STRUCT_TOP_BOTTOM:
3009                 case SEI_PIC_STRUCT_BOTTOM_TOP:
3010                     if (FIELD_OR_MBAFF_PICTURE)
3011                         cur->interlaced_frame = 1;
3012                     else
3013                         // try to flag soft telecine progressive
3014                         cur->interlaced_frame = h->prev_interlaced_frame;
3015                     break;
3016                 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
3017                 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
3018                     // Signal the possibility of telecined film externally (pic_struct 5,6)
3019                     // From these hints, let the applications decide if they apply deinterlacing.
3020                     cur->repeat_pict = 1;
3021                     break;
3022                 case SEI_PIC_STRUCT_FRAME_DOUBLING:
3023                     // Force progressive here, as doubling interlaced frame is a bad idea.
3024                     cur->repeat_pict = 2;
3025                     break;
3026                 case SEI_PIC_STRUCT_FRAME_TRIPLING:
3027                     cur->repeat_pict = 4;
3028                     break;
3029                 }
3030
3031                 if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
3032                     cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
3033             }else{
3034                 /* Derive interlacing flag from used decoding process. */
3035                 cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
3036             }
3037             h->prev_interlaced_frame = cur->interlaced_frame;
3038
3039             if (cur->field_poc[0] != cur->field_poc[1]){
3040                 /* Derive top_field_first from field pocs. */
3041                 cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
3042             }else{
3043                 if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
3044                     /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
3045                     if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
3046                       || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
3047                         cur->top_field_first = 1;
3048                     else
3049                         cur->top_field_first = 0;
3050                 }else{
3051                     /* Most likely progressive */
3052                     cur->top_field_first = 0;
3053                 }
3054             }
3055
3056         //FIXME do something with unavailable reference frames
3057
3058             /* Sort B-frames into display order */
3059
3060             if(h->sps.bitstream_restriction_flag
3061                && s->avctx->has_b_frames < h->sps.num_reorder_frames){
3062                 s->avctx->has_b_frames = h->sps.num_reorder_frames;
3063                 s->low_delay = 0;
3064             }
3065
3066             if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
3067                && !h->sps.bitstream_restriction_flag){
3068                 s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
3069                 s->low_delay= 0;
3070             }
3071
3072             pics = 0;
3073             while(h->delayed_pic[pics]) pics++;
3074
3075             assert(pics <= MAX_DELAYED_PIC_COUNT);
3076
3077             h->delayed_pic[pics++] = cur;
3078             if(cur->reference == 0)
3079                 cur->reference = DELAYED_PIC_REF;
3080
3081             out = h->delayed_pic[0];
3082             out_idx = 0;
3083             for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
3084                 if(h->delayed_pic[i]->poc < out->poc){
3085                     out = h->delayed_pic[i];
3086                     out_idx = i;
3087                 }
3088             if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
3089                 h->outputed_poc= INT_MIN;
3090             out_of_order = out->poc < h->outputed_poc;
3091
3092             if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
3093                 { }
3094             else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
3095                || (s->low_delay &&
3096                 ((h->outputed_poc != INT_MIN && out->poc > h->outputed_poc + 2)
3097                  || cur->pict_type == FF_B_TYPE)))
3098             {
3099                 s->low_delay = 0;
3100                 s->avctx->has_b_frames++;
3101             }
3102
3103             if(out_of_order || pics > s->avctx->has_b_frames){
3104                 out->reference &= ~DELAYED_PIC_REF;
3105                 for(i=out_idx; h->delayed_pic[i]; i++)
3106                     h->delayed_pic[i] = h->delayed_pic[i+1];
3107             }
3108             if(!out_of_order && pics > s->avctx->has_b_frames){
3109                 *data_size = sizeof(AVFrame);
3110
3111                 if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
3112                     h->outputed_poc = INT_MIN;
3113                 } else
3114                     h->outputed_poc = out->poc;
3115                 *pict= *(AVFrame*)out;
3116             }else{
3117                 av_log(avctx, AV_LOG_DEBUG, "no picture\n");
3118             }
3119         }
3120     }
3121
3122     assert(pict->data[0] || !*data_size);
3123     ff_print_debug_info(s, pict);
3124 //printf("out %d\n", (int)pict->data[0]);
3125
3126     return get_consumed_bytes(s, buf_index, buf_size);
3127 }
3128 #if 0
3129 static inline void fill_mb_avail(H264Context *h){
3130     MpegEncContext * const s = &h->s;
3131     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3132
3133     if(s->mb_y){
3134         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
3135         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
3136         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
3137     }else{
3138         h->mb_avail[0]=
3139         h->mb_avail[1]=
3140         h->mb_avail[2]= 0;
3141     }
3142     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
3143     h->mb_avail[4]= 1; //FIXME move out
3144     h->mb_avail[5]= 0; //FIXME move out
3145 }
3146 #endif
3147
3148 #ifdef TEST
3149 #undef printf
3150 #undef random
3151 #define COUNT 8000
3152 #define SIZE (COUNT*40)
3153 int main(void){
3154     int i;
3155     uint8_t temp[SIZE];
3156     PutBitContext pb;
3157     GetBitContext gb;
3158 //    int int_temp[10000];
3159     DSPContext dsp;
3160     AVCodecContext avctx;
3161
3162     dsputil_init(&dsp, &avctx);
3163
3164     init_put_bits(&pb, temp, SIZE);
3165     printf("testing unsigned exp golomb\n");
3166     for(i=0; i<COUNT; i++){
3167         START_TIMER
3168         set_ue_golomb(&pb, i);
3169         STOP_TIMER("set_ue_golomb");
3170     }
3171     flush_put_bits(&pb);
3172
3173     init_get_bits(&gb, temp, 8*SIZE);
3174     for(i=0; i<COUNT; i++){
3175         int j, s;
3176
3177         s= show_bits(&gb, 24);
3178
3179         START_TIMER
3180         j= get_ue_golomb(&gb);
3181         if(j != i){
3182             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
3183 //            return -1;
3184         }
3185         STOP_TIMER("get_ue_golomb");
3186     }
3187
3188
3189     init_put_bits(&pb, temp, SIZE);
3190     printf("testing signed exp golomb\n");
3191     for(i=0; i<COUNT; i++){
3192         START_TIMER
3193         set_se_golomb(&pb, i - COUNT/2);
3194         STOP_TIMER("set_se_golomb");
3195     }
3196     flush_put_bits(&pb);
3197
3198     init_get_bits(&gb, temp, 8*SIZE);
3199     for(i=0; i<COUNT; i++){
3200         int j, s;
3201
3202         s= show_bits(&gb, 24);
3203
3204         START_TIMER
3205         j= get_se_golomb(&gb);
3206         if(j != i - COUNT/2){
3207             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
3208 //            return -1;
3209         }
3210         STOP_TIMER("get_se_golomb");
3211     }
3212
3213 #if 0
3214     printf("testing 4x4 (I)DCT\n");
3215
3216     DCTELEM block[16];
3217     uint8_t src[16], ref[16];
3218     uint64_t error= 0, max_error=0;
3219
3220     for(i=0; i<COUNT; i++){
3221         int j;
3222 //        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
3223         for(j=0; j<16; j++){
3224             ref[j]= random()%255;
3225             src[j]= random()%255;
3226         }
3227
3228         h264_diff_dct_c(block, src, ref, 4);
3229
3230         //normalize
3231         for(j=0; j<16; j++){
3232 //            printf("%d ", block[j]);
3233             block[j]= block[j]*4;
3234             if(j&1) block[j]= (block[j]*4 + 2)/5;
3235             if(j&4) block[j]= (block[j]*4 + 2)/5;
3236         }
3237 //        printf("\n");
3238
3239         h->h264dsp.h264_idct_add(ref, block, 4);
3240 /*        for(j=0; j<16; j++){
3241             printf("%d ", ref[j]);
3242         }
3243         printf("\n");*/
3244
3245         for(j=0; j<16; j++){
3246             int diff= FFABS(src[j] - ref[j]);
3247
3248             error+= diff*diff;
3249             max_error= FFMAX(max_error, diff);
3250         }
3251     }
3252     printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
3253     printf("testing quantizer\n");
3254     for(qp=0; qp<52; qp++){
3255         for(i=0; i<16; i++)
3256             src1_block[i]= src2_block[i]= random()%255;
3257
3258     }
3259     printf("Testing NAL layer\n");
3260
3261     uint8_t bitstream[COUNT];
3262     uint8_t nal[COUNT*2];
3263     H264Context h;
3264     memset(&h, 0, sizeof(H264Context));
3265
3266     for(i=0; i<COUNT; i++){
3267         int zeros= i;
3268         int nal_length;
3269         int consumed;
3270         int out_length;
3271         uint8_t *out;
3272         int j;
3273
3274         for(j=0; j<COUNT; j++){
3275             bitstream[j]= (random() % 255) + 1;
3276         }
3277
3278         for(j=0; j<zeros; j++){
3279             int pos= random() % COUNT;
3280             while(bitstream[pos] == 0){
3281                 pos++;
3282                 pos %= COUNT;
3283             }
3284             bitstream[pos]=0;
3285         }
3286
3287         START_TIMER
3288
3289         nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
3290         if(nal_length<0){
3291             printf("encoding failed\n");
3292             return -1;
3293         }
3294
3295         out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
3296
3297         STOP_TIMER("NAL")
3298
3299         if(out_length != COUNT){
3300             printf("incorrect length %d %d\n", out_length, COUNT);
3301             return -1;
3302         }
3303
3304         if(consumed != nal_length){
3305             printf("incorrect consumed length %d %d\n", nal_length, consumed);
3306             return -1;
3307         }
3308
3309         if(memcmp(bitstream, out, COUNT)){
3310             printf("mismatch\n");
3311             return -1;
3312         }
3313     }
3314 #endif
3315
3316     printf("Testing RBSP\n");
3317
3318
3319     return 0;
3320 }
3321 #endif /* TEST */
3322
3323
3324 av_cold void ff_h264_free_context(H264Context *h)
3325 {
3326     int i;
3327
3328     free_tables(h); //FIXME cleanup init stuff perhaps
3329
3330     for(i = 0; i < MAX_SPS_COUNT; i++)
3331         av_freep(h->sps_buffers + i);
3332
3333     for(i = 0; i < MAX_PPS_COUNT; i++)
3334         av_freep(h->pps_buffers + i);
3335 }
3336
3337 av_cold int ff_h264_decode_end(AVCodecContext *avctx)
3338 {
3339     H264Context *h = avctx->priv_data;
3340     MpegEncContext *s = &h->s;
3341
3342     ff_h264_free_context(h);
3343
3344     MPV_common_end(s);
3345
3346 //    memset(h, 0, sizeof(H264Context));
3347
3348     return 0;
3349 }
3350
3351
3352 AVCodec h264_decoder = {
3353     "h264",
3354     AVMEDIA_TYPE_VIDEO,
3355     CODEC_ID_H264,
3356     sizeof(H264Context),
3357     ff_h264_decode_init,
3358     NULL,
3359     ff_h264_decode_end,
3360     decode_frame,
3361     /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3362     .flush= flush_dpb,
3363     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
3364 };
3365
3366 #if CONFIG_H264_VDPAU_DECODER
3367 AVCodec h264_vdpau_decoder = {
3368     "h264_vdpau",
3369     AVMEDIA_TYPE_VIDEO,
3370     CODEC_ID_H264,
3371     sizeof(H264Context),
3372     ff_h264_decode_init,
3373     NULL,
3374     ff_h264_decode_end,
3375     decode_frame,
3376     CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3377     .flush= flush_dpb,
3378     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
3379     .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
3380 };
3381 #endif