Dont allocate direct_table 8 times too large.
[ffmpeg.git] / libavcodec / h264.c
1 /*
2  * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * FFmpeg is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21
22 /**
23  * @file libavcodec/h264.c
24  * H.264 / AVC / MPEG4 part10 codec.
25  * @author Michael Niedermayer <michaelni@gmx.at>
26  */
27
28 #include "internal.h"
29 #include "dsputil.h"
30 #include "avcodec.h"
31 #include "mpegvideo.h"
32 #include "h264.h"
33 #include "h264data.h"
34 #include "h264_mvpred.h"
35 #include "h264_parser.h"
36 #include "golomb.h"
37 #include "mathops.h"
38 #include "rectangle.h"
39 #include "vdpau_internal.h"
40
41 #include "cabac.h"
42
43 //#undef NDEBUG
44 #include <assert.h>
45
46 static const uint8_t rem6[52]={
47 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
48 };
49
50 static const uint8_t div6[52]={
51 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8,
52 };
53
54 void ff_h264_write_back_intra_pred_mode(H264Context *h){
55     int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
56
57     AV_COPY32(mode, h->intra4x4_pred_mode_cache + 4 + 8*4);
58     mode[4]= h->intra4x4_pred_mode_cache[7+8*3];
59     mode[5]= h->intra4x4_pred_mode_cache[7+8*2];
60     mode[6]= h->intra4x4_pred_mode_cache[7+8*1];
61 }
62
63 /**
64  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
65  */
66 int ff_h264_check_intra4x4_pred_mode(H264Context *h){
67     MpegEncContext * const s = &h->s;
68     static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
69     static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
70     int i;
71
72     if(!(h->top_samples_available&0x8000)){
73         for(i=0; i<4; i++){
74             int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
75             if(status<0){
76                 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
77                 return -1;
78             } else if(status){
79                 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
80             }
81         }
82     }
83
84     if((h->left_samples_available&0x8888)!=0x8888){
85         static const int mask[4]={0x8000,0x2000,0x80,0x20};
86         for(i=0; i<4; i++){
87             if(!(h->left_samples_available&mask[i])){
88                 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
89                 if(status<0){
90                     av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
91                     return -1;
92                 } else if(status){
93                     h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
94                 }
95             }
96         }
97     }
98
99     return 0;
100 } //FIXME cleanup like ff_h264_check_intra_pred_mode
101
102 /**
103  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
104  */
105 int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
106     MpegEncContext * const s = &h->s;
107     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
108     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
109
110     if(mode > 6U) {
111         av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
112         return -1;
113     }
114
115     if(!(h->top_samples_available&0x8000)){
116         mode= top[ mode ];
117         if(mode<0){
118             av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
119             return -1;
120         }
121     }
122
123     if((h->left_samples_available&0x8080) != 0x8080){
124         mode= left[ mode ];
125         if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
126             mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
127         }
128         if(mode<0){
129             av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
130             return -1;
131         }
132     }
133
134     return mode;
135 }
136
137 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
138     int i, si, di;
139     uint8_t *dst;
140     int bufidx;
141
142 //    src[0]&0x80;                //forbidden bit
143     h->nal_ref_idc= src[0]>>5;
144     h->nal_unit_type= src[0]&0x1F;
145
146     src++; length--;
147 #if 0
148     for(i=0; i<length; i++)
149         printf("%2X ", src[i]);
150 #endif
151
152 #if HAVE_FAST_UNALIGNED
153 # if HAVE_FAST_64BIT
154 #   define RS 7
155     for(i=0; i+1<length; i+=9){
156         if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
157 # else
158 #   define RS 3
159     for(i=0; i+1<length; i+=5){
160         if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
161 # endif
162             continue;
163         if(i>0 && !src[i]) i--;
164         while(src[i]) i++;
165 #else
166 #   define RS 0
167     for(i=0; i+1<length; i+=2){
168         if(src[i]) continue;
169         if(i>0 && src[i-1]==0) i--;
170 #endif
171         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
172             if(src[i+2]!=3){
173                 /* startcode, so we must be past the end */
174                 length=i;
175             }
176             break;
177         }
178         i-= RS;
179     }
180
181     if(i>=length-1){ //no escaped 0
182         *dst_length= length;
183         *consumed= length+1; //+1 for the header
184         return src;
185     }
186
187     bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
188     av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
189     dst= h->rbsp_buffer[bufidx];
190
191     if (dst == NULL){
192         return NULL;
193     }
194
195 //printf("decoding esc\n");
196     memcpy(dst, src, i);
197     si=di=i;
198     while(si+2<length){
199         //remove escapes (very rare 1:2^22)
200         if(src[si+2]>3){
201             dst[di++]= src[si++];
202             dst[di++]= src[si++];
203         }else if(src[si]==0 && src[si+1]==0){
204             if(src[si+2]==3){ //escape
205                 dst[di++]= 0;
206                 dst[di++]= 0;
207                 si+=3;
208                 continue;
209             }else //next start code
210                 goto nsc;
211         }
212
213         dst[di++]= src[si++];
214     }
215     while(si<length)
216         dst[di++]= src[si++];
217 nsc:
218
219     memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
220
221     *dst_length= di;
222     *consumed= si + 1;//+1 for the header
223 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
224     return dst;
225 }
226
227 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
228     int v= *src;
229     int r;
230
231     tprintf(h->s.avctx, "rbsp trailing %X\n", v);
232
233     for(r=1; r<9; r++){
234         if(v&1) return r;
235         v>>=1;
236     }
237     return 0;
238 }
239
240 /**
241  * IDCT transforms the 16 dc values and dequantizes them.
242  * @param qp quantization parameter
243  */
244 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
245 #define stride 16
246     int i;
247     int temp[16]; //FIXME check if this is a good idea
248     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
249     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
250
251 //memset(block, 64, 2*256);
252 //return;
253     for(i=0; i<4; i++){
254         const int offset= y_offset[i];
255         const int z0= block[offset+stride*0] + block[offset+stride*4];
256         const int z1= block[offset+stride*0] - block[offset+stride*4];
257         const int z2= block[offset+stride*1] - block[offset+stride*5];
258         const int z3= block[offset+stride*1] + block[offset+stride*5];
259
260         temp[4*i+0]= z0+z3;
261         temp[4*i+1]= z1+z2;
262         temp[4*i+2]= z1-z2;
263         temp[4*i+3]= z0-z3;
264     }
265
266     for(i=0; i<4; i++){
267         const int offset= x_offset[i];
268         const int z0= temp[4*0+i] + temp[4*2+i];
269         const int z1= temp[4*0+i] - temp[4*2+i];
270         const int z2= temp[4*1+i] - temp[4*3+i];
271         const int z3= temp[4*1+i] + temp[4*3+i];
272
273         block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
274         block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
275         block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
276         block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
277     }
278 }
279
280 #if 0
281 /**
282  * DCT transforms the 16 dc values.
283  * @param qp quantization parameter ??? FIXME
284  */
285 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
286 //    const int qmul= dequant_coeff[qp][0];
287     int i;
288     int temp[16]; //FIXME check if this is a good idea
289     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
290     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
291
292     for(i=0; i<4; i++){
293         const int offset= y_offset[i];
294         const int z0= block[offset+stride*0] + block[offset+stride*4];
295         const int z1= block[offset+stride*0] - block[offset+stride*4];
296         const int z2= block[offset+stride*1] - block[offset+stride*5];
297         const int z3= block[offset+stride*1] + block[offset+stride*5];
298
299         temp[4*i+0]= z0+z3;
300         temp[4*i+1]= z1+z2;
301         temp[4*i+2]= z1-z2;
302         temp[4*i+3]= z0-z3;
303     }
304
305     for(i=0; i<4; i++){
306         const int offset= x_offset[i];
307         const int z0= temp[4*0+i] + temp[4*2+i];
308         const int z1= temp[4*0+i] - temp[4*2+i];
309         const int z2= temp[4*1+i] - temp[4*3+i];
310         const int z3= temp[4*1+i] + temp[4*3+i];
311
312         block[stride*0 +offset]= (z0 + z3)>>1;
313         block[stride*2 +offset]= (z1 + z2)>>1;
314         block[stride*8 +offset]= (z1 - z2)>>1;
315         block[stride*10+offset]= (z0 - z3)>>1;
316     }
317 }
318 #endif
319
320 #undef xStride
321 #undef stride
322
323 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
324     const int stride= 16*2;
325     const int xStride= 16;
326     int a,b,c,d,e;
327
328     a= block[stride*0 + xStride*0];
329     b= block[stride*0 + xStride*1];
330     c= block[stride*1 + xStride*0];
331     d= block[stride*1 + xStride*1];
332
333     e= a-b;
334     a= a+b;
335     b= c-d;
336     c= c+d;
337
338     block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
339     block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
340     block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
341     block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
342 }
343
344 #if 0
345 static void chroma_dc_dct_c(DCTELEM *block){
346     const int stride= 16*2;
347     const int xStride= 16;
348     int a,b,c,d,e;
349
350     a= block[stride*0 + xStride*0];
351     b= block[stride*0 + xStride*1];
352     c= block[stride*1 + xStride*0];
353     d= block[stride*1 + xStride*1];
354
355     e= a-b;
356     a= a+b;
357     b= c-d;
358     c= c+d;
359
360     block[stride*0 + xStride*0]= (a+c);
361     block[stride*0 + xStride*1]= (e+b);
362     block[stride*1 + xStride*0]= (a-c);
363     block[stride*1 + xStride*1]= (e-b);
364 }
365 #endif
366
367 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
368                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
369                            int src_x_offset, int src_y_offset,
370                            qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
371     MpegEncContext * const s = &h->s;
372     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
373     int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
374     const int luma_xy= (mx&3) + ((my&3)<<2);
375     uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
376     uint8_t * src_cb, * src_cr;
377     int extra_width= h->emu_edge_width;
378     int extra_height= h->emu_edge_height;
379     int emu=0;
380     const int full_mx= mx>>2;
381     const int full_my= my>>2;
382     const int pic_width  = 16*s->mb_width;
383     const int pic_height = 16*s->mb_height >> MB_FIELD;
384
385     if(mx&7) extra_width -= 3;
386     if(my&7) extra_height -= 3;
387
388     if(   full_mx < 0-extra_width
389        || full_my < 0-extra_height
390        || full_mx + 16/*FIXME*/ > pic_width + extra_width
391        || full_my + 16/*FIXME*/ > pic_height + extra_height){
392         ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
393             src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
394         emu=1;
395     }
396
397     qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
398     if(!square){
399         qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
400     }
401
402     if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
403
404     if(MB_FIELD){
405         // chroma offset when predicting from a field of opposite parity
406         my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
407         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
408     }
409     src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
410     src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
411
412     if(emu){
413         ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
414             src_cb= s->edge_emu_buffer;
415     }
416     chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
417
418     if(emu){
419         ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
420             src_cr= s->edge_emu_buffer;
421     }
422     chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
423 }
424
425 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
426                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
427                            int x_offset, int y_offset,
428                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
429                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
430                            int list0, int list1){
431     MpegEncContext * const s = &h->s;
432     qpel_mc_func *qpix_op=  qpix_put;
433     h264_chroma_mc_func chroma_op= chroma_put;
434
435     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
436     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
437     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
438     x_offset += 8*s->mb_x;
439     y_offset += 8*(s->mb_y >> MB_FIELD);
440
441     if(list0){
442         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
443         mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
444                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
445                            qpix_op, chroma_op);
446
447         qpix_op=  qpix_avg;
448         chroma_op= chroma_avg;
449     }
450
451     if(list1){
452         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
453         mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
454                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
455                            qpix_op, chroma_op);
456     }
457 }
458
459 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
460                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
461                            int x_offset, int y_offset,
462                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
463                            h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
464                            h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
465                            int list0, int list1){
466     MpegEncContext * const s = &h->s;
467
468     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
469     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
470     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
471     x_offset += 8*s->mb_x;
472     y_offset += 8*(s->mb_y >> MB_FIELD);
473
474     if(list0 && list1){
475         /* don't optimize for luma-only case, since B-frames usually
476          * use implicit weights => chroma too. */
477         uint8_t *tmp_cb = s->obmc_scratchpad;
478         uint8_t *tmp_cr = s->obmc_scratchpad + 8;
479         uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
480         int refn0 = h->ref_cache[0][ scan8[n] ];
481         int refn1 = h->ref_cache[1][ scan8[n] ];
482
483         mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
484                     dest_y, dest_cb, dest_cr,
485                     x_offset, y_offset, qpix_put, chroma_put);
486         mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
487                     tmp_y, tmp_cb, tmp_cr,
488                     x_offset, y_offset, qpix_put, chroma_put);
489
490         if(h->use_weight == 2){
491             int weight0 = h->implicit_weight[refn0][refn1];
492             int weight1 = 64 - weight0;
493             luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
494             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
495             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
496         }else{
497             luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
498                             h->luma_weight[0][refn0], h->luma_weight[1][refn1],
499                             h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
500             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
501                             h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
502                             h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
503             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
504                             h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
505                             h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
506         }
507     }else{
508         int list = list1 ? 1 : 0;
509         int refn = h->ref_cache[list][ scan8[n] ];
510         Picture *ref= &h->ref_list[list][refn];
511         mc_dir_part(h, ref, n, square, chroma_height, delta, list,
512                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
513                     qpix_put, chroma_put);
514
515         luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
516                        h->luma_weight[list][refn], h->luma_offset[list][refn]);
517         if(h->use_weight_chroma){
518             chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
519                              h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
520             chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
521                              h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
522         }
523     }
524 }
525
526 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
527                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
528                            int x_offset, int y_offset,
529                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
530                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
531                            h264_weight_func *weight_op, h264_biweight_func *weight_avg,
532                            int list0, int list1){
533     if((h->use_weight==2 && list0 && list1
534         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
535        || h->use_weight==1)
536         mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
537                          x_offset, y_offset, qpix_put, chroma_put,
538                          weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
539     else
540         mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
541                     x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
542 }
543
544 static inline void prefetch_motion(H264Context *h, int list){
545     /* fetch pixels for estimated mv 4 macroblocks ahead
546      * optimized for 64byte cache lines */
547     MpegEncContext * const s = &h->s;
548     const int refn = h->ref_cache[list][scan8[0]];
549     if(refn >= 0){
550         const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
551         const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
552         uint8_t **src= h->ref_list[list][refn].data;
553         int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
554         s->dsp.prefetch(src[0]+off, s->linesize, 4);
555         off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
556         s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
557     }
558 }
559
560 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
561                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
562                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
563                       h264_weight_func *weight_op, h264_biweight_func *weight_avg){
564     MpegEncContext * const s = &h->s;
565     const int mb_xy= h->mb_xy;
566     const int mb_type= s->current_picture.mb_type[mb_xy];
567
568     assert(IS_INTER(mb_type));
569
570     prefetch_motion(h, 0);
571
572     if(IS_16X16(mb_type)){
573         mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
574                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
575                 weight_op, weight_avg,
576                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
577     }else if(IS_16X8(mb_type)){
578         mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
579                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
580                 &weight_op[1], &weight_avg[1],
581                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
582         mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
583                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
584                 &weight_op[1], &weight_avg[1],
585                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
586     }else if(IS_8X16(mb_type)){
587         mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
588                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
589                 &weight_op[2], &weight_avg[2],
590                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
591         mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
592                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
593                 &weight_op[2], &weight_avg[2],
594                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
595     }else{
596         int i;
597
598         assert(IS_8X8(mb_type));
599
600         for(i=0; i<4; i++){
601             const int sub_mb_type= h->sub_mb_type[i];
602             const int n= 4*i;
603             int x_offset= (i&1)<<2;
604             int y_offset= (i&2)<<1;
605
606             if(IS_SUB_8X8(sub_mb_type)){
607                 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
608                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
609                     &weight_op[3], &weight_avg[3],
610                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
611             }else if(IS_SUB_8X4(sub_mb_type)){
612                 mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
613                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
614                     &weight_op[4], &weight_avg[4],
615                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
616                 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
617                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
618                     &weight_op[4], &weight_avg[4],
619                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
620             }else if(IS_SUB_4X8(sub_mb_type)){
621                 mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
622                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
623                     &weight_op[5], &weight_avg[5],
624                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
625                 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
626                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
627                     &weight_op[5], &weight_avg[5],
628                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
629             }else{
630                 int j;
631                 assert(IS_SUB_4X4(sub_mb_type));
632                 for(j=0; j<4; j++){
633                     int sub_x_offset= x_offset + 2*(j&1);
634                     int sub_y_offset= y_offset +   (j&2);
635                     mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
636                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
637                         &weight_op[6], &weight_avg[6],
638                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
639                 }
640             }
641         }
642     }
643
644     prefetch_motion(h, 1);
645 }
646
647
648 static void free_tables(H264Context *h){
649     int i;
650     H264Context *hx;
651     av_freep(&h->intra4x4_pred_mode);
652     av_freep(&h->chroma_pred_mode_table);
653     av_freep(&h->cbp_table);
654     av_freep(&h->mvd_table[0]);
655     av_freep(&h->mvd_table[1]);
656     av_freep(&h->direct_table);
657     av_freep(&h->non_zero_count);
658     av_freep(&h->slice_table_base);
659     h->slice_table= NULL;
660     av_freep(&h->list_counts);
661
662     av_freep(&h->mb2b_xy);
663     av_freep(&h->mb2br_xy);
664     av_freep(&h->mb2b8_xy);
665
666     for(i = 0; i < MAX_THREADS; i++) {
667         hx = h->thread_context[i];
668         if(!hx) continue;
669         av_freep(&hx->top_borders[1]);
670         av_freep(&hx->top_borders[0]);
671         av_freep(&hx->s.obmc_scratchpad);
672         av_freep(&hx->rbsp_buffer[1]);
673         av_freep(&hx->rbsp_buffer[0]);
674         hx->rbsp_buffer_size[0] = 0;
675         hx->rbsp_buffer_size[1] = 0;
676         if (i) av_freep(&h->thread_context[i]);
677     }
678 }
679
680 static void init_dequant8_coeff_table(H264Context *h){
681     int i,q,x;
682     const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
683     h->dequant8_coeff[0] = h->dequant8_buffer[0];
684     h->dequant8_coeff[1] = h->dequant8_buffer[1];
685
686     for(i=0; i<2; i++ ){
687         if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
688             h->dequant8_coeff[1] = h->dequant8_buffer[0];
689             break;
690         }
691
692         for(q=0; q<52; q++){
693             int shift = div6[q];
694             int idx = rem6[q];
695             for(x=0; x<64; x++)
696                 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
697                     ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
698                     h->pps.scaling_matrix8[i][x]) << shift;
699         }
700     }
701 }
702
703 static void init_dequant4_coeff_table(H264Context *h){
704     int i,j,q,x;
705     const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
706     for(i=0; i<6; i++ ){
707         h->dequant4_coeff[i] = h->dequant4_buffer[i];
708         for(j=0; j<i; j++){
709             if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
710                 h->dequant4_coeff[i] = h->dequant4_buffer[j];
711                 break;
712             }
713         }
714         if(j<i)
715             continue;
716
717         for(q=0; q<52; q++){
718             int shift = div6[q] + 2;
719             int idx = rem6[q];
720             for(x=0; x<16; x++)
721                 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
722                     ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
723                     h->pps.scaling_matrix4[i][x]) << shift;
724         }
725     }
726 }
727
728 static void init_dequant_tables(H264Context *h){
729     int i,x;
730     init_dequant4_coeff_table(h);
731     if(h->pps.transform_8x8_mode)
732         init_dequant8_coeff_table(h);
733     if(h->sps.transform_bypass){
734         for(i=0; i<6; i++)
735             for(x=0; x<16; x++)
736                 h->dequant4_coeff[i][0][x] = 1<<6;
737         if(h->pps.transform_8x8_mode)
738             for(i=0; i<2; i++)
739                 for(x=0; x<64; x++)
740                     h->dequant8_coeff[i][0][x] = 1<<6;
741     }
742 }
743
744
745 int ff_h264_alloc_tables(H264Context *h){
746     MpegEncContext * const s = &h->s;
747     const int big_mb_num= s->mb_stride * (s->mb_height+1);
748     int x,y;
749
750     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t), fail)
751
752     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 32 * sizeof(uint8_t), fail)
753     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
754     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
755
756     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
757     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*big_mb_num * sizeof(uint8_t), fail);
758     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*big_mb_num * sizeof(uint8_t), fail);
759     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
760     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
761
762     memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
763     h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
764
765     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy  , big_mb_num * sizeof(uint32_t), fail);
766     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
767     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b8_xy , big_mb_num * sizeof(uint32_t), fail);
768     for(y=0; y<s->mb_height; y++){
769         for(x=0; x<s->mb_width; x++){
770             const int mb_xy= x + y*s->mb_stride;
771             const int b_xy = 4*x + 4*y*h->b_stride;
772             const int b8_xy= 2*x + 2*y*h->b8_stride;
773
774             h->mb2b_xy [mb_xy]= b_xy;
775             h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
776             h->mb2b8_xy[mb_xy]= b8_xy;
777         }
778     }
779
780     s->obmc_scratchpad = NULL;
781
782     if(!h->dequant4_coeff[0])
783         init_dequant_tables(h);
784
785     return 0;
786 fail:
787     free_tables(h);
788     return -1;
789 }
790
791 /**
792  * Mimic alloc_tables(), but for every context thread.
793  */
794 static void clone_tables(H264Context *dst, H264Context *src){
795     dst->intra4x4_pred_mode       = src->intra4x4_pred_mode;
796     dst->non_zero_count           = src->non_zero_count;
797     dst->slice_table              = src->slice_table;
798     dst->cbp_table                = src->cbp_table;
799     dst->mb2b_xy                  = src->mb2b_xy;
800     dst->mb2br_xy                 = src->mb2br_xy;
801     dst->mb2b8_xy                 = src->mb2b8_xy;
802     dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
803     dst->mvd_table[0]             = src->mvd_table[0];
804     dst->mvd_table[1]             = src->mvd_table[1];
805     dst->direct_table             = src->direct_table;
806     dst->list_counts              = src->list_counts;
807
808     dst->s.obmc_scratchpad = NULL;
809     ff_h264_pred_init(&dst->hpc, src->s.codec_id);
810 }
811
812 /**
813  * Init context
814  * Allocate buffers which are not shared amongst multiple threads.
815  */
816 static int context_init(H264Context *h){
817     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
818     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
819
820     return 0;
821 fail:
822     return -1; // free_tables will clean up for us
823 }
824
825 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
826
827 static av_cold void common_init(H264Context *h){
828     MpegEncContext * const s = &h->s;
829
830     s->width = s->avctx->width;
831     s->height = s->avctx->height;
832     s->codec_id= s->avctx->codec->id;
833
834     ff_h264_pred_init(&h->hpc, s->codec_id);
835
836     h->dequant_coeff_pps= -1;
837     s->unrestricted_mv=1;
838     s->decode=1; //FIXME
839
840     dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
841
842     memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
843     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
844 }
845
846 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
847     H264Context *h= avctx->priv_data;
848     MpegEncContext * const s = &h->s;
849
850     MPV_decode_defaults(s);
851
852     s->avctx = avctx;
853     common_init(h);
854
855     s->out_format = FMT_H264;
856     s->workaround_bugs= avctx->workaround_bugs;
857
858     // set defaults
859 //    s->decode_mb= ff_h263_decode_mb;
860     s->quarter_sample = 1;
861     if(!avctx->has_b_frames)
862     s->low_delay= 1;
863
864     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
865
866     ff_h264_decode_init_vlc();
867
868     h->thread_context[0] = h;
869     h->outputed_poc = INT_MIN;
870     h->prev_poc_msb= 1<<16;
871     h->x264_build = -1;
872     ff_h264_reset_sei(h);
873     if(avctx->codec_id == CODEC_ID_H264){
874         if(avctx->ticks_per_frame == 1){
875             s->avctx->time_base.den *=2;
876         }
877         avctx->ticks_per_frame = 2;
878     }
879
880     if(avctx->extradata_size > 0 && avctx->extradata && *(char *)avctx->extradata == 1){
881         int i, cnt, nalsize;
882         unsigned char *p = avctx->extradata;
883
884         h->is_avc = 1;
885
886         if(avctx->extradata_size < 7) {
887             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
888             return -1;
889         }
890         /* sps and pps in the avcC always have length coded with 2 bytes,
891            so put a fake nal_length_size = 2 while parsing them */
892         h->nal_length_size = 2;
893         // Decode sps from avcC
894         cnt = *(p+5) & 0x1f; // Number of sps
895         p += 6;
896         for (i = 0; i < cnt; i++) {
897             nalsize = AV_RB16(p) + 2;
898             if(decode_nal_units(h, p, nalsize) < 0) {
899                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
900                 return -1;
901             }
902             p += nalsize;
903         }
904         // Decode pps from avcC
905         cnt = *(p++); // Number of pps
906         for (i = 0; i < cnt; i++) {
907             nalsize = AV_RB16(p) + 2;
908             if(decode_nal_units(h, p, nalsize)  != nalsize) {
909                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
910                 return -1;
911             }
912             p += nalsize;
913         }
914         // Now store right nal length size, that will be use to parse all other nals
915         h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
916     } else {
917         h->is_avc = 0;
918         if(decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) < 0)
919             return -1;
920     }
921     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
922         s->avctx->has_b_frames = h->sps.num_reorder_frames;
923         s->low_delay = 0;
924     }
925
926     return 0;
927 }
928
929 int ff_h264_frame_start(H264Context *h){
930     MpegEncContext * const s = &h->s;
931     int i;
932
933     if(MPV_frame_start(s, s->avctx) < 0)
934         return -1;
935     ff_er_frame_start(s);
936     /*
937      * MPV_frame_start uses pict_type to derive key_frame.
938      * This is incorrect for H.264; IDR markings must be used.
939      * Zero here; IDR markings per slice in frame or fields are ORed in later.
940      * See decode_nal_units().
941      */
942     s->current_picture_ptr->key_frame= 0;
943     s->current_picture_ptr->mmco_reset= 0;
944
945     assert(s->linesize && s->uvlinesize);
946
947     for(i=0; i<16; i++){
948         h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
949         h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
950     }
951     for(i=0; i<4; i++){
952         h->block_offset[16+i]=
953         h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
954         h->block_offset[24+16+i]=
955         h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
956     }
957
958     /* can't be in alloc_tables because linesize isn't known there.
959      * FIXME: redo bipred weight to not require extra buffer? */
960     for(i = 0; i < s->avctx->thread_count; i++)
961         if(!h->thread_context[i]->s.obmc_scratchpad)
962             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
963
964     /* some macroblocks will be accessed before they're available */
965     if(FRAME_MBAFF || s->avctx->thread_count > 1)
966         memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
967
968 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
969
970     // We mark the current picture as non-reference after allocating it, so
971     // that if we break out due to an error it can be released automatically
972     // in the next MPV_frame_start().
973     // SVQ3 as well as most other codecs have only last/next/current and thus
974     // get released even with set reference, besides SVQ3 and others do not
975     // mark frames as reference later "naturally".
976     if(s->codec_id != CODEC_ID_SVQ3)
977         s->current_picture_ptr->reference= 0;
978
979     s->current_picture_ptr->field_poc[0]=
980     s->current_picture_ptr->field_poc[1]= INT_MAX;
981     assert(s->current_picture_ptr->long_ref==0);
982
983     return 0;
984 }
985
986 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int simple){
987     MpegEncContext * const s = &h->s;
988     uint8_t *top_border;
989     int top_idx = 1;
990
991     src_y  -=   linesize;
992     src_cb -= uvlinesize;
993     src_cr -= uvlinesize;
994
995     if(!simple && FRAME_MBAFF){
996         if(s->mb_y&1){
997             if(!MB_MBAFF){
998                 top_border = h->top_borders[0][s->mb_x];
999                 AV_COPY128(top_border, src_y + 15*linesize);
1000                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1001                     AV_COPY64(top_border+16, src_cb+7*uvlinesize);
1002                     AV_COPY64(top_border+24, src_cr+7*uvlinesize);
1003                 }
1004             }
1005         }else if(MB_MBAFF){
1006             top_idx = 0;
1007         }else
1008             return;
1009     }
1010
1011     top_border = h->top_borders[top_idx][s->mb_x];
1012     // There are two lines saved, the line above the the top macroblock of a pair,
1013     // and the line above the bottom macroblock
1014     AV_COPY128(top_border, src_y + 16*linesize);
1015
1016     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1017         AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1018         AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1019     }
1020 }
1021
1022 static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg, int simple){
1023     MpegEncContext * const s = &h->s;
1024     int deblock_left;
1025     int deblock_top;
1026     int top_idx = 1;
1027     uint8_t *top_border_m1;
1028     uint8_t *top_border;
1029
1030     if(!simple && FRAME_MBAFF){
1031         if(s->mb_y&1){
1032             if(!MB_MBAFF)
1033                 return;
1034         }else{
1035             top_idx = MB_MBAFF ? 0 : 1;
1036         }
1037     }
1038
1039     if(h->deblocking_filter == 2) {
1040         deblock_left = h->left_type[0];
1041         deblock_top  = h->top_type;
1042     } else {
1043         deblock_left = (s->mb_x > 0);
1044         deblock_top =  (s->mb_y > !!MB_FIELD);
1045     }
1046
1047     src_y  -=   linesize + 1;
1048     src_cb -= uvlinesize + 1;
1049     src_cr -= uvlinesize + 1;
1050
1051     top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1052     top_border    = h->top_borders[top_idx][s->mb_x];
1053
1054 #define XCHG(a,b,xchg)\
1055 if (xchg) AV_SWAP64(b,a);\
1056 else      AV_COPY64(b,a);
1057
1058     if(deblock_top){
1059         if(deblock_left){
1060             XCHG(top_border_m1+8, src_y -7, 1);
1061         }
1062         XCHG(top_border+0, src_y +1, xchg);
1063         XCHG(top_border+8, src_y +9, 1);
1064         if(s->mb_x+1 < s->mb_width){
1065             XCHG(h->top_borders[top_idx][s->mb_x+1], src_y +17, 1);
1066         }
1067     }
1068
1069     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1070         if(deblock_top){
1071             if(deblock_left){
1072                 XCHG(top_border_m1+16, src_cb -7, 1);
1073                 XCHG(top_border_m1+24, src_cr -7, 1);
1074             }
1075             XCHG(top_border+16, src_cb+1, 1);
1076             XCHG(top_border+24, src_cr+1, 1);
1077         }
1078     }
1079 }
1080
1081 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
1082     MpegEncContext * const s = &h->s;
1083     const int mb_x= s->mb_x;
1084     const int mb_y= s->mb_y;
1085     const int mb_xy= h->mb_xy;
1086     const int mb_type= s->current_picture.mb_type[mb_xy];
1087     uint8_t  *dest_y, *dest_cb, *dest_cr;
1088     int linesize, uvlinesize /*dct_offset*/;
1089     int i;
1090     int *block_offset = &h->block_offset[0];
1091     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1092     /* is_h264 should always be true if SVQ3 is disabled. */
1093     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1094     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1095     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1096
1097     dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
1098     dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
1099     dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
1100
1101     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
1102     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
1103
1104     h->list_counts[mb_xy]= h->list_count;
1105
1106     if (!simple && MB_FIELD) {
1107         linesize   = h->mb_linesize   = s->linesize * 2;
1108         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1109         block_offset = &h->block_offset[24];
1110         if(mb_y&1){ //FIXME move out of this function?
1111             dest_y -= s->linesize*15;
1112             dest_cb-= s->uvlinesize*7;
1113             dest_cr-= s->uvlinesize*7;
1114         }
1115         if(FRAME_MBAFF) {
1116             int list;
1117             for(list=0; list<h->list_count; list++){
1118                 if(!USES_LIST(mb_type, list))
1119                     continue;
1120                 if(IS_16X16(mb_type)){
1121                     int8_t *ref = &h->ref_cache[list][scan8[0]];
1122                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1123                 }else{
1124                     for(i=0; i<16; i+=4){
1125                         int ref = h->ref_cache[list][scan8[i]];
1126                         if(ref >= 0)
1127                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1128                     }
1129                 }
1130             }
1131         }
1132     } else {
1133         linesize   = h->mb_linesize   = s->linesize;
1134         uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1135 //        dct_offset = s->linesize * 16;
1136     }
1137
1138     if (!simple && IS_INTRA_PCM(mb_type)) {
1139         for (i=0; i<16; i++) {
1140             memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1141         }
1142         for (i=0; i<8; i++) {
1143             memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
1144             memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
1145         }
1146     } else {
1147         if(IS_INTRA(mb_type)){
1148             if(h->deblocking_filter)
1149                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
1150
1151             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1152                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1153                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1154             }
1155
1156             if(IS_INTRA4x4(mb_type)){
1157                 if(simple || !s->encoding){
1158                     if(IS_8x8DCT(mb_type)){
1159                         if(transform_bypass){
1160                             idct_dc_add =
1161                             idct_add    = s->dsp.add_pixels8;
1162                         }else{
1163                             idct_dc_add = s->dsp.h264_idct8_dc_add;
1164                             idct_add    = s->dsp.h264_idct8_add;
1165                         }
1166                         for(i=0; i<16; i+=4){
1167                             uint8_t * const ptr= dest_y + block_offset[i];
1168                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1169                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1170                                 h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
1171                             }else{
1172                                 const int nnz = h->non_zero_count_cache[ scan8[i] ];
1173                                 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1174                                                             (h->topright_samples_available<<i)&0x4000, linesize);
1175                                 if(nnz){
1176                                     if(nnz == 1 && h->mb[i*16])
1177                                         idct_dc_add(ptr, h->mb + i*16, linesize);
1178                                     else
1179                                         idct_add   (ptr, h->mb + i*16, linesize);
1180                                 }
1181                             }
1182                         }
1183                     }else{
1184                         if(transform_bypass){
1185                             idct_dc_add =
1186                             idct_add    = s->dsp.add_pixels4;
1187                         }else{
1188                             idct_dc_add = s->dsp.h264_idct_dc_add;
1189                             idct_add    = s->dsp.h264_idct_add;
1190                         }
1191                         for(i=0; i<16; i++){
1192                             uint8_t * const ptr= dest_y + block_offset[i];
1193                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1194
1195                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1196                                 h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
1197                             }else{
1198                                 uint8_t *topright;
1199                                 int nnz, tr;
1200                                 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1201                                     const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1202                                     assert(mb_y || linesize <= block_offset[i]);
1203                                     if(!topright_avail){
1204                                         tr= ptr[3 - linesize]*0x01010101;
1205                                         topright= (uint8_t*) &tr;
1206                                     }else
1207                                         topright= ptr + 4 - linesize;
1208                                 }else
1209                                     topright= NULL;
1210
1211                                 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1212                                 nnz = h->non_zero_count_cache[ scan8[i] ];
1213                                 if(nnz){
1214                                     if(is_h264){
1215                                         if(nnz == 1 && h->mb[i*16])
1216                                             idct_dc_add(ptr, h->mb + i*16, linesize);
1217                                         else
1218                                             idct_add   (ptr, h->mb + i*16, linesize);
1219                                     }else
1220                                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1221                                 }
1222                             }
1223                         }
1224                     }
1225                 }
1226             }else{
1227                 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1228                 if(is_h264){
1229                     if(!transform_bypass)
1230                         h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
1231                 }else
1232                     ff_svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
1233             }
1234             if(h->deblocking_filter)
1235                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
1236         }else if(is_h264){
1237             hl_motion(h, dest_y, dest_cb, dest_cr,
1238                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1239                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1240                       s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
1241         }
1242
1243
1244         if(!IS_INTRA4x4(mb_type)){
1245             if(is_h264){
1246                 if(IS_INTRA16x16(mb_type)){
1247                     if(transform_bypass){
1248                         if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1249                             h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1250                         }else{
1251                             for(i=0; i<16; i++){
1252                                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1253                                     s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
1254                             }
1255                         }
1256                     }else{
1257                          s->dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1258                     }
1259                 }else if(h->cbp&15){
1260                     if(transform_bypass){
1261                         const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1262                         idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1263                         for(i=0; i<16; i+=di){
1264                             if(h->non_zero_count_cache[ scan8[i] ]){
1265                                 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
1266                             }
1267                         }
1268                     }else{
1269                         if(IS_8x8DCT(mb_type)){
1270                             s->dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1271                         }else{
1272                             s->dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1273                         }
1274                     }
1275                 }
1276             }else{
1277                 for(i=0; i<16; i++){
1278                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1279                         uint8_t * const ptr= dest_y + block_offset[i];
1280                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1281                     }
1282                 }
1283             }
1284         }
1285
1286         if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1287             uint8_t *dest[2] = {dest_cb, dest_cr};
1288             if(transform_bypass){
1289                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1290                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
1291                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
1292                 }else{
1293                     idct_add = s->dsp.add_pixels4;
1294                     for(i=16; i<16+8; i++){
1295                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1296                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1297                     }
1298                 }
1299             }else{
1300                 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp[0], h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
1301                 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp[1], h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
1302                 if(is_h264){
1303                     idct_add = s->dsp.h264_idct_add;
1304                     idct_dc_add = s->dsp.h264_idct_dc_add;
1305                     for(i=16; i<16+8; i++){
1306                         if(h->non_zero_count_cache[ scan8[i] ])
1307                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1308                         else if(h->mb[i*16])
1309                             idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1310                     }
1311                 }else{
1312                     for(i=16; i<16+8; i++){
1313                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1314                             uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1315                             ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
1316                         }
1317                     }
1318                 }
1319             }
1320         }
1321     }
1322     if(h->cbp || IS_INTRA(mb_type))
1323         s->dsp.clear_blocks(h->mb);
1324 }
1325
1326 /**
1327  * Process a macroblock; this case avoids checks for expensive uncommon cases.
1328  */
1329 static void hl_decode_mb_simple(H264Context *h){
1330     hl_decode_mb_internal(h, 1);
1331 }
1332
1333 /**
1334  * Process a macroblock; this handles edge cases, such as interlacing.
1335  */
1336 static void av_noinline hl_decode_mb_complex(H264Context *h){
1337     hl_decode_mb_internal(h, 0);
1338 }
1339
1340 void ff_h264_hl_decode_mb(H264Context *h){
1341     MpegEncContext * const s = &h->s;
1342     const int mb_xy= h->mb_xy;
1343     const int mb_type= s->current_picture.mb_type[mb_xy];
1344     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1345
1346     if (is_complex)
1347         hl_decode_mb_complex(h);
1348     else hl_decode_mb_simple(h);
1349 }
1350
1351 static int pred_weight_table(H264Context *h){
1352     MpegEncContext * const s = &h->s;
1353     int list, i;
1354     int luma_def, chroma_def;
1355
1356     h->use_weight= 0;
1357     h->use_weight_chroma= 0;
1358     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1359     h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1360     luma_def = 1<<h->luma_log2_weight_denom;
1361     chroma_def = 1<<h->chroma_log2_weight_denom;
1362
1363     for(list=0; list<2; list++){
1364         h->luma_weight_flag[list]   = 0;
1365         h->chroma_weight_flag[list] = 0;
1366         for(i=0; i<h->ref_count[list]; i++){
1367             int luma_weight_flag, chroma_weight_flag;
1368
1369             luma_weight_flag= get_bits1(&s->gb);
1370             if(luma_weight_flag){
1371                 h->luma_weight[list][i]= get_se_golomb(&s->gb);
1372                 h->luma_offset[list][i]= get_se_golomb(&s->gb);
1373                 if(   h->luma_weight[list][i] != luma_def
1374                    || h->luma_offset[list][i] != 0) {
1375                     h->use_weight= 1;
1376                     h->luma_weight_flag[list]= 1;
1377                 }
1378             }else{
1379                 h->luma_weight[list][i]= luma_def;
1380                 h->luma_offset[list][i]= 0;
1381             }
1382
1383             if(CHROMA){
1384                 chroma_weight_flag= get_bits1(&s->gb);
1385                 if(chroma_weight_flag){
1386                     int j;
1387                     for(j=0; j<2; j++){
1388                         h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
1389                         h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
1390                         if(   h->chroma_weight[list][i][j] != chroma_def
1391                            || h->chroma_offset[list][i][j] != 0) {
1392                             h->use_weight_chroma= 1;
1393                             h->chroma_weight_flag[list]= 1;
1394                         }
1395                     }
1396                 }else{
1397                     int j;
1398                     for(j=0; j<2; j++){
1399                         h->chroma_weight[list][i][j]= chroma_def;
1400                         h->chroma_offset[list][i][j]= 0;
1401                     }
1402                 }
1403             }
1404         }
1405         if(h->slice_type_nos != FF_B_TYPE) break;
1406     }
1407     h->use_weight= h->use_weight || h->use_weight_chroma;
1408     return 0;
1409 }
1410
1411 static void implicit_weight_table(H264Context *h){
1412     MpegEncContext * const s = &h->s;
1413     int ref0, ref1, i;
1414     int cur_poc = s->current_picture_ptr->poc;
1415
1416     for (i = 0; i < 2; i++) {
1417         h->luma_weight_flag[i]   = 0;
1418         h->chroma_weight_flag[i] = 0;
1419     }
1420
1421     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
1422        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1423         h->use_weight= 0;
1424         h->use_weight_chroma= 0;
1425         return;
1426     }
1427
1428     h->use_weight= 2;
1429     h->use_weight_chroma= 2;
1430     h->luma_log2_weight_denom= 5;
1431     h->chroma_log2_weight_denom= 5;
1432
1433     for(ref0=0; ref0 < h->ref_count[0]; ref0++){
1434         int poc0 = h->ref_list[0][ref0].poc;
1435         for(ref1=0; ref1 < h->ref_count[1]; ref1++){
1436             int poc1 = h->ref_list[1][ref1].poc;
1437             int td = av_clip(poc1 - poc0, -128, 127);
1438             if(td){
1439                 int tb = av_clip(cur_poc - poc0, -128, 127);
1440                 int tx = (16384 + (FFABS(td) >> 1)) / td;
1441                 int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
1442                 if(dist_scale_factor < -64 || dist_scale_factor > 128)
1443                     h->implicit_weight[ref0][ref1] = 32;
1444                 else
1445                     h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
1446             }else
1447                 h->implicit_weight[ref0][ref1] = 32;
1448         }
1449     }
1450 }
1451
1452 /**
1453  * instantaneous decoder refresh.
1454  */
1455 static void idr(H264Context *h){
1456     ff_h264_remove_all_refs(h);
1457     h->prev_frame_num= 0;
1458     h->prev_frame_num_offset= 0;
1459     h->prev_poc_msb=
1460     h->prev_poc_lsb= 0;
1461 }
1462
1463 /* forget old pics after a seek */
1464 static void flush_dpb(AVCodecContext *avctx){
1465     H264Context *h= avctx->priv_data;
1466     int i;
1467     for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1468         if(h->delayed_pic[i])
1469             h->delayed_pic[i]->reference= 0;
1470         h->delayed_pic[i]= NULL;
1471     }
1472     h->outputed_poc= INT_MIN;
1473     h->prev_interlaced_frame = 1;
1474     idr(h);
1475     if(h->s.current_picture_ptr)
1476         h->s.current_picture_ptr->reference= 0;
1477     h->s.first_field= 0;
1478     ff_h264_reset_sei(h);
1479     ff_mpeg_flush(avctx);
1480 }
1481
1482 static int init_poc(H264Context *h){
1483     MpegEncContext * const s = &h->s;
1484     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1485     int field_poc[2];
1486     Picture *cur = s->current_picture_ptr;
1487
1488     h->frame_num_offset= h->prev_frame_num_offset;
1489     if(h->frame_num < h->prev_frame_num)
1490         h->frame_num_offset += max_frame_num;
1491
1492     if(h->sps.poc_type==0){
1493         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1494
1495         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1496             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1497         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1498             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1499         else
1500             h->poc_msb = h->prev_poc_msb;
1501 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1502         field_poc[0] =
1503         field_poc[1] = h->poc_msb + h->poc_lsb;
1504         if(s->picture_structure == PICT_FRAME)
1505             field_poc[1] += h->delta_poc_bottom;
1506     }else if(h->sps.poc_type==1){
1507         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1508         int i;
1509
1510         if(h->sps.poc_cycle_length != 0)
1511             abs_frame_num = h->frame_num_offset + h->frame_num;
1512         else
1513             abs_frame_num = 0;
1514
1515         if(h->nal_ref_idc==0 && abs_frame_num > 0)
1516             abs_frame_num--;
1517
1518         expected_delta_per_poc_cycle = 0;
1519         for(i=0; i < h->sps.poc_cycle_length; i++)
1520             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1521
1522         if(abs_frame_num > 0){
1523             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1524             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1525
1526             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1527             for(i = 0; i <= frame_num_in_poc_cycle; i++)
1528                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1529         } else
1530             expectedpoc = 0;
1531
1532         if(h->nal_ref_idc == 0)
1533             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1534
1535         field_poc[0] = expectedpoc + h->delta_poc[0];
1536         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1537
1538         if(s->picture_structure == PICT_FRAME)
1539             field_poc[1] += h->delta_poc[1];
1540     }else{
1541         int poc= 2*(h->frame_num_offset + h->frame_num);
1542
1543         if(!h->nal_ref_idc)
1544             poc--;
1545
1546         field_poc[0]= poc;
1547         field_poc[1]= poc;
1548     }
1549
1550     if(s->picture_structure != PICT_BOTTOM_FIELD)
1551         s->current_picture_ptr->field_poc[0]= field_poc[0];
1552     if(s->picture_structure != PICT_TOP_FIELD)
1553         s->current_picture_ptr->field_poc[1]= field_poc[1];
1554     cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1555
1556     return 0;
1557 }
1558
1559
1560 /**
1561  * initialize scan tables
1562  */
1563 static void init_scan_tables(H264Context *h){
1564     MpegEncContext * const s = &h->s;
1565     int i;
1566     if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
1567         memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
1568         memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
1569     }else{
1570         for(i=0; i<16; i++){
1571 #define T(x) (x>>2) | ((x<<2) & 0xF)
1572             h->zigzag_scan[i] = T(zigzag_scan[i]);
1573             h-> field_scan[i] = T( field_scan[i]);
1574 #undef T
1575         }
1576     }
1577     if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
1578         memcpy(h->zigzag_scan8x8,       ff_zigzag_direct,     64*sizeof(uint8_t));
1579         memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
1580         memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
1581         memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
1582     }else{
1583         for(i=0; i<64; i++){
1584 #define T(x) (x>>3) | ((x&7)<<3)
1585             h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
1586             h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1587             h->field_scan8x8[i]        = T(field_scan8x8[i]);
1588             h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
1589 #undef T
1590         }
1591     }
1592     if(h->sps.transform_bypass){ //FIXME same ugly
1593         h->zigzag_scan_q0          = zigzag_scan;
1594         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
1595         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1596         h->field_scan_q0           = field_scan;
1597         h->field_scan8x8_q0        = field_scan8x8;
1598         h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
1599     }else{
1600         h->zigzag_scan_q0          = h->zigzag_scan;
1601         h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
1602         h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1603         h->field_scan_q0           = h->field_scan;
1604         h->field_scan8x8_q0        = h->field_scan8x8;
1605         h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
1606     }
1607 }
1608
1609 static void field_end(H264Context *h){
1610     MpegEncContext * const s = &h->s;
1611     AVCodecContext * const avctx= s->avctx;
1612     s->mb_y= 0;
1613
1614     s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1615     s->current_picture_ptr->pict_type= s->pict_type;
1616
1617     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1618         ff_vdpau_h264_set_reference_frames(s);
1619
1620     if(!s->dropable) {
1621         ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1622         h->prev_poc_msb= h->poc_msb;
1623         h->prev_poc_lsb= h->poc_lsb;
1624     }
1625     h->prev_frame_num_offset= h->frame_num_offset;
1626     h->prev_frame_num= h->frame_num;
1627
1628     if (avctx->hwaccel) {
1629         if (avctx->hwaccel->end_frame(avctx) < 0)
1630             av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1631     }
1632
1633     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1634         ff_vdpau_h264_picture_complete(s);
1635
1636     /*
1637      * FIXME: Error handling code does not seem to support interlaced
1638      * when slices span multiple rows
1639      * The ff_er_add_slice calls don't work right for bottom
1640      * fields; they cause massive erroneous error concealing
1641      * Error marking covers both fields (top and bottom).
1642      * This causes a mismatched s->error_count
1643      * and a bad error table. Further, the error count goes to
1644      * INT_MAX when called for bottom field, because mb_y is
1645      * past end by one (callers fault) and resync_mb_y != 0
1646      * causes problems for the first MB line, too.
1647      */
1648     if (!FIELD_PICTURE)
1649         ff_er_frame_end(s);
1650
1651     MPV_frame_end(s);
1652
1653     h->current_slice=0;
1654 }
1655
1656 /**
1657  * Replicates H264 "master" context to thread contexts.
1658  */
1659 static void clone_slice(H264Context *dst, H264Context *src)
1660 {
1661     memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
1662     dst->s.current_picture_ptr  = src->s.current_picture_ptr;
1663     dst->s.current_picture      = src->s.current_picture;
1664     dst->s.linesize             = src->s.linesize;
1665     dst->s.uvlinesize           = src->s.uvlinesize;
1666     dst->s.first_field          = src->s.first_field;
1667
1668     dst->prev_poc_msb           = src->prev_poc_msb;
1669     dst->prev_poc_lsb           = src->prev_poc_lsb;
1670     dst->prev_frame_num_offset  = src->prev_frame_num_offset;
1671     dst->prev_frame_num         = src->prev_frame_num;
1672     dst->short_ref_count        = src->short_ref_count;
1673
1674     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
1675     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
1676     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1677     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
1678
1679     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
1680     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
1681 }
1682
1683 /**
1684  * decodes a slice header.
1685  * This will also call MPV_common_init() and frame_start() as needed.
1686  *
1687  * @param h h264context
1688  * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1689  *
1690  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1691  */
1692 static int decode_slice_header(H264Context *h, H264Context *h0){
1693     MpegEncContext * const s = &h->s;
1694     MpegEncContext * const s0 = &h0->s;
1695     unsigned int first_mb_in_slice;
1696     unsigned int pps_id;
1697     int num_ref_idx_active_override_flag;
1698     unsigned int slice_type, tmp, i, j;
1699     int default_ref_list_done = 0;
1700     int last_pic_structure;
1701
1702     s->dropable= h->nal_ref_idc == 0;
1703
1704     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1705         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1706         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1707     }else{
1708         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1709         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1710     }
1711
1712     first_mb_in_slice= get_ue_golomb(&s->gb);
1713
1714     if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1715         if(h0->current_slice && FIELD_PICTURE){
1716             field_end(h);
1717         }
1718
1719         h0->current_slice = 0;
1720         if (!s0->first_field)
1721             s->current_picture_ptr= NULL;
1722     }
1723
1724     slice_type= get_ue_golomb_31(&s->gb);
1725     if(slice_type > 9){
1726         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);
1727         return -1;
1728     }
1729     if(slice_type > 4){
1730         slice_type -= 5;
1731         h->slice_type_fixed=1;
1732     }else
1733         h->slice_type_fixed=0;
1734
1735     slice_type= golomb_to_pict_type[ slice_type ];
1736     if (slice_type == FF_I_TYPE
1737         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
1738         default_ref_list_done = 1;
1739     }
1740     h->slice_type= slice_type;
1741     h->slice_type_nos= slice_type & 3;
1742
1743     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
1744
1745     pps_id= get_ue_golomb(&s->gb);
1746     if(pps_id>=MAX_PPS_COUNT){
1747         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
1748         return -1;
1749     }
1750     if(!h0->pps_buffers[pps_id]) {
1751         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
1752         return -1;
1753     }
1754     h->pps= *h0->pps_buffers[pps_id];
1755
1756     if(!h0->sps_buffers[h->pps.sps_id]) {
1757         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
1758         return -1;
1759     }
1760     h->sps = *h0->sps_buffers[h->pps.sps_id];
1761
1762     if(h == h0 && h->dequant_coeff_pps != pps_id){
1763         h->dequant_coeff_pps = pps_id;
1764         init_dequant_tables(h);
1765     }
1766
1767     s->mb_width= h->sps.mb_width;
1768     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1769
1770     h->b_stride=  s->mb_width*4;
1771     h->b8_stride= s->mb_width*2;
1772
1773     s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
1774     if(h->sps.frame_mbs_only_flag)
1775         s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
1776     else
1777         s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
1778
1779     if (s->context_initialized
1780         && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
1781         if(h != h0)
1782             return -1;   // width / height changed during parallelized decoding
1783         free_tables(h);
1784         flush_dpb(s->avctx);
1785         MPV_common_end(s);
1786     }
1787     if (!s->context_initialized) {
1788         if(h != h0)
1789             return -1;  // we cant (re-)initialize context during parallel decoding
1790
1791         avcodec_set_dimensions(s->avctx, s->width, s->height);
1792         s->avctx->sample_aspect_ratio= h->sps.sar;
1793         if(!s->avctx->sample_aspect_ratio.den)
1794             s->avctx->sample_aspect_ratio.den = 1;
1795
1796         if(h->sps.video_signal_type_present_flag){
1797             s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1798             if(h->sps.colour_description_present_flag){
1799                 s->avctx->color_primaries = h->sps.color_primaries;
1800                 s->avctx->color_trc       = h->sps.color_trc;
1801                 s->avctx->colorspace      = h->sps.colorspace;
1802             }
1803         }
1804
1805         if(h->sps.timing_info_present_flag){
1806             int64_t den= h->sps.time_scale;
1807             if(h->x264_build < 44U)
1808                 den *= 2;
1809             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
1810                       h->sps.num_units_in_tick, den, 1<<30);
1811         }
1812         s->avctx->pix_fmt = s->avctx->get_format(s->avctx, s->avctx->codec->pix_fmts);
1813         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1814
1815         if (MPV_common_init(s) < 0)
1816             return -1;
1817         s->first_field = 0;
1818         h->prev_interlaced_frame = 1;
1819
1820         init_scan_tables(h);
1821         ff_h264_alloc_tables(h);
1822
1823         for(i = 1; i < s->avctx->thread_count; i++) {
1824             H264Context *c;
1825             c = h->thread_context[i] = av_malloc(sizeof(H264Context));
1826             memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
1827             memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
1828             c->sps = h->sps;
1829             c->pps = h->pps;
1830             init_scan_tables(c);
1831             clone_tables(c, h);
1832         }
1833
1834         for(i = 0; i < s->avctx->thread_count; i++)
1835             if(context_init(h->thread_context[i]) < 0)
1836                 return -1;
1837     }
1838
1839     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1840
1841     h->mb_mbaff = 0;
1842     h->mb_aff_frame = 0;
1843     last_pic_structure = s0->picture_structure;
1844     if(h->sps.frame_mbs_only_flag){
1845         s->picture_structure= PICT_FRAME;
1846     }else{
1847         if(get_bits1(&s->gb)) { //field_pic_flag
1848             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
1849         } else {
1850             s->picture_structure= PICT_FRAME;
1851             h->mb_aff_frame = h->sps.mb_aff;
1852         }
1853     }
1854     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
1855
1856     if(h0->current_slice == 0){
1857         while(h->frame_num !=  h->prev_frame_num &&
1858               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
1859             av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
1860             if (ff_h264_frame_start(h) < 0)
1861                 return -1;
1862             h->prev_frame_num++;
1863             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
1864             s->current_picture_ptr->frame_num= h->prev_frame_num;
1865             ff_h264_execute_ref_pic_marking(h, NULL, 0);
1866         }
1867
1868         /* See if we have a decoded first field looking for a pair... */
1869         if (s0->first_field) {
1870             assert(s0->current_picture_ptr);
1871             assert(s0->current_picture_ptr->data[0]);
1872             assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
1873
1874             /* figure out if we have a complementary field pair */
1875             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
1876                 /*
1877                  * Previous field is unmatched. Don't display it, but let it
1878                  * remain for reference if marked as such.
1879                  */
1880                 s0->current_picture_ptr = NULL;
1881                 s0->first_field = FIELD_PICTURE;
1882
1883             } else {
1884                 if (h->nal_ref_idc &&
1885                         s0->current_picture_ptr->reference &&
1886                         s0->current_picture_ptr->frame_num != h->frame_num) {
1887                     /*
1888                      * This and previous field were reference, but had
1889                      * different frame_nums. Consider this field first in
1890                      * pair. Throw away previous field except for reference
1891                      * purposes.
1892                      */
1893                     s0->first_field = 1;
1894                     s0->current_picture_ptr = NULL;
1895
1896                 } else {
1897                     /* Second field in complementary pair */
1898                     s0->first_field = 0;
1899                 }
1900             }
1901
1902         } else {
1903             /* Frame or first field in a potentially complementary pair */
1904             assert(!s0->current_picture_ptr);
1905             s0->first_field = FIELD_PICTURE;
1906         }
1907
1908         if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
1909             s0->first_field = 0;
1910             return -1;
1911         }
1912     }
1913     if(h != h0)
1914         clone_slice(h, h0);
1915
1916     s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
1917
1918     assert(s->mb_num == s->mb_width * s->mb_height);
1919     if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
1920        first_mb_in_slice                    >= s->mb_num){
1921         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1922         return -1;
1923     }
1924     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
1925     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
1926     if (s->picture_structure == PICT_BOTTOM_FIELD)
1927         s->resync_mb_y = s->mb_y = s->mb_y + 1;
1928     assert(s->mb_y < s->mb_height);
1929
1930     if(s->picture_structure==PICT_FRAME){
1931         h->curr_pic_num=   h->frame_num;
1932         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
1933     }else{
1934         h->curr_pic_num= 2*h->frame_num + 1;
1935         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
1936     }
1937
1938     if(h->nal_unit_type == NAL_IDR_SLICE){
1939         get_ue_golomb(&s->gb); /* idr_pic_id */
1940     }
1941
1942     if(h->sps.poc_type==0){
1943         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
1944
1945         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
1946             h->delta_poc_bottom= get_se_golomb(&s->gb);
1947         }
1948     }
1949
1950     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
1951         h->delta_poc[0]= get_se_golomb(&s->gb);
1952
1953         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
1954             h->delta_poc[1]= get_se_golomb(&s->gb);
1955     }
1956
1957     init_poc(h);
1958
1959     if(h->pps.redundant_pic_cnt_present){
1960         h->redundant_pic_count= get_ue_golomb(&s->gb);
1961     }
1962
1963     //set defaults, might be overridden a few lines later
1964     h->ref_count[0]= h->pps.ref_count[0];
1965     h->ref_count[1]= h->pps.ref_count[1];
1966
1967     if(h->slice_type_nos != FF_I_TYPE){
1968         if(h->slice_type_nos == FF_B_TYPE){
1969             h->direct_spatial_mv_pred= get_bits1(&s->gb);
1970         }
1971         num_ref_idx_active_override_flag= get_bits1(&s->gb);
1972
1973         if(num_ref_idx_active_override_flag){
1974             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
1975             if(h->slice_type_nos==FF_B_TYPE)
1976                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
1977
1978             if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
1979                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
1980                 h->ref_count[0]= h->ref_count[1]= 1;
1981                 return -1;
1982             }
1983         }
1984         if(h->slice_type_nos == FF_B_TYPE)
1985             h->list_count= 2;
1986         else
1987             h->list_count= 1;
1988     }else
1989         h->list_count= 0;
1990
1991     if(!default_ref_list_done){
1992         ff_h264_fill_default_ref_list(h);
1993     }
1994
1995     if(h->slice_type_nos!=FF_I_TYPE && ff_h264_decode_ref_pic_list_reordering(h) < 0)
1996         return -1;
1997
1998     if(h->slice_type_nos!=FF_I_TYPE){
1999         s->last_picture_ptr= &h->ref_list[0][0];
2000         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
2001     }
2002     if(h->slice_type_nos==FF_B_TYPE){
2003         s->next_picture_ptr= &h->ref_list[1][0];
2004         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
2005     }
2006
2007     if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
2008        ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
2009         pred_weight_table(h);
2010     else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE)
2011         implicit_weight_table(h);
2012     else {
2013         h->use_weight = 0;
2014         for (i = 0; i < 2; i++) {
2015             h->luma_weight_flag[i]   = 0;
2016             h->chroma_weight_flag[i] = 0;
2017         }
2018     }
2019
2020     if(h->nal_ref_idc)
2021         ff_h264_decode_ref_pic_marking(h0, &s->gb);
2022
2023     if(FRAME_MBAFF)
2024         ff_h264_fill_mbaff_ref_list(h);
2025
2026     if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
2027         ff_h264_direct_dist_scale_factor(h);
2028     ff_h264_direct_ref_list_init(h);
2029
2030     if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
2031         tmp = get_ue_golomb_31(&s->gb);
2032         if(tmp > 2){
2033             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2034             return -1;
2035         }
2036         h->cabac_init_idc= tmp;
2037     }
2038
2039     h->last_qscale_diff = 0;
2040     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2041     if(tmp>51){
2042         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2043         return -1;
2044     }
2045     s->qscale= tmp;
2046     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2047     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2048     //FIXME qscale / qp ... stuff
2049     if(h->slice_type == FF_SP_TYPE){
2050         get_bits1(&s->gb); /* sp_for_switch_flag */
2051     }
2052     if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
2053         get_se_golomb(&s->gb); /* slice_qs_delta */
2054     }
2055
2056     h->deblocking_filter = 1;
2057     h->slice_alpha_c0_offset = 52;
2058     h->slice_beta_offset = 52;
2059     if( h->pps.deblocking_filter_parameters_present ) {
2060         tmp= get_ue_golomb_31(&s->gb);
2061         if(tmp > 2){
2062             av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2063             return -1;
2064         }
2065         h->deblocking_filter= tmp;
2066         if(h->deblocking_filter < 2)
2067             h->deblocking_filter^= 1; // 1<->0
2068
2069         if( h->deblocking_filter ) {
2070             h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2071             h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
2072             if(   h->slice_alpha_c0_offset > 104U
2073                || h->slice_beta_offset     > 104U){
2074                 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);
2075                 return -1;
2076             }
2077         }
2078     }
2079
2080     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
2081        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
2082        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
2083        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2084         h->deblocking_filter= 0;
2085
2086     if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2087         if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2088             /* Cheat slightly for speed:
2089                Do not bother to deblock across slices. */
2090             h->deblocking_filter = 2;
2091         } else {
2092             h0->max_contexts = 1;
2093             if(!h0->single_decode_warning) {
2094                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2095                 h0->single_decode_warning = 1;
2096             }
2097             if(h != h0)
2098                 return 1; // deblocking switched inside frame
2099         }
2100     }
2101     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]);
2102
2103 #if 0 //FMO
2104     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2105         slice_group_change_cycle= get_bits(&s->gb, ?);
2106 #endif
2107
2108     h0->last_slice_type = slice_type;
2109     h->slice_num = ++h0->current_slice;
2110     if(h->slice_num >= MAX_SLICES){
2111         av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2112     }
2113
2114     for(j=0; j<2; j++){
2115         int id_list[16];
2116         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2117         for(i=0; i<16; i++){
2118             id_list[i]= 60;
2119             if(h->ref_list[j][i].data[0]){
2120                 int k;
2121                 uint8_t *base= h->ref_list[j][i].base[0];
2122                 for(k=0; k<h->short_ref_count; k++)
2123                     if(h->short_ref[k]->base[0] == base){
2124                         id_list[i]= k;
2125                         break;
2126                     }
2127                 for(k=0; k<h->long_ref_count; k++)
2128                     if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2129                         id_list[i]= h->short_ref_count + k;
2130                         break;
2131                     }
2132             }
2133         }
2134
2135         ref2frm[0]=
2136         ref2frm[1]= -1;
2137         for(i=0; i<16; i++)
2138             ref2frm[i+2]= 4*id_list[i]
2139                           +(h->ref_list[j][i].reference&3);
2140         ref2frm[18+0]=
2141         ref2frm[18+1]= -1;
2142         for(i=16; i<48; i++)
2143             ref2frm[i+4]= 4*id_list[(i-16)>>1]
2144                           +(h->ref_list[j][i].reference&3);
2145     }
2146
2147     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
2148     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2149
2150     s->avctx->refs= h->sps.ref_frame_count;
2151
2152     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2153         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",
2154                h->slice_num,
2155                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2156                first_mb_in_slice,
2157                av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2158                pps_id, h->frame_num,
2159                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2160                h->ref_count[0], h->ref_count[1],
2161                s->qscale,
2162                h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
2163                h->use_weight,
2164                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2165                h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2166                );
2167     }
2168
2169     return 0;
2170 }
2171
2172 int ff_h264_get_slice_type(const H264Context *h)
2173 {
2174     switch (h->slice_type) {
2175     case FF_P_TYPE:  return 0;
2176     case FF_B_TYPE:  return 1;
2177     case FF_I_TYPE:  return 2;
2178     case FF_SP_TYPE: return 3;
2179     case FF_SI_TYPE: return 4;
2180     default:         return -1;
2181     }
2182 }
2183
2184 static void loop_filter(H264Context *h){
2185     MpegEncContext * const s = &h->s;
2186     uint8_t  *dest_y, *dest_cb, *dest_cr;
2187     int linesize, uvlinesize, mb_x, mb_y;
2188     const int end_mb_y= s->mb_y + FRAME_MBAFF;
2189     const int old_slice_type= h->slice_type;
2190
2191     if(h->deblocking_filter) {
2192         for(mb_x= 0; mb_x<s->mb_width; mb_x++){
2193             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
2194                 int mb_xy, mb_type;
2195                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2196                 h->slice_num= h->slice_table[mb_xy];
2197                 mb_type= s->current_picture.mb_type[mb_xy];
2198                 h->list_count= h->list_counts[mb_xy];
2199
2200                 if(FRAME_MBAFF)
2201                     h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2202
2203                 s->mb_x= mb_x;
2204                 s->mb_y= mb_y;
2205                 dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2206                 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2207                 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2208                     //FIXME simplify above
2209
2210                 if (MB_FIELD) {
2211                     linesize   = h->mb_linesize   = s->linesize * 2;
2212                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2213                     if(mb_y&1){ //FIXME move out of this function?
2214                         dest_y -= s->linesize*15;
2215                         dest_cb-= s->uvlinesize*7;
2216                         dest_cr-= s->uvlinesize*7;
2217                     }
2218                 } else {
2219                     linesize   = h->mb_linesize   = s->linesize;
2220                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2221                 }
2222                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2223                 if(fill_filter_caches(h, mb_type))
2224                     continue;
2225                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2226                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2227
2228                 if (FRAME_MBAFF) {
2229                     ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2230                 } else {
2231                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2232                 }
2233             }
2234         }
2235     }
2236     h->slice_type= old_slice_type;
2237     s->mb_x= 0;
2238     s->mb_y= end_mb_y - FRAME_MBAFF;
2239     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2240     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2241 }
2242
2243 static void predict_field_decoding_flag(H264Context *h){
2244     MpegEncContext * const s = &h->s;
2245     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2246     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
2247                 ? s->current_picture.mb_type[mb_xy-1]
2248                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
2249                 ? s->current_picture.mb_type[mb_xy-s->mb_stride]
2250                 : 0;
2251     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2252 }
2253
2254 static int decode_slice(struct AVCodecContext *avctx, void *arg){
2255     H264Context *h = *(void**)arg;
2256     MpegEncContext * const s = &h->s;
2257     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2258
2259     s->mb_skip_run= -1;
2260
2261     h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
2262                     (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
2263
2264     if( h->pps.cabac ) {
2265         /* realign */
2266         align_get_bits( &s->gb );
2267
2268         /* init cabac */
2269         ff_init_cabac_states( &h->cabac);
2270         ff_init_cabac_decoder( &h->cabac,
2271                                s->gb.buffer + get_bits_count(&s->gb)/8,
2272                                (get_bits_left(&s->gb) + 7)/8);
2273
2274         ff_h264_init_cabac_states(h);
2275
2276         for(;;){
2277 //START_TIMER
2278             int ret = ff_h264_decode_mb_cabac(h);
2279             int eos;
2280 //STOP_TIMER("decode_mb_cabac")
2281
2282             if(ret>=0) ff_h264_hl_decode_mb(h);
2283
2284             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
2285                 s->mb_y++;
2286
2287                 ret = ff_h264_decode_mb_cabac(h);
2288
2289                 if(ret>=0) ff_h264_hl_decode_mb(h);
2290                 s->mb_y--;
2291             }
2292             eos = get_cabac_terminate( &h->cabac );
2293
2294             if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
2295                 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);
2296                 return 0;
2297             }
2298             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2299                 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);
2300                 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);
2301                 return -1;
2302             }
2303
2304             if( ++s->mb_x >= s->mb_width ) {
2305                 s->mb_x = 0;
2306                 loop_filter(h);
2307                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2308                 ++s->mb_y;
2309                 if(FIELD_OR_MBAFF_PICTURE) {
2310                     ++s->mb_y;
2311                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2312                         predict_field_decoding_flag(h);
2313                 }
2314             }
2315
2316             if( eos || s->mb_y >= s->mb_height ) {
2317                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2318                 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);
2319                 return 0;
2320             }
2321         }
2322
2323     } else {
2324         for(;;){
2325             int ret = ff_h264_decode_mb_cavlc(h);
2326
2327             if(ret>=0) ff_h264_hl_decode_mb(h);
2328
2329             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
2330                 s->mb_y++;
2331                 ret = ff_h264_decode_mb_cavlc(h);
2332
2333                 if(ret>=0) ff_h264_hl_decode_mb(h);
2334                 s->mb_y--;
2335             }
2336
2337             if(ret<0){
2338                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2339                 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);
2340
2341                 return -1;
2342             }
2343
2344             if(++s->mb_x >= s->mb_width){
2345                 s->mb_x=0;
2346                 loop_filter(h);
2347                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2348                 ++s->mb_y;
2349                 if(FIELD_OR_MBAFF_PICTURE) {
2350                     ++s->mb_y;
2351                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2352                         predict_field_decoding_flag(h);
2353                 }
2354                 if(s->mb_y >= s->mb_height){
2355                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2356
2357                     if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
2358                         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);
2359
2360                         return 0;
2361                     }else{
2362                         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);
2363
2364                         return -1;
2365                     }
2366                 }
2367             }
2368
2369             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
2370                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2371                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2372                     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);
2373
2374                     return 0;
2375                 }else{
2376                     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);
2377
2378                     return -1;
2379                 }
2380             }
2381         }
2382     }
2383
2384 #if 0
2385     for(;s->mb_y < s->mb_height; s->mb_y++){
2386         for(;s->mb_x < s->mb_width; s->mb_x++){
2387             int ret= decode_mb(h);
2388
2389             ff_h264_hl_decode_mb(h);
2390
2391             if(ret<0){
2392                 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2393                 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);
2394
2395                 return -1;
2396             }
2397
2398             if(++s->mb_x >= s->mb_width){
2399                 s->mb_x=0;
2400                 if(++s->mb_y >= s->mb_height){
2401                     if(get_bits_count(s->gb) == s->gb.size_in_bits){
2402                         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);
2403
2404                         return 0;
2405                     }else{
2406                         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);
2407
2408                         return -1;
2409                     }
2410                 }
2411             }
2412
2413             if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2414                 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2415                     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);
2416
2417                     return 0;
2418                 }else{
2419                     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);
2420
2421                     return -1;
2422                 }
2423             }
2424         }
2425         s->mb_x=0;
2426         ff_draw_horiz_band(s, 16*s->mb_y, 16);
2427     }
2428 #endif
2429     return -1; //not reached
2430 }
2431
2432 /**
2433  * Call decode_slice() for each context.
2434  *
2435  * @param h h264 master context
2436  * @param context_count number of contexts to execute
2437  */
2438 static void execute_decode_slices(H264Context *h, int context_count){
2439     MpegEncContext * const s = &h->s;
2440     AVCodecContext * const avctx= s->avctx;
2441     H264Context *hx;
2442     int i;
2443
2444     if (s->avctx->hwaccel)
2445         return;
2446     if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2447         return;
2448     if(context_count == 1) {
2449         decode_slice(avctx, &h);
2450     } else {
2451         for(i = 1; i < context_count; i++) {
2452             hx = h->thread_context[i];
2453             hx->s.error_recognition = avctx->error_recognition;
2454             hx->s.error_count = 0;
2455         }
2456
2457         avctx->execute(avctx, (void *)decode_slice,
2458                        h->thread_context, NULL, context_count, sizeof(void*));
2459
2460         /* pull back stuff from slices to master context */
2461         hx = h->thread_context[context_count - 1];
2462         s->mb_x = hx->s.mb_x;
2463         s->mb_y = hx->s.mb_y;
2464         s->dropable = hx->s.dropable;
2465         s->picture_structure = hx->s.picture_structure;
2466         for(i = 1; i < context_count; i++)
2467             h->s.error_count += h->thread_context[i]->s.error_count;
2468     }
2469 }
2470
2471
2472 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
2473     MpegEncContext * const s = &h->s;
2474     AVCodecContext * const avctx= s->avctx;
2475     int buf_index=0;
2476     H264Context *hx; ///< thread context
2477     int context_count = 0;
2478     int next_avc= h->is_avc ? 0 : buf_size;
2479
2480     h->max_contexts = avctx->thread_count;
2481 #if 0
2482     int i;
2483     for(i=0; i<50; i++){
2484         av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
2485     }
2486 #endif
2487     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
2488         h->current_slice = 0;
2489         if (!s->first_field)
2490             s->current_picture_ptr= NULL;
2491         ff_h264_reset_sei(h);
2492     }
2493
2494     for(;;){
2495         int consumed;
2496         int dst_length;
2497         int bit_length;
2498         const uint8_t *ptr;
2499         int i, nalsize = 0;
2500         int err;
2501
2502         if(buf_index >= next_avc) {
2503             if(buf_index >= buf_size) break;
2504             nalsize = 0;
2505             for(i = 0; i < h->nal_length_size; i++)
2506                 nalsize = (nalsize << 8) | buf[buf_index++];
2507             if(nalsize <= 1 || nalsize > buf_size - buf_index){
2508                 if(nalsize == 1){
2509                     buf_index++;
2510                     continue;
2511                 }else{
2512                     av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2513                     break;
2514                 }
2515             }
2516             next_avc= buf_index + nalsize;
2517         } else {
2518             // start code prefix search
2519             for(; buf_index + 3 < next_avc; buf_index++){
2520                 // This should always succeed in the first iteration.
2521                 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2522                     break;
2523             }
2524
2525             if(buf_index+3 >= buf_size) break;
2526
2527             buf_index+=3;
2528             if(buf_index >= next_avc) continue;
2529         }
2530
2531         hx = h->thread_context[context_count];
2532
2533         ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
2534         if (ptr==NULL || dst_length < 0){
2535             return -1;
2536         }
2537         i= buf_index + consumed;
2538         if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
2539            buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
2540             s->workaround_bugs |= FF_BUG_TRUNCATED;
2541
2542         if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
2543         while(ptr[dst_length - 1] == 0 && dst_length > 0)
2544             dst_length--;
2545         }
2546         bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
2547
2548         if(s->avctx->debug&FF_DEBUG_STARTCODE){
2549             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);
2550         }
2551
2552         if (h->is_avc && (nalsize != consumed) && nalsize){
2553             av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
2554         }
2555
2556         buf_index += consumed;
2557
2558         if(  (s->hurry_up == 1 && h->nal_ref_idc  == 0) //FIXME do not discard SEI id
2559            ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0))
2560             continue;
2561
2562       again:
2563         err = 0;
2564         switch(hx->nal_unit_type){
2565         case NAL_IDR_SLICE:
2566             if (h->nal_unit_type != NAL_IDR_SLICE) {
2567                 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2568                 return -1;
2569             }
2570             idr(h); //FIXME ensure we don't loose some frames if there is reordering
2571         case NAL_SLICE:
2572             init_get_bits(&hx->s.gb, ptr, bit_length);
2573             hx->intra_gb_ptr=
2574             hx->inter_gb_ptr= &hx->s.gb;
2575             hx->s.data_partitioning = 0;
2576
2577             if((err = decode_slice_header(hx, h)))
2578                break;
2579
2580             avctx->profile = hx->sps.profile_idc;
2581             avctx->level   = hx->sps.level_idc;
2582
2583             if (s->avctx->hwaccel && h->current_slice == 1) {
2584                 if (s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
2585                     return -1;
2586             }
2587
2588             s->current_picture_ptr->key_frame |=
2589                     (hx->nal_unit_type == NAL_IDR_SLICE) ||
2590                     (h->sei_recovery_frame_cnt >= 0);
2591             if(hx->redundant_pic_count==0 && hx->s.hurry_up < 5
2592                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2593                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2594                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2595                && avctx->skip_frame < AVDISCARD_ALL){
2596                 if(avctx->hwaccel) {
2597                     if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
2598                         return -1;
2599                 }else
2600                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
2601                     static const uint8_t start_code[] = {0x00, 0x00, 0x01};
2602                     ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
2603                     ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
2604                 }else
2605                     context_count++;
2606             }
2607             break;
2608         case NAL_DPA:
2609             init_get_bits(&hx->s.gb, ptr, bit_length);
2610             hx->intra_gb_ptr=
2611             hx->inter_gb_ptr= NULL;
2612
2613             if ((err = decode_slice_header(hx, h)) < 0)
2614                 break;
2615
2616             avctx->profile = hx->sps.profile_idc;
2617             avctx->level   = hx->sps.level_idc;
2618
2619             hx->s.data_partitioning = 1;
2620
2621             break;
2622         case NAL_DPB:
2623             init_get_bits(&hx->intra_gb, ptr, bit_length);
2624             hx->intra_gb_ptr= &hx->intra_gb;
2625             break;
2626         case NAL_DPC:
2627             init_get_bits(&hx->inter_gb, ptr, bit_length);
2628             hx->inter_gb_ptr= &hx->inter_gb;
2629
2630             if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
2631                && s->context_initialized
2632                && s->hurry_up < 5
2633                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2634                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2635                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2636                && avctx->skip_frame < AVDISCARD_ALL)
2637                 context_count++;
2638             break;
2639         case NAL_SEI:
2640             init_get_bits(&s->gb, ptr, bit_length);
2641             ff_h264_decode_sei(h);
2642             break;
2643         case NAL_SPS:
2644             init_get_bits(&s->gb, ptr, bit_length);
2645             ff_h264_decode_seq_parameter_set(h);
2646
2647             if(s->flags& CODEC_FLAG_LOW_DELAY)
2648                 s->low_delay=1;
2649
2650             if(avctx->has_b_frames < 2)
2651                 avctx->has_b_frames= !s->low_delay;
2652             break;
2653         case NAL_PPS:
2654             init_get_bits(&s->gb, ptr, bit_length);
2655
2656             ff_h264_decode_picture_parameter_set(h, bit_length);
2657
2658             break;
2659         case NAL_AUD:
2660         case NAL_END_SEQUENCE:
2661         case NAL_END_STREAM:
2662         case NAL_FILLER_DATA:
2663         case NAL_SPS_EXT:
2664         case NAL_AUXILIARY_SLICE:
2665             break;
2666         default:
2667             av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
2668         }
2669
2670         if(context_count == h->max_contexts) {
2671             execute_decode_slices(h, context_count);
2672             context_count = 0;
2673         }
2674
2675         if (err < 0)
2676             av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
2677         else if(err == 1) {
2678             /* Slice could not be decoded in parallel mode, copy down
2679              * NAL unit stuff to context 0 and restart. Note that
2680              * rbsp_buffer is not transferred, but since we no longer
2681              * run in parallel mode this should not be an issue. */
2682             h->nal_unit_type = hx->nal_unit_type;
2683             h->nal_ref_idc   = hx->nal_ref_idc;
2684             hx = h;
2685             goto again;
2686         }
2687     }
2688     if(context_count)
2689         execute_decode_slices(h, context_count);
2690     return buf_index;
2691 }
2692
2693 /**
2694  * returns the number of bytes consumed for building the current frame
2695  */
2696 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
2697         if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
2698         if(pos+10>buf_size) pos=buf_size; // oops ;)
2699
2700         return pos;
2701 }
2702
2703 static int decode_frame(AVCodecContext *avctx,
2704                              void *data, int *data_size,
2705                              AVPacket *avpkt)
2706 {
2707     const uint8_t *buf = avpkt->data;
2708     int buf_size = avpkt->size;
2709     H264Context *h = avctx->priv_data;
2710     MpegEncContext *s = &h->s;
2711     AVFrame *pict = data;
2712     int buf_index;
2713
2714     s->flags= avctx->flags;
2715     s->flags2= avctx->flags2;
2716
2717    /* end of stream, output what is still in the buffers */
2718     if (buf_size == 0) {
2719         Picture *out;
2720         int i, out_idx;
2721
2722 //FIXME factorize this with the output code below
2723         out = h->delayed_pic[0];
2724         out_idx = 0;
2725         for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2726             if(h->delayed_pic[i]->poc < out->poc){
2727                 out = h->delayed_pic[i];
2728                 out_idx = i;
2729             }
2730
2731         for(i=out_idx; h->delayed_pic[i]; i++)
2732             h->delayed_pic[i] = h->delayed_pic[i+1];
2733
2734         if(out){
2735             *data_size = sizeof(AVFrame);
2736             *pict= *(AVFrame*)out;
2737         }
2738
2739         return 0;
2740     }
2741
2742     buf_index=decode_nal_units(h, buf, buf_size);
2743     if(buf_index < 0)
2744         return -1;
2745
2746     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
2747         if (avctx->skip_frame >= AVDISCARD_NONREF || s->hurry_up) return 0;
2748         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
2749         return -1;
2750     }
2751
2752     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
2753         Picture *out = s->current_picture_ptr;
2754         Picture *cur = s->current_picture_ptr;
2755         int i, pics, out_of_order, out_idx;
2756
2757         field_end(h);
2758
2759         if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
2760             /* Wait for second field. */
2761             *data_size = 0;
2762
2763         } else {
2764             cur->interlaced_frame = 0;
2765             cur->repeat_pict = 0;
2766
2767             /* Signal interlacing information externally. */
2768             /* Prioritize picture timing SEI information over used decoding process if it exists. */
2769
2770             if(h->sps.pic_struct_present_flag){
2771                 switch (h->sei_pic_struct)
2772                 {
2773                 case SEI_PIC_STRUCT_FRAME:
2774                     break;
2775                 case SEI_PIC_STRUCT_TOP_FIELD:
2776                 case SEI_PIC_STRUCT_BOTTOM_FIELD:
2777                     cur->interlaced_frame = 1;
2778                     break;
2779                 case SEI_PIC_STRUCT_TOP_BOTTOM:
2780                 case SEI_PIC_STRUCT_BOTTOM_TOP:
2781                     if (FIELD_OR_MBAFF_PICTURE)
2782                         cur->interlaced_frame = 1;
2783                     else
2784                         // try to flag soft telecine progressive
2785                         cur->interlaced_frame = h->prev_interlaced_frame;
2786                     break;
2787                 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
2788                 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
2789                     // Signal the possibility of telecined film externally (pic_struct 5,6)
2790                     // From these hints, let the applications decide if they apply deinterlacing.
2791                     cur->repeat_pict = 1;
2792                     break;
2793                 case SEI_PIC_STRUCT_FRAME_DOUBLING:
2794                     // Force progressive here, as doubling interlaced frame is a bad idea.
2795                     cur->repeat_pict = 2;
2796                     break;
2797                 case SEI_PIC_STRUCT_FRAME_TRIPLING:
2798                     cur->repeat_pict = 4;
2799                     break;
2800                 }
2801
2802                 if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
2803                     cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
2804             }else{
2805                 /* Derive interlacing flag from used decoding process. */
2806                 cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
2807             }
2808             h->prev_interlaced_frame = cur->interlaced_frame;
2809
2810             if (cur->field_poc[0] != cur->field_poc[1]){
2811                 /* Derive top_field_first from field pocs. */
2812                 cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
2813             }else{
2814                 if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
2815                     /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
2816                     if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
2817                       || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
2818                         cur->top_field_first = 1;
2819                     else
2820                         cur->top_field_first = 0;
2821                 }else{
2822                     /* Most likely progressive */
2823                     cur->top_field_first = 0;
2824                 }
2825             }
2826
2827         //FIXME do something with unavailable reference frames
2828
2829             /* Sort B-frames into display order */
2830
2831             if(h->sps.bitstream_restriction_flag
2832                && s->avctx->has_b_frames < h->sps.num_reorder_frames){
2833                 s->avctx->has_b_frames = h->sps.num_reorder_frames;
2834                 s->low_delay = 0;
2835             }
2836
2837             if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
2838                && !h->sps.bitstream_restriction_flag){
2839                 s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
2840                 s->low_delay= 0;
2841             }
2842
2843             pics = 0;
2844             while(h->delayed_pic[pics]) pics++;
2845
2846             assert(pics <= MAX_DELAYED_PIC_COUNT);
2847
2848             h->delayed_pic[pics++] = cur;
2849             if(cur->reference == 0)
2850                 cur->reference = DELAYED_PIC_REF;
2851
2852             out = h->delayed_pic[0];
2853             out_idx = 0;
2854             for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2855                 if(h->delayed_pic[i]->poc < out->poc){
2856                     out = h->delayed_pic[i];
2857                     out_idx = i;
2858                 }
2859             if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
2860                 h->outputed_poc= INT_MIN;
2861             out_of_order = out->poc < h->outputed_poc;
2862
2863             if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
2864                 { }
2865             else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
2866                || (s->low_delay &&
2867                 ((h->outputed_poc != INT_MIN && out->poc > h->outputed_poc + 2)
2868                  || cur->pict_type == FF_B_TYPE)))
2869             {
2870                 s->low_delay = 0;
2871                 s->avctx->has_b_frames++;
2872             }
2873
2874             if(out_of_order || pics > s->avctx->has_b_frames){
2875                 out->reference &= ~DELAYED_PIC_REF;
2876                 for(i=out_idx; h->delayed_pic[i]; i++)
2877                     h->delayed_pic[i] = h->delayed_pic[i+1];
2878             }
2879             if(!out_of_order && pics > s->avctx->has_b_frames){
2880                 *data_size = sizeof(AVFrame);
2881
2882                 if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
2883                     h->outputed_poc = INT_MIN;
2884                 } else
2885                     h->outputed_poc = out->poc;
2886                 *pict= *(AVFrame*)out;
2887             }else{
2888                 av_log(avctx, AV_LOG_DEBUG, "no picture\n");
2889             }
2890         }
2891     }
2892
2893     assert(pict->data[0] || !*data_size);
2894     ff_print_debug_info(s, pict);
2895 //printf("out %d\n", (int)pict->data[0]);
2896
2897     return get_consumed_bytes(s, buf_index, buf_size);
2898 }
2899 #if 0
2900 static inline void fill_mb_avail(H264Context *h){
2901     MpegEncContext * const s = &h->s;
2902     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2903
2904     if(s->mb_y){
2905         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
2906         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
2907         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
2908     }else{
2909         h->mb_avail[0]=
2910         h->mb_avail[1]=
2911         h->mb_avail[2]= 0;
2912     }
2913     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
2914     h->mb_avail[4]= 1; //FIXME move out
2915     h->mb_avail[5]= 0; //FIXME move out
2916 }
2917 #endif
2918
2919 #ifdef TEST
2920 #undef printf
2921 #undef random
2922 #define COUNT 8000
2923 #define SIZE (COUNT*40)
2924 int main(void){
2925     int i;
2926     uint8_t temp[SIZE];
2927     PutBitContext pb;
2928     GetBitContext gb;
2929 //    int int_temp[10000];
2930     DSPContext dsp;
2931     AVCodecContext avctx;
2932
2933     dsputil_init(&dsp, &avctx);
2934
2935     init_put_bits(&pb, temp, SIZE);
2936     printf("testing unsigned exp golomb\n");
2937     for(i=0; i<COUNT; i++){
2938         START_TIMER
2939         set_ue_golomb(&pb, i);
2940         STOP_TIMER("set_ue_golomb");
2941     }
2942     flush_put_bits(&pb);
2943
2944     init_get_bits(&gb, temp, 8*SIZE);
2945     for(i=0; i<COUNT; i++){
2946         int j, s;
2947
2948         s= show_bits(&gb, 24);
2949
2950         START_TIMER
2951         j= get_ue_golomb(&gb);
2952         if(j != i){
2953             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
2954 //            return -1;
2955         }
2956         STOP_TIMER("get_ue_golomb");
2957     }
2958
2959
2960     init_put_bits(&pb, temp, SIZE);
2961     printf("testing signed exp golomb\n");
2962     for(i=0; i<COUNT; i++){
2963         START_TIMER
2964         set_se_golomb(&pb, i - COUNT/2);
2965         STOP_TIMER("set_se_golomb");
2966     }
2967     flush_put_bits(&pb);
2968
2969     init_get_bits(&gb, temp, 8*SIZE);
2970     for(i=0; i<COUNT; i++){
2971         int j, s;
2972
2973         s= show_bits(&gb, 24);
2974
2975         START_TIMER
2976         j= get_se_golomb(&gb);
2977         if(j != i - COUNT/2){
2978             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
2979 //            return -1;
2980         }
2981         STOP_TIMER("get_se_golomb");
2982     }
2983
2984 #if 0
2985     printf("testing 4x4 (I)DCT\n");
2986
2987     DCTELEM block[16];
2988     uint8_t src[16], ref[16];
2989     uint64_t error= 0, max_error=0;
2990
2991     for(i=0; i<COUNT; i++){
2992         int j;
2993 //        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
2994         for(j=0; j<16; j++){
2995             ref[j]= random()%255;
2996             src[j]= random()%255;
2997         }
2998
2999         h264_diff_dct_c(block, src, ref, 4);
3000
3001         //normalize
3002         for(j=0; j<16; j++){
3003 //            printf("%d ", block[j]);
3004             block[j]= block[j]*4;
3005             if(j&1) block[j]= (block[j]*4 + 2)/5;
3006             if(j&4) block[j]= (block[j]*4 + 2)/5;
3007         }
3008 //        printf("\n");
3009
3010         s->dsp.h264_idct_add(ref, block, 4);
3011 /*        for(j=0; j<16; j++){
3012             printf("%d ", ref[j]);
3013         }
3014         printf("\n");*/
3015
3016         for(j=0; j<16; j++){
3017             int diff= FFABS(src[j] - ref[j]);
3018
3019             error+= diff*diff;
3020             max_error= FFMAX(max_error, diff);
3021         }
3022     }
3023     printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
3024     printf("testing quantizer\n");
3025     for(qp=0; qp<52; qp++){
3026         for(i=0; i<16; i++)
3027             src1_block[i]= src2_block[i]= random()%255;
3028
3029     }
3030     printf("Testing NAL layer\n");
3031
3032     uint8_t bitstream[COUNT];
3033     uint8_t nal[COUNT*2];
3034     H264Context h;
3035     memset(&h, 0, sizeof(H264Context));
3036
3037     for(i=0; i<COUNT; i++){
3038         int zeros= i;
3039         int nal_length;
3040         int consumed;
3041         int out_length;
3042         uint8_t *out;
3043         int j;
3044
3045         for(j=0; j<COUNT; j++){
3046             bitstream[j]= (random() % 255) + 1;
3047         }
3048
3049         for(j=0; j<zeros; j++){
3050             int pos= random() % COUNT;
3051             while(bitstream[pos] == 0){
3052                 pos++;
3053                 pos %= COUNT;
3054             }
3055             bitstream[pos]=0;
3056         }
3057
3058         START_TIMER
3059
3060         nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
3061         if(nal_length<0){
3062             printf("encoding failed\n");
3063             return -1;
3064         }
3065
3066         out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
3067
3068         STOP_TIMER("NAL")
3069
3070         if(out_length != COUNT){
3071             printf("incorrect length %d %d\n", out_length, COUNT);
3072             return -1;
3073         }
3074
3075         if(consumed != nal_length){
3076             printf("incorrect consumed length %d %d\n", nal_length, consumed);
3077             return -1;
3078         }
3079
3080         if(memcmp(bitstream, out, COUNT)){
3081             printf("mismatch\n");
3082             return -1;
3083         }
3084     }
3085 #endif
3086
3087     printf("Testing RBSP\n");
3088
3089
3090     return 0;
3091 }
3092 #endif /* TEST */
3093
3094
3095 av_cold void ff_h264_free_context(H264Context *h)
3096 {
3097     int i;
3098
3099     free_tables(h); //FIXME cleanup init stuff perhaps
3100
3101     for(i = 0; i < MAX_SPS_COUNT; i++)
3102         av_freep(h->sps_buffers + i);
3103
3104     for(i = 0; i < MAX_PPS_COUNT; i++)
3105         av_freep(h->pps_buffers + i);
3106 }
3107
3108 av_cold int ff_h264_decode_end(AVCodecContext *avctx)
3109 {
3110     H264Context *h = avctx->priv_data;
3111     MpegEncContext *s = &h->s;
3112
3113     ff_h264_free_context(h);
3114
3115     MPV_common_end(s);
3116
3117 //    memset(h, 0, sizeof(H264Context));
3118
3119     return 0;
3120 }
3121
3122
3123 AVCodec h264_decoder = {
3124     "h264",
3125     CODEC_TYPE_VIDEO,
3126     CODEC_ID_H264,
3127     sizeof(H264Context),
3128     ff_h264_decode_init,
3129     NULL,
3130     ff_h264_decode_end,
3131     decode_frame,
3132     /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3133     .flush= flush_dpb,
3134     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
3135     .pix_fmts= ff_hwaccel_pixfmt_list_420,
3136 };
3137
3138 #if CONFIG_H264_VDPAU_DECODER
3139 AVCodec h264_vdpau_decoder = {
3140     "h264_vdpau",
3141     CODEC_TYPE_VIDEO,
3142     CODEC_ID_H264,
3143     sizeof(H264Context),
3144     ff_h264_decode_init,
3145     NULL,
3146     ff_h264_decode_end,
3147     decode_frame,
3148     CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3149     .flush= flush_dpb,
3150     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
3151     .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
3152 };
3153 #endif