5185e31ebbda58f4c9d583985b4869ddc6640e4a
[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
665     for(i = 0; i < MAX_THREADS; i++) {
666         hx = h->thread_context[i];
667         if(!hx) continue;
668         av_freep(&hx->top_borders[1]);
669         av_freep(&hx->top_borders[0]);
670         av_freep(&hx->s.obmc_scratchpad);
671         av_freep(&hx->rbsp_buffer[1]);
672         av_freep(&hx->rbsp_buffer[0]);
673         hx->rbsp_buffer_size[0] = 0;
674         hx->rbsp_buffer_size[1] = 0;
675         if (i) av_freep(&h->thread_context[i]);
676     }
677 }
678
679 static void init_dequant8_coeff_table(H264Context *h){
680     int i,q,x;
681     const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
682     h->dequant8_coeff[0] = h->dequant8_buffer[0];
683     h->dequant8_coeff[1] = h->dequant8_buffer[1];
684
685     for(i=0; i<2; i++ ){
686         if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
687             h->dequant8_coeff[1] = h->dequant8_buffer[0];
688             break;
689         }
690
691         for(q=0; q<52; q++){
692             int shift = div6[q];
693             int idx = rem6[q];
694             for(x=0; x<64; x++)
695                 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
696                     ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
697                     h->pps.scaling_matrix8[i][x]) << shift;
698         }
699     }
700 }
701
702 static void init_dequant4_coeff_table(H264Context *h){
703     int i,j,q,x;
704     const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
705     for(i=0; i<6; i++ ){
706         h->dequant4_coeff[i] = h->dequant4_buffer[i];
707         for(j=0; j<i; j++){
708             if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
709                 h->dequant4_coeff[i] = h->dequant4_buffer[j];
710                 break;
711             }
712         }
713         if(j<i)
714             continue;
715
716         for(q=0; q<52; q++){
717             int shift = div6[q] + 2;
718             int idx = rem6[q];
719             for(x=0; x<16; x++)
720                 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
721                     ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
722                     h->pps.scaling_matrix4[i][x]) << shift;
723         }
724     }
725 }
726
727 static void init_dequant_tables(H264Context *h){
728     int i,x;
729     init_dequant4_coeff_table(h);
730     if(h->pps.transform_8x8_mode)
731         init_dequant8_coeff_table(h);
732     if(h->sps.transform_bypass){
733         for(i=0; i<6; i++)
734             for(x=0; x<16; x++)
735                 h->dequant4_coeff[i][0][x] = 1<<6;
736         if(h->pps.transform_8x8_mode)
737             for(i=0; i<2; i++)
738                 for(x=0; x<64; x++)
739                     h->dequant8_coeff[i][0][x] = 1<<6;
740     }
741 }
742
743
744 int ff_h264_alloc_tables(H264Context *h){
745     MpegEncContext * const s = &h->s;
746     const int big_mb_num= s->mb_stride * (s->mb_height+1);
747     int x,y;
748
749     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t), fail)
750
751     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 32 * sizeof(uint8_t), fail)
752     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
753     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
754
755     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
756     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*big_mb_num * sizeof(uint8_t), fail);
757     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*big_mb_num * sizeof(uint8_t), fail);
758     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
759     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
760
761     memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
762     h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
763
764     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy  , big_mb_num * sizeof(uint32_t), fail);
765     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
766     for(y=0; y<s->mb_height; y++){
767         for(x=0; x<s->mb_width; x++){
768             const int mb_xy= x + y*s->mb_stride;
769             const int b_xy = 4*x + 4*y*h->b_stride;
770
771             h->mb2b_xy [mb_xy]= b_xy;
772             h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
773         }
774     }
775
776     s->obmc_scratchpad = NULL;
777
778     if(!h->dequant4_coeff[0])
779         init_dequant_tables(h);
780
781     return 0;
782 fail:
783     free_tables(h);
784     return -1;
785 }
786
787 /**
788  * Mimic alloc_tables(), but for every context thread.
789  */
790 static void clone_tables(H264Context *dst, H264Context *src){
791     dst->intra4x4_pred_mode       = src->intra4x4_pred_mode;
792     dst->non_zero_count           = src->non_zero_count;
793     dst->slice_table              = src->slice_table;
794     dst->cbp_table                = src->cbp_table;
795     dst->mb2b_xy                  = src->mb2b_xy;
796     dst->mb2br_xy                 = src->mb2br_xy;
797     dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
798     dst->mvd_table[0]             = src->mvd_table[0];
799     dst->mvd_table[1]             = src->mvd_table[1];
800     dst->direct_table             = src->direct_table;
801     dst->list_counts              = src->list_counts;
802
803     dst->s.obmc_scratchpad = NULL;
804     ff_h264_pred_init(&dst->hpc, src->s.codec_id);
805 }
806
807 /**
808  * Init context
809  * Allocate buffers which are not shared amongst multiple threads.
810  */
811 static int context_init(H264Context *h){
812     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
813     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
814
815     return 0;
816 fail:
817     return -1; // free_tables will clean up for us
818 }
819
820 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
821
822 static av_cold void common_init(H264Context *h){
823     MpegEncContext * const s = &h->s;
824
825     s->width = s->avctx->width;
826     s->height = s->avctx->height;
827     s->codec_id= s->avctx->codec->id;
828
829     ff_h264_pred_init(&h->hpc, s->codec_id);
830
831     h->dequant_coeff_pps= -1;
832     s->unrestricted_mv=1;
833     s->decode=1; //FIXME
834
835     dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
836
837     memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
838     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
839 }
840
841 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
842     H264Context *h= avctx->priv_data;
843     MpegEncContext * const s = &h->s;
844
845     MPV_decode_defaults(s);
846
847     s->avctx = avctx;
848     common_init(h);
849
850     s->out_format = FMT_H264;
851     s->workaround_bugs= avctx->workaround_bugs;
852
853     // set defaults
854 //    s->decode_mb= ff_h263_decode_mb;
855     s->quarter_sample = 1;
856     if(!avctx->has_b_frames)
857     s->low_delay= 1;
858
859     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
860
861     ff_h264_decode_init_vlc();
862
863     h->thread_context[0] = h;
864     h->outputed_poc = INT_MIN;
865     h->prev_poc_msb= 1<<16;
866     h->x264_build = -1;
867     ff_h264_reset_sei(h);
868     if(avctx->codec_id == CODEC_ID_H264){
869         if(avctx->ticks_per_frame == 1){
870             s->avctx->time_base.den *=2;
871         }
872         avctx->ticks_per_frame = 2;
873     }
874
875     h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
876     h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
877
878
879     if(avctx->extradata_size > 0 && avctx->extradata && *(char *)avctx->extradata == 1){
880         int i, cnt, nalsize;
881         unsigned char *p = avctx->extradata;
882
883         h->is_avc = 1;
884
885         if(avctx->extradata_size < 7) {
886             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
887             return -1;
888         }
889         /* sps and pps in the avcC always have length coded with 2 bytes,
890            so put a fake nal_length_size = 2 while parsing them */
891         h->nal_length_size = 2;
892         // Decode sps from avcC
893         cnt = *(p+5) & 0x1f; // Number of sps
894         p += 6;
895         for (i = 0; i < cnt; i++) {
896             nalsize = AV_RB16(p) + 2;
897             if(decode_nal_units(h, p, nalsize) < 0) {
898                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
899                 return -1;
900             }
901             p += nalsize;
902         }
903         // Decode pps from avcC
904         cnt = *(p++); // Number of pps
905         for (i = 0; i < cnt; i++) {
906             nalsize = AV_RB16(p) + 2;
907             if(decode_nal_units(h, p, nalsize)  != nalsize) {
908                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
909                 return -1;
910             }
911             p += nalsize;
912         }
913         // Now store right nal length size, that will be use to parse all other nals
914         h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
915     } else {
916         h->is_avc = 0;
917         if(decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) < 0)
918             return -1;
919     }
920     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
921         s->avctx->has_b_frames = h->sps.num_reorder_frames;
922         s->low_delay = 0;
923     }
924
925     return 0;
926 }
927
928 int ff_h264_frame_start(H264Context *h){
929     MpegEncContext * const s = &h->s;
930     int i;
931
932     if(MPV_frame_start(s, s->avctx) < 0)
933         return -1;
934     ff_er_frame_start(s);
935     /*
936      * MPV_frame_start uses pict_type to derive key_frame.
937      * This is incorrect for H.264; IDR markings must be used.
938      * Zero here; IDR markings per slice in frame or fields are ORed in later.
939      * See decode_nal_units().
940      */
941     s->current_picture_ptr->key_frame= 0;
942     s->current_picture_ptr->mmco_reset= 0;
943
944     assert(s->linesize && s->uvlinesize);
945
946     for(i=0; i<16; i++){
947         h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
948         h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
949     }
950     for(i=0; i<4; i++){
951         h->block_offset[16+i]=
952         h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
953         h->block_offset[24+16+i]=
954         h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
955     }
956
957     /* can't be in alloc_tables because linesize isn't known there.
958      * FIXME: redo bipred weight to not require extra buffer? */
959     for(i = 0; i < s->avctx->thread_count; i++)
960         if(!h->thread_context[i]->s.obmc_scratchpad)
961             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
962
963     /* some macroblocks will be accessed before they're available */
964     if(FRAME_MBAFF || s->avctx->thread_count > 1)
965         memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
966
967 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
968
969     // We mark the current picture as non-reference after allocating it, so
970     // that if we break out due to an error it can be released automatically
971     // in the next MPV_frame_start().
972     // SVQ3 as well as most other codecs have only last/next/current and thus
973     // get released even with set reference, besides SVQ3 and others do not
974     // mark frames as reference later "naturally".
975     if(s->codec_id != CODEC_ID_SVQ3)
976         s->current_picture_ptr->reference= 0;
977
978     s->current_picture_ptr->field_poc[0]=
979     s->current_picture_ptr->field_poc[1]= INT_MAX;
980     assert(s->current_picture_ptr->long_ref==0);
981
982     return 0;
983 }
984
985 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){
986     MpegEncContext * const s = &h->s;
987     uint8_t *top_border;
988     int top_idx = 1;
989
990     src_y  -=   linesize;
991     src_cb -= uvlinesize;
992     src_cr -= uvlinesize;
993
994     if(!simple && FRAME_MBAFF){
995         if(s->mb_y&1){
996             if(!MB_MBAFF){
997                 top_border = h->top_borders[0][s->mb_x];
998                 AV_COPY128(top_border, src_y + 15*linesize);
999                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1000                     AV_COPY64(top_border+16, src_cb+7*uvlinesize);
1001                     AV_COPY64(top_border+24, src_cr+7*uvlinesize);
1002                 }
1003             }
1004         }else if(MB_MBAFF){
1005             top_idx = 0;
1006         }else
1007             return;
1008     }
1009
1010     top_border = h->top_borders[top_idx][s->mb_x];
1011     // There are two lines saved, the line above the the top macroblock of a pair,
1012     // and the line above the bottom macroblock
1013     AV_COPY128(top_border, src_y + 16*linesize);
1014
1015     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1016         AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1017         AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1018     }
1019 }
1020
1021 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){
1022     MpegEncContext * const s = &h->s;
1023     int deblock_left;
1024     int deblock_top;
1025     int top_idx = 1;
1026     uint8_t *top_border_m1;
1027     uint8_t *top_border;
1028
1029     if(!simple && FRAME_MBAFF){
1030         if(s->mb_y&1){
1031             if(!MB_MBAFF)
1032                 return;
1033         }else{
1034             top_idx = MB_MBAFF ? 0 : 1;
1035         }
1036     }
1037
1038     if(h->deblocking_filter == 2) {
1039         deblock_left = h->left_type[0];
1040         deblock_top  = h->top_type;
1041     } else {
1042         deblock_left = (s->mb_x > 0);
1043         deblock_top =  (s->mb_y > !!MB_FIELD);
1044     }
1045
1046     src_y  -=   linesize + 1;
1047     src_cb -= uvlinesize + 1;
1048     src_cr -= uvlinesize + 1;
1049
1050     top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1051     top_border    = h->top_borders[top_idx][s->mb_x];
1052
1053 #define XCHG(a,b,xchg)\
1054 if (xchg) AV_SWAP64(b,a);\
1055 else      AV_COPY64(b,a);
1056
1057     if(deblock_top){
1058         if(deblock_left){
1059             XCHG(top_border_m1+8, src_y -7, 1);
1060         }
1061         XCHG(top_border+0, src_y +1, xchg);
1062         XCHG(top_border+8, src_y +9, 1);
1063         if(s->mb_x+1 < s->mb_width){
1064             XCHG(h->top_borders[top_idx][s->mb_x+1], src_y +17, 1);
1065         }
1066     }
1067
1068     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1069         if(deblock_top){
1070             if(deblock_left){
1071                 XCHG(top_border_m1+16, src_cb -7, 1);
1072                 XCHG(top_border_m1+24, src_cr -7, 1);
1073             }
1074             XCHG(top_border+16, src_cb+1, 1);
1075             XCHG(top_border+24, src_cr+1, 1);
1076         }
1077     }
1078 }
1079
1080 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
1081     MpegEncContext * const s = &h->s;
1082     const int mb_x= s->mb_x;
1083     const int mb_y= s->mb_y;
1084     const int mb_xy= h->mb_xy;
1085     const int mb_type= s->current_picture.mb_type[mb_xy];
1086     uint8_t  *dest_y, *dest_cb, *dest_cr;
1087     int linesize, uvlinesize /*dct_offset*/;
1088     int i;
1089     int *block_offset = &h->block_offset[0];
1090     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1091     /* is_h264 should always be true if SVQ3 is disabled. */
1092     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1093     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1094     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1095
1096     dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
1097     dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
1098     dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
1099
1100     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
1101     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
1102
1103     h->list_counts[mb_xy]= h->list_count;
1104
1105     if (!simple && MB_FIELD) {
1106         linesize   = h->mb_linesize   = s->linesize * 2;
1107         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1108         block_offset = &h->block_offset[24];
1109         if(mb_y&1){ //FIXME move out of this function?
1110             dest_y -= s->linesize*15;
1111             dest_cb-= s->uvlinesize*7;
1112             dest_cr-= s->uvlinesize*7;
1113         }
1114         if(FRAME_MBAFF) {
1115             int list;
1116             for(list=0; list<h->list_count; list++){
1117                 if(!USES_LIST(mb_type, list))
1118                     continue;
1119                 if(IS_16X16(mb_type)){
1120                     int8_t *ref = &h->ref_cache[list][scan8[0]];
1121                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1122                 }else{
1123                     for(i=0; i<16; i+=4){
1124                         int ref = h->ref_cache[list][scan8[i]];
1125                         if(ref >= 0)
1126                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1127                     }
1128                 }
1129             }
1130         }
1131     } else {
1132         linesize   = h->mb_linesize   = s->linesize;
1133         uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1134 //        dct_offset = s->linesize * 16;
1135     }
1136
1137     if (!simple && IS_INTRA_PCM(mb_type)) {
1138         for (i=0; i<16; i++) {
1139             memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1140         }
1141         for (i=0; i<8; i++) {
1142             memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
1143             memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
1144         }
1145     } else {
1146         if(IS_INTRA(mb_type)){
1147             if(h->deblocking_filter)
1148                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
1149
1150             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1151                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1152                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1153             }
1154
1155             if(IS_INTRA4x4(mb_type)){
1156                 if(simple || !s->encoding){
1157                     if(IS_8x8DCT(mb_type)){
1158                         if(transform_bypass){
1159                             idct_dc_add =
1160                             idct_add    = s->dsp.add_pixels8;
1161                         }else{
1162                             idct_dc_add = s->dsp.h264_idct8_dc_add;
1163                             idct_add    = s->dsp.h264_idct8_add;
1164                         }
1165                         for(i=0; i<16; i+=4){
1166                             uint8_t * const ptr= dest_y + block_offset[i];
1167                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1168                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1169                                 h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
1170                             }else{
1171                                 const int nnz = h->non_zero_count_cache[ scan8[i] ];
1172                                 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1173                                                             (h->topright_samples_available<<i)&0x4000, linesize);
1174                                 if(nnz){
1175                                     if(nnz == 1 && h->mb[i*16])
1176                                         idct_dc_add(ptr, h->mb + i*16, linesize);
1177                                     else
1178                                         idct_add   (ptr, h->mb + i*16, linesize);
1179                                 }
1180                             }
1181                         }
1182                     }else{
1183                         if(transform_bypass){
1184                             idct_dc_add =
1185                             idct_add    = s->dsp.add_pixels4;
1186                         }else{
1187                             idct_dc_add = s->dsp.h264_idct_dc_add;
1188                             idct_add    = s->dsp.h264_idct_add;
1189                         }
1190                         for(i=0; i<16; i++){
1191                             uint8_t * const ptr= dest_y + block_offset[i];
1192                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1193
1194                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1195                                 h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
1196                             }else{
1197                                 uint8_t *topright;
1198                                 int nnz, tr;
1199                                 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1200                                     const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1201                                     assert(mb_y || linesize <= block_offset[i]);
1202                                     if(!topright_avail){
1203                                         tr= ptr[3 - linesize]*0x01010101;
1204                                         topright= (uint8_t*) &tr;
1205                                     }else
1206                                         topright= ptr + 4 - linesize;
1207                                 }else
1208                                     topright= NULL;
1209
1210                                 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1211                                 nnz = h->non_zero_count_cache[ scan8[i] ];
1212                                 if(nnz){
1213                                     if(is_h264){
1214                                         if(nnz == 1 && h->mb[i*16])
1215                                             idct_dc_add(ptr, h->mb + i*16, linesize);
1216                                         else
1217                                             idct_add   (ptr, h->mb + i*16, linesize);
1218                                     }else
1219                                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1220                                 }
1221                             }
1222                         }
1223                     }
1224                 }
1225             }else{
1226                 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1227                 if(is_h264){
1228                     if(!transform_bypass)
1229                         h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
1230                 }else
1231                     ff_svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
1232             }
1233             if(h->deblocking_filter)
1234                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
1235         }else if(is_h264){
1236             hl_motion(h, dest_y, dest_cb, dest_cr,
1237                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1238                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1239                       s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
1240         }
1241
1242
1243         if(!IS_INTRA4x4(mb_type)){
1244             if(is_h264){
1245                 if(IS_INTRA16x16(mb_type)){
1246                     if(transform_bypass){
1247                         if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1248                             h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1249                         }else{
1250                             for(i=0; i<16; i++){
1251                                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1252                                     s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
1253                             }
1254                         }
1255                     }else{
1256                          s->dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1257                     }
1258                 }else if(h->cbp&15){
1259                     if(transform_bypass){
1260                         const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1261                         idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1262                         for(i=0; i<16; i+=di){
1263                             if(h->non_zero_count_cache[ scan8[i] ]){
1264                                 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
1265                             }
1266                         }
1267                     }else{
1268                         if(IS_8x8DCT(mb_type)){
1269                             s->dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1270                         }else{
1271                             s->dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1272                         }
1273                     }
1274                 }
1275             }else{
1276                 for(i=0; i<16; i++){
1277                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1278                         uint8_t * const ptr= dest_y + block_offset[i];
1279                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1280                     }
1281                 }
1282             }
1283         }
1284
1285         if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1286             uint8_t *dest[2] = {dest_cb, dest_cr};
1287             if(transform_bypass){
1288                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1289                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
1290                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
1291                 }else{
1292                     idct_add = s->dsp.add_pixels4;
1293                     for(i=16; i<16+8; i++){
1294                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1295                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1296                     }
1297                 }
1298             }else{
1299                 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]);
1300                 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]);
1301                 if(is_h264){
1302                     idct_add = s->dsp.h264_idct_add;
1303                     idct_dc_add = s->dsp.h264_idct_dc_add;
1304                     for(i=16; i<16+8; i++){
1305                         if(h->non_zero_count_cache[ scan8[i] ])
1306                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1307                         else if(h->mb[i*16])
1308                             idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1309                     }
1310                 }else{
1311                     for(i=16; i<16+8; i++){
1312                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1313                             uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1314                             ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
1315                         }
1316                     }
1317                 }
1318             }
1319         }
1320     }
1321     if(h->cbp || IS_INTRA(mb_type))
1322         s->dsp.clear_blocks(h->mb);
1323 }
1324
1325 /**
1326  * Process a macroblock; this case avoids checks for expensive uncommon cases.
1327  */
1328 static void hl_decode_mb_simple(H264Context *h){
1329     hl_decode_mb_internal(h, 1);
1330 }
1331
1332 /**
1333  * Process a macroblock; this handles edge cases, such as interlacing.
1334  */
1335 static void av_noinline hl_decode_mb_complex(H264Context *h){
1336     hl_decode_mb_internal(h, 0);
1337 }
1338
1339 void ff_h264_hl_decode_mb(H264Context *h){
1340     MpegEncContext * const s = &h->s;
1341     const int mb_xy= h->mb_xy;
1342     const int mb_type= s->current_picture.mb_type[mb_xy];
1343     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1344
1345     if (is_complex)
1346         hl_decode_mb_complex(h);
1347     else hl_decode_mb_simple(h);
1348 }
1349
1350 static int pred_weight_table(H264Context *h){
1351     MpegEncContext * const s = &h->s;
1352     int list, i;
1353     int luma_def, chroma_def;
1354
1355     h->use_weight= 0;
1356     h->use_weight_chroma= 0;
1357     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1358     h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1359     luma_def = 1<<h->luma_log2_weight_denom;
1360     chroma_def = 1<<h->chroma_log2_weight_denom;
1361
1362     for(list=0; list<2; list++){
1363         h->luma_weight_flag[list]   = 0;
1364         h->chroma_weight_flag[list] = 0;
1365         for(i=0; i<h->ref_count[list]; i++){
1366             int luma_weight_flag, chroma_weight_flag;
1367
1368             luma_weight_flag= get_bits1(&s->gb);
1369             if(luma_weight_flag){
1370                 h->luma_weight[list][i]= get_se_golomb(&s->gb);
1371                 h->luma_offset[list][i]= get_se_golomb(&s->gb);
1372                 if(   h->luma_weight[list][i] != luma_def
1373                    || h->luma_offset[list][i] != 0) {
1374                     h->use_weight= 1;
1375                     h->luma_weight_flag[list]= 1;
1376                 }
1377             }else{
1378                 h->luma_weight[list][i]= luma_def;
1379                 h->luma_offset[list][i]= 0;
1380             }
1381
1382             if(CHROMA){
1383                 chroma_weight_flag= get_bits1(&s->gb);
1384                 if(chroma_weight_flag){
1385                     int j;
1386                     for(j=0; j<2; j++){
1387                         h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
1388                         h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
1389                         if(   h->chroma_weight[list][i][j] != chroma_def
1390                            || h->chroma_offset[list][i][j] != 0) {
1391                             h->use_weight_chroma= 1;
1392                             h->chroma_weight_flag[list]= 1;
1393                         }
1394                     }
1395                 }else{
1396                     int j;
1397                     for(j=0; j<2; j++){
1398                         h->chroma_weight[list][i][j]= chroma_def;
1399                         h->chroma_offset[list][i][j]= 0;
1400                     }
1401                 }
1402             }
1403         }
1404         if(h->slice_type_nos != FF_B_TYPE) break;
1405     }
1406     h->use_weight= h->use_weight || h->use_weight_chroma;
1407     return 0;
1408 }
1409
1410 static void implicit_weight_table(H264Context *h){
1411     MpegEncContext * const s = &h->s;
1412     int ref0, ref1, i;
1413     int cur_poc = s->current_picture_ptr->poc;
1414
1415     for (i = 0; i < 2; i++) {
1416         h->luma_weight_flag[i]   = 0;
1417         h->chroma_weight_flag[i] = 0;
1418     }
1419
1420     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
1421        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1422         h->use_weight= 0;
1423         h->use_weight_chroma= 0;
1424         return;
1425     }
1426
1427     h->use_weight= 2;
1428     h->use_weight_chroma= 2;
1429     h->luma_log2_weight_denom= 5;
1430     h->chroma_log2_weight_denom= 5;
1431
1432     for(ref0=0; ref0 < h->ref_count[0]; ref0++){
1433         int poc0 = h->ref_list[0][ref0].poc;
1434         for(ref1=0; ref1 < h->ref_count[1]; ref1++){
1435             int poc1 = h->ref_list[1][ref1].poc;
1436             int td = av_clip(poc1 - poc0, -128, 127);
1437             if(td){
1438                 int tb = av_clip(cur_poc - poc0, -128, 127);
1439                 int tx = (16384 + (FFABS(td) >> 1)) / td;
1440                 int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
1441                 if(dist_scale_factor < -64 || dist_scale_factor > 128)
1442                     h->implicit_weight[ref0][ref1] = 32;
1443                 else
1444                     h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
1445             }else
1446                 h->implicit_weight[ref0][ref1] = 32;
1447         }
1448     }
1449 }
1450
1451 /**
1452  * instantaneous decoder refresh.
1453  */
1454 static void idr(H264Context *h){
1455     ff_h264_remove_all_refs(h);
1456     h->prev_frame_num= 0;
1457     h->prev_frame_num_offset= 0;
1458     h->prev_poc_msb=
1459     h->prev_poc_lsb= 0;
1460 }
1461
1462 /* forget old pics after a seek */
1463 static void flush_dpb(AVCodecContext *avctx){
1464     H264Context *h= avctx->priv_data;
1465     int i;
1466     for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1467         if(h->delayed_pic[i])
1468             h->delayed_pic[i]->reference= 0;
1469         h->delayed_pic[i]= NULL;
1470     }
1471     h->outputed_poc= INT_MIN;
1472     h->prev_interlaced_frame = 1;
1473     idr(h);
1474     if(h->s.current_picture_ptr)
1475         h->s.current_picture_ptr->reference= 0;
1476     h->s.first_field= 0;
1477     ff_h264_reset_sei(h);
1478     ff_mpeg_flush(avctx);
1479 }
1480
1481 static int init_poc(H264Context *h){
1482     MpegEncContext * const s = &h->s;
1483     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1484     int field_poc[2];
1485     Picture *cur = s->current_picture_ptr;
1486
1487     h->frame_num_offset= h->prev_frame_num_offset;
1488     if(h->frame_num < h->prev_frame_num)
1489         h->frame_num_offset += max_frame_num;
1490
1491     if(h->sps.poc_type==0){
1492         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1493
1494         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1495             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1496         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1497             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1498         else
1499             h->poc_msb = h->prev_poc_msb;
1500 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1501         field_poc[0] =
1502         field_poc[1] = h->poc_msb + h->poc_lsb;
1503         if(s->picture_structure == PICT_FRAME)
1504             field_poc[1] += h->delta_poc_bottom;
1505     }else if(h->sps.poc_type==1){
1506         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1507         int i;
1508
1509         if(h->sps.poc_cycle_length != 0)
1510             abs_frame_num = h->frame_num_offset + h->frame_num;
1511         else
1512             abs_frame_num = 0;
1513
1514         if(h->nal_ref_idc==0 && abs_frame_num > 0)
1515             abs_frame_num--;
1516
1517         expected_delta_per_poc_cycle = 0;
1518         for(i=0; i < h->sps.poc_cycle_length; i++)
1519             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1520
1521         if(abs_frame_num > 0){
1522             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1523             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1524
1525             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1526             for(i = 0; i <= frame_num_in_poc_cycle; i++)
1527                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1528         } else
1529             expectedpoc = 0;
1530
1531         if(h->nal_ref_idc == 0)
1532             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1533
1534         field_poc[0] = expectedpoc + h->delta_poc[0];
1535         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1536
1537         if(s->picture_structure == PICT_FRAME)
1538             field_poc[1] += h->delta_poc[1];
1539     }else{
1540         int poc= 2*(h->frame_num_offset + h->frame_num);
1541
1542         if(!h->nal_ref_idc)
1543             poc--;
1544
1545         field_poc[0]= poc;
1546         field_poc[1]= poc;
1547     }
1548
1549     if(s->picture_structure != PICT_BOTTOM_FIELD)
1550         s->current_picture_ptr->field_poc[0]= field_poc[0];
1551     if(s->picture_structure != PICT_TOP_FIELD)
1552         s->current_picture_ptr->field_poc[1]= field_poc[1];
1553     cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1554
1555     return 0;
1556 }
1557
1558
1559 /**
1560  * initialize scan tables
1561  */
1562 static void init_scan_tables(H264Context *h){
1563     MpegEncContext * const s = &h->s;
1564     int i;
1565     if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
1566         memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
1567         memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
1568     }else{
1569         for(i=0; i<16; i++){
1570 #define T(x) (x>>2) | ((x<<2) & 0xF)
1571             h->zigzag_scan[i] = T(zigzag_scan[i]);
1572             h-> field_scan[i] = T( field_scan[i]);
1573 #undef T
1574         }
1575     }
1576     if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
1577         memcpy(h->zigzag_scan8x8,       ff_zigzag_direct,     64*sizeof(uint8_t));
1578         memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
1579         memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
1580         memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
1581     }else{
1582         for(i=0; i<64; i++){
1583 #define T(x) (x>>3) | ((x&7)<<3)
1584             h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
1585             h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1586             h->field_scan8x8[i]        = T(field_scan8x8[i]);
1587             h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
1588 #undef T
1589         }
1590     }
1591     if(h->sps.transform_bypass){ //FIXME same ugly
1592         h->zigzag_scan_q0          = zigzag_scan;
1593         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
1594         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1595         h->field_scan_q0           = field_scan;
1596         h->field_scan8x8_q0        = field_scan8x8;
1597         h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
1598     }else{
1599         h->zigzag_scan_q0          = h->zigzag_scan;
1600         h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
1601         h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1602         h->field_scan_q0           = h->field_scan;
1603         h->field_scan8x8_q0        = h->field_scan8x8;
1604         h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
1605     }
1606 }
1607
1608 static void field_end(H264Context *h){
1609     MpegEncContext * const s = &h->s;
1610     AVCodecContext * const avctx= s->avctx;
1611     s->mb_y= 0;
1612
1613     s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1614     s->current_picture_ptr->pict_type= s->pict_type;
1615
1616     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1617         ff_vdpau_h264_set_reference_frames(s);
1618
1619     if(!s->dropable) {
1620         ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1621         h->prev_poc_msb= h->poc_msb;
1622         h->prev_poc_lsb= h->poc_lsb;
1623     }
1624     h->prev_frame_num_offset= h->frame_num_offset;
1625     h->prev_frame_num= h->frame_num;
1626
1627     if (avctx->hwaccel) {
1628         if (avctx->hwaccel->end_frame(avctx) < 0)
1629             av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1630     }
1631
1632     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1633         ff_vdpau_h264_picture_complete(s);
1634
1635     /*
1636      * FIXME: Error handling code does not seem to support interlaced
1637      * when slices span multiple rows
1638      * The ff_er_add_slice calls don't work right for bottom
1639      * fields; they cause massive erroneous error concealing
1640      * Error marking covers both fields (top and bottom).
1641      * This causes a mismatched s->error_count
1642      * and a bad error table. Further, the error count goes to
1643      * INT_MAX when called for bottom field, because mb_y is
1644      * past end by one (callers fault) and resync_mb_y != 0
1645      * causes problems for the first MB line, too.
1646      */
1647     if (!FIELD_PICTURE)
1648         ff_er_frame_end(s);
1649
1650     MPV_frame_end(s);
1651
1652     h->current_slice=0;
1653 }
1654
1655 /**
1656  * Replicates H264 "master" context to thread contexts.
1657  */
1658 static void clone_slice(H264Context *dst, H264Context *src)
1659 {
1660     memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
1661     dst->s.current_picture_ptr  = src->s.current_picture_ptr;
1662     dst->s.current_picture      = src->s.current_picture;
1663     dst->s.linesize             = src->s.linesize;
1664     dst->s.uvlinesize           = src->s.uvlinesize;
1665     dst->s.first_field          = src->s.first_field;
1666
1667     dst->prev_poc_msb           = src->prev_poc_msb;
1668     dst->prev_poc_lsb           = src->prev_poc_lsb;
1669     dst->prev_frame_num_offset  = src->prev_frame_num_offset;
1670     dst->prev_frame_num         = src->prev_frame_num;
1671     dst->short_ref_count        = src->short_ref_count;
1672
1673     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
1674     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
1675     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1676     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
1677
1678     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
1679     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
1680 }
1681
1682 /**
1683  * decodes a slice header.
1684  * This will also call MPV_common_init() and frame_start() as needed.
1685  *
1686  * @param h h264context
1687  * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1688  *
1689  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1690  */
1691 static int decode_slice_header(H264Context *h, H264Context *h0){
1692     MpegEncContext * const s = &h->s;
1693     MpegEncContext * const s0 = &h0->s;
1694     unsigned int first_mb_in_slice;
1695     unsigned int pps_id;
1696     int num_ref_idx_active_override_flag;
1697     unsigned int slice_type, tmp, i, j;
1698     int default_ref_list_done = 0;
1699     int last_pic_structure;
1700
1701     s->dropable= h->nal_ref_idc == 0;
1702
1703     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1704         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1705         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1706     }else{
1707         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1708         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1709     }
1710
1711     first_mb_in_slice= get_ue_golomb(&s->gb);
1712
1713     if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1714         if(h0->current_slice && FIELD_PICTURE){
1715             field_end(h);
1716         }
1717
1718         h0->current_slice = 0;
1719         if (!s0->first_field)
1720             s->current_picture_ptr= NULL;
1721     }
1722
1723     slice_type= get_ue_golomb_31(&s->gb);
1724     if(slice_type > 9){
1725         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);
1726         return -1;
1727     }
1728     if(slice_type > 4){
1729         slice_type -= 5;
1730         h->slice_type_fixed=1;
1731     }else
1732         h->slice_type_fixed=0;
1733
1734     slice_type= golomb_to_pict_type[ slice_type ];
1735     if (slice_type == FF_I_TYPE
1736         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
1737         default_ref_list_done = 1;
1738     }
1739     h->slice_type= slice_type;
1740     h->slice_type_nos= slice_type & 3;
1741
1742     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
1743
1744     pps_id= get_ue_golomb(&s->gb);
1745     if(pps_id>=MAX_PPS_COUNT){
1746         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
1747         return -1;
1748     }
1749     if(!h0->pps_buffers[pps_id]) {
1750         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
1751         return -1;
1752     }
1753     h->pps= *h0->pps_buffers[pps_id];
1754
1755     if(!h0->sps_buffers[h->pps.sps_id]) {
1756         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
1757         return -1;
1758     }
1759     h->sps = *h0->sps_buffers[h->pps.sps_id];
1760
1761     if(h == h0 && h->dequant_coeff_pps != pps_id){
1762         h->dequant_coeff_pps = pps_id;
1763         init_dequant_tables(h);
1764     }
1765
1766     s->mb_width= h->sps.mb_width;
1767     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1768
1769     h->b_stride=  s->mb_width*4;
1770
1771     s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
1772     if(h->sps.frame_mbs_only_flag)
1773         s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
1774     else
1775         s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
1776
1777     if (s->context_initialized
1778         && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
1779         if(h != h0)
1780             return -1;   // width / height changed during parallelized decoding
1781         free_tables(h);
1782         flush_dpb(s->avctx);
1783         MPV_common_end(s);
1784     }
1785     if (!s->context_initialized) {
1786         if(h != h0)
1787             return -1;  // we cant (re-)initialize context during parallel decoding
1788
1789         avcodec_set_dimensions(s->avctx, s->width, s->height);
1790         s->avctx->sample_aspect_ratio= h->sps.sar;
1791         if(!s->avctx->sample_aspect_ratio.den)
1792             s->avctx->sample_aspect_ratio.den = 1;
1793
1794         if(h->sps.video_signal_type_present_flag){
1795             s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1796             if(h->sps.colour_description_present_flag){
1797                 s->avctx->color_primaries = h->sps.color_primaries;
1798                 s->avctx->color_trc       = h->sps.color_trc;
1799                 s->avctx->colorspace      = h->sps.colorspace;
1800             }
1801         }
1802
1803         if(h->sps.timing_info_present_flag){
1804             int64_t den= h->sps.time_scale;
1805             if(h->x264_build < 44U)
1806                 den *= 2;
1807             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
1808                       h->sps.num_units_in_tick, den, 1<<30);
1809         }
1810         s->avctx->pix_fmt = s->avctx->get_format(s->avctx, s->avctx->codec->pix_fmts);
1811         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1812
1813         if (MPV_common_init(s) < 0)
1814             return -1;
1815         s->first_field = 0;
1816         h->prev_interlaced_frame = 1;
1817
1818         init_scan_tables(h);
1819         ff_h264_alloc_tables(h);
1820
1821         for(i = 1; i < s->avctx->thread_count; i++) {
1822             H264Context *c;
1823             c = h->thread_context[i] = av_malloc(sizeof(H264Context));
1824             memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
1825             memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
1826             c->sps = h->sps;
1827             c->pps = h->pps;
1828             init_scan_tables(c);
1829             clone_tables(c, h);
1830         }
1831
1832         for(i = 0; i < s->avctx->thread_count; i++)
1833             if(context_init(h->thread_context[i]) < 0)
1834                 return -1;
1835     }
1836
1837     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1838
1839     h->mb_mbaff = 0;
1840     h->mb_aff_frame = 0;
1841     last_pic_structure = s0->picture_structure;
1842     if(h->sps.frame_mbs_only_flag){
1843         s->picture_structure= PICT_FRAME;
1844     }else{
1845         if(get_bits1(&s->gb)) { //field_pic_flag
1846             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
1847         } else {
1848             s->picture_structure= PICT_FRAME;
1849             h->mb_aff_frame = h->sps.mb_aff;
1850         }
1851     }
1852     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
1853
1854     if(h0->current_slice == 0){
1855         while(h->frame_num !=  h->prev_frame_num &&
1856               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
1857             av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
1858             if (ff_h264_frame_start(h) < 0)
1859                 return -1;
1860             h->prev_frame_num++;
1861             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
1862             s->current_picture_ptr->frame_num= h->prev_frame_num;
1863             ff_h264_execute_ref_pic_marking(h, NULL, 0);
1864         }
1865
1866         /* See if we have a decoded first field looking for a pair... */
1867         if (s0->first_field) {
1868             assert(s0->current_picture_ptr);
1869             assert(s0->current_picture_ptr->data[0]);
1870             assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
1871
1872             /* figure out if we have a complementary field pair */
1873             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
1874                 /*
1875                  * Previous field is unmatched. Don't display it, but let it
1876                  * remain for reference if marked as such.
1877                  */
1878                 s0->current_picture_ptr = NULL;
1879                 s0->first_field = FIELD_PICTURE;
1880
1881             } else {
1882                 if (h->nal_ref_idc &&
1883                         s0->current_picture_ptr->reference &&
1884                         s0->current_picture_ptr->frame_num != h->frame_num) {
1885                     /*
1886                      * This and previous field were reference, but had
1887                      * different frame_nums. Consider this field first in
1888                      * pair. Throw away previous field except for reference
1889                      * purposes.
1890                      */
1891                     s0->first_field = 1;
1892                     s0->current_picture_ptr = NULL;
1893
1894                 } else {
1895                     /* Second field in complementary pair */
1896                     s0->first_field = 0;
1897                 }
1898             }
1899
1900         } else {
1901             /* Frame or first field in a potentially complementary pair */
1902             assert(!s0->current_picture_ptr);
1903             s0->first_field = FIELD_PICTURE;
1904         }
1905
1906         if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
1907             s0->first_field = 0;
1908             return -1;
1909         }
1910     }
1911     if(h != h0)
1912         clone_slice(h, h0);
1913
1914     s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
1915
1916     assert(s->mb_num == s->mb_width * s->mb_height);
1917     if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
1918        first_mb_in_slice                    >= s->mb_num){
1919         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1920         return -1;
1921     }
1922     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
1923     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
1924     if (s->picture_structure == PICT_BOTTOM_FIELD)
1925         s->resync_mb_y = s->mb_y = s->mb_y + 1;
1926     assert(s->mb_y < s->mb_height);
1927
1928     if(s->picture_structure==PICT_FRAME){
1929         h->curr_pic_num=   h->frame_num;
1930         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
1931     }else{
1932         h->curr_pic_num= 2*h->frame_num + 1;
1933         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
1934     }
1935
1936     if(h->nal_unit_type == NAL_IDR_SLICE){
1937         get_ue_golomb(&s->gb); /* idr_pic_id */
1938     }
1939
1940     if(h->sps.poc_type==0){
1941         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
1942
1943         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
1944             h->delta_poc_bottom= get_se_golomb(&s->gb);
1945         }
1946     }
1947
1948     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
1949         h->delta_poc[0]= get_se_golomb(&s->gb);
1950
1951         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
1952             h->delta_poc[1]= get_se_golomb(&s->gb);
1953     }
1954
1955     init_poc(h);
1956
1957     if(h->pps.redundant_pic_cnt_present){
1958         h->redundant_pic_count= get_ue_golomb(&s->gb);
1959     }
1960
1961     //set defaults, might be overridden a few lines later
1962     h->ref_count[0]= h->pps.ref_count[0];
1963     h->ref_count[1]= h->pps.ref_count[1];
1964
1965     if(h->slice_type_nos != FF_I_TYPE){
1966         if(h->slice_type_nos == FF_B_TYPE){
1967             h->direct_spatial_mv_pred= get_bits1(&s->gb);
1968         }
1969         num_ref_idx_active_override_flag= get_bits1(&s->gb);
1970
1971         if(num_ref_idx_active_override_flag){
1972             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
1973             if(h->slice_type_nos==FF_B_TYPE)
1974                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
1975
1976             if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
1977                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
1978                 h->ref_count[0]= h->ref_count[1]= 1;
1979                 return -1;
1980             }
1981         }
1982         if(h->slice_type_nos == FF_B_TYPE)
1983             h->list_count= 2;
1984         else
1985             h->list_count= 1;
1986     }else
1987         h->list_count= 0;
1988
1989     if(!default_ref_list_done){
1990         ff_h264_fill_default_ref_list(h);
1991     }
1992
1993     if(h->slice_type_nos!=FF_I_TYPE && ff_h264_decode_ref_pic_list_reordering(h) < 0)
1994         return -1;
1995
1996     if(h->slice_type_nos!=FF_I_TYPE){
1997         s->last_picture_ptr= &h->ref_list[0][0];
1998         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
1999     }
2000     if(h->slice_type_nos==FF_B_TYPE){
2001         s->next_picture_ptr= &h->ref_list[1][0];
2002         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
2003     }
2004
2005     if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
2006        ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
2007         pred_weight_table(h);
2008     else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE)
2009         implicit_weight_table(h);
2010     else {
2011         h->use_weight = 0;
2012         for (i = 0; i < 2; i++) {
2013             h->luma_weight_flag[i]   = 0;
2014             h->chroma_weight_flag[i] = 0;
2015         }
2016     }
2017
2018     if(h->nal_ref_idc)
2019         ff_h264_decode_ref_pic_marking(h0, &s->gb);
2020
2021     if(FRAME_MBAFF)
2022         ff_h264_fill_mbaff_ref_list(h);
2023
2024     if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
2025         ff_h264_direct_dist_scale_factor(h);
2026     ff_h264_direct_ref_list_init(h);
2027
2028     if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
2029         tmp = get_ue_golomb_31(&s->gb);
2030         if(tmp > 2){
2031             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2032             return -1;
2033         }
2034         h->cabac_init_idc= tmp;
2035     }
2036
2037     h->last_qscale_diff = 0;
2038     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2039     if(tmp>51){
2040         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2041         return -1;
2042     }
2043     s->qscale= tmp;
2044     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2045     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2046     //FIXME qscale / qp ... stuff
2047     if(h->slice_type == FF_SP_TYPE){
2048         get_bits1(&s->gb); /* sp_for_switch_flag */
2049     }
2050     if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
2051         get_se_golomb(&s->gb); /* slice_qs_delta */
2052     }
2053
2054     h->deblocking_filter = 1;
2055     h->slice_alpha_c0_offset = 52;
2056     h->slice_beta_offset = 52;
2057     if( h->pps.deblocking_filter_parameters_present ) {
2058         tmp= get_ue_golomb_31(&s->gb);
2059         if(tmp > 2){
2060             av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2061             return -1;
2062         }
2063         h->deblocking_filter= tmp;
2064         if(h->deblocking_filter < 2)
2065             h->deblocking_filter^= 1; // 1<->0
2066
2067         if( h->deblocking_filter ) {
2068             h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2069             h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
2070             if(   h->slice_alpha_c0_offset > 104U
2071                || h->slice_beta_offset     > 104U){
2072                 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);
2073                 return -1;
2074             }
2075         }
2076     }
2077
2078     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
2079        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
2080        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
2081        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2082         h->deblocking_filter= 0;
2083
2084     if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2085         if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2086             /* Cheat slightly for speed:
2087                Do not bother to deblock across slices. */
2088             h->deblocking_filter = 2;
2089         } else {
2090             h0->max_contexts = 1;
2091             if(!h0->single_decode_warning) {
2092                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2093                 h0->single_decode_warning = 1;
2094             }
2095             if(h != h0)
2096                 return 1; // deblocking switched inside frame
2097         }
2098     }
2099     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]);
2100
2101 #if 0 //FMO
2102     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2103         slice_group_change_cycle= get_bits(&s->gb, ?);
2104 #endif
2105
2106     h0->last_slice_type = slice_type;
2107     h->slice_num = ++h0->current_slice;
2108     if(h->slice_num >= MAX_SLICES){
2109         av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2110     }
2111
2112     for(j=0; j<2; j++){
2113         int id_list[16];
2114         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2115         for(i=0; i<16; i++){
2116             id_list[i]= 60;
2117             if(h->ref_list[j][i].data[0]){
2118                 int k;
2119                 uint8_t *base= h->ref_list[j][i].base[0];
2120                 for(k=0; k<h->short_ref_count; k++)
2121                     if(h->short_ref[k]->base[0] == base){
2122                         id_list[i]= k;
2123                         break;
2124                     }
2125                 for(k=0; k<h->long_ref_count; k++)
2126                     if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2127                         id_list[i]= h->short_ref_count + k;
2128                         break;
2129                     }
2130             }
2131         }
2132
2133         ref2frm[0]=
2134         ref2frm[1]= -1;
2135         for(i=0; i<16; i++)
2136             ref2frm[i+2]= 4*id_list[i]
2137                           +(h->ref_list[j][i].reference&3);
2138         ref2frm[18+0]=
2139         ref2frm[18+1]= -1;
2140         for(i=16; i<48; i++)
2141             ref2frm[i+4]= 4*id_list[(i-16)>>1]
2142                           +(h->ref_list[j][i].reference&3);
2143     }
2144
2145     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
2146     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2147
2148     s->avctx->refs= h->sps.ref_frame_count;
2149
2150     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2151         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",
2152                h->slice_num,
2153                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2154                first_mb_in_slice,
2155                av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2156                pps_id, h->frame_num,
2157                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2158                h->ref_count[0], h->ref_count[1],
2159                s->qscale,
2160                h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
2161                h->use_weight,
2162                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2163                h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2164                );
2165     }
2166
2167     return 0;
2168 }
2169
2170 int ff_h264_get_slice_type(const H264Context *h)
2171 {
2172     switch (h->slice_type) {
2173     case FF_P_TYPE:  return 0;
2174     case FF_B_TYPE:  return 1;
2175     case FF_I_TYPE:  return 2;
2176     case FF_SP_TYPE: return 3;
2177     case FF_SI_TYPE: return 4;
2178     default:         return -1;
2179     }
2180 }
2181
2182 static void loop_filter(H264Context *h){
2183     MpegEncContext * const s = &h->s;
2184     uint8_t  *dest_y, *dest_cb, *dest_cr;
2185     int linesize, uvlinesize, mb_x, mb_y;
2186     const int end_mb_y= s->mb_y + FRAME_MBAFF;
2187     const int old_slice_type= h->slice_type;
2188
2189     if(h->deblocking_filter) {
2190         for(mb_x= 0; mb_x<s->mb_width; mb_x++){
2191             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
2192                 int mb_xy, mb_type;
2193                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2194                 h->slice_num= h->slice_table[mb_xy];
2195                 mb_type= s->current_picture.mb_type[mb_xy];
2196                 h->list_count= h->list_counts[mb_xy];
2197
2198                 if(FRAME_MBAFF)
2199                     h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2200
2201                 s->mb_x= mb_x;
2202                 s->mb_y= mb_y;
2203                 dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2204                 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2205                 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2206                     //FIXME simplify above
2207
2208                 if (MB_FIELD) {
2209                     linesize   = h->mb_linesize   = s->linesize * 2;
2210                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2211                     if(mb_y&1){ //FIXME move out of this function?
2212                         dest_y -= s->linesize*15;
2213                         dest_cb-= s->uvlinesize*7;
2214                         dest_cr-= s->uvlinesize*7;
2215                     }
2216                 } else {
2217                     linesize   = h->mb_linesize   = s->linesize;
2218                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2219                 }
2220                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2221                 if(fill_filter_caches(h, mb_type))
2222                     continue;
2223                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2224                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2225
2226                 if (FRAME_MBAFF) {
2227                     ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2228                 } else {
2229                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2230                 }
2231             }
2232         }
2233     }
2234     h->slice_type= old_slice_type;
2235     s->mb_x= 0;
2236     s->mb_y= end_mb_y - FRAME_MBAFF;
2237     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2238     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2239 }
2240
2241 static void predict_field_decoding_flag(H264Context *h){
2242     MpegEncContext * const s = &h->s;
2243     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2244     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
2245                 ? s->current_picture.mb_type[mb_xy-1]
2246                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
2247                 ? s->current_picture.mb_type[mb_xy-s->mb_stride]
2248                 : 0;
2249     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2250 }
2251
2252 static int decode_slice(struct AVCodecContext *avctx, void *arg){
2253     H264Context *h = *(void**)arg;
2254     MpegEncContext * const s = &h->s;
2255     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2256
2257     s->mb_skip_run= -1;
2258
2259     h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
2260                     (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
2261
2262     if( h->pps.cabac ) {
2263         /* realign */
2264         align_get_bits( &s->gb );
2265
2266         /* init cabac */
2267         ff_init_cabac_states( &h->cabac);
2268         ff_init_cabac_decoder( &h->cabac,
2269                                s->gb.buffer + get_bits_count(&s->gb)/8,
2270                                (get_bits_left(&s->gb) + 7)/8);
2271
2272         ff_h264_init_cabac_states(h);
2273
2274         for(;;){
2275 //START_TIMER
2276             int ret = ff_h264_decode_mb_cabac(h);
2277             int eos;
2278 //STOP_TIMER("decode_mb_cabac")
2279
2280             if(ret>=0) ff_h264_hl_decode_mb(h);
2281
2282             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
2283                 s->mb_y++;
2284
2285                 ret = ff_h264_decode_mb_cabac(h);
2286
2287                 if(ret>=0) ff_h264_hl_decode_mb(h);
2288                 s->mb_y--;
2289             }
2290             eos = get_cabac_terminate( &h->cabac );
2291
2292             if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
2293                 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);
2294                 return 0;
2295             }
2296             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2297                 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);
2298                 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);
2299                 return -1;
2300             }
2301
2302             if( ++s->mb_x >= s->mb_width ) {
2303                 s->mb_x = 0;
2304                 loop_filter(h);
2305                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2306                 ++s->mb_y;
2307                 if(FIELD_OR_MBAFF_PICTURE) {
2308                     ++s->mb_y;
2309                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2310                         predict_field_decoding_flag(h);
2311                 }
2312             }
2313
2314             if( eos || s->mb_y >= s->mb_height ) {
2315                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2316                 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);
2317                 return 0;
2318             }
2319         }
2320
2321     } else {
2322         for(;;){
2323             int ret = ff_h264_decode_mb_cavlc(h);
2324
2325             if(ret>=0) ff_h264_hl_decode_mb(h);
2326
2327             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
2328                 s->mb_y++;
2329                 ret = ff_h264_decode_mb_cavlc(h);
2330
2331                 if(ret>=0) ff_h264_hl_decode_mb(h);
2332                 s->mb_y--;
2333             }
2334
2335             if(ret<0){
2336                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2337                 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);
2338
2339                 return -1;
2340             }
2341
2342             if(++s->mb_x >= s->mb_width){
2343                 s->mb_x=0;
2344                 loop_filter(h);
2345                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2346                 ++s->mb_y;
2347                 if(FIELD_OR_MBAFF_PICTURE) {
2348                     ++s->mb_y;
2349                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2350                         predict_field_decoding_flag(h);
2351                 }
2352                 if(s->mb_y >= s->mb_height){
2353                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2354
2355                     if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
2356                         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);
2357
2358                         return 0;
2359                     }else{
2360                         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);
2361
2362                         return -1;
2363                     }
2364                 }
2365             }
2366
2367             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
2368                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2369                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2370                     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);
2371
2372                     return 0;
2373                 }else{
2374                     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);
2375
2376                     return -1;
2377                 }
2378             }
2379         }
2380     }
2381
2382 #if 0
2383     for(;s->mb_y < s->mb_height; s->mb_y++){
2384         for(;s->mb_x < s->mb_width; s->mb_x++){
2385             int ret= decode_mb(h);
2386
2387             ff_h264_hl_decode_mb(h);
2388
2389             if(ret<0){
2390                 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2391                 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);
2392
2393                 return -1;
2394             }
2395
2396             if(++s->mb_x >= s->mb_width){
2397                 s->mb_x=0;
2398                 if(++s->mb_y >= s->mb_height){
2399                     if(get_bits_count(s->gb) == s->gb.size_in_bits){
2400                         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);
2401
2402                         return 0;
2403                     }else{
2404                         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);
2405
2406                         return -1;
2407                     }
2408                 }
2409             }
2410
2411             if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2412                 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2413                     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);
2414
2415                     return 0;
2416                 }else{
2417                     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);
2418
2419                     return -1;
2420                 }
2421             }
2422         }
2423         s->mb_x=0;
2424         ff_draw_horiz_band(s, 16*s->mb_y, 16);
2425     }
2426 #endif
2427     return -1; //not reached
2428 }
2429
2430 /**
2431  * Call decode_slice() for each context.
2432  *
2433  * @param h h264 master context
2434  * @param context_count number of contexts to execute
2435  */
2436 static void execute_decode_slices(H264Context *h, int context_count){
2437     MpegEncContext * const s = &h->s;
2438     AVCodecContext * const avctx= s->avctx;
2439     H264Context *hx;
2440     int i;
2441
2442     if (s->avctx->hwaccel)
2443         return;
2444     if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2445         return;
2446     if(context_count == 1) {
2447         decode_slice(avctx, &h);
2448     } else {
2449         for(i = 1; i < context_count; i++) {
2450             hx = h->thread_context[i];
2451             hx->s.error_recognition = avctx->error_recognition;
2452             hx->s.error_count = 0;
2453         }
2454
2455         avctx->execute(avctx, (void *)decode_slice,
2456                        h->thread_context, NULL, context_count, sizeof(void*));
2457
2458         /* pull back stuff from slices to master context */
2459         hx = h->thread_context[context_count - 1];
2460         s->mb_x = hx->s.mb_x;
2461         s->mb_y = hx->s.mb_y;
2462         s->dropable = hx->s.dropable;
2463         s->picture_structure = hx->s.picture_structure;
2464         for(i = 1; i < context_count; i++)
2465             h->s.error_count += h->thread_context[i]->s.error_count;
2466     }
2467 }
2468
2469
2470 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
2471     MpegEncContext * const s = &h->s;
2472     AVCodecContext * const avctx= s->avctx;
2473     int buf_index=0;
2474     H264Context *hx; ///< thread context
2475     int context_count = 0;
2476     int next_avc= h->is_avc ? 0 : buf_size;
2477
2478     h->max_contexts = avctx->thread_count;
2479 #if 0
2480     int i;
2481     for(i=0; i<50; i++){
2482         av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
2483     }
2484 #endif
2485     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
2486         h->current_slice = 0;
2487         if (!s->first_field)
2488             s->current_picture_ptr= NULL;
2489         ff_h264_reset_sei(h);
2490     }
2491
2492     for(;;){
2493         int consumed;
2494         int dst_length;
2495         int bit_length;
2496         const uint8_t *ptr;
2497         int i, nalsize = 0;
2498         int err;
2499
2500         if(buf_index >= next_avc) {
2501             if(buf_index >= buf_size) break;
2502             nalsize = 0;
2503             for(i = 0; i < h->nal_length_size; i++)
2504                 nalsize = (nalsize << 8) | buf[buf_index++];
2505             if(nalsize <= 1 || nalsize > buf_size - buf_index){
2506                 if(nalsize == 1){
2507                     buf_index++;
2508                     continue;
2509                 }else{
2510                     av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2511                     break;
2512                 }
2513             }
2514             next_avc= buf_index + nalsize;
2515         } else {
2516             // start code prefix search
2517             for(; buf_index + 3 < next_avc; buf_index++){
2518                 // This should always succeed in the first iteration.
2519                 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2520                     break;
2521             }
2522
2523             if(buf_index+3 >= buf_size) break;
2524
2525             buf_index+=3;
2526             if(buf_index >= next_avc) continue;
2527         }
2528
2529         hx = h->thread_context[context_count];
2530
2531         ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
2532         if (ptr==NULL || dst_length < 0){
2533             return -1;
2534         }
2535         i= buf_index + consumed;
2536         if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
2537            buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
2538             s->workaround_bugs |= FF_BUG_TRUNCATED;
2539
2540         if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
2541         while(ptr[dst_length - 1] == 0 && dst_length > 0)
2542             dst_length--;
2543         }
2544         bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
2545
2546         if(s->avctx->debug&FF_DEBUG_STARTCODE){
2547             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);
2548         }
2549
2550         if (h->is_avc && (nalsize != consumed) && nalsize){
2551             av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
2552         }
2553
2554         buf_index += consumed;
2555
2556         if(  (s->hurry_up == 1 && h->nal_ref_idc  == 0) //FIXME do not discard SEI id
2557            ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0))
2558             continue;
2559
2560       again:
2561         err = 0;
2562         switch(hx->nal_unit_type){
2563         case NAL_IDR_SLICE:
2564             if (h->nal_unit_type != NAL_IDR_SLICE) {
2565                 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2566                 return -1;
2567             }
2568             idr(h); //FIXME ensure we don't loose some frames if there is reordering
2569         case NAL_SLICE:
2570             init_get_bits(&hx->s.gb, ptr, bit_length);
2571             hx->intra_gb_ptr=
2572             hx->inter_gb_ptr= &hx->s.gb;
2573             hx->s.data_partitioning = 0;
2574
2575             if((err = decode_slice_header(hx, h)))
2576                break;
2577
2578             avctx->profile = hx->sps.profile_idc;
2579             avctx->level   = hx->sps.level_idc;
2580
2581             if (s->avctx->hwaccel && h->current_slice == 1) {
2582                 if (s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
2583                     return -1;
2584             }
2585
2586             s->current_picture_ptr->key_frame |=
2587                     (hx->nal_unit_type == NAL_IDR_SLICE) ||
2588                     (h->sei_recovery_frame_cnt >= 0);
2589             if(hx->redundant_pic_count==0 && hx->s.hurry_up < 5
2590                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2591                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2592                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2593                && avctx->skip_frame < AVDISCARD_ALL){
2594                 if(avctx->hwaccel) {
2595                     if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
2596                         return -1;
2597                 }else
2598                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
2599                     static const uint8_t start_code[] = {0x00, 0x00, 0x01};
2600                     ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
2601                     ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
2602                 }else
2603                     context_count++;
2604             }
2605             break;
2606         case NAL_DPA:
2607             init_get_bits(&hx->s.gb, ptr, bit_length);
2608             hx->intra_gb_ptr=
2609             hx->inter_gb_ptr= NULL;
2610
2611             if ((err = decode_slice_header(hx, h)) < 0)
2612                 break;
2613
2614             avctx->profile = hx->sps.profile_idc;
2615             avctx->level   = hx->sps.level_idc;
2616
2617             hx->s.data_partitioning = 1;
2618
2619             break;
2620         case NAL_DPB:
2621             init_get_bits(&hx->intra_gb, ptr, bit_length);
2622             hx->intra_gb_ptr= &hx->intra_gb;
2623             break;
2624         case NAL_DPC:
2625             init_get_bits(&hx->inter_gb, ptr, bit_length);
2626             hx->inter_gb_ptr= &hx->inter_gb;
2627
2628             if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
2629                && s->context_initialized
2630                && s->hurry_up < 5
2631                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2632                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2633                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2634                && avctx->skip_frame < AVDISCARD_ALL)
2635                 context_count++;
2636             break;
2637         case NAL_SEI:
2638             init_get_bits(&s->gb, ptr, bit_length);
2639             ff_h264_decode_sei(h);
2640             break;
2641         case NAL_SPS:
2642             init_get_bits(&s->gb, ptr, bit_length);
2643             ff_h264_decode_seq_parameter_set(h);
2644
2645             if(s->flags& CODEC_FLAG_LOW_DELAY)
2646                 s->low_delay=1;
2647
2648             if(avctx->has_b_frames < 2)
2649                 avctx->has_b_frames= !s->low_delay;
2650             break;
2651         case NAL_PPS:
2652             init_get_bits(&s->gb, ptr, bit_length);
2653
2654             ff_h264_decode_picture_parameter_set(h, bit_length);
2655
2656             break;
2657         case NAL_AUD:
2658         case NAL_END_SEQUENCE:
2659         case NAL_END_STREAM:
2660         case NAL_FILLER_DATA:
2661         case NAL_SPS_EXT:
2662         case NAL_AUXILIARY_SLICE:
2663             break;
2664         default:
2665             av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
2666         }
2667
2668         if(context_count == h->max_contexts) {
2669             execute_decode_slices(h, context_count);
2670             context_count = 0;
2671         }
2672
2673         if (err < 0)
2674             av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
2675         else if(err == 1) {
2676             /* Slice could not be decoded in parallel mode, copy down
2677              * NAL unit stuff to context 0 and restart. Note that
2678              * rbsp_buffer is not transferred, but since we no longer
2679              * run in parallel mode this should not be an issue. */
2680             h->nal_unit_type = hx->nal_unit_type;
2681             h->nal_ref_idc   = hx->nal_ref_idc;
2682             hx = h;
2683             goto again;
2684         }
2685     }
2686     if(context_count)
2687         execute_decode_slices(h, context_count);
2688     return buf_index;
2689 }
2690
2691 /**
2692  * returns the number of bytes consumed for building the current frame
2693  */
2694 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
2695         if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
2696         if(pos+10>buf_size) pos=buf_size; // oops ;)
2697
2698         return pos;
2699 }
2700
2701 static int decode_frame(AVCodecContext *avctx,
2702                              void *data, int *data_size,
2703                              AVPacket *avpkt)
2704 {
2705     const uint8_t *buf = avpkt->data;
2706     int buf_size = avpkt->size;
2707     H264Context *h = avctx->priv_data;
2708     MpegEncContext *s = &h->s;
2709     AVFrame *pict = data;
2710     int buf_index;
2711
2712     s->flags= avctx->flags;
2713     s->flags2= avctx->flags2;
2714
2715    /* end of stream, output what is still in the buffers */
2716     if (buf_size == 0) {
2717         Picture *out;
2718         int i, out_idx;
2719
2720 //FIXME factorize this with the output code below
2721         out = h->delayed_pic[0];
2722         out_idx = 0;
2723         for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2724             if(h->delayed_pic[i]->poc < out->poc){
2725                 out = h->delayed_pic[i];
2726                 out_idx = i;
2727             }
2728
2729         for(i=out_idx; h->delayed_pic[i]; i++)
2730             h->delayed_pic[i] = h->delayed_pic[i+1];
2731
2732         if(out){
2733             *data_size = sizeof(AVFrame);
2734             *pict= *(AVFrame*)out;
2735         }
2736
2737         return 0;
2738     }
2739
2740     buf_index=decode_nal_units(h, buf, buf_size);
2741     if(buf_index < 0)
2742         return -1;
2743
2744     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
2745         if (avctx->skip_frame >= AVDISCARD_NONREF || s->hurry_up) return 0;
2746         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
2747         return -1;
2748     }
2749
2750     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
2751         Picture *out = s->current_picture_ptr;
2752         Picture *cur = s->current_picture_ptr;
2753         int i, pics, out_of_order, out_idx;
2754
2755         field_end(h);
2756
2757         if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
2758             /* Wait for second field. */
2759             *data_size = 0;
2760
2761         } else {
2762             cur->interlaced_frame = 0;
2763             cur->repeat_pict = 0;
2764
2765             /* Signal interlacing information externally. */
2766             /* Prioritize picture timing SEI information over used decoding process if it exists. */
2767
2768             if(h->sps.pic_struct_present_flag){
2769                 switch (h->sei_pic_struct)
2770                 {
2771                 case SEI_PIC_STRUCT_FRAME:
2772                     break;
2773                 case SEI_PIC_STRUCT_TOP_FIELD:
2774                 case SEI_PIC_STRUCT_BOTTOM_FIELD:
2775                     cur->interlaced_frame = 1;
2776                     break;
2777                 case SEI_PIC_STRUCT_TOP_BOTTOM:
2778                 case SEI_PIC_STRUCT_BOTTOM_TOP:
2779                     if (FIELD_OR_MBAFF_PICTURE)
2780                         cur->interlaced_frame = 1;
2781                     else
2782                         // try to flag soft telecine progressive
2783                         cur->interlaced_frame = h->prev_interlaced_frame;
2784                     break;
2785                 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
2786                 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
2787                     // Signal the possibility of telecined film externally (pic_struct 5,6)
2788                     // From these hints, let the applications decide if they apply deinterlacing.
2789                     cur->repeat_pict = 1;
2790                     break;
2791                 case SEI_PIC_STRUCT_FRAME_DOUBLING:
2792                     // Force progressive here, as doubling interlaced frame is a bad idea.
2793                     cur->repeat_pict = 2;
2794                     break;
2795                 case SEI_PIC_STRUCT_FRAME_TRIPLING:
2796                     cur->repeat_pict = 4;
2797                     break;
2798                 }
2799
2800                 if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
2801                     cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
2802             }else{
2803                 /* Derive interlacing flag from used decoding process. */
2804                 cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
2805             }
2806             h->prev_interlaced_frame = cur->interlaced_frame;
2807
2808             if (cur->field_poc[0] != cur->field_poc[1]){
2809                 /* Derive top_field_first from field pocs. */
2810                 cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
2811             }else{
2812                 if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
2813                     /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
2814                     if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
2815                       || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
2816                         cur->top_field_first = 1;
2817                     else
2818                         cur->top_field_first = 0;
2819                 }else{
2820                     /* Most likely progressive */
2821                     cur->top_field_first = 0;
2822                 }
2823             }
2824
2825         //FIXME do something with unavailable reference frames
2826
2827             /* Sort B-frames into display order */
2828
2829             if(h->sps.bitstream_restriction_flag
2830                && s->avctx->has_b_frames < h->sps.num_reorder_frames){
2831                 s->avctx->has_b_frames = h->sps.num_reorder_frames;
2832                 s->low_delay = 0;
2833             }
2834
2835             if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
2836                && !h->sps.bitstream_restriction_flag){
2837                 s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
2838                 s->low_delay= 0;
2839             }
2840
2841             pics = 0;
2842             while(h->delayed_pic[pics]) pics++;
2843
2844             assert(pics <= MAX_DELAYED_PIC_COUNT);
2845
2846             h->delayed_pic[pics++] = cur;
2847             if(cur->reference == 0)
2848                 cur->reference = DELAYED_PIC_REF;
2849
2850             out = h->delayed_pic[0];
2851             out_idx = 0;
2852             for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2853                 if(h->delayed_pic[i]->poc < out->poc){
2854                     out = h->delayed_pic[i];
2855                     out_idx = i;
2856                 }
2857             if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
2858                 h->outputed_poc= INT_MIN;
2859             out_of_order = out->poc < h->outputed_poc;
2860
2861             if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
2862                 { }
2863             else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
2864                || (s->low_delay &&
2865                 ((h->outputed_poc != INT_MIN && out->poc > h->outputed_poc + 2)
2866                  || cur->pict_type == FF_B_TYPE)))
2867             {
2868                 s->low_delay = 0;
2869                 s->avctx->has_b_frames++;
2870             }
2871
2872             if(out_of_order || pics > s->avctx->has_b_frames){
2873                 out->reference &= ~DELAYED_PIC_REF;
2874                 for(i=out_idx; h->delayed_pic[i]; i++)
2875                     h->delayed_pic[i] = h->delayed_pic[i+1];
2876             }
2877             if(!out_of_order && pics > s->avctx->has_b_frames){
2878                 *data_size = sizeof(AVFrame);
2879
2880                 if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
2881                     h->outputed_poc = INT_MIN;
2882                 } else
2883                     h->outputed_poc = out->poc;
2884                 *pict= *(AVFrame*)out;
2885             }else{
2886                 av_log(avctx, AV_LOG_DEBUG, "no picture\n");
2887             }
2888         }
2889     }
2890
2891     assert(pict->data[0] || !*data_size);
2892     ff_print_debug_info(s, pict);
2893 //printf("out %d\n", (int)pict->data[0]);
2894
2895     return get_consumed_bytes(s, buf_index, buf_size);
2896 }
2897 #if 0
2898 static inline void fill_mb_avail(H264Context *h){
2899     MpegEncContext * const s = &h->s;
2900     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2901
2902     if(s->mb_y){
2903         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
2904         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
2905         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
2906     }else{
2907         h->mb_avail[0]=
2908         h->mb_avail[1]=
2909         h->mb_avail[2]= 0;
2910     }
2911     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
2912     h->mb_avail[4]= 1; //FIXME move out
2913     h->mb_avail[5]= 0; //FIXME move out
2914 }
2915 #endif
2916
2917 #ifdef TEST
2918 #undef printf
2919 #undef random
2920 #define COUNT 8000
2921 #define SIZE (COUNT*40)
2922 int main(void){
2923     int i;
2924     uint8_t temp[SIZE];
2925     PutBitContext pb;
2926     GetBitContext gb;
2927 //    int int_temp[10000];
2928     DSPContext dsp;
2929     AVCodecContext avctx;
2930
2931     dsputil_init(&dsp, &avctx);
2932
2933     init_put_bits(&pb, temp, SIZE);
2934     printf("testing unsigned exp golomb\n");
2935     for(i=0; i<COUNT; i++){
2936         START_TIMER
2937         set_ue_golomb(&pb, i);
2938         STOP_TIMER("set_ue_golomb");
2939     }
2940     flush_put_bits(&pb);
2941
2942     init_get_bits(&gb, temp, 8*SIZE);
2943     for(i=0; i<COUNT; i++){
2944         int j, s;
2945
2946         s= show_bits(&gb, 24);
2947
2948         START_TIMER
2949         j= get_ue_golomb(&gb);
2950         if(j != i){
2951             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
2952 //            return -1;
2953         }
2954         STOP_TIMER("get_ue_golomb");
2955     }
2956
2957
2958     init_put_bits(&pb, temp, SIZE);
2959     printf("testing signed exp golomb\n");
2960     for(i=0; i<COUNT; i++){
2961         START_TIMER
2962         set_se_golomb(&pb, i - COUNT/2);
2963         STOP_TIMER("set_se_golomb");
2964     }
2965     flush_put_bits(&pb);
2966
2967     init_get_bits(&gb, temp, 8*SIZE);
2968     for(i=0; i<COUNT; i++){
2969         int j, s;
2970
2971         s= show_bits(&gb, 24);
2972
2973         START_TIMER
2974         j= get_se_golomb(&gb);
2975         if(j != i - COUNT/2){
2976             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
2977 //            return -1;
2978         }
2979         STOP_TIMER("get_se_golomb");
2980     }
2981
2982 #if 0
2983     printf("testing 4x4 (I)DCT\n");
2984
2985     DCTELEM block[16];
2986     uint8_t src[16], ref[16];
2987     uint64_t error= 0, max_error=0;
2988
2989     for(i=0; i<COUNT; i++){
2990         int j;
2991 //        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
2992         for(j=0; j<16; j++){
2993             ref[j]= random()%255;
2994             src[j]= random()%255;
2995         }
2996
2997         h264_diff_dct_c(block, src, ref, 4);
2998
2999         //normalize
3000         for(j=0; j<16; j++){
3001 //            printf("%d ", block[j]);
3002             block[j]= block[j]*4;
3003             if(j&1) block[j]= (block[j]*4 + 2)/5;
3004             if(j&4) block[j]= (block[j]*4 + 2)/5;
3005         }
3006 //        printf("\n");
3007
3008         s->dsp.h264_idct_add(ref, block, 4);
3009 /*        for(j=0; j<16; j++){
3010             printf("%d ", ref[j]);
3011         }
3012         printf("\n");*/
3013
3014         for(j=0; j<16; j++){
3015             int diff= FFABS(src[j] - ref[j]);
3016
3017             error+= diff*diff;
3018             max_error= FFMAX(max_error, diff);
3019         }
3020     }
3021     printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
3022     printf("testing quantizer\n");
3023     for(qp=0; qp<52; qp++){
3024         for(i=0; i<16; i++)
3025             src1_block[i]= src2_block[i]= random()%255;
3026
3027     }
3028     printf("Testing NAL layer\n");
3029
3030     uint8_t bitstream[COUNT];
3031     uint8_t nal[COUNT*2];
3032     H264Context h;
3033     memset(&h, 0, sizeof(H264Context));
3034
3035     for(i=0; i<COUNT; i++){
3036         int zeros= i;
3037         int nal_length;
3038         int consumed;
3039         int out_length;
3040         uint8_t *out;
3041         int j;
3042
3043         for(j=0; j<COUNT; j++){
3044             bitstream[j]= (random() % 255) + 1;
3045         }
3046
3047         for(j=0; j<zeros; j++){
3048             int pos= random() % COUNT;
3049             while(bitstream[pos] == 0){
3050                 pos++;
3051                 pos %= COUNT;
3052             }
3053             bitstream[pos]=0;
3054         }
3055
3056         START_TIMER
3057
3058         nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
3059         if(nal_length<0){
3060             printf("encoding failed\n");
3061             return -1;
3062         }
3063
3064         out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
3065
3066         STOP_TIMER("NAL")
3067
3068         if(out_length != COUNT){
3069             printf("incorrect length %d %d\n", out_length, COUNT);
3070             return -1;
3071         }
3072
3073         if(consumed != nal_length){
3074             printf("incorrect consumed length %d %d\n", nal_length, consumed);
3075             return -1;
3076         }
3077
3078         if(memcmp(bitstream, out, COUNT)){
3079             printf("mismatch\n");
3080             return -1;
3081         }
3082     }
3083 #endif
3084
3085     printf("Testing RBSP\n");
3086
3087
3088     return 0;
3089 }
3090 #endif /* TEST */
3091
3092
3093 av_cold void ff_h264_free_context(H264Context *h)
3094 {
3095     int i;
3096
3097     free_tables(h); //FIXME cleanup init stuff perhaps
3098
3099     for(i = 0; i < MAX_SPS_COUNT; i++)
3100         av_freep(h->sps_buffers + i);
3101
3102     for(i = 0; i < MAX_PPS_COUNT; i++)
3103         av_freep(h->pps_buffers + i);
3104 }
3105
3106 av_cold int ff_h264_decode_end(AVCodecContext *avctx)
3107 {
3108     H264Context *h = avctx->priv_data;
3109     MpegEncContext *s = &h->s;
3110
3111     ff_h264_free_context(h);
3112
3113     MPV_common_end(s);
3114
3115 //    memset(h, 0, sizeof(H264Context));
3116
3117     return 0;
3118 }
3119
3120
3121 AVCodec h264_decoder = {
3122     "h264",
3123     CODEC_TYPE_VIDEO,
3124     CODEC_ID_H264,
3125     sizeof(H264Context),
3126     ff_h264_decode_init,
3127     NULL,
3128     ff_h264_decode_end,
3129     decode_frame,
3130     /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3131     .flush= flush_dpb,
3132     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
3133     .pix_fmts= ff_hwaccel_pixfmt_list_420,
3134 };
3135
3136 #if CONFIG_H264_VDPAU_DECODER
3137 AVCodec h264_vdpau_decoder = {
3138     "h264_vdpau",
3139     CODEC_TYPE_VIDEO,
3140     CODEC_ID_H264,
3141     sizeof(H264Context),
3142     ff_h264_decode_init,
3143     NULL,
3144     ff_h264_decode_end,
3145     decode_frame,
3146     CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3147     .flush= flush_dpb,
3148     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
3149     .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
3150 };
3151 #endif