604c29e273c74f060d693cdb9034d43a95200c98
[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     const int mb_xy= h->mb_xy;
56
57     h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
58     h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
59     h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
60     h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
61     h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
62     h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
63     h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
64 }
65
66 /**
67  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
68  */
69 int ff_h264_check_intra4x4_pred_mode(H264Context *h){
70     MpegEncContext * const s = &h->s;
71     static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
72     static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
73     int i;
74
75     if(!(h->top_samples_available&0x8000)){
76         for(i=0; i<4; i++){
77             int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
78             if(status<0){
79                 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);
80                 return -1;
81             } else if(status){
82                 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
83             }
84         }
85     }
86
87     if((h->left_samples_available&0x8888)!=0x8888){
88         static const int mask[4]={0x8000,0x2000,0x80,0x20};
89         for(i=0; i<4; i++){
90             if(!(h->left_samples_available&mask[i])){
91                 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
92                 if(status<0){
93                     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);
94                     return -1;
95                 } else if(status){
96                     h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
97                 }
98             }
99         }
100     }
101
102     return 0;
103 } //FIXME cleanup like ff_h264_check_intra_pred_mode
104
105 /**
106  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
107  */
108 int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
109     MpegEncContext * const s = &h->s;
110     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
111     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
112
113     if(mode > 6U) {
114         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);
115         return -1;
116     }
117
118     if(!(h->top_samples_available&0x8000)){
119         mode= top[ mode ];
120         if(mode<0){
121             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);
122             return -1;
123         }
124     }
125
126     if((h->left_samples_available&0x8080) != 0x8080){
127         mode= left[ mode ];
128         if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
129             mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
130         }
131         if(mode<0){
132             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);
133             return -1;
134         }
135     }
136
137     return mode;
138 }
139
140 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
141     int i, si, di;
142     uint8_t *dst;
143     int bufidx;
144
145 //    src[0]&0x80;                //forbidden bit
146     h->nal_ref_idc= src[0]>>5;
147     h->nal_unit_type= src[0]&0x1F;
148
149     src++; length--;
150 #if 0
151     for(i=0; i<length; i++)
152         printf("%2X ", src[i]);
153 #endif
154
155 #if HAVE_FAST_UNALIGNED
156 # if HAVE_FAST_64BIT
157 #   define RS 7
158     for(i=0; i+1<length; i+=9){
159         if(!((~*(const uint64_t*)(src+i) & (*(const uint64_t*)(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
160 # else
161 #   define RS 3
162     for(i=0; i+1<length; i+=5){
163         if(!((~*(const uint32_t*)(src+i) & (*(const uint32_t*)(src+i) - 0x01000101U)) & 0x80008080U))
164 # endif
165             continue;
166         if(i>0 && !src[i]) i--;
167         while(src[i]) i++;
168 #else
169 #   define RS 0
170     for(i=0; i+1<length; i+=2){
171         if(src[i]) continue;
172         if(i>0 && src[i-1]==0) i--;
173 #endif
174         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
175             if(src[i+2]!=3){
176                 /* startcode, so we must be past the end */
177                 length=i;
178             }
179             break;
180         }
181         i-= RS;
182     }
183
184     if(i>=length-1){ //no escaped 0
185         *dst_length= length;
186         *consumed= length+1; //+1 for the header
187         return src;
188     }
189
190     bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
191     av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
192     dst= h->rbsp_buffer[bufidx];
193
194     if (dst == NULL){
195         return NULL;
196     }
197
198 //printf("decoding esc\n");
199     memcpy(dst, src, i);
200     si=di=i;
201     while(si+2<length){
202         //remove escapes (very rare 1:2^22)
203         if(src[si+2]>3){
204             dst[di++]= src[si++];
205             dst[di++]= src[si++];
206         }else if(src[si]==0 && src[si+1]==0){
207             if(src[si+2]==3){ //escape
208                 dst[di++]= 0;
209                 dst[di++]= 0;
210                 si+=3;
211                 continue;
212             }else //next start code
213                 goto nsc;
214         }
215
216         dst[di++]= src[si++];
217     }
218     while(si<length)
219         dst[di++]= src[si++];
220 nsc:
221
222     memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
223
224     *dst_length= di;
225     *consumed= si + 1;//+1 for the header
226 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
227     return dst;
228 }
229
230 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
231     int v= *src;
232     int r;
233
234     tprintf(h->s.avctx, "rbsp trailing %X\n", v);
235
236     for(r=1; r<9; r++){
237         if(v&1) return r;
238         v>>=1;
239     }
240     return 0;
241 }
242
243 /**
244  * IDCT transforms the 16 dc values and dequantizes them.
245  * @param qp quantization parameter
246  */
247 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
248 #define stride 16
249     int i;
250     int temp[16]; //FIXME check if this is a good idea
251     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
252     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
253
254 //memset(block, 64, 2*256);
255 //return;
256     for(i=0; i<4; i++){
257         const int offset= y_offset[i];
258         const int z0= block[offset+stride*0] + block[offset+stride*4];
259         const int z1= block[offset+stride*0] - block[offset+stride*4];
260         const int z2= block[offset+stride*1] - block[offset+stride*5];
261         const int z3= block[offset+stride*1] + block[offset+stride*5];
262
263         temp[4*i+0]= z0+z3;
264         temp[4*i+1]= z1+z2;
265         temp[4*i+2]= z1-z2;
266         temp[4*i+3]= z0-z3;
267     }
268
269     for(i=0; i<4; i++){
270         const int offset= x_offset[i];
271         const int z0= temp[4*0+i] + temp[4*2+i];
272         const int z1= temp[4*0+i] - temp[4*2+i];
273         const int z2= temp[4*1+i] - temp[4*3+i];
274         const int z3= temp[4*1+i] + temp[4*3+i];
275
276         block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
277         block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
278         block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
279         block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
280     }
281 }
282
283 #if 0
284 /**
285  * DCT transforms the 16 dc values.
286  * @param qp quantization parameter ??? FIXME
287  */
288 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
289 //    const int qmul= dequant_coeff[qp][0];
290     int i;
291     int temp[16]; //FIXME check if this is a good idea
292     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
293     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
294
295     for(i=0; i<4; i++){
296         const int offset= y_offset[i];
297         const int z0= block[offset+stride*0] + block[offset+stride*4];
298         const int z1= block[offset+stride*0] - block[offset+stride*4];
299         const int z2= block[offset+stride*1] - block[offset+stride*5];
300         const int z3= block[offset+stride*1] + block[offset+stride*5];
301
302         temp[4*i+0]= z0+z3;
303         temp[4*i+1]= z1+z2;
304         temp[4*i+2]= z1-z2;
305         temp[4*i+3]= z0-z3;
306     }
307
308     for(i=0; i<4; i++){
309         const int offset= x_offset[i];
310         const int z0= temp[4*0+i] + temp[4*2+i];
311         const int z1= temp[4*0+i] - temp[4*2+i];
312         const int z2= temp[4*1+i] - temp[4*3+i];
313         const int z3= temp[4*1+i] + temp[4*3+i];
314
315         block[stride*0 +offset]= (z0 + z3)>>1;
316         block[stride*2 +offset]= (z1 + z2)>>1;
317         block[stride*8 +offset]= (z1 - z2)>>1;
318         block[stride*10+offset]= (z0 - z3)>>1;
319     }
320 }
321 #endif
322
323 #undef xStride
324 #undef stride
325
326 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
327     const int stride= 16*2;
328     const int xStride= 16;
329     int a,b,c,d,e;
330
331     a= block[stride*0 + xStride*0];
332     b= block[stride*0 + xStride*1];
333     c= block[stride*1 + xStride*0];
334     d= block[stride*1 + xStride*1];
335
336     e= a-b;
337     a= a+b;
338     b= c-d;
339     c= c+d;
340
341     block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
342     block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
343     block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
344     block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
345 }
346
347 #if 0
348 static void chroma_dc_dct_c(DCTELEM *block){
349     const int stride= 16*2;
350     const int xStride= 16;
351     int a,b,c,d,e;
352
353     a= block[stride*0 + xStride*0];
354     b= block[stride*0 + xStride*1];
355     c= block[stride*1 + xStride*0];
356     d= block[stride*1 + xStride*1];
357
358     e= a-b;
359     a= a+b;
360     b= c-d;
361     c= c+d;
362
363     block[stride*0 + xStride*0]= (a+c);
364     block[stride*0 + xStride*1]= (e+b);
365     block[stride*1 + xStride*0]= (a-c);
366     block[stride*1 + xStride*1]= (e-b);
367 }
368 #endif
369
370 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
371                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
372                            int src_x_offset, int src_y_offset,
373                            qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
374     MpegEncContext * const s = &h->s;
375     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
376     int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
377     const int luma_xy= (mx&3) + ((my&3)<<2);
378     uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
379     uint8_t * src_cb, * src_cr;
380     int extra_width= h->emu_edge_width;
381     int extra_height= h->emu_edge_height;
382     int emu=0;
383     const int full_mx= mx>>2;
384     const int full_my= my>>2;
385     const int pic_width  = 16*s->mb_width;
386     const int pic_height = 16*s->mb_height >> MB_FIELD;
387
388     if(mx&7) extra_width -= 3;
389     if(my&7) extra_height -= 3;
390
391     if(   full_mx < 0-extra_width
392        || full_my < 0-extra_height
393        || full_mx + 16/*FIXME*/ > pic_width + extra_width
394        || full_my + 16/*FIXME*/ > pic_height + extra_height){
395         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);
396             src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
397         emu=1;
398     }
399
400     qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
401     if(!square){
402         qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
403     }
404
405     if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
406
407     if(MB_FIELD){
408         // chroma offset when predicting from a field of opposite parity
409         my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
410         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
411     }
412     src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
413     src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
414
415     if(emu){
416         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);
417             src_cb= s->edge_emu_buffer;
418     }
419     chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
420
421     if(emu){
422         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);
423             src_cr= s->edge_emu_buffer;
424     }
425     chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
426 }
427
428 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
429                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
430                            int x_offset, int y_offset,
431                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
432                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
433                            int list0, int list1){
434     MpegEncContext * const s = &h->s;
435     qpel_mc_func *qpix_op=  qpix_put;
436     h264_chroma_mc_func chroma_op= chroma_put;
437
438     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
439     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
440     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
441     x_offset += 8*s->mb_x;
442     y_offset += 8*(s->mb_y >> MB_FIELD);
443
444     if(list0){
445         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
446         mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
447                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
448                            qpix_op, chroma_op);
449
450         qpix_op=  qpix_avg;
451         chroma_op= chroma_avg;
452     }
453
454     if(list1){
455         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
456         mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
457                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
458                            qpix_op, chroma_op);
459     }
460 }
461
462 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
463                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
464                            int x_offset, int y_offset,
465                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
466                            h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
467                            h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
468                            int list0, int list1){
469     MpegEncContext * const s = &h->s;
470
471     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
472     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
473     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
474     x_offset += 8*s->mb_x;
475     y_offset += 8*(s->mb_y >> MB_FIELD);
476
477     if(list0 && list1){
478         /* don't optimize for luma-only case, since B-frames usually
479          * use implicit weights => chroma too. */
480         uint8_t *tmp_cb = s->obmc_scratchpad;
481         uint8_t *tmp_cr = s->obmc_scratchpad + 8;
482         uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
483         int refn0 = h->ref_cache[0][ scan8[n] ];
484         int refn1 = h->ref_cache[1][ scan8[n] ];
485
486         mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
487                     dest_y, dest_cb, dest_cr,
488                     x_offset, y_offset, qpix_put, chroma_put);
489         mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
490                     tmp_y, tmp_cb, tmp_cr,
491                     x_offset, y_offset, qpix_put, chroma_put);
492
493         if(h->use_weight == 2){
494             int weight0 = h->implicit_weight[refn0][refn1];
495             int weight1 = 64 - weight0;
496             luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
497             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
498             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
499         }else{
500             luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
501                             h->luma_weight[0][refn0], h->luma_weight[1][refn1],
502                             h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
503             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
504                             h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
505                             h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
506             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
507                             h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
508                             h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
509         }
510     }else{
511         int list = list1 ? 1 : 0;
512         int refn = h->ref_cache[list][ scan8[n] ];
513         Picture *ref= &h->ref_list[list][refn];
514         mc_dir_part(h, ref, n, square, chroma_height, delta, list,
515                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
516                     qpix_put, chroma_put);
517
518         luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
519                        h->luma_weight[list][refn], h->luma_offset[list][refn]);
520         if(h->use_weight_chroma){
521             chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
522                              h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
523             chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
524                              h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
525         }
526     }
527 }
528
529 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
530                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
531                            int x_offset, int y_offset,
532                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
533                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
534                            h264_weight_func *weight_op, h264_biweight_func *weight_avg,
535                            int list0, int list1){
536     if((h->use_weight==2 && list0 && list1
537         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
538        || h->use_weight==1)
539         mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
540                          x_offset, y_offset, qpix_put, chroma_put,
541                          weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
542     else
543         mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
544                     x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
545 }
546
547 static inline void prefetch_motion(H264Context *h, int list){
548     /* fetch pixels for estimated mv 4 macroblocks ahead
549      * optimized for 64byte cache lines */
550     MpegEncContext * const s = &h->s;
551     const int refn = h->ref_cache[list][scan8[0]];
552     if(refn >= 0){
553         const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
554         const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
555         uint8_t **src= h->ref_list[list][refn].data;
556         int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
557         s->dsp.prefetch(src[0]+off, s->linesize, 4);
558         off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
559         s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
560     }
561 }
562
563 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
564                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
565                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
566                       h264_weight_func *weight_op, h264_biweight_func *weight_avg){
567     MpegEncContext * const s = &h->s;
568     const int mb_xy= h->mb_xy;
569     const int mb_type= s->current_picture.mb_type[mb_xy];
570
571     assert(IS_INTER(mb_type));
572
573     prefetch_motion(h, 0);
574
575     if(IS_16X16(mb_type)){
576         mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
577                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
578                 weight_op, weight_avg,
579                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
580     }else if(IS_16X8(mb_type)){
581         mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
582                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
583                 &weight_op[1], &weight_avg[1],
584                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
585         mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
586                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
587                 &weight_op[1], &weight_avg[1],
588                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
589     }else if(IS_8X16(mb_type)){
590         mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
591                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
592                 &weight_op[2], &weight_avg[2],
593                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
594         mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
595                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
596                 &weight_op[2], &weight_avg[2],
597                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
598     }else{
599         int i;
600
601         assert(IS_8X8(mb_type));
602
603         for(i=0; i<4; i++){
604             const int sub_mb_type= h->sub_mb_type[i];
605             const int n= 4*i;
606             int x_offset= (i&1)<<2;
607             int y_offset= (i&2)<<1;
608
609             if(IS_SUB_8X8(sub_mb_type)){
610                 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
611                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
612                     &weight_op[3], &weight_avg[3],
613                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
614             }else if(IS_SUB_8X4(sub_mb_type)){
615                 mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
616                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
617                     &weight_op[4], &weight_avg[4],
618                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
619                 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
620                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
621                     &weight_op[4], &weight_avg[4],
622                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
623             }else if(IS_SUB_4X8(sub_mb_type)){
624                 mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
625                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
626                     &weight_op[5], &weight_avg[5],
627                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
628                 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
629                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
630                     &weight_op[5], &weight_avg[5],
631                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
632             }else{
633                 int j;
634                 assert(IS_SUB_4X4(sub_mb_type));
635                 for(j=0; j<4; j++){
636                     int sub_x_offset= x_offset + 2*(j&1);
637                     int sub_y_offset= y_offset +   (j&2);
638                     mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
639                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
640                         &weight_op[6], &weight_avg[6],
641                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
642                 }
643             }
644         }
645     }
646
647     prefetch_motion(h, 1);
648 }
649
650
651 static void free_tables(H264Context *h){
652     int i;
653     H264Context *hx;
654     av_freep(&h->intra4x4_pred_mode);
655     av_freep(&h->chroma_pred_mode_table);
656     av_freep(&h->cbp_table);
657     av_freep(&h->mvd_table[0]);
658     av_freep(&h->mvd_table[1]);
659     av_freep(&h->direct_table);
660     av_freep(&h->non_zero_count);
661     av_freep(&h->slice_table_base);
662     h->slice_table= NULL;
663     av_freep(&h->list_counts);
664
665     av_freep(&h->mb2b_xy);
666     av_freep(&h->mb2b8_xy);
667
668     for(i = 0; i < MAX_THREADS; i++) {
669         hx = h->thread_context[i];
670         if(!hx) continue;
671         av_freep(&hx->top_borders[1]);
672         av_freep(&hx->top_borders[0]);
673         av_freep(&hx->s.obmc_scratchpad);
674         av_freep(&hx->rbsp_buffer[1]);
675         av_freep(&hx->rbsp_buffer[0]);
676         hx->rbsp_buffer_size[0] = 0;
677         hx->rbsp_buffer_size[1] = 0;
678         if (i) av_freep(&h->thread_context[i]);
679     }
680 }
681
682 static void init_dequant8_coeff_table(H264Context *h){
683     int i,q,x;
684     const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
685     h->dequant8_coeff[0] = h->dequant8_buffer[0];
686     h->dequant8_coeff[1] = h->dequant8_buffer[1];
687
688     for(i=0; i<2; i++ ){
689         if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
690             h->dequant8_coeff[1] = h->dequant8_buffer[0];
691             break;
692         }
693
694         for(q=0; q<52; q++){
695             int shift = div6[q];
696             int idx = rem6[q];
697             for(x=0; x<64; x++)
698                 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
699                     ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
700                     h->pps.scaling_matrix8[i][x]) << shift;
701         }
702     }
703 }
704
705 static void init_dequant4_coeff_table(H264Context *h){
706     int i,j,q,x;
707     const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
708     for(i=0; i<6; i++ ){
709         h->dequant4_coeff[i] = h->dequant4_buffer[i];
710         for(j=0; j<i; j++){
711             if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
712                 h->dequant4_coeff[i] = h->dequant4_buffer[j];
713                 break;
714             }
715         }
716         if(j<i)
717             continue;
718
719         for(q=0; q<52; q++){
720             int shift = div6[q] + 2;
721             int idx = rem6[q];
722             for(x=0; x<16; x++)
723                 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
724                     ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
725                     h->pps.scaling_matrix4[i][x]) << shift;
726         }
727     }
728 }
729
730 static void init_dequant_tables(H264Context *h){
731     int i,x;
732     init_dequant4_coeff_table(h);
733     if(h->pps.transform_8x8_mode)
734         init_dequant8_coeff_table(h);
735     if(h->sps.transform_bypass){
736         for(i=0; i<6; i++)
737             for(x=0; x<16; x++)
738                 h->dequant4_coeff[i][0][x] = 1<<6;
739         if(h->pps.transform_8x8_mode)
740             for(i=0; i<2; i++)
741                 for(x=0; x<64; x++)
742                     h->dequant8_coeff[i][0][x] = 1<<6;
743     }
744 }
745
746
747 int ff_h264_alloc_tables(H264Context *h){
748     MpegEncContext * const s = &h->s;
749     const int big_mb_num= s->mb_stride * (s->mb_height+1);
750     int x,y;
751
752     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t), fail)
753
754     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 32 * sizeof(uint8_t), fail)
755     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
756     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
757
758     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
759     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t), fail);
760     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t), fail);
761     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 32*big_mb_num * sizeof(uint8_t) , fail);
762     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
763
764     memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
765     h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
766
767     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy  , big_mb_num * sizeof(uint32_t), fail);
768     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b8_xy , big_mb_num * sizeof(uint32_t), fail);
769     for(y=0; y<s->mb_height; y++){
770         for(x=0; x<s->mb_width; x++){
771             const int mb_xy= x + y*s->mb_stride;
772             const int b_xy = 4*x + 4*y*h->b_stride;
773             const int b8_xy= 2*x + 2*y*h->b8_stride;
774
775             h->mb2b_xy [mb_xy]= b_xy;
776             h->mb2b8_xy[mb_xy]= b8_xy;
777         }
778     }
779
780     s->obmc_scratchpad = NULL;
781
782     if(!h->dequant4_coeff[0])
783         init_dequant_tables(h);
784
785     return 0;
786 fail:
787     free_tables(h);
788     return -1;
789 }
790
791 /**
792  * Mimic alloc_tables(), but for every context thread.
793  */
794 static void clone_tables(H264Context *dst, H264Context *src){
795     dst->intra4x4_pred_mode       = src->intra4x4_pred_mode;
796     dst->non_zero_count           = src->non_zero_count;
797     dst->slice_table              = src->slice_table;
798     dst->cbp_table                = src->cbp_table;
799     dst->mb2b_xy                  = src->mb2b_xy;
800     dst->mb2b8_xy                 = src->mb2b8_xy;
801     dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
802     dst->mvd_table[0]             = src->mvd_table[0];
803     dst->mvd_table[1]             = src->mvd_table[1];
804     dst->direct_table             = src->direct_table;
805     dst->list_counts              = src->list_counts;
806
807     dst->s.obmc_scratchpad = NULL;
808     ff_h264_pred_init(&dst->hpc, src->s.codec_id);
809 }
810
811 /**
812  * Init context
813  * Allocate buffers which are not shared amongst multiple threads.
814  */
815 static int context_init(H264Context *h){
816     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
817     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
818
819     return 0;
820 fail:
821     return -1; // free_tables will clean up for us
822 }
823
824 static av_cold void common_init(H264Context *h){
825     MpegEncContext * const s = &h->s;
826
827     s->width = s->avctx->width;
828     s->height = s->avctx->height;
829     s->codec_id= s->avctx->codec->id;
830
831     ff_h264_pred_init(&h->hpc, s->codec_id);
832
833     h->dequant_coeff_pps= -1;
834     s->unrestricted_mv=1;
835     s->decode=1; //FIXME
836
837     dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
838
839     memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
840     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
841 }
842
843 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
844     H264Context *h= avctx->priv_data;
845     MpegEncContext * const s = &h->s;
846
847     MPV_decode_defaults(s);
848
849     s->avctx = avctx;
850     common_init(h);
851
852     s->out_format = FMT_H264;
853     s->workaround_bugs= avctx->workaround_bugs;
854
855     // set defaults
856 //    s->decode_mb= ff_h263_decode_mb;
857     s->quarter_sample = 1;
858     if(!avctx->has_b_frames)
859     s->low_delay= 1;
860
861     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
862
863     ff_h264_decode_init_vlc();
864
865     if(avctx->extradata_size > 0 && avctx->extradata &&
866        *(char *)avctx->extradata == 1){
867         h->is_avc = 1;
868         h->got_avcC = 0;
869     } else {
870         h->is_avc = 0;
871     }
872
873     h->thread_context[0] = h;
874     h->outputed_poc = INT_MIN;
875     h->prev_poc_msb= 1<<16;
876     h->x264_build = -1;
877     ff_h264_reset_sei(h);
878     if(avctx->codec_id == CODEC_ID_H264){
879         if(avctx->ticks_per_frame == 1){
880             s->avctx->time_base.den *=2;
881         }
882         avctx->ticks_per_frame = 2;
883     }
884     return 0;
885 }
886
887 int ff_h264_frame_start(H264Context *h){
888     MpegEncContext * const s = &h->s;
889     int i;
890
891     if(MPV_frame_start(s, s->avctx) < 0)
892         return -1;
893     ff_er_frame_start(s);
894     /*
895      * MPV_frame_start uses pict_type to derive key_frame.
896      * This is incorrect for H.264; IDR markings must be used.
897      * Zero here; IDR markings per slice in frame or fields are ORed in later.
898      * See decode_nal_units().
899      */
900     s->current_picture_ptr->key_frame= 0;
901     s->current_picture_ptr->mmco_reset= 0;
902
903     assert(s->linesize && s->uvlinesize);
904
905     for(i=0; i<16; i++){
906         h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
907         h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
908     }
909     for(i=0; i<4; i++){
910         h->block_offset[16+i]=
911         h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
912         h->block_offset[24+16+i]=
913         h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
914     }
915
916     /* can't be in alloc_tables because linesize isn't known there.
917      * FIXME: redo bipred weight to not require extra buffer? */
918     for(i = 0; i < s->avctx->thread_count; i++)
919         if(!h->thread_context[i]->s.obmc_scratchpad)
920             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
921
922     /* some macroblocks will be accessed before they're available */
923     if(FRAME_MBAFF || s->avctx->thread_count > 1)
924         memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
925
926 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
927
928     // We mark the current picture as non-reference after allocating it, so
929     // that if we break out due to an error it can be released automatically
930     // in the next MPV_frame_start().
931     // SVQ3 as well as most other codecs have only last/next/current and thus
932     // get released even with set reference, besides SVQ3 and others do not
933     // mark frames as reference later "naturally".
934     if(s->codec_id != CODEC_ID_SVQ3)
935         s->current_picture_ptr->reference= 0;
936
937     s->current_picture_ptr->field_poc[0]=
938     s->current_picture_ptr->field_poc[1]= INT_MAX;
939     assert(s->current_picture_ptr->long_ref==0);
940
941     return 0;
942 }
943
944 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){
945     MpegEncContext * const s = &h->s;
946     uint8_t *top_border;
947     int top_idx = 1;
948
949     src_y  -=   linesize;
950     src_cb -= uvlinesize;
951     src_cr -= uvlinesize;
952
953     if(!simple && FRAME_MBAFF){
954         if(s->mb_y&1){
955             if(!MB_MBAFF){
956                 top_border = h->top_borders[0][s->mb_x];
957                 AV_COPY128(top_border, src_y + 15*linesize);
958                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
959                     AV_COPY64(top_border+16, src_cb+7*uvlinesize);
960                     AV_COPY64(top_border+24, src_cr+7*uvlinesize);
961                 }
962             }
963         }else if(MB_MBAFF){
964             top_idx = 0;
965         }else
966             return;
967     }
968
969     top_border = h->top_borders[top_idx][s->mb_x];
970     // There are two lines saved, the line above the the top macroblock of a pair,
971     // and the line above the bottom macroblock
972     AV_COPY128(top_border, src_y + 16*linesize);
973
974     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
975         AV_COPY64(top_border+16, src_cb+8*uvlinesize);
976         AV_COPY64(top_border+24, src_cr+8*uvlinesize);
977     }
978 }
979
980 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){
981     MpegEncContext * const s = &h->s;
982     int deblock_left;
983     int deblock_top;
984     int mb_xy;
985     int top_idx = 1;
986     uint8_t *top_border_m1;
987     uint8_t *top_border;
988
989     if(!simple && FRAME_MBAFF){
990         if(s->mb_y&1){
991             if(!MB_MBAFF)
992                 return;
993         }else{
994             top_idx = MB_MBAFF ? 0 : 1;
995         }
996     }
997
998     if(h->deblocking_filter == 2) {
999         mb_xy = h->mb_xy;
1000         deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
1001         deblock_top  = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
1002     } else {
1003         deblock_left = (s->mb_x > 0);
1004         deblock_top =  (s->mb_y > !!MB_FIELD);
1005     }
1006
1007     src_y  -=   linesize + 1;
1008     src_cb -= uvlinesize + 1;
1009     src_cr -= uvlinesize + 1;
1010
1011     top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1012     top_border    = h->top_borders[top_idx][s->mb_x];
1013
1014 #define XCHG(a,b,xchg)\
1015 if (xchg) AV_SWAP64(b,a);\
1016 else      AV_COPY64(b,a);
1017
1018     if(deblock_top){
1019         if(deblock_left){
1020             XCHG(top_border_m1+8, src_y -7, 1);
1021         }
1022         XCHG(top_border+0, src_y +1, xchg);
1023         XCHG(top_border+8, src_y +9, 1);
1024         if(s->mb_x+1 < s->mb_width){
1025             XCHG(h->top_borders[top_idx][s->mb_x+1], src_y +17, 1);
1026         }
1027     }
1028
1029     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1030         if(deblock_top){
1031             if(deblock_left){
1032                 XCHG(top_border_m1+16, src_cb -7, 1);
1033                 XCHG(top_border_m1+24, src_cr -7, 1);
1034             }
1035             XCHG(top_border+16, src_cb+1, 1);
1036             XCHG(top_border+24, src_cr+1, 1);
1037         }
1038     }
1039 }
1040
1041 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
1042     MpegEncContext * const s = &h->s;
1043     const int mb_x= s->mb_x;
1044     const int mb_y= s->mb_y;
1045     const int mb_xy= h->mb_xy;
1046     const int mb_type= s->current_picture.mb_type[mb_xy];
1047     uint8_t  *dest_y, *dest_cb, *dest_cr;
1048     int linesize, uvlinesize /*dct_offset*/;
1049     int i;
1050     int *block_offset = &h->block_offset[0];
1051     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1052     /* is_h264 should always be true if SVQ3 is disabled. */
1053     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1054     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1055     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1056
1057     dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
1058     dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
1059     dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
1060
1061     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
1062     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
1063
1064     h->list_counts[mb_xy]= h->list_count;
1065
1066     if (!simple && MB_FIELD) {
1067         linesize   = h->mb_linesize   = s->linesize * 2;
1068         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1069         block_offset = &h->block_offset[24];
1070         if(mb_y&1){ //FIXME move out of this function?
1071             dest_y -= s->linesize*15;
1072             dest_cb-= s->uvlinesize*7;
1073             dest_cr-= s->uvlinesize*7;
1074         }
1075         if(FRAME_MBAFF) {
1076             int list;
1077             for(list=0; list<h->list_count; list++){
1078                 if(!USES_LIST(mb_type, list))
1079                     continue;
1080                 if(IS_16X16(mb_type)){
1081                     int8_t *ref = &h->ref_cache[list][scan8[0]];
1082                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1083                 }else{
1084                     for(i=0; i<16; i+=4){
1085                         int ref = h->ref_cache[list][scan8[i]];
1086                         if(ref >= 0)
1087                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1088                     }
1089                 }
1090             }
1091         }
1092     } else {
1093         linesize   = h->mb_linesize   = s->linesize;
1094         uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1095 //        dct_offset = s->linesize * 16;
1096     }
1097
1098     if (!simple && IS_INTRA_PCM(mb_type)) {
1099         for (i=0; i<16; i++) {
1100             memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1101         }
1102         for (i=0; i<8; i++) {
1103             memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
1104             memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
1105         }
1106     } else {
1107         if(IS_INTRA(mb_type)){
1108             if(h->deblocking_filter)
1109                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
1110
1111             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1112                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1113                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1114             }
1115
1116             if(IS_INTRA4x4(mb_type)){
1117                 if(simple || !s->encoding){
1118                     if(IS_8x8DCT(mb_type)){
1119                         if(transform_bypass){
1120                             idct_dc_add =
1121                             idct_add    = s->dsp.add_pixels8;
1122                         }else{
1123                             idct_dc_add = s->dsp.h264_idct8_dc_add;
1124                             idct_add    = s->dsp.h264_idct8_add;
1125                         }
1126                         for(i=0; i<16; i+=4){
1127                             uint8_t * const ptr= dest_y + block_offset[i];
1128                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1129                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1130                                 h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
1131                             }else{
1132                                 const int nnz = h->non_zero_count_cache[ scan8[i] ];
1133                                 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1134                                                             (h->topright_samples_available<<i)&0x4000, linesize);
1135                                 if(nnz){
1136                                     if(nnz == 1 && h->mb[i*16])
1137                                         idct_dc_add(ptr, h->mb + i*16, linesize);
1138                                     else
1139                                         idct_add   (ptr, h->mb + i*16, linesize);
1140                                 }
1141                             }
1142                         }
1143                     }else{
1144                         if(transform_bypass){
1145                             idct_dc_add =
1146                             idct_add    = s->dsp.add_pixels4;
1147                         }else{
1148                             idct_dc_add = s->dsp.h264_idct_dc_add;
1149                             idct_add    = s->dsp.h264_idct_add;
1150                         }
1151                         for(i=0; i<16; i++){
1152                             uint8_t * const ptr= dest_y + block_offset[i];
1153                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1154
1155                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1156                                 h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
1157                             }else{
1158                                 uint8_t *topright;
1159                                 int nnz, tr;
1160                                 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1161                                     const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1162                                     assert(mb_y || linesize <= block_offset[i]);
1163                                     if(!topright_avail){
1164                                         tr= ptr[3 - linesize]*0x01010101;
1165                                         topright= (uint8_t*) &tr;
1166                                     }else
1167                                         topright= ptr + 4 - linesize;
1168                                 }else
1169                                     topright= NULL;
1170
1171                                 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1172                                 nnz = h->non_zero_count_cache[ scan8[i] ];
1173                                 if(nnz){
1174                                     if(is_h264){
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                                     }else
1180                                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1181                                 }
1182                             }
1183                         }
1184                     }
1185                 }
1186             }else{
1187                 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1188                 if(is_h264){
1189                     if(!transform_bypass)
1190                         h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
1191                 }else
1192                     ff_svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
1193             }
1194             if(h->deblocking_filter)
1195                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
1196         }else if(is_h264){
1197             hl_motion(h, dest_y, dest_cb, dest_cr,
1198                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1199                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1200                       s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
1201         }
1202
1203
1204         if(!IS_INTRA4x4(mb_type)){
1205             if(is_h264){
1206                 if(IS_INTRA16x16(mb_type)){
1207                     if(transform_bypass){
1208                         if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1209                             h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1210                         }else{
1211                             for(i=0; i<16; i++){
1212                                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1213                                     s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
1214                             }
1215                         }
1216                     }else{
1217                          s->dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1218                     }
1219                 }else if(h->cbp&15){
1220                     if(transform_bypass){
1221                         const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1222                         idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1223                         for(i=0; i<16; i+=di){
1224                             if(h->non_zero_count_cache[ scan8[i] ]){
1225                                 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
1226                             }
1227                         }
1228                     }else{
1229                         if(IS_8x8DCT(mb_type)){
1230                             s->dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1231                         }else{
1232                             s->dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1233                         }
1234                     }
1235                 }
1236             }else{
1237                 for(i=0; i<16; i++){
1238                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1239                         uint8_t * const ptr= dest_y + block_offset[i];
1240                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1241                     }
1242                 }
1243             }
1244         }
1245
1246         if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1247             uint8_t *dest[2] = {dest_cb, dest_cr};
1248             if(transform_bypass){
1249                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1250                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
1251                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
1252                 }else{
1253                     idct_add = s->dsp.add_pixels4;
1254                     for(i=16; i<16+8; i++){
1255                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1256                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1257                     }
1258                 }
1259             }else{
1260                 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]);
1261                 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]);
1262                 if(is_h264){
1263                     idct_add = s->dsp.h264_idct_add;
1264                     idct_dc_add = s->dsp.h264_idct_dc_add;
1265                     for(i=16; i<16+8; i++){
1266                         if(h->non_zero_count_cache[ scan8[i] ])
1267                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1268                         else if(h->mb[i*16])
1269                             idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1270                     }
1271                 }else{
1272                     for(i=16; i<16+8; i++){
1273                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1274                             uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1275                             ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
1276                         }
1277                     }
1278                 }
1279             }
1280         }
1281     }
1282     if(h->cbp || IS_INTRA(mb_type))
1283         s->dsp.clear_blocks(h->mb);
1284 }
1285
1286 /**
1287  * Process a macroblock; this case avoids checks for expensive uncommon cases.
1288  */
1289 static void hl_decode_mb_simple(H264Context *h){
1290     hl_decode_mb_internal(h, 1);
1291 }
1292
1293 /**
1294  * Process a macroblock; this handles edge cases, such as interlacing.
1295  */
1296 static void av_noinline hl_decode_mb_complex(H264Context *h){
1297     hl_decode_mb_internal(h, 0);
1298 }
1299
1300 void ff_h264_hl_decode_mb(H264Context *h){
1301     MpegEncContext * const s = &h->s;
1302     const int mb_xy= h->mb_xy;
1303     const int mb_type= s->current_picture.mb_type[mb_xy];
1304     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1305
1306     if (is_complex)
1307         hl_decode_mb_complex(h);
1308     else hl_decode_mb_simple(h);
1309 }
1310
1311 static int pred_weight_table(H264Context *h){
1312     MpegEncContext * const s = &h->s;
1313     int list, i;
1314     int luma_def, chroma_def;
1315
1316     h->use_weight= 0;
1317     h->use_weight_chroma= 0;
1318     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1319     h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1320     luma_def = 1<<h->luma_log2_weight_denom;
1321     chroma_def = 1<<h->chroma_log2_weight_denom;
1322
1323     for(list=0; list<2; list++){
1324         h->luma_weight_flag[list]   = 0;
1325         h->chroma_weight_flag[list] = 0;
1326         for(i=0; i<h->ref_count[list]; i++){
1327             int luma_weight_flag, chroma_weight_flag;
1328
1329             luma_weight_flag= get_bits1(&s->gb);
1330             if(luma_weight_flag){
1331                 h->luma_weight[list][i]= get_se_golomb(&s->gb);
1332                 h->luma_offset[list][i]= get_se_golomb(&s->gb);
1333                 if(   h->luma_weight[list][i] != luma_def
1334                    || h->luma_offset[list][i] != 0) {
1335                     h->use_weight= 1;
1336                     h->luma_weight_flag[list]= 1;
1337                 }
1338             }else{
1339                 h->luma_weight[list][i]= luma_def;
1340                 h->luma_offset[list][i]= 0;
1341             }
1342
1343             if(CHROMA){
1344                 chroma_weight_flag= get_bits1(&s->gb);
1345                 if(chroma_weight_flag){
1346                     int j;
1347                     for(j=0; j<2; j++){
1348                         h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
1349                         h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
1350                         if(   h->chroma_weight[list][i][j] != chroma_def
1351                            || h->chroma_offset[list][i][j] != 0) {
1352                             h->use_weight_chroma= 1;
1353                             h->chroma_weight_flag[list]= 1;
1354                         }
1355                     }
1356                 }else{
1357                     int j;
1358                     for(j=0; j<2; j++){
1359                         h->chroma_weight[list][i][j]= chroma_def;
1360                         h->chroma_offset[list][i][j]= 0;
1361                     }
1362                 }
1363             }
1364         }
1365         if(h->slice_type_nos != FF_B_TYPE) break;
1366     }
1367     h->use_weight= h->use_weight || h->use_weight_chroma;
1368     return 0;
1369 }
1370
1371 static void implicit_weight_table(H264Context *h){
1372     MpegEncContext * const s = &h->s;
1373     int ref0, ref1, i;
1374     int cur_poc = s->current_picture_ptr->poc;
1375
1376     for (i = 0; i < 2; i++) {
1377         h->luma_weight_flag[i]   = 0;
1378         h->chroma_weight_flag[i] = 0;
1379     }
1380
1381     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
1382        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1383         h->use_weight= 0;
1384         h->use_weight_chroma= 0;
1385         return;
1386     }
1387
1388     h->use_weight= 2;
1389     h->use_weight_chroma= 2;
1390     h->luma_log2_weight_denom= 5;
1391     h->chroma_log2_weight_denom= 5;
1392
1393     for(ref0=0; ref0 < h->ref_count[0]; ref0++){
1394         int poc0 = h->ref_list[0][ref0].poc;
1395         for(ref1=0; ref1 < h->ref_count[1]; ref1++){
1396             int poc1 = h->ref_list[1][ref1].poc;
1397             int td = av_clip(poc1 - poc0, -128, 127);
1398             if(td){
1399                 int tb = av_clip(cur_poc - poc0, -128, 127);
1400                 int tx = (16384 + (FFABS(td) >> 1)) / td;
1401                 int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
1402                 if(dist_scale_factor < -64 || dist_scale_factor > 128)
1403                     h->implicit_weight[ref0][ref1] = 32;
1404                 else
1405                     h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
1406             }else
1407                 h->implicit_weight[ref0][ref1] = 32;
1408         }
1409     }
1410 }
1411
1412 /**
1413  * instantaneous decoder refresh.
1414  */
1415 static void idr(H264Context *h){
1416     ff_h264_remove_all_refs(h);
1417     h->prev_frame_num= 0;
1418     h->prev_frame_num_offset= 0;
1419     h->prev_poc_msb=
1420     h->prev_poc_lsb= 0;
1421 }
1422
1423 /* forget old pics after a seek */
1424 static void flush_dpb(AVCodecContext *avctx){
1425     H264Context *h= avctx->priv_data;
1426     int i;
1427     for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1428         if(h->delayed_pic[i])
1429             h->delayed_pic[i]->reference= 0;
1430         h->delayed_pic[i]= NULL;
1431     }
1432     h->outputed_poc= INT_MIN;
1433     h->prev_interlaced_frame = 1;
1434     idr(h);
1435     if(h->s.current_picture_ptr)
1436         h->s.current_picture_ptr->reference= 0;
1437     h->s.first_field= 0;
1438     ff_h264_reset_sei(h);
1439     ff_mpeg_flush(avctx);
1440 }
1441
1442 static int init_poc(H264Context *h){
1443     MpegEncContext * const s = &h->s;
1444     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1445     int field_poc[2];
1446     Picture *cur = s->current_picture_ptr;
1447
1448     h->frame_num_offset= h->prev_frame_num_offset;
1449     if(h->frame_num < h->prev_frame_num)
1450         h->frame_num_offset += max_frame_num;
1451
1452     if(h->sps.poc_type==0){
1453         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1454
1455         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1456             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1457         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1458             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1459         else
1460             h->poc_msb = h->prev_poc_msb;
1461 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1462         field_poc[0] =
1463         field_poc[1] = h->poc_msb + h->poc_lsb;
1464         if(s->picture_structure == PICT_FRAME)
1465             field_poc[1] += h->delta_poc_bottom;
1466     }else if(h->sps.poc_type==1){
1467         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1468         int i;
1469
1470         if(h->sps.poc_cycle_length != 0)
1471             abs_frame_num = h->frame_num_offset + h->frame_num;
1472         else
1473             abs_frame_num = 0;
1474
1475         if(h->nal_ref_idc==0 && abs_frame_num > 0)
1476             abs_frame_num--;
1477
1478         expected_delta_per_poc_cycle = 0;
1479         for(i=0; i < h->sps.poc_cycle_length; i++)
1480             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1481
1482         if(abs_frame_num > 0){
1483             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1484             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1485
1486             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1487             for(i = 0; i <= frame_num_in_poc_cycle; i++)
1488                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1489         } else
1490             expectedpoc = 0;
1491
1492         if(h->nal_ref_idc == 0)
1493             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1494
1495         field_poc[0] = expectedpoc + h->delta_poc[0];
1496         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1497
1498         if(s->picture_structure == PICT_FRAME)
1499             field_poc[1] += h->delta_poc[1];
1500     }else{
1501         int poc= 2*(h->frame_num_offset + h->frame_num);
1502
1503         if(!h->nal_ref_idc)
1504             poc--;
1505
1506         field_poc[0]= poc;
1507         field_poc[1]= poc;
1508     }
1509
1510     if(s->picture_structure != PICT_BOTTOM_FIELD)
1511         s->current_picture_ptr->field_poc[0]= field_poc[0];
1512     if(s->picture_structure != PICT_TOP_FIELD)
1513         s->current_picture_ptr->field_poc[1]= field_poc[1];
1514     cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1515
1516     return 0;
1517 }
1518
1519
1520 /**
1521  * initialize scan tables
1522  */
1523 static void init_scan_tables(H264Context *h){
1524     MpegEncContext * const s = &h->s;
1525     int i;
1526     if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
1527         memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
1528         memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
1529     }else{
1530         for(i=0; i<16; i++){
1531 #define T(x) (x>>2) | ((x<<2) & 0xF)
1532             h->zigzag_scan[i] = T(zigzag_scan[i]);
1533             h-> field_scan[i] = T( field_scan[i]);
1534 #undef T
1535         }
1536     }
1537     if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
1538         memcpy(h->zigzag_scan8x8,       ff_zigzag_direct,     64*sizeof(uint8_t));
1539         memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
1540         memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
1541         memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
1542     }else{
1543         for(i=0; i<64; i++){
1544 #define T(x) (x>>3) | ((x&7)<<3)
1545             h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
1546             h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1547             h->field_scan8x8[i]        = T(field_scan8x8[i]);
1548             h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
1549 #undef T
1550         }
1551     }
1552     if(h->sps.transform_bypass){ //FIXME same ugly
1553         h->zigzag_scan_q0          = zigzag_scan;
1554         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
1555         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1556         h->field_scan_q0           = field_scan;
1557         h->field_scan8x8_q0        = field_scan8x8;
1558         h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
1559     }else{
1560         h->zigzag_scan_q0          = h->zigzag_scan;
1561         h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
1562         h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1563         h->field_scan_q0           = h->field_scan;
1564         h->field_scan8x8_q0        = h->field_scan8x8;
1565         h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
1566     }
1567 }
1568
1569 static void field_end(H264Context *h){
1570     MpegEncContext * const s = &h->s;
1571     AVCodecContext * const avctx= s->avctx;
1572     s->mb_y= 0;
1573
1574     s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1575     s->current_picture_ptr->pict_type= s->pict_type;
1576
1577     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1578         ff_vdpau_h264_set_reference_frames(s);
1579
1580     if(!s->dropable) {
1581         ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1582         h->prev_poc_msb= h->poc_msb;
1583         h->prev_poc_lsb= h->poc_lsb;
1584     }
1585     h->prev_frame_num_offset= h->frame_num_offset;
1586     h->prev_frame_num= h->frame_num;
1587
1588     if (avctx->hwaccel) {
1589         if (avctx->hwaccel->end_frame(avctx) < 0)
1590             av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1591     }
1592
1593     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1594         ff_vdpau_h264_picture_complete(s);
1595
1596     /*
1597      * FIXME: Error handling code does not seem to support interlaced
1598      * when slices span multiple rows
1599      * The ff_er_add_slice calls don't work right for bottom
1600      * fields; they cause massive erroneous error concealing
1601      * Error marking covers both fields (top and bottom).
1602      * This causes a mismatched s->error_count
1603      * and a bad error table. Further, the error count goes to
1604      * INT_MAX when called for bottom field, because mb_y is
1605      * past end by one (callers fault) and resync_mb_y != 0
1606      * causes problems for the first MB line, too.
1607      */
1608     if (!FIELD_PICTURE)
1609         ff_er_frame_end(s);
1610
1611     MPV_frame_end(s);
1612
1613     h->current_slice=0;
1614 }
1615
1616 /**
1617  * Replicates H264 "master" context to thread contexts.
1618  */
1619 static void clone_slice(H264Context *dst, H264Context *src)
1620 {
1621     memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
1622     dst->s.current_picture_ptr  = src->s.current_picture_ptr;
1623     dst->s.current_picture      = src->s.current_picture;
1624     dst->s.linesize             = src->s.linesize;
1625     dst->s.uvlinesize           = src->s.uvlinesize;
1626     dst->s.first_field          = src->s.first_field;
1627
1628     dst->prev_poc_msb           = src->prev_poc_msb;
1629     dst->prev_poc_lsb           = src->prev_poc_lsb;
1630     dst->prev_frame_num_offset  = src->prev_frame_num_offset;
1631     dst->prev_frame_num         = src->prev_frame_num;
1632     dst->short_ref_count        = src->short_ref_count;
1633
1634     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
1635     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
1636     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1637     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
1638
1639     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
1640     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
1641 }
1642
1643 /**
1644  * decodes a slice header.
1645  * This will also call MPV_common_init() and frame_start() as needed.
1646  *
1647  * @param h h264context
1648  * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1649  *
1650  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1651  */
1652 static int decode_slice_header(H264Context *h, H264Context *h0){
1653     MpegEncContext * const s = &h->s;
1654     MpegEncContext * const s0 = &h0->s;
1655     unsigned int first_mb_in_slice;
1656     unsigned int pps_id;
1657     int num_ref_idx_active_override_flag;
1658     unsigned int slice_type, tmp, i, j;
1659     int default_ref_list_done = 0;
1660     int last_pic_structure;
1661
1662     s->dropable= h->nal_ref_idc == 0;
1663
1664     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1665         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1666         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1667     }else{
1668         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1669         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1670     }
1671
1672     first_mb_in_slice= get_ue_golomb(&s->gb);
1673
1674     if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1675         if(h0->current_slice && FIELD_PICTURE){
1676             field_end(h);
1677         }
1678
1679         h0->current_slice = 0;
1680         if (!s0->first_field)
1681             s->current_picture_ptr= NULL;
1682     }
1683
1684     slice_type= get_ue_golomb_31(&s->gb);
1685     if(slice_type > 9){
1686         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);
1687         return -1;
1688     }
1689     if(slice_type > 4){
1690         slice_type -= 5;
1691         h->slice_type_fixed=1;
1692     }else
1693         h->slice_type_fixed=0;
1694
1695     slice_type= golomb_to_pict_type[ slice_type ];
1696     if (slice_type == FF_I_TYPE
1697         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
1698         default_ref_list_done = 1;
1699     }
1700     h->slice_type= slice_type;
1701     h->slice_type_nos= slice_type & 3;
1702
1703     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
1704
1705     pps_id= get_ue_golomb(&s->gb);
1706     if(pps_id>=MAX_PPS_COUNT){
1707         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
1708         return -1;
1709     }
1710     if(!h0->pps_buffers[pps_id]) {
1711         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
1712         return -1;
1713     }
1714     h->pps= *h0->pps_buffers[pps_id];
1715
1716     if(!h0->sps_buffers[h->pps.sps_id]) {
1717         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
1718         return -1;
1719     }
1720     h->sps = *h0->sps_buffers[h->pps.sps_id];
1721
1722     if(h == h0 && h->dequant_coeff_pps != pps_id){
1723         h->dequant_coeff_pps = pps_id;
1724         init_dequant_tables(h);
1725     }
1726
1727     s->mb_width= h->sps.mb_width;
1728     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1729
1730     h->b_stride=  s->mb_width*4;
1731     h->b8_stride= s->mb_width*2;
1732
1733     s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
1734     if(h->sps.frame_mbs_only_flag)
1735         s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
1736     else
1737         s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
1738
1739     if (s->context_initialized
1740         && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
1741         if(h != h0)
1742             return -1;   // width / height changed during parallelized decoding
1743         free_tables(h);
1744         flush_dpb(s->avctx);
1745         MPV_common_end(s);
1746     }
1747     if (!s->context_initialized) {
1748         if(h != h0)
1749             return -1;  // we cant (re-)initialize context during parallel decoding
1750
1751         avcodec_set_dimensions(s->avctx, s->width, s->height);
1752         s->avctx->sample_aspect_ratio= h->sps.sar;
1753         if(!s->avctx->sample_aspect_ratio.den)
1754             s->avctx->sample_aspect_ratio.den = 1;
1755
1756         if(h->sps.video_signal_type_present_flag){
1757             s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1758             if(h->sps.colour_description_present_flag){
1759                 s->avctx->color_primaries = h->sps.color_primaries;
1760                 s->avctx->color_trc       = h->sps.color_trc;
1761                 s->avctx->colorspace      = h->sps.colorspace;
1762             }
1763         }
1764
1765         if(h->sps.timing_info_present_flag){
1766             int64_t den= h->sps.time_scale;
1767             if(h->x264_build < 44U)
1768                 den *= 2;
1769             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
1770                       h->sps.num_units_in_tick, den, 1<<30);
1771         }
1772         s->avctx->pix_fmt = s->avctx->get_format(s->avctx, s->avctx->codec->pix_fmts);
1773         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1774
1775         if (MPV_common_init(s) < 0)
1776             return -1;
1777         s->first_field = 0;
1778         h->prev_interlaced_frame = 1;
1779
1780         init_scan_tables(h);
1781         ff_h264_alloc_tables(h);
1782
1783         for(i = 1; i < s->avctx->thread_count; i++) {
1784             H264Context *c;
1785             c = h->thread_context[i] = av_malloc(sizeof(H264Context));
1786             memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
1787             memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
1788             c->sps = h->sps;
1789             c->pps = h->pps;
1790             init_scan_tables(c);
1791             clone_tables(c, h);
1792         }
1793
1794         for(i = 0; i < s->avctx->thread_count; i++)
1795             if(context_init(h->thread_context[i]) < 0)
1796                 return -1;
1797     }
1798
1799     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1800
1801     h->mb_mbaff = 0;
1802     h->mb_aff_frame = 0;
1803     last_pic_structure = s0->picture_structure;
1804     if(h->sps.frame_mbs_only_flag){
1805         s->picture_structure= PICT_FRAME;
1806     }else{
1807         if(get_bits1(&s->gb)) { //field_pic_flag
1808             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
1809         } else {
1810             s->picture_structure= PICT_FRAME;
1811             h->mb_aff_frame = h->sps.mb_aff;
1812         }
1813     }
1814     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
1815
1816     if(h0->current_slice == 0){
1817         while(h->frame_num !=  h->prev_frame_num &&
1818               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
1819             av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
1820             if (ff_h264_frame_start(h) < 0)
1821                 return -1;
1822             h->prev_frame_num++;
1823             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
1824             s->current_picture_ptr->frame_num= h->prev_frame_num;
1825             ff_h264_execute_ref_pic_marking(h, NULL, 0);
1826         }
1827
1828         /* See if we have a decoded first field looking for a pair... */
1829         if (s0->first_field) {
1830             assert(s0->current_picture_ptr);
1831             assert(s0->current_picture_ptr->data[0]);
1832             assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
1833
1834             /* figure out if we have a complementary field pair */
1835             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
1836                 /*
1837                  * Previous field is unmatched. Don't display it, but let it
1838                  * remain for reference if marked as such.
1839                  */
1840                 s0->current_picture_ptr = NULL;
1841                 s0->first_field = FIELD_PICTURE;
1842
1843             } else {
1844                 if (h->nal_ref_idc &&
1845                         s0->current_picture_ptr->reference &&
1846                         s0->current_picture_ptr->frame_num != h->frame_num) {
1847                     /*
1848                      * This and previous field were reference, but had
1849                      * different frame_nums. Consider this field first in
1850                      * pair. Throw away previous field except for reference
1851                      * purposes.
1852                      */
1853                     s0->first_field = 1;
1854                     s0->current_picture_ptr = NULL;
1855
1856                 } else {
1857                     /* Second field in complementary pair */
1858                     s0->first_field = 0;
1859                 }
1860             }
1861
1862         } else {
1863             /* Frame or first field in a potentially complementary pair */
1864             assert(!s0->current_picture_ptr);
1865             s0->first_field = FIELD_PICTURE;
1866         }
1867
1868         if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
1869             s0->first_field = 0;
1870             return -1;
1871         }
1872     }
1873     if(h != h0)
1874         clone_slice(h, h0);
1875
1876     s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
1877
1878     assert(s->mb_num == s->mb_width * s->mb_height);
1879     if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
1880        first_mb_in_slice                    >= s->mb_num){
1881         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1882         return -1;
1883     }
1884     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
1885     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
1886     if (s->picture_structure == PICT_BOTTOM_FIELD)
1887         s->resync_mb_y = s->mb_y = s->mb_y + 1;
1888     assert(s->mb_y < s->mb_height);
1889
1890     if(s->picture_structure==PICT_FRAME){
1891         h->curr_pic_num=   h->frame_num;
1892         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
1893     }else{
1894         h->curr_pic_num= 2*h->frame_num + 1;
1895         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
1896     }
1897
1898     if(h->nal_unit_type == NAL_IDR_SLICE){
1899         get_ue_golomb(&s->gb); /* idr_pic_id */
1900     }
1901
1902     if(h->sps.poc_type==0){
1903         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
1904
1905         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
1906             h->delta_poc_bottom= get_se_golomb(&s->gb);
1907         }
1908     }
1909
1910     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
1911         h->delta_poc[0]= get_se_golomb(&s->gb);
1912
1913         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
1914             h->delta_poc[1]= get_se_golomb(&s->gb);
1915     }
1916
1917     init_poc(h);
1918
1919     if(h->pps.redundant_pic_cnt_present){
1920         h->redundant_pic_count= get_ue_golomb(&s->gb);
1921     }
1922
1923     //set defaults, might be overridden a few lines later
1924     h->ref_count[0]= h->pps.ref_count[0];
1925     h->ref_count[1]= h->pps.ref_count[1];
1926
1927     if(h->slice_type_nos != FF_I_TYPE){
1928         if(h->slice_type_nos == FF_B_TYPE){
1929             h->direct_spatial_mv_pred= get_bits1(&s->gb);
1930         }
1931         num_ref_idx_active_override_flag= get_bits1(&s->gb);
1932
1933         if(num_ref_idx_active_override_flag){
1934             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
1935             if(h->slice_type_nos==FF_B_TYPE)
1936                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
1937
1938             if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
1939                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
1940                 h->ref_count[0]= h->ref_count[1]= 1;
1941                 return -1;
1942             }
1943         }
1944         if(h->slice_type_nos == FF_B_TYPE)
1945             h->list_count= 2;
1946         else
1947             h->list_count= 1;
1948     }else
1949         h->list_count= 0;
1950
1951     if(!default_ref_list_done){
1952         ff_h264_fill_default_ref_list(h);
1953     }
1954
1955     if(h->slice_type_nos!=FF_I_TYPE && ff_h264_decode_ref_pic_list_reordering(h) < 0)
1956         return -1;
1957
1958     if(h->slice_type_nos!=FF_I_TYPE){
1959         s->last_picture_ptr= &h->ref_list[0][0];
1960         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
1961     }
1962     if(h->slice_type_nos==FF_B_TYPE){
1963         s->next_picture_ptr= &h->ref_list[1][0];
1964         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
1965     }
1966
1967     if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
1968        ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
1969         pred_weight_table(h);
1970     else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE)
1971         implicit_weight_table(h);
1972     else {
1973         h->use_weight = 0;
1974         for (i = 0; i < 2; i++) {
1975             h->luma_weight_flag[i]   = 0;
1976             h->chroma_weight_flag[i] = 0;
1977         }
1978     }
1979
1980     if(h->nal_ref_idc)
1981         ff_h264_decode_ref_pic_marking(h0, &s->gb);
1982
1983     if(FRAME_MBAFF)
1984         ff_h264_fill_mbaff_ref_list(h);
1985
1986     if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
1987         ff_h264_direct_dist_scale_factor(h);
1988     ff_h264_direct_ref_list_init(h);
1989
1990     if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
1991         tmp = get_ue_golomb_31(&s->gb);
1992         if(tmp > 2){
1993             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
1994             return -1;
1995         }
1996         h->cabac_init_idc= tmp;
1997     }
1998
1999     h->last_qscale_diff = 0;
2000     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2001     if(tmp>51){
2002         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2003         return -1;
2004     }
2005     s->qscale= tmp;
2006     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2007     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2008     //FIXME qscale / qp ... stuff
2009     if(h->slice_type == FF_SP_TYPE){
2010         get_bits1(&s->gb); /* sp_for_switch_flag */
2011     }
2012     if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
2013         get_se_golomb(&s->gb); /* slice_qs_delta */
2014     }
2015
2016     h->deblocking_filter = 1;
2017     h->slice_alpha_c0_offset = 52;
2018     h->slice_beta_offset = 52;
2019     if( h->pps.deblocking_filter_parameters_present ) {
2020         tmp= get_ue_golomb_31(&s->gb);
2021         if(tmp > 2){
2022             av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2023             return -1;
2024         }
2025         h->deblocking_filter= tmp;
2026         if(h->deblocking_filter < 2)
2027             h->deblocking_filter^= 1; // 1<->0
2028
2029         if( h->deblocking_filter ) {
2030             h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2031             h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
2032             if(   h->slice_alpha_c0_offset > 104U
2033                || h->slice_beta_offset     > 104U){
2034                 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);
2035                 return -1;
2036             }
2037         }
2038     }
2039
2040     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
2041        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
2042        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
2043        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2044         h->deblocking_filter= 0;
2045
2046     if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2047         if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2048             /* Cheat slightly for speed:
2049                Do not bother to deblock across slices. */
2050             h->deblocking_filter = 2;
2051         } else {
2052             h0->max_contexts = 1;
2053             if(!h0->single_decode_warning) {
2054                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2055                 h0->single_decode_warning = 1;
2056             }
2057             if(h != h0)
2058                 return 1; // deblocking switched inside frame
2059         }
2060     }
2061     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]);
2062
2063 #if 0 //FMO
2064     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2065         slice_group_change_cycle= get_bits(&s->gb, ?);
2066 #endif
2067
2068     h0->last_slice_type = slice_type;
2069     h->slice_num = ++h0->current_slice;
2070     if(h->slice_num >= MAX_SLICES){
2071         av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2072     }
2073
2074     for(j=0; j<2; j++){
2075         int id_list[16];
2076         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2077         for(i=0; i<16; i++){
2078             id_list[i]= 60;
2079             if(h->ref_list[j][i].data[0]){
2080                 int k;
2081                 uint8_t *base= h->ref_list[j][i].base[0];
2082                 for(k=0; k<h->short_ref_count; k++)
2083                     if(h->short_ref[k]->base[0] == base){
2084                         id_list[i]= k;
2085                         break;
2086                     }
2087                 for(k=0; k<h->long_ref_count; k++)
2088                     if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2089                         id_list[i]= h->short_ref_count + k;
2090                         break;
2091                     }
2092             }
2093         }
2094
2095         ref2frm[0]=
2096         ref2frm[1]= -1;
2097         for(i=0; i<16; i++)
2098             ref2frm[i+2]= 4*id_list[i]
2099                           +(h->ref_list[j][i].reference&3);
2100         ref2frm[18+0]=
2101         ref2frm[18+1]= -1;
2102         for(i=16; i<48; i++)
2103             ref2frm[i+4]= 4*id_list[(i-16)>>1]
2104                           +(h->ref_list[j][i].reference&3);
2105     }
2106
2107     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
2108     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2109
2110     s->avctx->refs= h->sps.ref_frame_count;
2111
2112     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2113         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",
2114                h->slice_num,
2115                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2116                first_mb_in_slice,
2117                av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2118                pps_id, h->frame_num,
2119                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2120                h->ref_count[0], h->ref_count[1],
2121                s->qscale,
2122                h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
2123                h->use_weight,
2124                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2125                h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2126                );
2127     }
2128
2129     return 0;
2130 }
2131
2132 int ff_h264_get_slice_type(const H264Context *h)
2133 {
2134     switch (h->slice_type) {
2135     case FF_P_TYPE:  return 0;
2136     case FF_B_TYPE:  return 1;
2137     case FF_I_TYPE:  return 2;
2138     case FF_SP_TYPE: return 3;
2139     case FF_SI_TYPE: return 4;
2140     default:         return -1;
2141     }
2142 }
2143
2144 static void loop_filter(H264Context *h){
2145     MpegEncContext * const s = &h->s;
2146     uint8_t  *dest_y, *dest_cb, *dest_cr;
2147     int linesize, uvlinesize, mb_x, mb_y;
2148     const int end_mb_y= s->mb_y + FRAME_MBAFF;
2149     const int old_slice_type= h->slice_type;
2150
2151     if(h->deblocking_filter) {
2152         for(mb_x= 0; mb_x<s->mb_width; mb_x++){
2153             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
2154                 int mb_xy, mb_type;
2155                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2156                 h->slice_num= h->slice_table[mb_xy];
2157                 mb_type= s->current_picture.mb_type[mb_xy];
2158                 h->list_count= h->list_counts[mb_xy];
2159
2160                 if(FRAME_MBAFF)
2161                     h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2162
2163                 s->mb_x= mb_x;
2164                 s->mb_y= mb_y;
2165                 dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2166                 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2167                 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2168                     //FIXME simplify above
2169
2170                 if (MB_FIELD) {
2171                     linesize   = h->mb_linesize   = s->linesize * 2;
2172                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2173                     if(mb_y&1){ //FIXME move out of this function?
2174                         dest_y -= s->linesize*15;
2175                         dest_cb-= s->uvlinesize*7;
2176                         dest_cr-= s->uvlinesize*7;
2177                     }
2178                 } else {
2179                     linesize   = h->mb_linesize   = s->linesize;
2180                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2181                 }
2182                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2183                 if(fill_filter_caches(h, mb_type))
2184                     continue;
2185                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2186                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2187
2188                 if (FRAME_MBAFF) {
2189                     ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2190                 } else {
2191                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2192                 }
2193             }
2194         }
2195     }
2196     h->slice_type= old_slice_type;
2197     s->mb_x= 0;
2198     s->mb_y= end_mb_y - FRAME_MBAFF;
2199     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2200     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2201 }
2202
2203 static int decode_slice(struct AVCodecContext *avctx, void *arg){
2204     H264Context *h = *(void**)arg;
2205     MpegEncContext * const s = &h->s;
2206     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2207
2208     s->mb_skip_run= -1;
2209
2210     h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
2211                     (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
2212
2213     if( h->pps.cabac ) {
2214         /* realign */
2215         align_get_bits( &s->gb );
2216
2217         /* init cabac */
2218         ff_init_cabac_states( &h->cabac);
2219         ff_init_cabac_decoder( &h->cabac,
2220                                s->gb.buffer + get_bits_count(&s->gb)/8,
2221                                (get_bits_left(&s->gb) + 7)/8);
2222
2223         ff_h264_init_cabac_states(h);
2224
2225         for(;;){
2226 //START_TIMER
2227             int ret = ff_h264_decode_mb_cabac(h);
2228             int eos;
2229 //STOP_TIMER("decode_mb_cabac")
2230
2231             if(ret>=0) ff_h264_hl_decode_mb(h);
2232
2233             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
2234                 s->mb_y++;
2235
2236                 ret = ff_h264_decode_mb_cabac(h);
2237
2238                 if(ret>=0) ff_h264_hl_decode_mb(h);
2239                 s->mb_y--;
2240             }
2241             eos = get_cabac_terminate( &h->cabac );
2242
2243             if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
2244                 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);
2245                 return 0;
2246             }
2247             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2248                 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);
2249                 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);
2250                 return -1;
2251             }
2252
2253             if( ++s->mb_x >= s->mb_width ) {
2254                 s->mb_x = 0;
2255                 loop_filter(h);
2256                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2257                 ++s->mb_y;
2258                 if(FIELD_OR_MBAFF_PICTURE) {
2259                     ++s->mb_y;
2260                 }
2261             }
2262
2263             if( eos || s->mb_y >= s->mb_height ) {
2264                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2265                 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);
2266                 return 0;
2267             }
2268         }
2269
2270     } else {
2271         for(;;){
2272             int ret = ff_h264_decode_mb_cavlc(h);
2273
2274             if(ret>=0) ff_h264_hl_decode_mb(h);
2275
2276             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
2277                 s->mb_y++;
2278                 ret = ff_h264_decode_mb_cavlc(h);
2279
2280                 if(ret>=0) ff_h264_hl_decode_mb(h);
2281                 s->mb_y--;
2282             }
2283
2284             if(ret<0){
2285                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2286                 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);
2287
2288                 return -1;
2289             }
2290
2291             if(++s->mb_x >= s->mb_width){
2292                 s->mb_x=0;
2293                 loop_filter(h);
2294                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2295                 ++s->mb_y;
2296                 if(FIELD_OR_MBAFF_PICTURE) {
2297                     ++s->mb_y;
2298                 }
2299                 if(s->mb_y >= s->mb_height){
2300                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2301
2302                     if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
2303                         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);
2304
2305                         return 0;
2306                     }else{
2307                         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);
2308
2309                         return -1;
2310                     }
2311                 }
2312             }
2313
2314             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
2315                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2316                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2317                     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);
2318
2319                     return 0;
2320                 }else{
2321                     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);
2322
2323                     return -1;
2324                 }
2325             }
2326         }
2327     }
2328
2329 #if 0
2330     for(;s->mb_y < s->mb_height; s->mb_y++){
2331         for(;s->mb_x < s->mb_width; s->mb_x++){
2332             int ret= decode_mb(h);
2333
2334             ff_h264_hl_decode_mb(h);
2335
2336             if(ret<0){
2337                 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2338                 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);
2339
2340                 return -1;
2341             }
2342
2343             if(++s->mb_x >= s->mb_width){
2344                 s->mb_x=0;
2345                 if(++s->mb_y >= s->mb_height){
2346                     if(get_bits_count(s->gb) == s->gb.size_in_bits){
2347                         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);
2348
2349                         return 0;
2350                     }else{
2351                         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);
2352
2353                         return -1;
2354                     }
2355                 }
2356             }
2357
2358             if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2359                 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2360                     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);
2361
2362                     return 0;
2363                 }else{
2364                     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);
2365
2366                     return -1;
2367                 }
2368             }
2369         }
2370         s->mb_x=0;
2371         ff_draw_horiz_band(s, 16*s->mb_y, 16);
2372     }
2373 #endif
2374     return -1; //not reached
2375 }
2376
2377 /**
2378  * Call decode_slice() for each context.
2379  *
2380  * @param h h264 master context
2381  * @param context_count number of contexts to execute
2382  */
2383 static void execute_decode_slices(H264Context *h, int context_count){
2384     MpegEncContext * const s = &h->s;
2385     AVCodecContext * const avctx= s->avctx;
2386     H264Context *hx;
2387     int i;
2388
2389     if (s->avctx->hwaccel)
2390         return;
2391     if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2392         return;
2393     if(context_count == 1) {
2394         decode_slice(avctx, &h);
2395     } else {
2396         for(i = 1; i < context_count; i++) {
2397             hx = h->thread_context[i];
2398             hx->s.error_recognition = avctx->error_recognition;
2399             hx->s.error_count = 0;
2400         }
2401
2402         avctx->execute(avctx, (void *)decode_slice,
2403                        h->thread_context, NULL, context_count, sizeof(void*));
2404
2405         /* pull back stuff from slices to master context */
2406         hx = h->thread_context[context_count - 1];
2407         s->mb_x = hx->s.mb_x;
2408         s->mb_y = hx->s.mb_y;
2409         s->dropable = hx->s.dropable;
2410         s->picture_structure = hx->s.picture_structure;
2411         for(i = 1; i < context_count; i++)
2412             h->s.error_count += h->thread_context[i]->s.error_count;
2413     }
2414 }
2415
2416
2417 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
2418     MpegEncContext * const s = &h->s;
2419     AVCodecContext * const avctx= s->avctx;
2420     int buf_index=0;
2421     H264Context *hx; ///< thread context
2422     int context_count = 0;
2423     int next_avc= h->is_avc ? 0 : buf_size;
2424
2425     h->max_contexts = avctx->thread_count;
2426 #if 0
2427     int i;
2428     for(i=0; i<50; i++){
2429         av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
2430     }
2431 #endif
2432     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
2433         h->current_slice = 0;
2434         if (!s->first_field)
2435             s->current_picture_ptr= NULL;
2436         ff_h264_reset_sei(h);
2437     }
2438
2439     for(;;){
2440         int consumed;
2441         int dst_length;
2442         int bit_length;
2443         const uint8_t *ptr;
2444         int i, nalsize = 0;
2445         int err;
2446
2447         if(buf_index >= next_avc) {
2448             if(buf_index >= buf_size) break;
2449             nalsize = 0;
2450             for(i = 0; i < h->nal_length_size; i++)
2451                 nalsize = (nalsize << 8) | buf[buf_index++];
2452             if(nalsize <= 1 || nalsize > buf_size - buf_index){
2453                 if(nalsize == 1){
2454                     buf_index++;
2455                     continue;
2456                 }else{
2457                     av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2458                     break;
2459                 }
2460             }
2461             next_avc= buf_index + nalsize;
2462         } else {
2463             // start code prefix search
2464             for(; buf_index + 3 < next_avc; buf_index++){
2465                 // This should always succeed in the first iteration.
2466                 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2467                     break;
2468             }
2469
2470             if(buf_index+3 >= buf_size) break;
2471
2472             buf_index+=3;
2473             if(buf_index >= next_avc) continue;
2474         }
2475
2476         hx = h->thread_context[context_count];
2477
2478         ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
2479         if (ptr==NULL || dst_length < 0){
2480             return -1;
2481         }
2482         i= buf_index + consumed;
2483         if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
2484            buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
2485             s->workaround_bugs |= FF_BUG_TRUNCATED;
2486
2487         if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
2488         while(ptr[dst_length - 1] == 0 && dst_length > 0)
2489             dst_length--;
2490         }
2491         bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
2492
2493         if(s->avctx->debug&FF_DEBUG_STARTCODE){
2494             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);
2495         }
2496
2497         if (h->is_avc && (nalsize != consumed) && nalsize){
2498             av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
2499         }
2500
2501         buf_index += consumed;
2502
2503         if(  (s->hurry_up == 1 && h->nal_ref_idc  == 0) //FIXME do not discard SEI id
2504            ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0))
2505             continue;
2506
2507       again:
2508         err = 0;
2509         switch(hx->nal_unit_type){
2510         case NAL_IDR_SLICE:
2511             if (h->nal_unit_type != NAL_IDR_SLICE) {
2512                 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2513                 return -1;
2514             }
2515             idr(h); //FIXME ensure we don't loose some frames if there is reordering
2516         case NAL_SLICE:
2517             init_get_bits(&hx->s.gb, ptr, bit_length);
2518             hx->intra_gb_ptr=
2519             hx->inter_gb_ptr= &hx->s.gb;
2520             hx->s.data_partitioning = 0;
2521
2522             if((err = decode_slice_header(hx, h)))
2523                break;
2524
2525             avctx->profile = hx->sps.profile_idc;
2526             avctx->level   = hx->sps.level_idc;
2527
2528             if (s->avctx->hwaccel && h->current_slice == 1) {
2529                 if (s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
2530                     return -1;
2531             }
2532
2533             s->current_picture_ptr->key_frame |=
2534                     (hx->nal_unit_type == NAL_IDR_SLICE) ||
2535                     (h->sei_recovery_frame_cnt >= 0);
2536             if(hx->redundant_pic_count==0 && hx->s.hurry_up < 5
2537                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2538                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2539                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2540                && avctx->skip_frame < AVDISCARD_ALL){
2541                 if(avctx->hwaccel) {
2542                     if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
2543                         return -1;
2544                 }else
2545                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
2546                     static const uint8_t start_code[] = {0x00, 0x00, 0x01};
2547                     ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
2548                     ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
2549                 }else
2550                     context_count++;
2551             }
2552             break;
2553         case NAL_DPA:
2554             init_get_bits(&hx->s.gb, ptr, bit_length);
2555             hx->intra_gb_ptr=
2556             hx->inter_gb_ptr= NULL;
2557
2558             if ((err = decode_slice_header(hx, h)) < 0)
2559                 break;
2560
2561             avctx->profile = hx->sps.profile_idc;
2562             avctx->level   = hx->sps.level_idc;
2563
2564             hx->s.data_partitioning = 1;
2565
2566             break;
2567         case NAL_DPB:
2568             init_get_bits(&hx->intra_gb, ptr, bit_length);
2569             hx->intra_gb_ptr= &hx->intra_gb;
2570             break;
2571         case NAL_DPC:
2572             init_get_bits(&hx->inter_gb, ptr, bit_length);
2573             hx->inter_gb_ptr= &hx->inter_gb;
2574
2575             if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
2576                && s->context_initialized
2577                && s->hurry_up < 5
2578                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2579                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2580                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2581                && avctx->skip_frame < AVDISCARD_ALL)
2582                 context_count++;
2583             break;
2584         case NAL_SEI:
2585             init_get_bits(&s->gb, ptr, bit_length);
2586             ff_h264_decode_sei(h);
2587             break;
2588         case NAL_SPS:
2589             init_get_bits(&s->gb, ptr, bit_length);
2590             ff_h264_decode_seq_parameter_set(h);
2591
2592             if(s->flags& CODEC_FLAG_LOW_DELAY)
2593                 s->low_delay=1;
2594
2595             if(avctx->has_b_frames < 2)
2596                 avctx->has_b_frames= !s->low_delay;
2597             break;
2598         case NAL_PPS:
2599             init_get_bits(&s->gb, ptr, bit_length);
2600
2601             ff_h264_decode_picture_parameter_set(h, bit_length);
2602
2603             break;
2604         case NAL_AUD:
2605         case NAL_END_SEQUENCE:
2606         case NAL_END_STREAM:
2607         case NAL_FILLER_DATA:
2608         case NAL_SPS_EXT:
2609         case NAL_AUXILIARY_SLICE:
2610             break;
2611         default:
2612             av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
2613         }
2614
2615         if(context_count == h->max_contexts) {
2616             execute_decode_slices(h, context_count);
2617             context_count = 0;
2618         }
2619
2620         if (err < 0)
2621             av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
2622         else if(err == 1) {
2623             /* Slice could not be decoded in parallel mode, copy down
2624              * NAL unit stuff to context 0 and restart. Note that
2625              * rbsp_buffer is not transferred, but since we no longer
2626              * run in parallel mode this should not be an issue. */
2627             h->nal_unit_type = hx->nal_unit_type;
2628             h->nal_ref_idc   = hx->nal_ref_idc;
2629             hx = h;
2630             goto again;
2631         }
2632     }
2633     if(context_count)
2634         execute_decode_slices(h, context_count);
2635     return buf_index;
2636 }
2637
2638 /**
2639  * returns the number of bytes consumed for building the current frame
2640  */
2641 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
2642         if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
2643         if(pos+10>buf_size) pos=buf_size; // oops ;)
2644
2645         return pos;
2646 }
2647
2648 static int decode_frame(AVCodecContext *avctx,
2649                              void *data, int *data_size,
2650                              AVPacket *avpkt)
2651 {
2652     const uint8_t *buf = avpkt->data;
2653     int buf_size = avpkt->size;
2654     H264Context *h = avctx->priv_data;
2655     MpegEncContext *s = &h->s;
2656     AVFrame *pict = data;
2657     int buf_index;
2658
2659     s->flags= avctx->flags;
2660     s->flags2= avctx->flags2;
2661
2662    /* end of stream, output what is still in the buffers */
2663     if (buf_size == 0) {
2664         Picture *out;
2665         int i, out_idx;
2666
2667 //FIXME factorize this with the output code below
2668         out = h->delayed_pic[0];
2669         out_idx = 0;
2670         for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2671             if(h->delayed_pic[i]->poc < out->poc){
2672                 out = h->delayed_pic[i];
2673                 out_idx = i;
2674             }
2675
2676         for(i=out_idx; h->delayed_pic[i]; i++)
2677             h->delayed_pic[i] = h->delayed_pic[i+1];
2678
2679         if(out){
2680             *data_size = sizeof(AVFrame);
2681             *pict= *(AVFrame*)out;
2682         }
2683
2684         return 0;
2685     }
2686
2687     if(h->is_avc && !h->got_avcC) {
2688         int i, cnt, nalsize;
2689         unsigned char *p = avctx->extradata;
2690         if(avctx->extradata_size < 7) {
2691             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
2692             return -1;
2693         }
2694         if(*p != 1) {
2695             av_log(avctx, AV_LOG_ERROR, "Unknown avcC version %d\n", *p);
2696             return -1;
2697         }
2698         /* sps and pps in the avcC always have length coded with 2 bytes,
2699            so put a fake nal_length_size = 2 while parsing them */
2700         h->nal_length_size = 2;
2701         // Decode sps from avcC
2702         cnt = *(p+5) & 0x1f; // Number of sps
2703         p += 6;
2704         for (i = 0; i < cnt; i++) {
2705             nalsize = AV_RB16(p) + 2;
2706             if(decode_nal_units(h, p, nalsize) < 0) {
2707                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
2708                 return -1;
2709             }
2710             p += nalsize;
2711         }
2712         // Decode pps from avcC
2713         cnt = *(p++); // Number of pps
2714         for (i = 0; i < cnt; i++) {
2715             nalsize = AV_RB16(p) + 2;
2716             if(decode_nal_units(h, p, nalsize)  != nalsize) {
2717                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
2718                 return -1;
2719             }
2720             p += nalsize;
2721         }
2722         // Now store right nal length size, that will be use to parse all other nals
2723         h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
2724         // Do not reparse avcC
2725         h->got_avcC = 1;
2726     }
2727
2728     if(!h->got_avcC && !h->is_avc && s->avctx->extradata_size){
2729         if(decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) < 0)
2730             return -1;
2731         h->got_avcC = 1;
2732     }
2733
2734     buf_index=decode_nal_units(h, buf, buf_size);
2735     if(buf_index < 0)
2736         return -1;
2737
2738     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
2739         if (avctx->skip_frame >= AVDISCARD_NONREF || s->hurry_up) return 0;
2740         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
2741         return -1;
2742     }
2743
2744     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
2745         Picture *out = s->current_picture_ptr;
2746         Picture *cur = s->current_picture_ptr;
2747         int i, pics, out_of_order, out_idx;
2748
2749         field_end(h);
2750
2751         if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
2752             /* Wait for second field. */
2753             *data_size = 0;
2754
2755         } else {
2756             cur->interlaced_frame = 0;
2757             cur->repeat_pict = 0;
2758
2759             /* Signal interlacing information externally. */
2760             /* Prioritize picture timing SEI information over used decoding process if it exists. */
2761
2762             if(h->sps.pic_struct_present_flag){
2763                 switch (h->sei_pic_struct)
2764                 {
2765                 case SEI_PIC_STRUCT_FRAME:
2766                     break;
2767                 case SEI_PIC_STRUCT_TOP_FIELD:
2768                 case SEI_PIC_STRUCT_BOTTOM_FIELD:
2769                     cur->interlaced_frame = 1;
2770                     break;
2771                 case SEI_PIC_STRUCT_TOP_BOTTOM:
2772                 case SEI_PIC_STRUCT_BOTTOM_TOP:
2773                     if (FIELD_OR_MBAFF_PICTURE)
2774                         cur->interlaced_frame = 1;
2775                     else
2776                         // try to flag soft telecine progressive
2777                         cur->interlaced_frame = h->prev_interlaced_frame;
2778                     break;
2779                 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
2780                 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
2781                     // Signal the possibility of telecined film externally (pic_struct 5,6)
2782                     // From these hints, let the applications decide if they apply deinterlacing.
2783                     cur->repeat_pict = 1;
2784                     break;
2785                 case SEI_PIC_STRUCT_FRAME_DOUBLING:
2786                     // Force progressive here, as doubling interlaced frame is a bad idea.
2787                     cur->repeat_pict = 2;
2788                     break;
2789                 case SEI_PIC_STRUCT_FRAME_TRIPLING:
2790                     cur->repeat_pict = 4;
2791                     break;
2792                 }
2793
2794                 if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
2795                     cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
2796             }else{
2797                 /* Derive interlacing flag from used decoding process. */
2798                 cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
2799             }
2800             h->prev_interlaced_frame = cur->interlaced_frame;
2801
2802             if (cur->field_poc[0] != cur->field_poc[1]){
2803                 /* Derive top_field_first from field pocs. */
2804                 cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
2805             }else{
2806                 if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
2807                     /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
2808                     if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
2809                       || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
2810                         cur->top_field_first = 1;
2811                     else
2812                         cur->top_field_first = 0;
2813                 }else{
2814                     /* Most likely progressive */
2815                     cur->top_field_first = 0;
2816                 }
2817             }
2818
2819         //FIXME do something with unavailable reference frames
2820
2821             /* Sort B-frames into display order */
2822
2823             if(h->sps.bitstream_restriction_flag
2824                && s->avctx->has_b_frames < h->sps.num_reorder_frames){
2825                 s->avctx->has_b_frames = h->sps.num_reorder_frames;
2826                 s->low_delay = 0;
2827             }
2828
2829             if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
2830                && !h->sps.bitstream_restriction_flag){
2831                 s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
2832                 s->low_delay= 0;
2833             }
2834
2835             pics = 0;
2836             while(h->delayed_pic[pics]) pics++;
2837
2838             assert(pics <= MAX_DELAYED_PIC_COUNT);
2839
2840             h->delayed_pic[pics++] = cur;
2841             if(cur->reference == 0)
2842                 cur->reference = DELAYED_PIC_REF;
2843
2844             out = h->delayed_pic[0];
2845             out_idx = 0;
2846             for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2847                 if(h->delayed_pic[i]->poc < out->poc){
2848                     out = h->delayed_pic[i];
2849                     out_idx = i;
2850                 }
2851             if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
2852                 h->outputed_poc= INT_MIN;
2853             out_of_order = out->poc < h->outputed_poc;
2854
2855             if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
2856                 { }
2857             else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
2858                || (s->low_delay &&
2859                 ((h->outputed_poc != INT_MIN && out->poc > h->outputed_poc + 2)
2860                  || cur->pict_type == FF_B_TYPE)))
2861             {
2862                 s->low_delay = 0;
2863                 s->avctx->has_b_frames++;
2864             }
2865
2866             if(out_of_order || pics > s->avctx->has_b_frames){
2867                 out->reference &= ~DELAYED_PIC_REF;
2868                 for(i=out_idx; h->delayed_pic[i]; i++)
2869                     h->delayed_pic[i] = h->delayed_pic[i+1];
2870             }
2871             if(!out_of_order && pics > s->avctx->has_b_frames){
2872                 *data_size = sizeof(AVFrame);
2873
2874                 if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
2875                     h->outputed_poc = INT_MIN;
2876                 } else
2877                     h->outputed_poc = out->poc;
2878                 *pict= *(AVFrame*)out;
2879             }else{
2880                 av_log(avctx, AV_LOG_DEBUG, "no picture\n");
2881             }
2882         }
2883     }
2884
2885     assert(pict->data[0] || !*data_size);
2886     ff_print_debug_info(s, pict);
2887 //printf("out %d\n", (int)pict->data[0]);
2888
2889     return get_consumed_bytes(s, buf_index, buf_size);
2890 }
2891 #if 0
2892 static inline void fill_mb_avail(H264Context *h){
2893     MpegEncContext * const s = &h->s;
2894     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2895
2896     if(s->mb_y){
2897         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
2898         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
2899         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
2900     }else{
2901         h->mb_avail[0]=
2902         h->mb_avail[1]=
2903         h->mb_avail[2]= 0;
2904     }
2905     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
2906     h->mb_avail[4]= 1; //FIXME move out
2907     h->mb_avail[5]= 0; //FIXME move out
2908 }
2909 #endif
2910
2911 #ifdef TEST
2912 #undef printf
2913 #undef random
2914 #define COUNT 8000
2915 #define SIZE (COUNT*40)
2916 int main(void){
2917     int i;
2918     uint8_t temp[SIZE];
2919     PutBitContext pb;
2920     GetBitContext gb;
2921 //    int int_temp[10000];
2922     DSPContext dsp;
2923     AVCodecContext avctx;
2924
2925     dsputil_init(&dsp, &avctx);
2926
2927     init_put_bits(&pb, temp, SIZE);
2928     printf("testing unsigned exp golomb\n");
2929     for(i=0; i<COUNT; i++){
2930         START_TIMER
2931         set_ue_golomb(&pb, i);
2932         STOP_TIMER("set_ue_golomb");
2933     }
2934     flush_put_bits(&pb);
2935
2936     init_get_bits(&gb, temp, 8*SIZE);
2937     for(i=0; i<COUNT; i++){
2938         int j, s;
2939
2940         s= show_bits(&gb, 24);
2941
2942         START_TIMER
2943         j= get_ue_golomb(&gb);
2944         if(j != i){
2945             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
2946 //            return -1;
2947         }
2948         STOP_TIMER("get_ue_golomb");
2949     }
2950
2951
2952     init_put_bits(&pb, temp, SIZE);
2953     printf("testing signed exp golomb\n");
2954     for(i=0; i<COUNT; i++){
2955         START_TIMER
2956         set_se_golomb(&pb, i - COUNT/2);
2957         STOP_TIMER("set_se_golomb");
2958     }
2959     flush_put_bits(&pb);
2960
2961     init_get_bits(&gb, temp, 8*SIZE);
2962     for(i=0; i<COUNT; i++){
2963         int j, s;
2964
2965         s= show_bits(&gb, 24);
2966
2967         START_TIMER
2968         j= get_se_golomb(&gb);
2969         if(j != i - COUNT/2){
2970             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
2971 //            return -1;
2972         }
2973         STOP_TIMER("get_se_golomb");
2974     }
2975
2976 #if 0
2977     printf("testing 4x4 (I)DCT\n");
2978
2979     DCTELEM block[16];
2980     uint8_t src[16], ref[16];
2981     uint64_t error= 0, max_error=0;
2982
2983     for(i=0; i<COUNT; i++){
2984         int j;
2985 //        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
2986         for(j=0; j<16; j++){
2987             ref[j]= random()%255;
2988             src[j]= random()%255;
2989         }
2990
2991         h264_diff_dct_c(block, src, ref, 4);
2992
2993         //normalize
2994         for(j=0; j<16; j++){
2995 //            printf("%d ", block[j]);
2996             block[j]= block[j]*4;
2997             if(j&1) block[j]= (block[j]*4 + 2)/5;
2998             if(j&4) block[j]= (block[j]*4 + 2)/5;
2999         }
3000 //        printf("\n");
3001
3002         s->dsp.h264_idct_add(ref, block, 4);
3003 /*        for(j=0; j<16; j++){
3004             printf("%d ", ref[j]);
3005         }
3006         printf("\n");*/
3007
3008         for(j=0; j<16; j++){
3009             int diff= FFABS(src[j] - ref[j]);
3010
3011             error+= diff*diff;
3012             max_error= FFMAX(max_error, diff);
3013         }
3014     }
3015     printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
3016     printf("testing quantizer\n");
3017     for(qp=0; qp<52; qp++){
3018         for(i=0; i<16; i++)
3019             src1_block[i]= src2_block[i]= random()%255;
3020
3021     }
3022     printf("Testing NAL layer\n");
3023
3024     uint8_t bitstream[COUNT];
3025     uint8_t nal[COUNT*2];
3026     H264Context h;
3027     memset(&h, 0, sizeof(H264Context));
3028
3029     for(i=0; i<COUNT; i++){
3030         int zeros= i;
3031         int nal_length;
3032         int consumed;
3033         int out_length;
3034         uint8_t *out;
3035         int j;
3036
3037         for(j=0; j<COUNT; j++){
3038             bitstream[j]= (random() % 255) + 1;
3039         }
3040
3041         for(j=0; j<zeros; j++){
3042             int pos= random() % COUNT;
3043             while(bitstream[pos] == 0){
3044                 pos++;
3045                 pos %= COUNT;
3046             }
3047             bitstream[pos]=0;
3048         }
3049
3050         START_TIMER
3051
3052         nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
3053         if(nal_length<0){
3054             printf("encoding failed\n");
3055             return -1;
3056         }
3057
3058         out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
3059
3060         STOP_TIMER("NAL")
3061
3062         if(out_length != COUNT){
3063             printf("incorrect length %d %d\n", out_length, COUNT);
3064             return -1;
3065         }
3066
3067         if(consumed != nal_length){
3068             printf("incorrect consumed length %d %d\n", nal_length, consumed);
3069             return -1;
3070         }
3071
3072         if(memcmp(bitstream, out, COUNT)){
3073             printf("mismatch\n");
3074             return -1;
3075         }
3076     }
3077 #endif
3078
3079     printf("Testing RBSP\n");
3080
3081
3082     return 0;
3083 }
3084 #endif /* TEST */
3085
3086
3087 av_cold void ff_h264_free_context(H264Context *h)
3088 {
3089     int i;
3090
3091     free_tables(h); //FIXME cleanup init stuff perhaps
3092
3093     for(i = 0; i < MAX_SPS_COUNT; i++)
3094         av_freep(h->sps_buffers + i);
3095
3096     for(i = 0; i < MAX_PPS_COUNT; i++)
3097         av_freep(h->pps_buffers + i);
3098 }
3099
3100 av_cold int ff_h264_decode_end(AVCodecContext *avctx)
3101 {
3102     H264Context *h = avctx->priv_data;
3103     MpegEncContext *s = &h->s;
3104
3105     ff_h264_free_context(h);
3106
3107     MPV_common_end(s);
3108
3109 //    memset(h, 0, sizeof(H264Context));
3110
3111     return 0;
3112 }
3113
3114
3115 AVCodec h264_decoder = {
3116     "h264",
3117     CODEC_TYPE_VIDEO,
3118     CODEC_ID_H264,
3119     sizeof(H264Context),
3120     ff_h264_decode_init,
3121     NULL,
3122     ff_h264_decode_end,
3123     decode_frame,
3124     /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3125     .flush= flush_dpb,
3126     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
3127     .pix_fmts= ff_hwaccel_pixfmt_list_420,
3128 };
3129
3130 #if CONFIG_H264_VDPAU_DECODER
3131 AVCodec h264_vdpau_decoder = {
3132     "h264_vdpau",
3133     CODEC_TYPE_VIDEO,
3134     CODEC_ID_H264,
3135     sizeof(H264Context),
3136     ff_h264_decode_init,
3137     NULL,
3138     ff_h264_decode_end,
3139     decode_frame,
3140     CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3141     .flush= flush_dpb,
3142     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
3143     .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
3144 };
3145 #endif