Simplify deblock_left/top condition for deblocking_filter=2
[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         deblock_left = h->left_type[0];
1000         deblock_top  = h->top_type;
1001     } else {
1002         deblock_left = (s->mb_x > 0);
1003         deblock_top =  (s->mb_y > !!MB_FIELD);
1004     }
1005
1006     src_y  -=   linesize + 1;
1007     src_cb -= uvlinesize + 1;
1008     src_cr -= uvlinesize + 1;
1009
1010     top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1011     top_border    = h->top_borders[top_idx][s->mb_x];
1012
1013 #define XCHG(a,b,xchg)\
1014 if (xchg) AV_SWAP64(b,a);\
1015 else      AV_COPY64(b,a);
1016
1017     if(deblock_top){
1018         if(deblock_left){
1019             XCHG(top_border_m1+8, src_y -7, 1);
1020         }
1021         XCHG(top_border+0, src_y +1, xchg);
1022         XCHG(top_border+8, src_y +9, 1);
1023         if(s->mb_x+1 < s->mb_width){
1024             XCHG(h->top_borders[top_idx][s->mb_x+1], src_y +17, 1);
1025         }
1026     }
1027
1028     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1029         if(deblock_top){
1030             if(deblock_left){
1031                 XCHG(top_border_m1+16, src_cb -7, 1);
1032                 XCHG(top_border_m1+24, src_cr -7, 1);
1033             }
1034             XCHG(top_border+16, src_cb+1, 1);
1035             XCHG(top_border+24, src_cr+1, 1);
1036         }
1037     }
1038 }
1039
1040 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
1041     MpegEncContext * const s = &h->s;
1042     const int mb_x= s->mb_x;
1043     const int mb_y= s->mb_y;
1044     const int mb_xy= h->mb_xy;
1045     const int mb_type= s->current_picture.mb_type[mb_xy];
1046     uint8_t  *dest_y, *dest_cb, *dest_cr;
1047     int linesize, uvlinesize /*dct_offset*/;
1048     int i;
1049     int *block_offset = &h->block_offset[0];
1050     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1051     /* is_h264 should always be true if SVQ3 is disabled. */
1052     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1053     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1054     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1055
1056     dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
1057     dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
1058     dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
1059
1060     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
1061     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
1062
1063     h->list_counts[mb_xy]= h->list_count;
1064
1065     if (!simple && MB_FIELD) {
1066         linesize   = h->mb_linesize   = s->linesize * 2;
1067         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1068         block_offset = &h->block_offset[24];
1069         if(mb_y&1){ //FIXME move out of this function?
1070             dest_y -= s->linesize*15;
1071             dest_cb-= s->uvlinesize*7;
1072             dest_cr-= s->uvlinesize*7;
1073         }
1074         if(FRAME_MBAFF) {
1075             int list;
1076             for(list=0; list<h->list_count; list++){
1077                 if(!USES_LIST(mb_type, list))
1078                     continue;
1079                 if(IS_16X16(mb_type)){
1080                     int8_t *ref = &h->ref_cache[list][scan8[0]];
1081                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1082                 }else{
1083                     for(i=0; i<16; i+=4){
1084                         int ref = h->ref_cache[list][scan8[i]];
1085                         if(ref >= 0)
1086                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1087                     }
1088                 }
1089             }
1090         }
1091     } else {
1092         linesize   = h->mb_linesize   = s->linesize;
1093         uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1094 //        dct_offset = s->linesize * 16;
1095     }
1096
1097     if (!simple && IS_INTRA_PCM(mb_type)) {
1098         for (i=0; i<16; i++) {
1099             memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1100         }
1101         for (i=0; i<8; i++) {
1102             memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
1103             memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
1104         }
1105     } else {
1106         if(IS_INTRA(mb_type)){
1107             if(h->deblocking_filter)
1108                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
1109
1110             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1111                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1112                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1113             }
1114
1115             if(IS_INTRA4x4(mb_type)){
1116                 if(simple || !s->encoding){
1117                     if(IS_8x8DCT(mb_type)){
1118                         if(transform_bypass){
1119                             idct_dc_add =
1120                             idct_add    = s->dsp.add_pixels8;
1121                         }else{
1122                             idct_dc_add = s->dsp.h264_idct8_dc_add;
1123                             idct_add    = s->dsp.h264_idct8_add;
1124                         }
1125                         for(i=0; i<16; i+=4){
1126                             uint8_t * const ptr= dest_y + block_offset[i];
1127                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1128                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1129                                 h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
1130                             }else{
1131                                 const int nnz = h->non_zero_count_cache[ scan8[i] ];
1132                                 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1133                                                             (h->topright_samples_available<<i)&0x4000, linesize);
1134                                 if(nnz){
1135                                     if(nnz == 1 && h->mb[i*16])
1136                                         idct_dc_add(ptr, h->mb + i*16, linesize);
1137                                     else
1138                                         idct_add   (ptr, h->mb + i*16, linesize);
1139                                 }
1140                             }
1141                         }
1142                     }else{
1143                         if(transform_bypass){
1144                             idct_dc_add =
1145                             idct_add    = s->dsp.add_pixels4;
1146                         }else{
1147                             idct_dc_add = s->dsp.h264_idct_dc_add;
1148                             idct_add    = s->dsp.h264_idct_add;
1149                         }
1150                         for(i=0; i<16; i++){
1151                             uint8_t * const ptr= dest_y + block_offset[i];
1152                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1153
1154                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1155                                 h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
1156                             }else{
1157                                 uint8_t *topright;
1158                                 int nnz, tr;
1159                                 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1160                                     const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1161                                     assert(mb_y || linesize <= block_offset[i]);
1162                                     if(!topright_avail){
1163                                         tr= ptr[3 - linesize]*0x01010101;
1164                                         topright= (uint8_t*) &tr;
1165                                     }else
1166                                         topright= ptr + 4 - linesize;
1167                                 }else
1168                                     topright= NULL;
1169
1170                                 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1171                                 nnz = h->non_zero_count_cache[ scan8[i] ];
1172                                 if(nnz){
1173                                     if(is_h264){
1174                                         if(nnz == 1 && h->mb[i*16])
1175                                             idct_dc_add(ptr, h->mb + i*16, linesize);
1176                                         else
1177                                             idct_add   (ptr, h->mb + i*16, linesize);
1178                                     }else
1179                                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1180                                 }
1181                             }
1182                         }
1183                     }
1184                 }
1185             }else{
1186                 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1187                 if(is_h264){
1188                     if(!transform_bypass)
1189                         h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
1190                 }else
1191                     ff_svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
1192             }
1193             if(h->deblocking_filter)
1194                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
1195         }else if(is_h264){
1196             hl_motion(h, dest_y, dest_cb, dest_cr,
1197                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1198                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1199                       s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
1200         }
1201
1202
1203         if(!IS_INTRA4x4(mb_type)){
1204             if(is_h264){
1205                 if(IS_INTRA16x16(mb_type)){
1206                     if(transform_bypass){
1207                         if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1208                             h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1209                         }else{
1210                             for(i=0; i<16; i++){
1211                                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1212                                     s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
1213                             }
1214                         }
1215                     }else{
1216                          s->dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1217                     }
1218                 }else if(h->cbp&15){
1219                     if(transform_bypass){
1220                         const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1221                         idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1222                         for(i=0; i<16; i+=di){
1223                             if(h->non_zero_count_cache[ scan8[i] ]){
1224                                 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
1225                             }
1226                         }
1227                     }else{
1228                         if(IS_8x8DCT(mb_type)){
1229                             s->dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1230                         }else{
1231                             s->dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1232                         }
1233                     }
1234                 }
1235             }else{
1236                 for(i=0; i<16; i++){
1237                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1238                         uint8_t * const ptr= dest_y + block_offset[i];
1239                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1240                     }
1241                 }
1242             }
1243         }
1244
1245         if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1246             uint8_t *dest[2] = {dest_cb, dest_cr};
1247             if(transform_bypass){
1248                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1249                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
1250                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
1251                 }else{
1252                     idct_add = s->dsp.add_pixels4;
1253                     for(i=16; i<16+8; i++){
1254                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1255                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1256                     }
1257                 }
1258             }else{
1259                 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]);
1260                 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]);
1261                 if(is_h264){
1262                     idct_add = s->dsp.h264_idct_add;
1263                     idct_dc_add = s->dsp.h264_idct_dc_add;
1264                     for(i=16; i<16+8; i++){
1265                         if(h->non_zero_count_cache[ scan8[i] ])
1266                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1267                         else if(h->mb[i*16])
1268                             idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1269                     }
1270                 }else{
1271                     for(i=16; i<16+8; i++){
1272                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1273                             uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1274                             ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
1275                         }
1276                     }
1277                 }
1278             }
1279         }
1280     }
1281     if(h->cbp || IS_INTRA(mb_type))
1282         s->dsp.clear_blocks(h->mb);
1283 }
1284
1285 /**
1286  * Process a macroblock; this case avoids checks for expensive uncommon cases.
1287  */
1288 static void hl_decode_mb_simple(H264Context *h){
1289     hl_decode_mb_internal(h, 1);
1290 }
1291
1292 /**
1293  * Process a macroblock; this handles edge cases, such as interlacing.
1294  */
1295 static void av_noinline hl_decode_mb_complex(H264Context *h){
1296     hl_decode_mb_internal(h, 0);
1297 }
1298
1299 void ff_h264_hl_decode_mb(H264Context *h){
1300     MpegEncContext * const s = &h->s;
1301     const int mb_xy= h->mb_xy;
1302     const int mb_type= s->current_picture.mb_type[mb_xy];
1303     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1304
1305     if (is_complex)
1306         hl_decode_mb_complex(h);
1307     else hl_decode_mb_simple(h);
1308 }
1309
1310 static int pred_weight_table(H264Context *h){
1311     MpegEncContext * const s = &h->s;
1312     int list, i;
1313     int luma_def, chroma_def;
1314
1315     h->use_weight= 0;
1316     h->use_weight_chroma= 0;
1317     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1318     h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1319     luma_def = 1<<h->luma_log2_weight_denom;
1320     chroma_def = 1<<h->chroma_log2_weight_denom;
1321
1322     for(list=0; list<2; list++){
1323         h->luma_weight_flag[list]   = 0;
1324         h->chroma_weight_flag[list] = 0;
1325         for(i=0; i<h->ref_count[list]; i++){
1326             int luma_weight_flag, chroma_weight_flag;
1327
1328             luma_weight_flag= get_bits1(&s->gb);
1329             if(luma_weight_flag){
1330                 h->luma_weight[list][i]= get_se_golomb(&s->gb);
1331                 h->luma_offset[list][i]= get_se_golomb(&s->gb);
1332                 if(   h->luma_weight[list][i] != luma_def
1333                    || h->luma_offset[list][i] != 0) {
1334                     h->use_weight= 1;
1335                     h->luma_weight_flag[list]= 1;
1336                 }
1337             }else{
1338                 h->luma_weight[list][i]= luma_def;
1339                 h->luma_offset[list][i]= 0;
1340             }
1341
1342             if(CHROMA){
1343                 chroma_weight_flag= get_bits1(&s->gb);
1344                 if(chroma_weight_flag){
1345                     int j;
1346                     for(j=0; j<2; j++){
1347                         h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
1348                         h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
1349                         if(   h->chroma_weight[list][i][j] != chroma_def
1350                            || h->chroma_offset[list][i][j] != 0) {
1351                             h->use_weight_chroma= 1;
1352                             h->chroma_weight_flag[list]= 1;
1353                         }
1354                     }
1355                 }else{
1356                     int j;
1357                     for(j=0; j<2; j++){
1358                         h->chroma_weight[list][i][j]= chroma_def;
1359                         h->chroma_offset[list][i][j]= 0;
1360                     }
1361                 }
1362             }
1363         }
1364         if(h->slice_type_nos != FF_B_TYPE) break;
1365     }
1366     h->use_weight= h->use_weight || h->use_weight_chroma;
1367     return 0;
1368 }
1369
1370 static void implicit_weight_table(H264Context *h){
1371     MpegEncContext * const s = &h->s;
1372     int ref0, ref1, i;
1373     int cur_poc = s->current_picture_ptr->poc;
1374
1375     for (i = 0; i < 2; i++) {
1376         h->luma_weight_flag[i]   = 0;
1377         h->chroma_weight_flag[i] = 0;
1378     }
1379
1380     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
1381        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1382         h->use_weight= 0;
1383         h->use_weight_chroma= 0;
1384         return;
1385     }
1386
1387     h->use_weight= 2;
1388     h->use_weight_chroma= 2;
1389     h->luma_log2_weight_denom= 5;
1390     h->chroma_log2_weight_denom= 5;
1391
1392     for(ref0=0; ref0 < h->ref_count[0]; ref0++){
1393         int poc0 = h->ref_list[0][ref0].poc;
1394         for(ref1=0; ref1 < h->ref_count[1]; ref1++){
1395             int poc1 = h->ref_list[1][ref1].poc;
1396             int td = av_clip(poc1 - poc0, -128, 127);
1397             if(td){
1398                 int tb = av_clip(cur_poc - poc0, -128, 127);
1399                 int tx = (16384 + (FFABS(td) >> 1)) / td;
1400                 int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
1401                 if(dist_scale_factor < -64 || dist_scale_factor > 128)
1402                     h->implicit_weight[ref0][ref1] = 32;
1403                 else
1404                     h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
1405             }else
1406                 h->implicit_weight[ref0][ref1] = 32;
1407         }
1408     }
1409 }
1410
1411 /**
1412  * instantaneous decoder refresh.
1413  */
1414 static void idr(H264Context *h){
1415     ff_h264_remove_all_refs(h);
1416     h->prev_frame_num= 0;
1417     h->prev_frame_num_offset= 0;
1418     h->prev_poc_msb=
1419     h->prev_poc_lsb= 0;
1420 }
1421
1422 /* forget old pics after a seek */
1423 static void flush_dpb(AVCodecContext *avctx){
1424     H264Context *h= avctx->priv_data;
1425     int i;
1426     for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1427         if(h->delayed_pic[i])
1428             h->delayed_pic[i]->reference= 0;
1429         h->delayed_pic[i]= NULL;
1430     }
1431     h->outputed_poc= INT_MIN;
1432     h->prev_interlaced_frame = 1;
1433     idr(h);
1434     if(h->s.current_picture_ptr)
1435         h->s.current_picture_ptr->reference= 0;
1436     h->s.first_field= 0;
1437     ff_h264_reset_sei(h);
1438     ff_mpeg_flush(avctx);
1439 }
1440
1441 static int init_poc(H264Context *h){
1442     MpegEncContext * const s = &h->s;
1443     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1444     int field_poc[2];
1445     Picture *cur = s->current_picture_ptr;
1446
1447     h->frame_num_offset= h->prev_frame_num_offset;
1448     if(h->frame_num < h->prev_frame_num)
1449         h->frame_num_offset += max_frame_num;
1450
1451     if(h->sps.poc_type==0){
1452         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1453
1454         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1455             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1456         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1457             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1458         else
1459             h->poc_msb = h->prev_poc_msb;
1460 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1461         field_poc[0] =
1462         field_poc[1] = h->poc_msb + h->poc_lsb;
1463         if(s->picture_structure == PICT_FRAME)
1464             field_poc[1] += h->delta_poc_bottom;
1465     }else if(h->sps.poc_type==1){
1466         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1467         int i;
1468
1469         if(h->sps.poc_cycle_length != 0)
1470             abs_frame_num = h->frame_num_offset + h->frame_num;
1471         else
1472             abs_frame_num = 0;
1473
1474         if(h->nal_ref_idc==0 && abs_frame_num > 0)
1475             abs_frame_num--;
1476
1477         expected_delta_per_poc_cycle = 0;
1478         for(i=0; i < h->sps.poc_cycle_length; i++)
1479             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1480
1481         if(abs_frame_num > 0){
1482             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1483             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1484
1485             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1486             for(i = 0; i <= frame_num_in_poc_cycle; i++)
1487                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1488         } else
1489             expectedpoc = 0;
1490
1491         if(h->nal_ref_idc == 0)
1492             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1493
1494         field_poc[0] = expectedpoc + h->delta_poc[0];
1495         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1496
1497         if(s->picture_structure == PICT_FRAME)
1498             field_poc[1] += h->delta_poc[1];
1499     }else{
1500         int poc= 2*(h->frame_num_offset + h->frame_num);
1501
1502         if(!h->nal_ref_idc)
1503             poc--;
1504
1505         field_poc[0]= poc;
1506         field_poc[1]= poc;
1507     }
1508
1509     if(s->picture_structure != PICT_BOTTOM_FIELD)
1510         s->current_picture_ptr->field_poc[0]= field_poc[0];
1511     if(s->picture_structure != PICT_TOP_FIELD)
1512         s->current_picture_ptr->field_poc[1]= field_poc[1];
1513     cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1514
1515     return 0;
1516 }
1517
1518
1519 /**
1520  * initialize scan tables
1521  */
1522 static void init_scan_tables(H264Context *h){
1523     MpegEncContext * const s = &h->s;
1524     int i;
1525     if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
1526         memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
1527         memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
1528     }else{
1529         for(i=0; i<16; i++){
1530 #define T(x) (x>>2) | ((x<<2) & 0xF)
1531             h->zigzag_scan[i] = T(zigzag_scan[i]);
1532             h-> field_scan[i] = T( field_scan[i]);
1533 #undef T
1534         }
1535     }
1536     if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
1537         memcpy(h->zigzag_scan8x8,       ff_zigzag_direct,     64*sizeof(uint8_t));
1538         memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
1539         memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
1540         memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
1541     }else{
1542         for(i=0; i<64; i++){
1543 #define T(x) (x>>3) | ((x&7)<<3)
1544             h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
1545             h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1546             h->field_scan8x8[i]        = T(field_scan8x8[i]);
1547             h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
1548 #undef T
1549         }
1550     }
1551     if(h->sps.transform_bypass){ //FIXME same ugly
1552         h->zigzag_scan_q0          = zigzag_scan;
1553         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
1554         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1555         h->field_scan_q0           = field_scan;
1556         h->field_scan8x8_q0        = field_scan8x8;
1557         h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
1558     }else{
1559         h->zigzag_scan_q0          = h->zigzag_scan;
1560         h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
1561         h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1562         h->field_scan_q0           = h->field_scan;
1563         h->field_scan8x8_q0        = h->field_scan8x8;
1564         h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
1565     }
1566 }
1567
1568 static void field_end(H264Context *h){
1569     MpegEncContext * const s = &h->s;
1570     AVCodecContext * const avctx= s->avctx;
1571     s->mb_y= 0;
1572
1573     s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1574     s->current_picture_ptr->pict_type= s->pict_type;
1575
1576     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1577         ff_vdpau_h264_set_reference_frames(s);
1578
1579     if(!s->dropable) {
1580         ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1581         h->prev_poc_msb= h->poc_msb;
1582         h->prev_poc_lsb= h->poc_lsb;
1583     }
1584     h->prev_frame_num_offset= h->frame_num_offset;
1585     h->prev_frame_num= h->frame_num;
1586
1587     if (avctx->hwaccel) {
1588         if (avctx->hwaccel->end_frame(avctx) < 0)
1589             av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1590     }
1591
1592     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1593         ff_vdpau_h264_picture_complete(s);
1594
1595     /*
1596      * FIXME: Error handling code does not seem to support interlaced
1597      * when slices span multiple rows
1598      * The ff_er_add_slice calls don't work right for bottom
1599      * fields; they cause massive erroneous error concealing
1600      * Error marking covers both fields (top and bottom).
1601      * This causes a mismatched s->error_count
1602      * and a bad error table. Further, the error count goes to
1603      * INT_MAX when called for bottom field, because mb_y is
1604      * past end by one (callers fault) and resync_mb_y != 0
1605      * causes problems for the first MB line, too.
1606      */
1607     if (!FIELD_PICTURE)
1608         ff_er_frame_end(s);
1609
1610     MPV_frame_end(s);
1611
1612     h->current_slice=0;
1613 }
1614
1615 /**
1616  * Replicates H264 "master" context to thread contexts.
1617  */
1618 static void clone_slice(H264Context *dst, H264Context *src)
1619 {
1620     memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
1621     dst->s.current_picture_ptr  = src->s.current_picture_ptr;
1622     dst->s.current_picture      = src->s.current_picture;
1623     dst->s.linesize             = src->s.linesize;
1624     dst->s.uvlinesize           = src->s.uvlinesize;
1625     dst->s.first_field          = src->s.first_field;
1626
1627     dst->prev_poc_msb           = src->prev_poc_msb;
1628     dst->prev_poc_lsb           = src->prev_poc_lsb;
1629     dst->prev_frame_num_offset  = src->prev_frame_num_offset;
1630     dst->prev_frame_num         = src->prev_frame_num;
1631     dst->short_ref_count        = src->short_ref_count;
1632
1633     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
1634     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
1635     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1636     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
1637
1638     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
1639     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
1640 }
1641
1642 /**
1643  * decodes a slice header.
1644  * This will also call MPV_common_init() and frame_start() as needed.
1645  *
1646  * @param h h264context
1647  * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1648  *
1649  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1650  */
1651 static int decode_slice_header(H264Context *h, H264Context *h0){
1652     MpegEncContext * const s = &h->s;
1653     MpegEncContext * const s0 = &h0->s;
1654     unsigned int first_mb_in_slice;
1655     unsigned int pps_id;
1656     int num_ref_idx_active_override_flag;
1657     unsigned int slice_type, tmp, i, j;
1658     int default_ref_list_done = 0;
1659     int last_pic_structure;
1660
1661     s->dropable= h->nal_ref_idc == 0;
1662
1663     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1664         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1665         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1666     }else{
1667         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1668         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1669     }
1670
1671     first_mb_in_slice= get_ue_golomb(&s->gb);
1672
1673     if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1674         if(h0->current_slice && FIELD_PICTURE){
1675             field_end(h);
1676         }
1677
1678         h0->current_slice = 0;
1679         if (!s0->first_field)
1680             s->current_picture_ptr= NULL;
1681     }
1682
1683     slice_type= get_ue_golomb_31(&s->gb);
1684     if(slice_type > 9){
1685         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);
1686         return -1;
1687     }
1688     if(slice_type > 4){
1689         slice_type -= 5;
1690         h->slice_type_fixed=1;
1691     }else
1692         h->slice_type_fixed=0;
1693
1694     slice_type= golomb_to_pict_type[ slice_type ];
1695     if (slice_type == FF_I_TYPE
1696         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
1697         default_ref_list_done = 1;
1698     }
1699     h->slice_type= slice_type;
1700     h->slice_type_nos= slice_type & 3;
1701
1702     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
1703
1704     pps_id= get_ue_golomb(&s->gb);
1705     if(pps_id>=MAX_PPS_COUNT){
1706         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
1707         return -1;
1708     }
1709     if(!h0->pps_buffers[pps_id]) {
1710         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
1711         return -1;
1712     }
1713     h->pps= *h0->pps_buffers[pps_id];
1714
1715     if(!h0->sps_buffers[h->pps.sps_id]) {
1716         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
1717         return -1;
1718     }
1719     h->sps = *h0->sps_buffers[h->pps.sps_id];
1720
1721     if(h == h0 && h->dequant_coeff_pps != pps_id){
1722         h->dequant_coeff_pps = pps_id;
1723         init_dequant_tables(h);
1724     }
1725
1726     s->mb_width= h->sps.mb_width;
1727     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1728
1729     h->b_stride=  s->mb_width*4;
1730     h->b8_stride= s->mb_width*2;
1731
1732     s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
1733     if(h->sps.frame_mbs_only_flag)
1734         s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
1735     else
1736         s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
1737
1738     if (s->context_initialized
1739         && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
1740         if(h != h0)
1741             return -1;   // width / height changed during parallelized decoding
1742         free_tables(h);
1743         flush_dpb(s->avctx);
1744         MPV_common_end(s);
1745     }
1746     if (!s->context_initialized) {
1747         if(h != h0)
1748             return -1;  // we cant (re-)initialize context during parallel decoding
1749
1750         avcodec_set_dimensions(s->avctx, s->width, s->height);
1751         s->avctx->sample_aspect_ratio= h->sps.sar;
1752         if(!s->avctx->sample_aspect_ratio.den)
1753             s->avctx->sample_aspect_ratio.den = 1;
1754
1755         if(h->sps.video_signal_type_present_flag){
1756             s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1757             if(h->sps.colour_description_present_flag){
1758                 s->avctx->color_primaries = h->sps.color_primaries;
1759                 s->avctx->color_trc       = h->sps.color_trc;
1760                 s->avctx->colorspace      = h->sps.colorspace;
1761             }
1762         }
1763
1764         if(h->sps.timing_info_present_flag){
1765             int64_t den= h->sps.time_scale;
1766             if(h->x264_build < 44U)
1767                 den *= 2;
1768             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
1769                       h->sps.num_units_in_tick, den, 1<<30);
1770         }
1771         s->avctx->pix_fmt = s->avctx->get_format(s->avctx, s->avctx->codec->pix_fmts);
1772         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1773
1774         if (MPV_common_init(s) < 0)
1775             return -1;
1776         s->first_field = 0;
1777         h->prev_interlaced_frame = 1;
1778
1779         init_scan_tables(h);
1780         ff_h264_alloc_tables(h);
1781
1782         for(i = 1; i < s->avctx->thread_count; i++) {
1783             H264Context *c;
1784             c = h->thread_context[i] = av_malloc(sizeof(H264Context));
1785             memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
1786             memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
1787             c->sps = h->sps;
1788             c->pps = h->pps;
1789             init_scan_tables(c);
1790             clone_tables(c, h);
1791         }
1792
1793         for(i = 0; i < s->avctx->thread_count; i++)
1794             if(context_init(h->thread_context[i]) < 0)
1795                 return -1;
1796     }
1797
1798     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1799
1800     h->mb_mbaff = 0;
1801     h->mb_aff_frame = 0;
1802     last_pic_structure = s0->picture_structure;
1803     if(h->sps.frame_mbs_only_flag){
1804         s->picture_structure= PICT_FRAME;
1805     }else{
1806         if(get_bits1(&s->gb)) { //field_pic_flag
1807             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
1808         } else {
1809             s->picture_structure= PICT_FRAME;
1810             h->mb_aff_frame = h->sps.mb_aff;
1811         }
1812     }
1813     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
1814
1815     if(h0->current_slice == 0){
1816         while(h->frame_num !=  h->prev_frame_num &&
1817               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
1818             av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
1819             if (ff_h264_frame_start(h) < 0)
1820                 return -1;
1821             h->prev_frame_num++;
1822             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
1823             s->current_picture_ptr->frame_num= h->prev_frame_num;
1824             ff_h264_execute_ref_pic_marking(h, NULL, 0);
1825         }
1826
1827         /* See if we have a decoded first field looking for a pair... */
1828         if (s0->first_field) {
1829             assert(s0->current_picture_ptr);
1830             assert(s0->current_picture_ptr->data[0]);
1831             assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
1832
1833             /* figure out if we have a complementary field pair */
1834             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
1835                 /*
1836                  * Previous field is unmatched. Don't display it, but let it
1837                  * remain for reference if marked as such.
1838                  */
1839                 s0->current_picture_ptr = NULL;
1840                 s0->first_field = FIELD_PICTURE;
1841
1842             } else {
1843                 if (h->nal_ref_idc &&
1844                         s0->current_picture_ptr->reference &&
1845                         s0->current_picture_ptr->frame_num != h->frame_num) {
1846                     /*
1847                      * This and previous field were reference, but had
1848                      * different frame_nums. Consider this field first in
1849                      * pair. Throw away previous field except for reference
1850                      * purposes.
1851                      */
1852                     s0->first_field = 1;
1853                     s0->current_picture_ptr = NULL;
1854
1855                 } else {
1856                     /* Second field in complementary pair */
1857                     s0->first_field = 0;
1858                 }
1859             }
1860
1861         } else {
1862             /* Frame or first field in a potentially complementary pair */
1863             assert(!s0->current_picture_ptr);
1864             s0->first_field = FIELD_PICTURE;
1865         }
1866
1867         if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
1868             s0->first_field = 0;
1869             return -1;
1870         }
1871     }
1872     if(h != h0)
1873         clone_slice(h, h0);
1874
1875     s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
1876
1877     assert(s->mb_num == s->mb_width * s->mb_height);
1878     if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
1879        first_mb_in_slice                    >= s->mb_num){
1880         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1881         return -1;
1882     }
1883     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
1884     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
1885     if (s->picture_structure == PICT_BOTTOM_FIELD)
1886         s->resync_mb_y = s->mb_y = s->mb_y + 1;
1887     assert(s->mb_y < s->mb_height);
1888
1889     if(s->picture_structure==PICT_FRAME){
1890         h->curr_pic_num=   h->frame_num;
1891         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
1892     }else{
1893         h->curr_pic_num= 2*h->frame_num + 1;
1894         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
1895     }
1896
1897     if(h->nal_unit_type == NAL_IDR_SLICE){
1898         get_ue_golomb(&s->gb); /* idr_pic_id */
1899     }
1900
1901     if(h->sps.poc_type==0){
1902         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
1903
1904         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
1905             h->delta_poc_bottom= get_se_golomb(&s->gb);
1906         }
1907     }
1908
1909     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
1910         h->delta_poc[0]= get_se_golomb(&s->gb);
1911
1912         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
1913             h->delta_poc[1]= get_se_golomb(&s->gb);
1914     }
1915
1916     init_poc(h);
1917
1918     if(h->pps.redundant_pic_cnt_present){
1919         h->redundant_pic_count= get_ue_golomb(&s->gb);
1920     }
1921
1922     //set defaults, might be overridden a few lines later
1923     h->ref_count[0]= h->pps.ref_count[0];
1924     h->ref_count[1]= h->pps.ref_count[1];
1925
1926     if(h->slice_type_nos != FF_I_TYPE){
1927         if(h->slice_type_nos == FF_B_TYPE){
1928             h->direct_spatial_mv_pred= get_bits1(&s->gb);
1929         }
1930         num_ref_idx_active_override_flag= get_bits1(&s->gb);
1931
1932         if(num_ref_idx_active_override_flag){
1933             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
1934             if(h->slice_type_nos==FF_B_TYPE)
1935                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
1936
1937             if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
1938                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
1939                 h->ref_count[0]= h->ref_count[1]= 1;
1940                 return -1;
1941             }
1942         }
1943         if(h->slice_type_nos == FF_B_TYPE)
1944             h->list_count= 2;
1945         else
1946             h->list_count= 1;
1947     }else
1948         h->list_count= 0;
1949
1950     if(!default_ref_list_done){
1951         ff_h264_fill_default_ref_list(h);
1952     }
1953
1954     if(h->slice_type_nos!=FF_I_TYPE && ff_h264_decode_ref_pic_list_reordering(h) < 0)
1955         return -1;
1956
1957     if(h->slice_type_nos!=FF_I_TYPE){
1958         s->last_picture_ptr= &h->ref_list[0][0];
1959         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
1960     }
1961     if(h->slice_type_nos==FF_B_TYPE){
1962         s->next_picture_ptr= &h->ref_list[1][0];
1963         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
1964     }
1965
1966     if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
1967        ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
1968         pred_weight_table(h);
1969     else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE)
1970         implicit_weight_table(h);
1971     else {
1972         h->use_weight = 0;
1973         for (i = 0; i < 2; i++) {
1974             h->luma_weight_flag[i]   = 0;
1975             h->chroma_weight_flag[i] = 0;
1976         }
1977     }
1978
1979     if(h->nal_ref_idc)
1980         ff_h264_decode_ref_pic_marking(h0, &s->gb);
1981
1982     if(FRAME_MBAFF)
1983         ff_h264_fill_mbaff_ref_list(h);
1984
1985     if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
1986         ff_h264_direct_dist_scale_factor(h);
1987     ff_h264_direct_ref_list_init(h);
1988
1989     if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
1990         tmp = get_ue_golomb_31(&s->gb);
1991         if(tmp > 2){
1992             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
1993             return -1;
1994         }
1995         h->cabac_init_idc= tmp;
1996     }
1997
1998     h->last_qscale_diff = 0;
1999     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2000     if(tmp>51){
2001         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2002         return -1;
2003     }
2004     s->qscale= tmp;
2005     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2006     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2007     //FIXME qscale / qp ... stuff
2008     if(h->slice_type == FF_SP_TYPE){
2009         get_bits1(&s->gb); /* sp_for_switch_flag */
2010     }
2011     if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
2012         get_se_golomb(&s->gb); /* slice_qs_delta */
2013     }
2014
2015     h->deblocking_filter = 1;
2016     h->slice_alpha_c0_offset = 52;
2017     h->slice_beta_offset = 52;
2018     if( h->pps.deblocking_filter_parameters_present ) {
2019         tmp= get_ue_golomb_31(&s->gb);
2020         if(tmp > 2){
2021             av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2022             return -1;
2023         }
2024         h->deblocking_filter= tmp;
2025         if(h->deblocking_filter < 2)
2026             h->deblocking_filter^= 1; // 1<->0
2027
2028         if( h->deblocking_filter ) {
2029             h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2030             h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
2031             if(   h->slice_alpha_c0_offset > 104U
2032                || h->slice_beta_offset     > 104U){
2033                 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);
2034                 return -1;
2035             }
2036         }
2037     }
2038
2039     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
2040        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
2041        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
2042        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2043         h->deblocking_filter= 0;
2044
2045     if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2046         if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2047             /* Cheat slightly for speed:
2048                Do not bother to deblock across slices. */
2049             h->deblocking_filter = 2;
2050         } else {
2051             h0->max_contexts = 1;
2052             if(!h0->single_decode_warning) {
2053                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2054                 h0->single_decode_warning = 1;
2055             }
2056             if(h != h0)
2057                 return 1; // deblocking switched inside frame
2058         }
2059     }
2060     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]);
2061
2062 #if 0 //FMO
2063     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2064         slice_group_change_cycle= get_bits(&s->gb, ?);
2065 #endif
2066
2067     h0->last_slice_type = slice_type;
2068     h->slice_num = ++h0->current_slice;
2069     if(h->slice_num >= MAX_SLICES){
2070         av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2071     }
2072
2073     for(j=0; j<2; j++){
2074         int id_list[16];
2075         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2076         for(i=0; i<16; i++){
2077             id_list[i]= 60;
2078             if(h->ref_list[j][i].data[0]){
2079                 int k;
2080                 uint8_t *base= h->ref_list[j][i].base[0];
2081                 for(k=0; k<h->short_ref_count; k++)
2082                     if(h->short_ref[k]->base[0] == base){
2083                         id_list[i]= k;
2084                         break;
2085                     }
2086                 for(k=0; k<h->long_ref_count; k++)
2087                     if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2088                         id_list[i]= h->short_ref_count + k;
2089                         break;
2090                     }
2091             }
2092         }
2093
2094         ref2frm[0]=
2095         ref2frm[1]= -1;
2096         for(i=0; i<16; i++)
2097             ref2frm[i+2]= 4*id_list[i]
2098                           +(h->ref_list[j][i].reference&3);
2099         ref2frm[18+0]=
2100         ref2frm[18+1]= -1;
2101         for(i=16; i<48; i++)
2102             ref2frm[i+4]= 4*id_list[(i-16)>>1]
2103                           +(h->ref_list[j][i].reference&3);
2104     }
2105
2106     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
2107     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2108
2109     s->avctx->refs= h->sps.ref_frame_count;
2110
2111     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2112         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",
2113                h->slice_num,
2114                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2115                first_mb_in_slice,
2116                av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2117                pps_id, h->frame_num,
2118                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2119                h->ref_count[0], h->ref_count[1],
2120                s->qscale,
2121                h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
2122                h->use_weight,
2123                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2124                h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2125                );
2126     }
2127
2128     return 0;
2129 }
2130
2131 int ff_h264_get_slice_type(const H264Context *h)
2132 {
2133     switch (h->slice_type) {
2134     case FF_P_TYPE:  return 0;
2135     case FF_B_TYPE:  return 1;
2136     case FF_I_TYPE:  return 2;
2137     case FF_SP_TYPE: return 3;
2138     case FF_SI_TYPE: return 4;
2139     default:         return -1;
2140     }
2141 }
2142
2143 static void loop_filter(H264Context *h){
2144     MpegEncContext * const s = &h->s;
2145     uint8_t  *dest_y, *dest_cb, *dest_cr;
2146     int linesize, uvlinesize, mb_x, mb_y;
2147     const int end_mb_y= s->mb_y + FRAME_MBAFF;
2148     const int old_slice_type= h->slice_type;
2149
2150     if(h->deblocking_filter) {
2151         for(mb_x= 0; mb_x<s->mb_width; mb_x++){
2152             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
2153                 int mb_xy, mb_type;
2154                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2155                 h->slice_num= h->slice_table[mb_xy];
2156                 mb_type= s->current_picture.mb_type[mb_xy];
2157                 h->list_count= h->list_counts[mb_xy];
2158
2159                 if(FRAME_MBAFF)
2160                     h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2161
2162                 s->mb_x= mb_x;
2163                 s->mb_y= mb_y;
2164                 dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2165                 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2166                 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2167                     //FIXME simplify above
2168
2169                 if (MB_FIELD) {
2170                     linesize   = h->mb_linesize   = s->linesize * 2;
2171                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2172                     if(mb_y&1){ //FIXME move out of this function?
2173                         dest_y -= s->linesize*15;
2174                         dest_cb-= s->uvlinesize*7;
2175                         dest_cr-= s->uvlinesize*7;
2176                     }
2177                 } else {
2178                     linesize   = h->mb_linesize   = s->linesize;
2179                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2180                 }
2181                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2182                 if(fill_filter_caches(h, mb_type))
2183                     continue;
2184                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2185                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2186
2187                 if (FRAME_MBAFF) {
2188                     ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2189                 } else {
2190                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2191                 }
2192             }
2193         }
2194     }
2195     h->slice_type= old_slice_type;
2196     s->mb_x= 0;
2197     s->mb_y= end_mb_y - FRAME_MBAFF;
2198     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2199     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2200 }
2201
2202 static void predict_field_decoding_flag(H264Context *h){
2203     MpegEncContext * const s = &h->s;
2204     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2205     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
2206                 ? s->current_picture.mb_type[mb_xy-1]
2207                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
2208                 ? s->current_picture.mb_type[mb_xy-s->mb_stride]
2209                 : 0;
2210     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2211 }
2212
2213 static int decode_slice(struct AVCodecContext *avctx, void *arg){
2214     H264Context *h = *(void**)arg;
2215     MpegEncContext * const s = &h->s;
2216     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2217
2218     s->mb_skip_run= -1;
2219
2220     h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
2221                     (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
2222
2223     if( h->pps.cabac ) {
2224         /* realign */
2225         align_get_bits( &s->gb );
2226
2227         /* init cabac */
2228         ff_init_cabac_states( &h->cabac);
2229         ff_init_cabac_decoder( &h->cabac,
2230                                s->gb.buffer + get_bits_count(&s->gb)/8,
2231                                (get_bits_left(&s->gb) + 7)/8);
2232
2233         ff_h264_init_cabac_states(h);
2234
2235         for(;;){
2236 //START_TIMER
2237             int ret = ff_h264_decode_mb_cabac(h);
2238             int eos;
2239 //STOP_TIMER("decode_mb_cabac")
2240
2241             if(ret>=0) ff_h264_hl_decode_mb(h);
2242
2243             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
2244                 s->mb_y++;
2245
2246                 ret = ff_h264_decode_mb_cabac(h);
2247
2248                 if(ret>=0) ff_h264_hl_decode_mb(h);
2249                 s->mb_y--;
2250             }
2251             eos = get_cabac_terminate( &h->cabac );
2252
2253             if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
2254                 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);
2255                 return 0;
2256             }
2257             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2258                 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);
2259                 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);
2260                 return -1;
2261             }
2262
2263             if( ++s->mb_x >= s->mb_width ) {
2264                 s->mb_x = 0;
2265                 loop_filter(h);
2266                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2267                 ++s->mb_y;
2268                 if(FIELD_OR_MBAFF_PICTURE) {
2269                     ++s->mb_y;
2270                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2271                         predict_field_decoding_flag(h);
2272                 }
2273             }
2274
2275             if( eos || s->mb_y >= s->mb_height ) {
2276                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2277                 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);
2278                 return 0;
2279             }
2280         }
2281
2282     } else {
2283         for(;;){
2284             int ret = ff_h264_decode_mb_cavlc(h);
2285
2286             if(ret>=0) ff_h264_hl_decode_mb(h);
2287
2288             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
2289                 s->mb_y++;
2290                 ret = ff_h264_decode_mb_cavlc(h);
2291
2292                 if(ret>=0) ff_h264_hl_decode_mb(h);
2293                 s->mb_y--;
2294             }
2295
2296             if(ret<0){
2297                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2298                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
2299
2300                 return -1;
2301             }
2302
2303             if(++s->mb_x >= s->mb_width){
2304                 s->mb_x=0;
2305                 loop_filter(h);
2306                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2307                 ++s->mb_y;
2308                 if(FIELD_OR_MBAFF_PICTURE) {
2309                     ++s->mb_y;
2310                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2311                         predict_field_decoding_flag(h);
2312                 }
2313                 if(s->mb_y >= s->mb_height){
2314                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2315
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_END|DC_END|MV_END)&part_mask);
2322
2323                         return -1;
2324                     }
2325                 }
2326             }
2327
2328             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
2329                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2330                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2331                     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);
2332
2333                     return 0;
2334                 }else{
2335                     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);
2336
2337                     return -1;
2338                 }
2339             }
2340         }
2341     }
2342
2343 #if 0
2344     for(;s->mb_y < s->mb_height; s->mb_y++){
2345         for(;s->mb_x < s->mb_width; s->mb_x++){
2346             int ret= decode_mb(h);
2347
2348             ff_h264_hl_decode_mb(h);
2349
2350             if(ret<0){
2351                 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2352                 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);
2353
2354                 return -1;
2355             }
2356
2357             if(++s->mb_x >= s->mb_width){
2358                 s->mb_x=0;
2359                 if(++s->mb_y >= s->mb_height){
2360                     if(get_bits_count(s->gb) == s->gb.size_in_bits){
2361                         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);
2362
2363                         return 0;
2364                     }else{
2365                         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);
2366
2367                         return -1;
2368                     }
2369                 }
2370             }
2371
2372             if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2373                 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2374                     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);
2375
2376                     return 0;
2377                 }else{
2378                     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);
2379
2380                     return -1;
2381                 }
2382             }
2383         }
2384         s->mb_x=0;
2385         ff_draw_horiz_band(s, 16*s->mb_y, 16);
2386     }
2387 #endif
2388     return -1; //not reached
2389 }
2390
2391 /**
2392  * Call decode_slice() for each context.
2393  *
2394  * @param h h264 master context
2395  * @param context_count number of contexts to execute
2396  */
2397 static void execute_decode_slices(H264Context *h, int context_count){
2398     MpegEncContext * const s = &h->s;
2399     AVCodecContext * const avctx= s->avctx;
2400     H264Context *hx;
2401     int i;
2402
2403     if (s->avctx->hwaccel)
2404         return;
2405     if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2406         return;
2407     if(context_count == 1) {
2408         decode_slice(avctx, &h);
2409     } else {
2410         for(i = 1; i < context_count; i++) {
2411             hx = h->thread_context[i];
2412             hx->s.error_recognition = avctx->error_recognition;
2413             hx->s.error_count = 0;
2414         }
2415
2416         avctx->execute(avctx, (void *)decode_slice,
2417                        h->thread_context, NULL, context_count, sizeof(void*));
2418
2419         /* pull back stuff from slices to master context */
2420         hx = h->thread_context[context_count - 1];
2421         s->mb_x = hx->s.mb_x;
2422         s->mb_y = hx->s.mb_y;
2423         s->dropable = hx->s.dropable;
2424         s->picture_structure = hx->s.picture_structure;
2425         for(i = 1; i < context_count; i++)
2426             h->s.error_count += h->thread_context[i]->s.error_count;
2427     }
2428 }
2429
2430
2431 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
2432     MpegEncContext * const s = &h->s;
2433     AVCodecContext * const avctx= s->avctx;
2434     int buf_index=0;
2435     H264Context *hx; ///< thread context
2436     int context_count = 0;
2437     int next_avc= h->is_avc ? 0 : buf_size;
2438
2439     h->max_contexts = avctx->thread_count;
2440 #if 0
2441     int i;
2442     for(i=0; i<50; i++){
2443         av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
2444     }
2445 #endif
2446     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
2447         h->current_slice = 0;
2448         if (!s->first_field)
2449             s->current_picture_ptr= NULL;
2450         ff_h264_reset_sei(h);
2451     }
2452
2453     for(;;){
2454         int consumed;
2455         int dst_length;
2456         int bit_length;
2457         const uint8_t *ptr;
2458         int i, nalsize = 0;
2459         int err;
2460
2461         if(buf_index >= next_avc) {
2462             if(buf_index >= buf_size) break;
2463             nalsize = 0;
2464             for(i = 0; i < h->nal_length_size; i++)
2465                 nalsize = (nalsize << 8) | buf[buf_index++];
2466             if(nalsize <= 1 || nalsize > buf_size - buf_index){
2467                 if(nalsize == 1){
2468                     buf_index++;
2469                     continue;
2470                 }else{
2471                     av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2472                     break;
2473                 }
2474             }
2475             next_avc= buf_index + nalsize;
2476         } else {
2477             // start code prefix search
2478             for(; buf_index + 3 < next_avc; buf_index++){
2479                 // This should always succeed in the first iteration.
2480                 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2481                     break;
2482             }
2483
2484             if(buf_index+3 >= buf_size) break;
2485
2486             buf_index+=3;
2487             if(buf_index >= next_avc) continue;
2488         }
2489
2490         hx = h->thread_context[context_count];
2491
2492         ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
2493         if (ptr==NULL || dst_length < 0){
2494             return -1;
2495         }
2496         i= buf_index + consumed;
2497         if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
2498            buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
2499             s->workaround_bugs |= FF_BUG_TRUNCATED;
2500
2501         if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
2502         while(ptr[dst_length - 1] == 0 && dst_length > 0)
2503             dst_length--;
2504         }
2505         bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
2506
2507         if(s->avctx->debug&FF_DEBUG_STARTCODE){
2508             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);
2509         }
2510
2511         if (h->is_avc && (nalsize != consumed) && nalsize){
2512             av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
2513         }
2514
2515         buf_index += consumed;
2516
2517         if(  (s->hurry_up == 1 && h->nal_ref_idc  == 0) //FIXME do not discard SEI id
2518            ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0))
2519             continue;
2520
2521       again:
2522         err = 0;
2523         switch(hx->nal_unit_type){
2524         case NAL_IDR_SLICE:
2525             if (h->nal_unit_type != NAL_IDR_SLICE) {
2526                 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2527                 return -1;
2528             }
2529             idr(h); //FIXME ensure we don't loose some frames if there is reordering
2530         case NAL_SLICE:
2531             init_get_bits(&hx->s.gb, ptr, bit_length);
2532             hx->intra_gb_ptr=
2533             hx->inter_gb_ptr= &hx->s.gb;
2534             hx->s.data_partitioning = 0;
2535
2536             if((err = decode_slice_header(hx, h)))
2537                break;
2538
2539             avctx->profile = hx->sps.profile_idc;
2540             avctx->level   = hx->sps.level_idc;
2541
2542             if (s->avctx->hwaccel && h->current_slice == 1) {
2543                 if (s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
2544                     return -1;
2545             }
2546
2547             s->current_picture_ptr->key_frame |=
2548                     (hx->nal_unit_type == NAL_IDR_SLICE) ||
2549                     (h->sei_recovery_frame_cnt >= 0);
2550             if(hx->redundant_pic_count==0 && hx->s.hurry_up < 5
2551                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2552                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2553                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2554                && avctx->skip_frame < AVDISCARD_ALL){
2555                 if(avctx->hwaccel) {
2556                     if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
2557                         return -1;
2558                 }else
2559                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
2560                     static const uint8_t start_code[] = {0x00, 0x00, 0x01};
2561                     ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
2562                     ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
2563                 }else
2564                     context_count++;
2565             }
2566             break;
2567         case NAL_DPA:
2568             init_get_bits(&hx->s.gb, ptr, bit_length);
2569             hx->intra_gb_ptr=
2570             hx->inter_gb_ptr= NULL;
2571
2572             if ((err = decode_slice_header(hx, h)) < 0)
2573                 break;
2574
2575             avctx->profile = hx->sps.profile_idc;
2576             avctx->level   = hx->sps.level_idc;
2577
2578             hx->s.data_partitioning = 1;
2579
2580             break;
2581         case NAL_DPB:
2582             init_get_bits(&hx->intra_gb, ptr, bit_length);
2583             hx->intra_gb_ptr= &hx->intra_gb;
2584             break;
2585         case NAL_DPC:
2586             init_get_bits(&hx->inter_gb, ptr, bit_length);
2587             hx->inter_gb_ptr= &hx->inter_gb;
2588
2589             if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
2590                && s->context_initialized
2591                && s->hurry_up < 5
2592                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2593                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2594                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2595                && avctx->skip_frame < AVDISCARD_ALL)
2596                 context_count++;
2597             break;
2598         case NAL_SEI:
2599             init_get_bits(&s->gb, ptr, bit_length);
2600             ff_h264_decode_sei(h);
2601             break;
2602         case NAL_SPS:
2603             init_get_bits(&s->gb, ptr, bit_length);
2604             ff_h264_decode_seq_parameter_set(h);
2605
2606             if(s->flags& CODEC_FLAG_LOW_DELAY)
2607                 s->low_delay=1;
2608
2609             if(avctx->has_b_frames < 2)
2610                 avctx->has_b_frames= !s->low_delay;
2611             break;
2612         case NAL_PPS:
2613             init_get_bits(&s->gb, ptr, bit_length);
2614
2615             ff_h264_decode_picture_parameter_set(h, bit_length);
2616
2617             break;
2618         case NAL_AUD:
2619         case NAL_END_SEQUENCE:
2620         case NAL_END_STREAM:
2621         case NAL_FILLER_DATA:
2622         case NAL_SPS_EXT:
2623         case NAL_AUXILIARY_SLICE:
2624             break;
2625         default:
2626             av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
2627         }
2628
2629         if(context_count == h->max_contexts) {
2630             execute_decode_slices(h, context_count);
2631             context_count = 0;
2632         }
2633
2634         if (err < 0)
2635             av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
2636         else if(err == 1) {
2637             /* Slice could not be decoded in parallel mode, copy down
2638              * NAL unit stuff to context 0 and restart. Note that
2639              * rbsp_buffer is not transferred, but since we no longer
2640              * run in parallel mode this should not be an issue. */
2641             h->nal_unit_type = hx->nal_unit_type;
2642             h->nal_ref_idc   = hx->nal_ref_idc;
2643             hx = h;
2644             goto again;
2645         }
2646     }
2647     if(context_count)
2648         execute_decode_slices(h, context_count);
2649     return buf_index;
2650 }
2651
2652 /**
2653  * returns the number of bytes consumed for building the current frame
2654  */
2655 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
2656         if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
2657         if(pos+10>buf_size) pos=buf_size; // oops ;)
2658
2659         return pos;
2660 }
2661
2662 static int decode_frame(AVCodecContext *avctx,
2663                              void *data, int *data_size,
2664                              AVPacket *avpkt)
2665 {
2666     const uint8_t *buf = avpkt->data;
2667     int buf_size = avpkt->size;
2668     H264Context *h = avctx->priv_data;
2669     MpegEncContext *s = &h->s;
2670     AVFrame *pict = data;
2671     int buf_index;
2672
2673     s->flags= avctx->flags;
2674     s->flags2= avctx->flags2;
2675
2676    /* end of stream, output what is still in the buffers */
2677     if (buf_size == 0) {
2678         Picture *out;
2679         int i, out_idx;
2680
2681 //FIXME factorize this with the output code below
2682         out = h->delayed_pic[0];
2683         out_idx = 0;
2684         for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2685             if(h->delayed_pic[i]->poc < out->poc){
2686                 out = h->delayed_pic[i];
2687                 out_idx = i;
2688             }
2689
2690         for(i=out_idx; h->delayed_pic[i]; i++)
2691             h->delayed_pic[i] = h->delayed_pic[i+1];
2692
2693         if(out){
2694             *data_size = sizeof(AVFrame);
2695             *pict= *(AVFrame*)out;
2696         }
2697
2698         return 0;
2699     }
2700
2701     if(h->is_avc && !h->got_avcC) {
2702         int i, cnt, nalsize;
2703         unsigned char *p = avctx->extradata;
2704         if(avctx->extradata_size < 7) {
2705             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
2706             return -1;
2707         }
2708         if(*p != 1) {
2709             av_log(avctx, AV_LOG_ERROR, "Unknown avcC version %d\n", *p);
2710             return -1;
2711         }
2712         /* sps and pps in the avcC always have length coded with 2 bytes,
2713            so put a fake nal_length_size = 2 while parsing them */
2714         h->nal_length_size = 2;
2715         // Decode sps from avcC
2716         cnt = *(p+5) & 0x1f; // Number of sps
2717         p += 6;
2718         for (i = 0; i < cnt; i++) {
2719             nalsize = AV_RB16(p) + 2;
2720             if(decode_nal_units(h, p, nalsize) < 0) {
2721                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
2722                 return -1;
2723             }
2724             p += nalsize;
2725         }
2726         // Decode pps from avcC
2727         cnt = *(p++); // Number of pps
2728         for (i = 0; i < cnt; i++) {
2729             nalsize = AV_RB16(p) + 2;
2730             if(decode_nal_units(h, p, nalsize)  != nalsize) {
2731                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
2732                 return -1;
2733             }
2734             p += nalsize;
2735         }
2736         // Now store right nal length size, that will be use to parse all other nals
2737         h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
2738         // Do not reparse avcC
2739         h->got_avcC = 1;
2740     }
2741
2742     if(!h->got_avcC && !h->is_avc && s->avctx->extradata_size){
2743         if(decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) < 0)
2744             return -1;
2745         h->got_avcC = 1;
2746     }
2747
2748     buf_index=decode_nal_units(h, buf, buf_size);
2749     if(buf_index < 0)
2750         return -1;
2751
2752     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
2753         if (avctx->skip_frame >= AVDISCARD_NONREF || s->hurry_up) return 0;
2754         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
2755         return -1;
2756     }
2757
2758     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
2759         Picture *out = s->current_picture_ptr;
2760         Picture *cur = s->current_picture_ptr;
2761         int i, pics, out_of_order, out_idx;
2762
2763         field_end(h);
2764
2765         if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
2766             /* Wait for second field. */
2767             *data_size = 0;
2768
2769         } else {
2770             cur->interlaced_frame = 0;
2771             cur->repeat_pict = 0;
2772
2773             /* Signal interlacing information externally. */
2774             /* Prioritize picture timing SEI information over used decoding process if it exists. */
2775
2776             if(h->sps.pic_struct_present_flag){
2777                 switch (h->sei_pic_struct)
2778                 {
2779                 case SEI_PIC_STRUCT_FRAME:
2780                     break;
2781                 case SEI_PIC_STRUCT_TOP_FIELD:
2782                 case SEI_PIC_STRUCT_BOTTOM_FIELD:
2783                     cur->interlaced_frame = 1;
2784                     break;
2785                 case SEI_PIC_STRUCT_TOP_BOTTOM:
2786                 case SEI_PIC_STRUCT_BOTTOM_TOP:
2787                     if (FIELD_OR_MBAFF_PICTURE)
2788                         cur->interlaced_frame = 1;
2789                     else
2790                         // try to flag soft telecine progressive
2791                         cur->interlaced_frame = h->prev_interlaced_frame;
2792                     break;
2793                 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
2794                 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
2795                     // Signal the possibility of telecined film externally (pic_struct 5,6)
2796                     // From these hints, let the applications decide if they apply deinterlacing.
2797                     cur->repeat_pict = 1;
2798                     break;
2799                 case SEI_PIC_STRUCT_FRAME_DOUBLING:
2800                     // Force progressive here, as doubling interlaced frame is a bad idea.
2801                     cur->repeat_pict = 2;
2802                     break;
2803                 case SEI_PIC_STRUCT_FRAME_TRIPLING:
2804                     cur->repeat_pict = 4;
2805                     break;
2806                 }
2807
2808                 if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
2809                     cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
2810             }else{
2811                 /* Derive interlacing flag from used decoding process. */
2812                 cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
2813             }
2814             h->prev_interlaced_frame = cur->interlaced_frame;
2815
2816             if (cur->field_poc[0] != cur->field_poc[1]){
2817                 /* Derive top_field_first from field pocs. */
2818                 cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
2819             }else{
2820                 if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
2821                     /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
2822                     if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
2823                       || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
2824                         cur->top_field_first = 1;
2825                     else
2826                         cur->top_field_first = 0;
2827                 }else{
2828                     /* Most likely progressive */
2829                     cur->top_field_first = 0;
2830                 }
2831             }
2832
2833         //FIXME do something with unavailable reference frames
2834
2835             /* Sort B-frames into display order */
2836
2837             if(h->sps.bitstream_restriction_flag
2838                && s->avctx->has_b_frames < h->sps.num_reorder_frames){
2839                 s->avctx->has_b_frames = h->sps.num_reorder_frames;
2840                 s->low_delay = 0;
2841             }
2842
2843             if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
2844                && !h->sps.bitstream_restriction_flag){
2845                 s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
2846                 s->low_delay= 0;
2847             }
2848
2849             pics = 0;
2850             while(h->delayed_pic[pics]) pics++;
2851
2852             assert(pics <= MAX_DELAYED_PIC_COUNT);
2853
2854             h->delayed_pic[pics++] = cur;
2855             if(cur->reference == 0)
2856                 cur->reference = DELAYED_PIC_REF;
2857
2858             out = h->delayed_pic[0];
2859             out_idx = 0;
2860             for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2861                 if(h->delayed_pic[i]->poc < out->poc){
2862                     out = h->delayed_pic[i];
2863                     out_idx = i;
2864                 }
2865             if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
2866                 h->outputed_poc= INT_MIN;
2867             out_of_order = out->poc < h->outputed_poc;
2868
2869             if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
2870                 { }
2871             else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
2872                || (s->low_delay &&
2873                 ((h->outputed_poc != INT_MIN && out->poc > h->outputed_poc + 2)
2874                  || cur->pict_type == FF_B_TYPE)))
2875             {
2876                 s->low_delay = 0;
2877                 s->avctx->has_b_frames++;
2878             }
2879
2880             if(out_of_order || pics > s->avctx->has_b_frames){
2881                 out->reference &= ~DELAYED_PIC_REF;
2882                 for(i=out_idx; h->delayed_pic[i]; i++)
2883                     h->delayed_pic[i] = h->delayed_pic[i+1];
2884             }
2885             if(!out_of_order && pics > s->avctx->has_b_frames){
2886                 *data_size = sizeof(AVFrame);
2887
2888                 if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
2889                     h->outputed_poc = INT_MIN;
2890                 } else
2891                     h->outputed_poc = out->poc;
2892                 *pict= *(AVFrame*)out;
2893             }else{
2894                 av_log(avctx, AV_LOG_DEBUG, "no picture\n");
2895             }
2896         }
2897     }
2898
2899     assert(pict->data[0] || !*data_size);
2900     ff_print_debug_info(s, pict);
2901 //printf("out %d\n", (int)pict->data[0]);
2902
2903     return get_consumed_bytes(s, buf_index, buf_size);
2904 }
2905 #if 0
2906 static inline void fill_mb_avail(H264Context *h){
2907     MpegEncContext * const s = &h->s;
2908     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2909
2910     if(s->mb_y){
2911         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
2912         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
2913         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
2914     }else{
2915         h->mb_avail[0]=
2916         h->mb_avail[1]=
2917         h->mb_avail[2]= 0;
2918     }
2919     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
2920     h->mb_avail[4]= 1; //FIXME move out
2921     h->mb_avail[5]= 0; //FIXME move out
2922 }
2923 #endif
2924
2925 #ifdef TEST
2926 #undef printf
2927 #undef random
2928 #define COUNT 8000
2929 #define SIZE (COUNT*40)
2930 int main(void){
2931     int i;
2932     uint8_t temp[SIZE];
2933     PutBitContext pb;
2934     GetBitContext gb;
2935 //    int int_temp[10000];
2936     DSPContext dsp;
2937     AVCodecContext avctx;
2938
2939     dsputil_init(&dsp, &avctx);
2940
2941     init_put_bits(&pb, temp, SIZE);
2942     printf("testing unsigned exp golomb\n");
2943     for(i=0; i<COUNT; i++){
2944         START_TIMER
2945         set_ue_golomb(&pb, i);
2946         STOP_TIMER("set_ue_golomb");
2947     }
2948     flush_put_bits(&pb);
2949
2950     init_get_bits(&gb, temp, 8*SIZE);
2951     for(i=0; i<COUNT; i++){
2952         int j, s;
2953
2954         s= show_bits(&gb, 24);
2955
2956         START_TIMER
2957         j= get_ue_golomb(&gb);
2958         if(j != i){
2959             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
2960 //            return -1;
2961         }
2962         STOP_TIMER("get_ue_golomb");
2963     }
2964
2965
2966     init_put_bits(&pb, temp, SIZE);
2967     printf("testing signed exp golomb\n");
2968     for(i=0; i<COUNT; i++){
2969         START_TIMER
2970         set_se_golomb(&pb, i - COUNT/2);
2971         STOP_TIMER("set_se_golomb");
2972     }
2973     flush_put_bits(&pb);
2974
2975     init_get_bits(&gb, temp, 8*SIZE);
2976     for(i=0; i<COUNT; i++){
2977         int j, s;
2978
2979         s= show_bits(&gb, 24);
2980
2981         START_TIMER
2982         j= get_se_golomb(&gb);
2983         if(j != i - COUNT/2){
2984             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
2985 //            return -1;
2986         }
2987         STOP_TIMER("get_se_golomb");
2988     }
2989
2990 #if 0
2991     printf("testing 4x4 (I)DCT\n");
2992
2993     DCTELEM block[16];
2994     uint8_t src[16], ref[16];
2995     uint64_t error= 0, max_error=0;
2996
2997     for(i=0; i<COUNT; i++){
2998         int j;
2999 //        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
3000         for(j=0; j<16; j++){
3001             ref[j]= random()%255;
3002             src[j]= random()%255;
3003         }
3004
3005         h264_diff_dct_c(block, src, ref, 4);
3006
3007         //normalize
3008         for(j=0; j<16; j++){
3009 //            printf("%d ", block[j]);
3010             block[j]= block[j]*4;
3011             if(j&1) block[j]= (block[j]*4 + 2)/5;
3012             if(j&4) block[j]= (block[j]*4 + 2)/5;
3013         }
3014 //        printf("\n");
3015
3016         s->dsp.h264_idct_add(ref, block, 4);
3017 /*        for(j=0; j<16; j++){
3018             printf("%d ", ref[j]);
3019         }
3020         printf("\n");*/
3021
3022         for(j=0; j<16; j++){
3023             int diff= FFABS(src[j] - ref[j]);
3024
3025             error+= diff*diff;
3026             max_error= FFMAX(max_error, diff);
3027         }
3028     }
3029     printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
3030     printf("testing quantizer\n");
3031     for(qp=0; qp<52; qp++){
3032         for(i=0; i<16; i++)
3033             src1_block[i]= src2_block[i]= random()%255;
3034
3035     }
3036     printf("Testing NAL layer\n");
3037
3038     uint8_t bitstream[COUNT];
3039     uint8_t nal[COUNT*2];
3040     H264Context h;
3041     memset(&h, 0, sizeof(H264Context));
3042
3043     for(i=0; i<COUNT; i++){
3044         int zeros= i;
3045         int nal_length;
3046         int consumed;
3047         int out_length;
3048         uint8_t *out;
3049         int j;
3050
3051         for(j=0; j<COUNT; j++){
3052             bitstream[j]= (random() % 255) + 1;
3053         }
3054
3055         for(j=0; j<zeros; j++){
3056             int pos= random() % COUNT;
3057             while(bitstream[pos] == 0){
3058                 pos++;
3059                 pos %= COUNT;
3060             }
3061             bitstream[pos]=0;
3062         }
3063
3064         START_TIMER
3065
3066         nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
3067         if(nal_length<0){
3068             printf("encoding failed\n");
3069             return -1;
3070         }
3071
3072         out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
3073
3074         STOP_TIMER("NAL")
3075
3076         if(out_length != COUNT){
3077             printf("incorrect length %d %d\n", out_length, COUNT);
3078             return -1;
3079         }
3080
3081         if(consumed != nal_length){
3082             printf("incorrect consumed length %d %d\n", nal_length, consumed);
3083             return -1;
3084         }
3085
3086         if(memcmp(bitstream, out, COUNT)){
3087             printf("mismatch\n");
3088             return -1;
3089         }
3090     }
3091 #endif
3092
3093     printf("Testing RBSP\n");
3094
3095
3096     return 0;
3097 }
3098 #endif /* TEST */
3099
3100
3101 av_cold void ff_h264_free_context(H264Context *h)
3102 {
3103     int i;
3104
3105     free_tables(h); //FIXME cleanup init stuff perhaps
3106
3107     for(i = 0; i < MAX_SPS_COUNT; i++)
3108         av_freep(h->sps_buffers + i);
3109
3110     for(i = 0; i < MAX_PPS_COUNT; i++)
3111         av_freep(h->pps_buffers + i);
3112 }
3113
3114 av_cold int ff_h264_decode_end(AVCodecContext *avctx)
3115 {
3116     H264Context *h = avctx->priv_data;
3117     MpegEncContext *s = &h->s;
3118
3119     ff_h264_free_context(h);
3120
3121     MPV_common_end(s);
3122
3123 //    memset(h, 0, sizeof(H264Context));
3124
3125     return 0;
3126 }
3127
3128
3129 AVCodec h264_decoder = {
3130     "h264",
3131     CODEC_TYPE_VIDEO,
3132     CODEC_ID_H264,
3133     sizeof(H264Context),
3134     ff_h264_decode_init,
3135     NULL,
3136     ff_h264_decode_end,
3137     decode_frame,
3138     /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3139     .flush= flush_dpb,
3140     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
3141     .pix_fmts= ff_hwaccel_pixfmt_list_420,
3142 };
3143
3144 #if CONFIG_H264_VDPAU_DECODER
3145 AVCodec h264_vdpau_decoder = {
3146     "h264_vdpau",
3147     CODEC_TYPE_VIDEO,
3148     CODEC_ID_H264,
3149     sizeof(H264Context),
3150     ff_h264_decode_init,
3151     NULL,
3152     ff_h264_decode_end,
3153     decode_frame,
3154     CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3155     .flush= flush_dpb,
3156     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
3157     .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
3158 };
3159 #endif