cb9bd68fea0f3b202236e3ebc4acf74eba9eea64
[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 #if ARCH_X86
43 #include "x86/h264_i386.h"
44 #endif
45
46 //#undef NDEBUG
47 #include <assert.h>
48
49 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
50 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
51
52 static const uint8_t rem6[52]={
53 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,
54 };
55
56 static const uint8_t div6[52]={
57 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,
58 };
59
60 void ff_h264_write_back_intra_pred_mode(H264Context *h){
61     const int mb_xy= h->mb_xy;
62
63     h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
64     h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
65     h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
66     h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
67     h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
68     h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
69     h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
70 }
71
72 /**
73  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
74  */
75 int ff_h264_check_intra4x4_pred_mode(H264Context *h){
76     MpegEncContext * const s = &h->s;
77     static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
78     static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
79     int i;
80
81     if(!(h->top_samples_available&0x8000)){
82         for(i=0; i<4; i++){
83             int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
84             if(status<0){
85                 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
86                 return -1;
87             } else if(status){
88                 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
89             }
90         }
91     }
92
93     if((h->left_samples_available&0x8888)!=0x8888){
94         static const int mask[4]={0x8000,0x2000,0x80,0x20};
95         for(i=0; i<4; i++){
96             if(!(h->left_samples_available&mask[i])){
97                 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
98                 if(status<0){
99                     av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
100                     return -1;
101                 } else if(status){
102                     h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
103                 }
104             }
105         }
106     }
107
108     return 0;
109 } //FIXME cleanup like ff_h264_check_intra_pred_mode
110
111 /**
112  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
113  */
114 int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
115     MpegEncContext * const s = &h->s;
116     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
117     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
118
119     if(mode > 6U) {
120         av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
121         return -1;
122     }
123
124     if(!(h->top_samples_available&0x8000)){
125         mode= top[ mode ];
126         if(mode<0){
127             av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
128             return -1;
129         }
130     }
131
132     if((h->left_samples_available&0x8080) != 0x8080){
133         mode= left[ mode ];
134         if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
135             mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
136         }
137         if(mode<0){
138             av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
139             return -1;
140         }
141     }
142
143     return mode;
144 }
145
146 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
147     int i, si, di;
148     uint8_t *dst;
149     int bufidx;
150
151 //    src[0]&0x80;                //forbidden bit
152     h->nal_ref_idc= src[0]>>5;
153     h->nal_unit_type= src[0]&0x1F;
154
155     src++; length--;
156 #if 0
157     for(i=0; i<length; i++)
158         printf("%2X ", src[i]);
159 #endif
160
161 #if HAVE_FAST_UNALIGNED
162 # if HAVE_FAST_64BIT
163 #   define RS 7
164     for(i=0; i+1<length; i+=9){
165         if(!((~*(const uint64_t*)(src+i) & (*(const uint64_t*)(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
166 # else
167 #   define RS 3
168     for(i=0; i+1<length; i+=5){
169         if(!((~*(const uint32_t*)(src+i) & (*(const uint32_t*)(src+i) - 0x01000101U)) & 0x80008080U))
170 # endif
171             continue;
172         if(i>0 && !src[i]) i--;
173         while(src[i]) i++;
174 #else
175 #   define RS 0
176     for(i=0; i+1<length; i+=2){
177         if(src[i]) continue;
178         if(i>0 && src[i-1]==0) i--;
179 #endif
180         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
181             if(src[i+2]!=3){
182                 /* startcode, so we must be past the end */
183                 length=i;
184             }
185             break;
186         }
187         i-= RS;
188     }
189
190     if(i>=length-1){ //no escaped 0
191         *dst_length= length;
192         *consumed= length+1; //+1 for the header
193         return src;
194     }
195
196     bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
197     av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
198     dst= h->rbsp_buffer[bufidx];
199
200     if (dst == NULL){
201         return NULL;
202     }
203
204 //printf("decoding esc\n");
205     memcpy(dst, src, i);
206     si=di=i;
207     while(si+2<length){
208         //remove escapes (very rare 1:2^22)
209         if(src[si+2]>3){
210             dst[di++]= src[si++];
211             dst[di++]= src[si++];
212         }else if(src[si]==0 && src[si+1]==0){
213             if(src[si+2]==3){ //escape
214                 dst[di++]= 0;
215                 dst[di++]= 0;
216                 si+=3;
217                 continue;
218             }else //next start code
219                 goto nsc;
220         }
221
222         dst[di++]= src[si++];
223     }
224     while(si<length)
225         dst[di++]= src[si++];
226 nsc:
227
228     memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
229
230     *dst_length= di;
231     *consumed= si + 1;//+1 for the header
232 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
233     return dst;
234 }
235
236 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
237     int v= *src;
238     int r;
239
240     tprintf(h->s.avctx, "rbsp trailing %X\n", v);
241
242     for(r=1; r<9; r++){
243         if(v&1) return r;
244         v>>=1;
245     }
246     return 0;
247 }
248
249 /**
250  * IDCT transforms the 16 dc values and dequantizes them.
251  * @param qp quantization parameter
252  */
253 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
254 #define stride 16
255     int i;
256     int temp[16]; //FIXME check if this is a good idea
257     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
258     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
259
260 //memset(block, 64, 2*256);
261 //return;
262     for(i=0; i<4; i++){
263         const int offset= y_offset[i];
264         const int z0= block[offset+stride*0] + block[offset+stride*4];
265         const int z1= block[offset+stride*0] - block[offset+stride*4];
266         const int z2= block[offset+stride*1] - block[offset+stride*5];
267         const int z3= block[offset+stride*1] + block[offset+stride*5];
268
269         temp[4*i+0]= z0+z3;
270         temp[4*i+1]= z1+z2;
271         temp[4*i+2]= z1-z2;
272         temp[4*i+3]= z0-z3;
273     }
274
275     for(i=0; i<4; i++){
276         const int offset= x_offset[i];
277         const int z0= temp[4*0+i] + temp[4*2+i];
278         const int z1= temp[4*0+i] - temp[4*2+i];
279         const int z2= temp[4*1+i] - temp[4*3+i];
280         const int z3= temp[4*1+i] + temp[4*3+i];
281
282         block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
283         block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
284         block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
285         block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
286     }
287 }
288
289 #if 0
290 /**
291  * DCT transforms the 16 dc values.
292  * @param qp quantization parameter ??? FIXME
293  */
294 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
295 //    const int qmul= dequant_coeff[qp][0];
296     int i;
297     int temp[16]; //FIXME check if this is a good idea
298     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
299     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
300
301     for(i=0; i<4; i++){
302         const int offset= y_offset[i];
303         const int z0= block[offset+stride*0] + block[offset+stride*4];
304         const int z1= block[offset+stride*0] - block[offset+stride*4];
305         const int z2= block[offset+stride*1] - block[offset+stride*5];
306         const int z3= block[offset+stride*1] + block[offset+stride*5];
307
308         temp[4*i+0]= z0+z3;
309         temp[4*i+1]= z1+z2;
310         temp[4*i+2]= z1-z2;
311         temp[4*i+3]= z0-z3;
312     }
313
314     for(i=0; i<4; i++){
315         const int offset= x_offset[i];
316         const int z0= temp[4*0+i] + temp[4*2+i];
317         const int z1= temp[4*0+i] - temp[4*2+i];
318         const int z2= temp[4*1+i] - temp[4*3+i];
319         const int z3= temp[4*1+i] + temp[4*3+i];
320
321         block[stride*0 +offset]= (z0 + z3)>>1;
322         block[stride*2 +offset]= (z1 + z2)>>1;
323         block[stride*8 +offset]= (z1 - z2)>>1;
324         block[stride*10+offset]= (z0 - z3)>>1;
325     }
326 }
327 #endif
328
329 #undef xStride
330 #undef stride
331
332 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
333     const int stride= 16*2;
334     const int xStride= 16;
335     int a,b,c,d,e;
336
337     a= block[stride*0 + xStride*0];
338     b= block[stride*0 + xStride*1];
339     c= block[stride*1 + xStride*0];
340     d= block[stride*1 + xStride*1];
341
342     e= a-b;
343     a= a+b;
344     b= c-d;
345     c= c+d;
346
347     block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
348     block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
349     block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
350     block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
351 }
352
353 #if 0
354 static void chroma_dc_dct_c(DCTELEM *block){
355     const int stride= 16*2;
356     const int xStride= 16;
357     int a,b,c,d,e;
358
359     a= block[stride*0 + xStride*0];
360     b= block[stride*0 + xStride*1];
361     c= block[stride*1 + xStride*0];
362     d= block[stride*1 + xStride*1];
363
364     e= a-b;
365     a= a+b;
366     b= c-d;
367     c= c+d;
368
369     block[stride*0 + xStride*0]= (a+c);
370     block[stride*0 + xStride*1]= (e+b);
371     block[stride*1 + xStride*0]= (a-c);
372     block[stride*1 + xStride*1]= (e-b);
373 }
374 #endif
375
376 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
377                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
378                            int src_x_offset, int src_y_offset,
379                            qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
380     MpegEncContext * const s = &h->s;
381     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
382     int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
383     const int luma_xy= (mx&3) + ((my&3)<<2);
384     uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
385     uint8_t * src_cb, * src_cr;
386     int extra_width= h->emu_edge_width;
387     int extra_height= h->emu_edge_height;
388     int emu=0;
389     const int full_mx= mx>>2;
390     const int full_my= my>>2;
391     const int pic_width  = 16*s->mb_width;
392     const int pic_height = 16*s->mb_height >> MB_FIELD;
393
394     if(mx&7) extra_width -= 3;
395     if(my&7) extra_height -= 3;
396
397     if(   full_mx < 0-extra_width
398        || full_my < 0-extra_height
399        || full_mx + 16/*FIXME*/ > pic_width + extra_width
400        || full_my + 16/*FIXME*/ > pic_height + extra_height){
401         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);
402             src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
403         emu=1;
404     }
405
406     qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
407     if(!square){
408         qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
409     }
410
411     if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
412
413     if(MB_FIELD){
414         // chroma offset when predicting from a field of opposite parity
415         my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
416         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
417     }
418     src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
419     src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
420
421     if(emu){
422         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);
423             src_cb= s->edge_emu_buffer;
424     }
425     chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
426
427     if(emu){
428         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);
429             src_cr= s->edge_emu_buffer;
430     }
431     chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
432 }
433
434 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
435                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
436                            int x_offset, int y_offset,
437                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
438                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
439                            int list0, int list1){
440     MpegEncContext * const s = &h->s;
441     qpel_mc_func *qpix_op=  qpix_put;
442     h264_chroma_mc_func chroma_op= chroma_put;
443
444     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
445     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
446     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
447     x_offset += 8*s->mb_x;
448     y_offset += 8*(s->mb_y >> MB_FIELD);
449
450     if(list0){
451         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
452         mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
453                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
454                            qpix_op, chroma_op);
455
456         qpix_op=  qpix_avg;
457         chroma_op= chroma_avg;
458     }
459
460     if(list1){
461         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
462         mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
463                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
464                            qpix_op, chroma_op);
465     }
466 }
467
468 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
469                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
470                            int x_offset, int y_offset,
471                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
472                            h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
473                            h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
474                            int list0, int list1){
475     MpegEncContext * const s = &h->s;
476
477     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
478     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
479     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
480     x_offset += 8*s->mb_x;
481     y_offset += 8*(s->mb_y >> MB_FIELD);
482
483     if(list0 && list1){
484         /* don't optimize for luma-only case, since B-frames usually
485          * use implicit weights => chroma too. */
486         uint8_t *tmp_cb = s->obmc_scratchpad;
487         uint8_t *tmp_cr = s->obmc_scratchpad + 8;
488         uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
489         int refn0 = h->ref_cache[0][ scan8[n] ];
490         int refn1 = h->ref_cache[1][ scan8[n] ];
491
492         mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
493                     dest_y, dest_cb, dest_cr,
494                     x_offset, y_offset, qpix_put, chroma_put);
495         mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
496                     tmp_y, tmp_cb, tmp_cr,
497                     x_offset, y_offset, qpix_put, chroma_put);
498
499         if(h->use_weight == 2){
500             int weight0 = h->implicit_weight[refn0][refn1];
501             int weight1 = 64 - weight0;
502             luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
503             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
504             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
505         }else{
506             luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
507                             h->luma_weight[0][refn0], h->luma_weight[1][refn1],
508                             h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
509             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
510                             h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
511                             h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
512             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
513                             h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
514                             h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
515         }
516     }else{
517         int list = list1 ? 1 : 0;
518         int refn = h->ref_cache[list][ scan8[n] ];
519         Picture *ref= &h->ref_list[list][refn];
520         mc_dir_part(h, ref, n, square, chroma_height, delta, list,
521                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
522                     qpix_put, chroma_put);
523
524         luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
525                        h->luma_weight[list][refn], h->luma_offset[list][refn]);
526         if(h->use_weight_chroma){
527             chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
528                              h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
529             chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
530                              h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
531         }
532     }
533 }
534
535 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
536                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
537                            int x_offset, int y_offset,
538                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
539                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
540                            h264_weight_func *weight_op, h264_biweight_func *weight_avg,
541                            int list0, int list1){
542     if((h->use_weight==2 && list0 && list1
543         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
544        || h->use_weight==1)
545         mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
546                          x_offset, y_offset, qpix_put, chroma_put,
547                          weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
548     else
549         mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
550                     x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
551 }
552
553 static inline void prefetch_motion(H264Context *h, int list){
554     /* fetch pixels for estimated mv 4 macroblocks ahead
555      * optimized for 64byte cache lines */
556     MpegEncContext * const s = &h->s;
557     const int refn = h->ref_cache[list][scan8[0]];
558     if(refn >= 0){
559         const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
560         const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
561         uint8_t **src= h->ref_list[list][refn].data;
562         int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
563         s->dsp.prefetch(src[0]+off, s->linesize, 4);
564         off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
565         s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
566     }
567 }
568
569 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
570                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
571                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
572                       h264_weight_func *weight_op, h264_biweight_func *weight_avg){
573     MpegEncContext * const s = &h->s;
574     const int mb_xy= h->mb_xy;
575     const int mb_type= s->current_picture.mb_type[mb_xy];
576
577     assert(IS_INTER(mb_type));
578
579     prefetch_motion(h, 0);
580
581     if(IS_16X16(mb_type)){
582         mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
583                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
584                 &weight_op[0], &weight_avg[0],
585                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
586     }else if(IS_16X8(mb_type)){
587         mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
588                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
589                 &weight_op[1], &weight_avg[1],
590                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
591         mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
592                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
593                 &weight_op[1], &weight_avg[1],
594                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
595     }else if(IS_8X16(mb_type)){
596         mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
597                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
598                 &weight_op[2], &weight_avg[2],
599                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
600         mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
601                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
602                 &weight_op[2], &weight_avg[2],
603                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
604     }else{
605         int i;
606
607         assert(IS_8X8(mb_type));
608
609         for(i=0; i<4; i++){
610             const int sub_mb_type= h->sub_mb_type[i];
611             const int n= 4*i;
612             int x_offset= (i&1)<<2;
613             int y_offset= (i&2)<<1;
614
615             if(IS_SUB_8X8(sub_mb_type)){
616                 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
617                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
618                     &weight_op[3], &weight_avg[3],
619                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
620             }else if(IS_SUB_8X4(sub_mb_type)){
621                 mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
622                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
623                     &weight_op[4], &weight_avg[4],
624                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
625                 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
626                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
627                     &weight_op[4], &weight_avg[4],
628                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
629             }else if(IS_SUB_4X8(sub_mb_type)){
630                 mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
631                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
632                     &weight_op[5], &weight_avg[5],
633                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
634                 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
635                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
636                     &weight_op[5], &weight_avg[5],
637                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
638             }else{
639                 int j;
640                 assert(IS_SUB_4X4(sub_mb_type));
641                 for(j=0; j<4; j++){
642                     int sub_x_offset= x_offset + 2*(j&1);
643                     int sub_y_offset= y_offset +   (j&2);
644                     mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
645                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
646                         &weight_op[6], &weight_avg[6],
647                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
648                 }
649             }
650         }
651     }
652
653     prefetch_motion(h, 1);
654 }
655
656
657 static void free_tables(H264Context *h){
658     int i;
659     H264Context *hx;
660     av_freep(&h->intra4x4_pred_mode);
661     av_freep(&h->chroma_pred_mode_table);
662     av_freep(&h->cbp_table);
663     av_freep(&h->mvd_table[0]);
664     av_freep(&h->mvd_table[1]);
665     av_freep(&h->direct_table);
666     av_freep(&h->non_zero_count);
667     av_freep(&h->slice_table_base);
668     h->slice_table= NULL;
669
670     av_freep(&h->mb2b_xy);
671     av_freep(&h->mb2b8_xy);
672
673     for(i = 0; i < MAX_THREADS; i++) {
674         hx = h->thread_context[i];
675         if(!hx) continue;
676         av_freep(&hx->top_borders[1]);
677         av_freep(&hx->top_borders[0]);
678         av_freep(&hx->s.obmc_scratchpad);
679         av_freep(&hx->rbsp_buffer[1]);
680         av_freep(&hx->rbsp_buffer[0]);
681         hx->rbsp_buffer_size[0] = 0;
682         hx->rbsp_buffer_size[1] = 0;
683         if (i) av_freep(&h->thread_context[i]);
684     }
685 }
686
687 static void init_dequant8_coeff_table(H264Context *h){
688     int i,q,x;
689     const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
690     h->dequant8_coeff[0] = h->dequant8_buffer[0];
691     h->dequant8_coeff[1] = h->dequant8_buffer[1];
692
693     for(i=0; i<2; i++ ){
694         if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
695             h->dequant8_coeff[1] = h->dequant8_buffer[0];
696             break;
697         }
698
699         for(q=0; q<52; q++){
700             int shift = div6[q];
701             int idx = rem6[q];
702             for(x=0; x<64; x++)
703                 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
704                     ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
705                     h->pps.scaling_matrix8[i][x]) << shift;
706         }
707     }
708 }
709
710 static void init_dequant4_coeff_table(H264Context *h){
711     int i,j,q,x;
712     const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
713     for(i=0; i<6; i++ ){
714         h->dequant4_coeff[i] = h->dequant4_buffer[i];
715         for(j=0; j<i; j++){
716             if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
717                 h->dequant4_coeff[i] = h->dequant4_buffer[j];
718                 break;
719             }
720         }
721         if(j<i)
722             continue;
723
724         for(q=0; q<52; q++){
725             int shift = div6[q] + 2;
726             int idx = rem6[q];
727             for(x=0; x<16; x++)
728                 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
729                     ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
730                     h->pps.scaling_matrix4[i][x]) << shift;
731         }
732     }
733 }
734
735 static void init_dequant_tables(H264Context *h){
736     int i,x;
737     init_dequant4_coeff_table(h);
738     if(h->pps.transform_8x8_mode)
739         init_dequant8_coeff_table(h);
740     if(h->sps.transform_bypass){
741         for(i=0; i<6; i++)
742             for(x=0; x<16; x++)
743                 h->dequant4_coeff[i][0][x] = 1<<6;
744         if(h->pps.transform_8x8_mode)
745             for(i=0; i<2; i++)
746                 for(x=0; x<64; x++)
747                     h->dequant8_coeff[i][0][x] = 1<<6;
748     }
749 }
750
751
752 int ff_h264_alloc_tables(H264Context *h){
753     MpegEncContext * const s = &h->s;
754     const int big_mb_num= s->mb_stride * (s->mb_height+1);
755     int x,y;
756
757     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t), fail)
758
759     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t), fail)
760     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
761     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
762
763     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
764     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t), fail);
765     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t), fail);
766     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 32*big_mb_num * sizeof(uint8_t) , fail);
767
768     memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
769     h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
770
771     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy  , big_mb_num * sizeof(uint32_t), fail);
772     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b8_xy , big_mb_num * sizeof(uint32_t), fail);
773     for(y=0; y<s->mb_height; y++){
774         for(x=0; x<s->mb_width; x++){
775             const int mb_xy= x + y*s->mb_stride;
776             const int b_xy = 4*x + 4*y*h->b_stride;
777             const int b8_xy= 2*x + 2*y*h->b8_stride;
778
779             h->mb2b_xy [mb_xy]= b_xy;
780             h->mb2b8_xy[mb_xy]= b8_xy;
781         }
782     }
783
784     s->obmc_scratchpad = NULL;
785
786     if(!h->dequant4_coeff[0])
787         init_dequant_tables(h);
788
789     return 0;
790 fail:
791     free_tables(h);
792     return -1;
793 }
794
795 /**
796  * Mimic alloc_tables(), but for every context thread.
797  */
798 static void clone_tables(H264Context *dst, H264Context *src){
799     dst->intra4x4_pred_mode       = src->intra4x4_pred_mode;
800     dst->non_zero_count           = src->non_zero_count;
801     dst->slice_table              = src->slice_table;
802     dst->cbp_table                = src->cbp_table;
803     dst->mb2b_xy                  = src->mb2b_xy;
804     dst->mb2b8_xy                 = src->mb2b8_xy;
805     dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
806     dst->mvd_table[0]             = src->mvd_table[0];
807     dst->mvd_table[1]             = src->mvd_table[1];
808     dst->direct_table             = src->direct_table;
809
810     dst->s.obmc_scratchpad = NULL;
811     ff_h264_pred_init(&dst->hpc, src->s.codec_id);
812 }
813
814 /**
815  * Init context
816  * Allocate buffers which are not shared amongst multiple threads.
817  */
818 static int context_init(H264Context *h){
819     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
820     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
821
822     return 0;
823 fail:
824     return -1; // free_tables will clean up for us
825 }
826
827 static av_cold void common_init(H264Context *h){
828     MpegEncContext * const s = &h->s;
829
830     s->width = s->avctx->width;
831     s->height = s->avctx->height;
832     s->codec_id= s->avctx->codec->id;
833
834     ff_h264_pred_init(&h->hpc, s->codec_id);
835
836     h->dequant_coeff_pps= -1;
837     s->unrestricted_mv=1;
838     s->decode=1; //FIXME
839
840     dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
841
842     memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
843     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
844 }
845
846 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
847     H264Context *h= avctx->priv_data;
848     MpegEncContext * const s = &h->s;
849
850     MPV_decode_defaults(s);
851
852     s->avctx = avctx;
853     common_init(h);
854
855     s->out_format = FMT_H264;
856     s->workaround_bugs= avctx->workaround_bugs;
857
858     // set defaults
859 //    s->decode_mb= ff_h263_decode_mb;
860     s->quarter_sample = 1;
861     if(!avctx->has_b_frames)
862     s->low_delay= 1;
863
864     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
865
866     ff_h264_decode_init_vlc();
867
868     if(avctx->extradata_size > 0 && avctx->extradata &&
869        *(char *)avctx->extradata == 1){
870         h->is_avc = 1;
871         h->got_avcC = 0;
872     } else {
873         h->is_avc = 0;
874     }
875
876     h->thread_context[0] = h;
877     h->outputed_poc = INT_MIN;
878     h->prev_poc_msb= 1<<16;
879     ff_h264_reset_sei(h);
880     if(avctx->codec_id == CODEC_ID_H264){
881         if(avctx->ticks_per_frame == 1){
882             s->avctx->time_base.den *=2;
883         }
884         avctx->ticks_per_frame = 2;
885     }
886     return 0;
887 }
888
889 int ff_h264_frame_start(H264Context *h){
890     MpegEncContext * const s = &h->s;
891     int i;
892
893     if(MPV_frame_start(s, s->avctx) < 0)
894         return -1;
895     ff_er_frame_start(s);
896     /*
897      * MPV_frame_start uses pict_type to derive key_frame.
898      * This is incorrect for H.264; IDR markings must be used.
899      * Zero here; IDR markings per slice in frame or fields are ORed in later.
900      * See decode_nal_units().
901      */
902     s->current_picture_ptr->key_frame= 0;
903     s->current_picture_ptr->mmco_reset= 0;
904
905     assert(s->linesize && s->uvlinesize);
906
907     for(i=0; i<16; i++){
908         h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
909         h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
910     }
911     for(i=0; i<4; i++){
912         h->block_offset[16+i]=
913         h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
914         h->block_offset[24+16+i]=
915         h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
916     }
917
918     /* can't be in alloc_tables because linesize isn't known there.
919      * FIXME: redo bipred weight to not require extra buffer? */
920     for(i = 0; i < s->avctx->thread_count; i++)
921         if(!h->thread_context[i]->s.obmc_scratchpad)
922             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
923
924     /* some macroblocks will be accessed before they're available */
925     if(FRAME_MBAFF || s->avctx->thread_count > 1)
926         memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
927
928 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
929
930     // We mark the current picture as non-reference after allocating it, so
931     // that if we break out due to an error it can be released automatically
932     // in the next MPV_frame_start().
933     // SVQ3 as well as most other codecs have only last/next/current and thus
934     // get released even with set reference, besides SVQ3 and others do not
935     // mark frames as reference later "naturally".
936     if(s->codec_id != CODEC_ID_SVQ3)
937         s->current_picture_ptr->reference= 0;
938
939     s->current_picture_ptr->field_poc[0]=
940     s->current_picture_ptr->field_poc[1]= INT_MAX;
941     assert(s->current_picture_ptr->long_ref==0);
942
943     return 0;
944 }
945
946 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){
947     MpegEncContext * const s = &h->s;
948     int i;
949     int step    = 1;
950     int offset  = 1;
951     int uvoffset= 1;
952     int top_idx = 1;
953     int skiplast= 0;
954
955     src_y  -=   linesize;
956     src_cb -= uvlinesize;
957     src_cr -= uvlinesize;
958
959     if(!simple && FRAME_MBAFF){
960         if(s->mb_y&1){
961             offset  = MB_MBAFF ? 1 : 17;
962             uvoffset= MB_MBAFF ? 1 : 9;
963             if(!MB_MBAFF){
964                 *(uint64_t*)(h->top_borders[0][s->mb_x]+ 0)= *(uint64_t*)(src_y +  15*linesize);
965                 *(uint64_t*)(h->top_borders[0][s->mb_x]+ 8)= *(uint64_t*)(src_y +8+15*linesize);
966                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
967                     *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+7*uvlinesize);
968                     *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+7*uvlinesize);
969                 }
970             }
971         }else{
972             if(!MB_MBAFF){
973                 h->left_border[0]= h->top_borders[0][s->mb_x][15];
974                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
975                     h->left_border[34   ]= h->top_borders[0][s->mb_x][16+7  ];
976                     h->left_border[34+18]= h->top_borders[0][s->mb_x][16+8+7];
977                 }
978                 skiplast= 1;
979             }
980             offset  =
981             uvoffset=
982             top_idx = MB_MBAFF ? 0 : 1;
983         }
984         step= MB_MBAFF ? 2 : 1;
985     }
986
987     // There are two lines saved, the line above the the top macroblock of a pair,
988     // and the line above the bottom macroblock
989     h->left_border[offset]= h->top_borders[top_idx][s->mb_x][15];
990     for(i=1; i<17 - skiplast; i++){
991         h->left_border[offset+i*step]= src_y[15+i*  linesize];
992     }
993
994     *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
995     *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
996
997     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
998         h->left_border[uvoffset+34   ]= h->top_borders[top_idx][s->mb_x][16+7];
999         h->left_border[uvoffset+34+18]= h->top_borders[top_idx][s->mb_x][24+7];
1000         for(i=1; i<9 - skiplast; i++){
1001             h->left_border[uvoffset+34   +i*step]= src_cb[7+i*uvlinesize];
1002             h->left_border[uvoffset+34+18+i*step]= src_cr[7+i*uvlinesize];
1003         }
1004         *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
1005         *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
1006     }
1007 }
1008
1009 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){
1010     MpegEncContext * const s = &h->s;
1011     int temp8, i;
1012     uint64_t temp64;
1013     int deblock_left;
1014     int deblock_top;
1015     int mb_xy;
1016     int step    = 1;
1017     int offset  = 1;
1018     int uvoffset= 1;
1019     int top_idx = 1;
1020
1021     if(!simple && FRAME_MBAFF){
1022         if(s->mb_y&1){
1023             offset  = MB_MBAFF ? 1 : 17;
1024             uvoffset= MB_MBAFF ? 1 : 9;
1025         }else{
1026             offset  =
1027             uvoffset=
1028             top_idx = MB_MBAFF ? 0 : 1;
1029         }
1030         step= MB_MBAFF ? 2 : 1;
1031     }
1032
1033     if(h->deblocking_filter == 2) {
1034         mb_xy = h->mb_xy;
1035         deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
1036         deblock_top  = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
1037     } else {
1038         deblock_left = (s->mb_x > 0);
1039         deblock_top =  (s->mb_y > !!MB_FIELD);
1040     }
1041
1042     src_y  -=   linesize + 1;
1043     src_cb -= uvlinesize + 1;
1044     src_cr -= uvlinesize + 1;
1045
1046 #define XCHG(a,b,t,xchg)\
1047 t= a;\
1048 if(xchg)\
1049     a= b;\
1050 b= t;
1051
1052     if(deblock_left){
1053         for(i = !deblock_top; i<16; i++){
1054             XCHG(h->left_border[offset+i*step], src_y [i*  linesize], temp8, xchg);
1055         }
1056         XCHG(h->left_border[offset+i*step], src_y [i*  linesize], temp8, 1);
1057     }
1058
1059     if(deblock_top){
1060         XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
1061         XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
1062         if(s->mb_x+1 < s->mb_width){
1063             XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
1064         }
1065     }
1066
1067     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1068         if(deblock_left){
1069             for(i = !deblock_top; i<8; i++){
1070                 XCHG(h->left_border[uvoffset+34   +i*step], src_cb[i*uvlinesize], temp8, xchg);
1071                 XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, xchg);
1072             }
1073             XCHG(h->left_border[uvoffset+34   +i*step], src_cb[i*uvlinesize], temp8, 1);
1074             XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, 1);
1075         }
1076         if(deblock_top){
1077             XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
1078             XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
1079         }
1080     }
1081 }
1082
1083 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
1084     MpegEncContext * const s = &h->s;
1085     const int mb_x= s->mb_x;
1086     const int mb_y= s->mb_y;
1087     const int mb_xy= h->mb_xy;
1088     const int mb_type= s->current_picture.mb_type[mb_xy];
1089     uint8_t  *dest_y, *dest_cb, *dest_cr;
1090     int linesize, uvlinesize /*dct_offset*/;
1091     int i;
1092     int *block_offset = &h->block_offset[0];
1093     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1094     /* is_h264 should always be true if SVQ3 is disabled. */
1095     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1096     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1097     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1098
1099     dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
1100     dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
1101     dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
1102
1103     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
1104     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
1105
1106     if (!simple && MB_FIELD) {
1107         linesize   = h->mb_linesize   = s->linesize * 2;
1108         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1109         block_offset = &h->block_offset[24];
1110         if(mb_y&1){ //FIXME move out of this function?
1111             dest_y -= s->linesize*15;
1112             dest_cb-= s->uvlinesize*7;
1113             dest_cr-= s->uvlinesize*7;
1114         }
1115         if(FRAME_MBAFF) {
1116             int list;
1117             for(list=0; list<h->list_count; list++){
1118                 if(!USES_LIST(mb_type, list))
1119                     continue;
1120                 if(IS_16X16(mb_type)){
1121                     int8_t *ref = &h->ref_cache[list][scan8[0]];
1122                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1123                 }else{
1124                     for(i=0; i<16; i+=4){
1125                         int ref = h->ref_cache[list][scan8[i]];
1126                         if(ref >= 0)
1127                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1128                     }
1129                 }
1130             }
1131         }
1132     } else {
1133         linesize   = h->mb_linesize   = s->linesize;
1134         uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1135 //        dct_offset = s->linesize * 16;
1136     }
1137
1138     if (!simple && IS_INTRA_PCM(mb_type)) {
1139         for (i=0; i<16; i++) {
1140             memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1141         }
1142         for (i=0; i<8; i++) {
1143             memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
1144             memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
1145         }
1146     } else {
1147         if(IS_INTRA(mb_type)){
1148             if(h->deblocking_filter)
1149                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
1150
1151             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1152                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1153                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1154             }
1155
1156             if(IS_INTRA4x4(mb_type)){
1157                 if(simple || !s->encoding){
1158                     if(IS_8x8DCT(mb_type)){
1159                         if(transform_bypass){
1160                             idct_dc_add =
1161                             idct_add    = s->dsp.add_pixels8;
1162                         }else{
1163                             idct_dc_add = s->dsp.h264_idct8_dc_add;
1164                             idct_add    = s->dsp.h264_idct8_add;
1165                         }
1166                         for(i=0; i<16; i+=4){
1167                             uint8_t * const ptr= dest_y + block_offset[i];
1168                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1169                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1170                                 h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
1171                             }else{
1172                                 const int nnz = h->non_zero_count_cache[ scan8[i] ];
1173                                 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1174                                                             (h->topright_samples_available<<i)&0x4000, linesize);
1175                                 if(nnz){
1176                                     if(nnz == 1 && h->mb[i*16])
1177                                         idct_dc_add(ptr, h->mb + i*16, linesize);
1178                                     else
1179                                         idct_add   (ptr, h->mb + i*16, linesize);
1180                                 }
1181                             }
1182                         }
1183                     }else{
1184                         if(transform_bypass){
1185                             idct_dc_add =
1186                             idct_add    = s->dsp.add_pixels4;
1187                         }else{
1188                             idct_dc_add = s->dsp.h264_idct_dc_add;
1189                             idct_add    = s->dsp.h264_idct_add;
1190                         }
1191                         for(i=0; i<16; i++){
1192                             uint8_t * const ptr= dest_y + block_offset[i];
1193                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1194
1195                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1196                                 h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
1197                             }else{
1198                                 uint8_t *topright;
1199                                 int nnz, tr;
1200                                 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1201                                     const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1202                                     assert(mb_y || linesize <= block_offset[i]);
1203                                     if(!topright_avail){
1204                                         tr= ptr[3 - linesize]*0x01010101;
1205                                         topright= (uint8_t*) &tr;
1206                                     }else
1207                                         topright= ptr + 4 - linesize;
1208                                 }else
1209                                     topright= NULL;
1210
1211                                 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1212                                 nnz = h->non_zero_count_cache[ scan8[i] ];
1213                                 if(nnz){
1214                                     if(is_h264){
1215                                         if(nnz == 1 && h->mb[i*16])
1216                                             idct_dc_add(ptr, h->mb + i*16, linesize);
1217                                         else
1218                                             idct_add   (ptr, h->mb + i*16, linesize);
1219                                     }else
1220                                         svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1221                                 }
1222                             }
1223                         }
1224                     }
1225                 }
1226             }else{
1227                 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1228                 if(is_h264){
1229                     if(!transform_bypass)
1230                         h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
1231                 }else
1232                     svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
1233             }
1234             if(h->deblocking_filter)
1235                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
1236         }else if(is_h264){
1237             hl_motion(h, dest_y, dest_cb, dest_cr,
1238                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1239                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1240                       s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
1241         }
1242
1243
1244         if(!IS_INTRA4x4(mb_type)){
1245             if(is_h264){
1246                 if(IS_INTRA16x16(mb_type)){
1247                     if(transform_bypass){
1248                         if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1249                             h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1250                         }else{
1251                             for(i=0; i<16; i++){
1252                                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1253                                     s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
1254                             }
1255                         }
1256                     }else{
1257                          s->dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1258                     }
1259                 }else if(h->cbp&15){
1260                     if(transform_bypass){
1261                         const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1262                         idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1263                         for(i=0; i<16; i+=di){
1264                             if(h->non_zero_count_cache[ scan8[i] ]){
1265                                 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
1266                             }
1267                         }
1268                     }else{
1269                         if(IS_8x8DCT(mb_type)){
1270                             s->dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1271                         }else{
1272                             s->dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1273                         }
1274                     }
1275                 }
1276             }else{
1277                 for(i=0; i<16; i++){
1278                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1279                         uint8_t * const ptr= dest_y + block_offset[i];
1280                         svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1281                     }
1282                 }
1283             }
1284         }
1285
1286         if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1287             uint8_t *dest[2] = {dest_cb, dest_cr};
1288             if(transform_bypass){
1289                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1290                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
1291                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
1292                 }else{
1293                     idct_add = s->dsp.add_pixels4;
1294                     for(i=16; i<16+8; i++){
1295                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1296                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1297                     }
1298                 }
1299             }else{
1300                 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp[0], h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
1301                 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp[1], h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
1302                 if(is_h264){
1303                     idct_add = s->dsp.h264_idct_add;
1304                     idct_dc_add = s->dsp.h264_idct_dc_add;
1305                     for(i=16; i<16+8; i++){
1306                         if(h->non_zero_count_cache[ scan8[i] ])
1307                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1308                         else if(h->mb[i*16])
1309                             idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1310                     }
1311                 }else{
1312                     for(i=16; i<16+8; i++){
1313                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1314                             uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1315                             svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
1316                         }
1317                     }
1318                 }
1319             }
1320         }
1321     }
1322     if(h->cbp || IS_INTRA(mb_type))
1323         s->dsp.clear_blocks(h->mb);
1324
1325     if(h->deblocking_filter) {
1326         backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, simple);
1327         fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
1328         h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
1329         h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
1330         if (!simple && FRAME_MBAFF) {
1331             ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
1332         } else {
1333             ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
1334         }
1335     }
1336 }
1337
1338 /**
1339  * Process a macroblock; this case avoids checks for expensive uncommon cases.
1340  */
1341 static void hl_decode_mb_simple(H264Context *h){
1342     hl_decode_mb_internal(h, 1);
1343 }
1344
1345 /**
1346  * Process a macroblock; this handles edge cases, such as interlacing.
1347  */
1348 static void av_noinline hl_decode_mb_complex(H264Context *h){
1349     hl_decode_mb_internal(h, 0);
1350 }
1351
1352 void ff_h264_hl_decode_mb(H264Context *h){
1353     MpegEncContext * const s = &h->s;
1354     const int mb_xy= h->mb_xy;
1355     const int mb_type= s->current_picture.mb_type[mb_xy];
1356     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1357
1358     if (is_complex)
1359         hl_decode_mb_complex(h);
1360     else hl_decode_mb_simple(h);
1361 }
1362
1363 static int pred_weight_table(H264Context *h){
1364     MpegEncContext * const s = &h->s;
1365     int list, i;
1366     int luma_def, chroma_def;
1367
1368     h->use_weight= 0;
1369     h->use_weight_chroma= 0;
1370     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1371     h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1372     luma_def = 1<<h->luma_log2_weight_denom;
1373     chroma_def = 1<<h->chroma_log2_weight_denom;
1374
1375     for(list=0; list<2; list++){
1376         h->luma_weight_flag[list]   = 0;
1377         h->chroma_weight_flag[list] = 0;
1378         for(i=0; i<h->ref_count[list]; i++){
1379             int luma_weight_flag, chroma_weight_flag;
1380
1381             luma_weight_flag= get_bits1(&s->gb);
1382             if(luma_weight_flag){
1383                 h->luma_weight[list][i]= get_se_golomb(&s->gb);
1384                 h->luma_offset[list][i]= get_se_golomb(&s->gb);
1385                 if(   h->luma_weight[list][i] != luma_def
1386                    || h->luma_offset[list][i] != 0) {
1387                     h->use_weight= 1;
1388                     h->luma_weight_flag[list]= 1;
1389                 }
1390             }else{
1391                 h->luma_weight[list][i]= luma_def;
1392                 h->luma_offset[list][i]= 0;
1393             }
1394
1395             if(CHROMA){
1396                 chroma_weight_flag= get_bits1(&s->gb);
1397                 if(chroma_weight_flag){
1398                     int j;
1399                     for(j=0; j<2; j++){
1400                         h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
1401                         h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
1402                         if(   h->chroma_weight[list][i][j] != chroma_def
1403                            || h->chroma_offset[list][i][j] != 0) {
1404                             h->use_weight_chroma= 1;
1405                             h->chroma_weight_flag[list]= 1;
1406                         }
1407                     }
1408                 }else{
1409                     int j;
1410                     for(j=0; j<2; j++){
1411                         h->chroma_weight[list][i][j]= chroma_def;
1412                         h->chroma_offset[list][i][j]= 0;
1413                     }
1414                 }
1415             }
1416         }
1417         if(h->slice_type_nos != FF_B_TYPE) break;
1418     }
1419     h->use_weight= h->use_weight || h->use_weight_chroma;
1420     return 0;
1421 }
1422
1423 static void implicit_weight_table(H264Context *h){
1424     MpegEncContext * const s = &h->s;
1425     int ref0, ref1, i;
1426     int cur_poc = s->current_picture_ptr->poc;
1427
1428     for (i = 0; i < 2; i++) {
1429         h->luma_weight_flag[i]   = 0;
1430         h->chroma_weight_flag[i] = 0;
1431     }
1432
1433     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
1434        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1435         h->use_weight= 0;
1436         h->use_weight_chroma= 0;
1437         return;
1438     }
1439
1440     h->use_weight= 2;
1441     h->use_weight_chroma= 2;
1442     h->luma_log2_weight_denom= 5;
1443     h->chroma_log2_weight_denom= 5;
1444
1445     for(ref0=0; ref0 < h->ref_count[0]; ref0++){
1446         int poc0 = h->ref_list[0][ref0].poc;
1447         for(ref1=0; ref1 < h->ref_count[1]; ref1++){
1448             int poc1 = h->ref_list[1][ref1].poc;
1449             int td = av_clip(poc1 - poc0, -128, 127);
1450             if(td){
1451                 int tb = av_clip(cur_poc - poc0, -128, 127);
1452                 int tx = (16384 + (FFABS(td) >> 1)) / td;
1453                 int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
1454                 if(dist_scale_factor < -64 || dist_scale_factor > 128)
1455                     h->implicit_weight[ref0][ref1] = 32;
1456                 else
1457                     h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
1458             }else
1459                 h->implicit_weight[ref0][ref1] = 32;
1460         }
1461     }
1462 }
1463
1464 /**
1465  * instantaneous decoder refresh.
1466  */
1467 static void idr(H264Context *h){
1468     ff_h264_remove_all_refs(h);
1469     h->prev_frame_num= 0;
1470     h->prev_frame_num_offset= 0;
1471     h->prev_poc_msb=
1472     h->prev_poc_lsb= 0;
1473 }
1474
1475 /* forget old pics after a seek */
1476 static void flush_dpb(AVCodecContext *avctx){
1477     H264Context *h= avctx->priv_data;
1478     int i;
1479     for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1480         if(h->delayed_pic[i])
1481             h->delayed_pic[i]->reference= 0;
1482         h->delayed_pic[i]= NULL;
1483     }
1484     h->outputed_poc= INT_MIN;
1485     h->prev_interlaced_frame = 1;
1486     idr(h);
1487     if(h->s.current_picture_ptr)
1488         h->s.current_picture_ptr->reference= 0;
1489     h->s.first_field= 0;
1490     ff_h264_reset_sei(h);
1491     ff_mpeg_flush(avctx);
1492 }
1493
1494 static int init_poc(H264Context *h){
1495     MpegEncContext * const s = &h->s;
1496     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1497     int field_poc[2];
1498     Picture *cur = s->current_picture_ptr;
1499
1500     h->frame_num_offset= h->prev_frame_num_offset;
1501     if(h->frame_num < h->prev_frame_num)
1502         h->frame_num_offset += max_frame_num;
1503
1504     if(h->sps.poc_type==0){
1505         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1506
1507         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1508             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1509         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1510             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1511         else
1512             h->poc_msb = h->prev_poc_msb;
1513 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1514         field_poc[0] =
1515         field_poc[1] = h->poc_msb + h->poc_lsb;
1516         if(s->picture_structure == PICT_FRAME)
1517             field_poc[1] += h->delta_poc_bottom;
1518     }else if(h->sps.poc_type==1){
1519         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1520         int i;
1521
1522         if(h->sps.poc_cycle_length != 0)
1523             abs_frame_num = h->frame_num_offset + h->frame_num;
1524         else
1525             abs_frame_num = 0;
1526
1527         if(h->nal_ref_idc==0 && abs_frame_num > 0)
1528             abs_frame_num--;
1529
1530         expected_delta_per_poc_cycle = 0;
1531         for(i=0; i < h->sps.poc_cycle_length; i++)
1532             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1533
1534         if(abs_frame_num > 0){
1535             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1536             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1537
1538             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1539             for(i = 0; i <= frame_num_in_poc_cycle; i++)
1540                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1541         } else
1542             expectedpoc = 0;
1543
1544         if(h->nal_ref_idc == 0)
1545             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1546
1547         field_poc[0] = expectedpoc + h->delta_poc[0];
1548         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1549
1550         if(s->picture_structure == PICT_FRAME)
1551             field_poc[1] += h->delta_poc[1];
1552     }else{
1553         int poc= 2*(h->frame_num_offset + h->frame_num);
1554
1555         if(!h->nal_ref_idc)
1556             poc--;
1557
1558         field_poc[0]= poc;
1559         field_poc[1]= poc;
1560     }
1561
1562     if(s->picture_structure != PICT_BOTTOM_FIELD)
1563         s->current_picture_ptr->field_poc[0]= field_poc[0];
1564     if(s->picture_structure != PICT_TOP_FIELD)
1565         s->current_picture_ptr->field_poc[1]= field_poc[1];
1566     cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1567
1568     return 0;
1569 }
1570
1571
1572 /**
1573  * initialize scan tables
1574  */
1575 static void init_scan_tables(H264Context *h){
1576     MpegEncContext * const s = &h->s;
1577     int i;
1578     if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
1579         memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
1580         memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
1581     }else{
1582         for(i=0; i<16; i++){
1583 #define T(x) (x>>2) | ((x<<2) & 0xF)
1584             h->zigzag_scan[i] = T(zigzag_scan[i]);
1585             h-> field_scan[i] = T( field_scan[i]);
1586 #undef T
1587         }
1588     }
1589     if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
1590         memcpy(h->zigzag_scan8x8,       ff_zigzag_direct,     64*sizeof(uint8_t));
1591         memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
1592         memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
1593         memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
1594     }else{
1595         for(i=0; i<64; i++){
1596 #define T(x) (x>>3) | ((x&7)<<3)
1597             h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
1598             h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1599             h->field_scan8x8[i]        = T(field_scan8x8[i]);
1600             h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
1601 #undef T
1602         }
1603     }
1604     if(h->sps.transform_bypass){ //FIXME same ugly
1605         h->zigzag_scan_q0          = zigzag_scan;
1606         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
1607         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1608         h->field_scan_q0           = field_scan;
1609         h->field_scan8x8_q0        = field_scan8x8;
1610         h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
1611     }else{
1612         h->zigzag_scan_q0          = h->zigzag_scan;
1613         h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
1614         h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1615         h->field_scan_q0           = h->field_scan;
1616         h->field_scan8x8_q0        = h->field_scan8x8;
1617         h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
1618     }
1619 }
1620
1621 static void field_end(H264Context *h){
1622     MpegEncContext * const s = &h->s;
1623     AVCodecContext * const avctx= s->avctx;
1624     s->mb_y= 0;
1625
1626     s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1627     s->current_picture_ptr->pict_type= s->pict_type;
1628
1629     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1630         ff_vdpau_h264_set_reference_frames(s);
1631
1632     if(!s->dropable) {
1633         ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1634         h->prev_poc_msb= h->poc_msb;
1635         h->prev_poc_lsb= h->poc_lsb;
1636     }
1637     h->prev_frame_num_offset= h->frame_num_offset;
1638     h->prev_frame_num= h->frame_num;
1639
1640     if (avctx->hwaccel) {
1641         if (avctx->hwaccel->end_frame(avctx) < 0)
1642             av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1643     }
1644
1645     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1646         ff_vdpau_h264_picture_complete(s);
1647
1648     /*
1649      * FIXME: Error handling code does not seem to support interlaced
1650      * when slices span multiple rows
1651      * The ff_er_add_slice calls don't work right for bottom
1652      * fields; they cause massive erroneous error concealing
1653      * Error marking covers both fields (top and bottom).
1654      * This causes a mismatched s->error_count
1655      * and a bad error table. Further, the error count goes to
1656      * INT_MAX when called for bottom field, because mb_y is
1657      * past end by one (callers fault) and resync_mb_y != 0
1658      * causes problems for the first MB line, too.
1659      */
1660     if (!FIELD_PICTURE)
1661         ff_er_frame_end(s);
1662
1663     MPV_frame_end(s);
1664
1665     h->current_slice=0;
1666 }
1667
1668 /**
1669  * Replicates H264 "master" context to thread contexts.
1670  */
1671 static void clone_slice(H264Context *dst, H264Context *src)
1672 {
1673     memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
1674     dst->s.current_picture_ptr  = src->s.current_picture_ptr;
1675     dst->s.current_picture      = src->s.current_picture;
1676     dst->s.linesize             = src->s.linesize;
1677     dst->s.uvlinesize           = src->s.uvlinesize;
1678     dst->s.first_field          = src->s.first_field;
1679
1680     dst->prev_poc_msb           = src->prev_poc_msb;
1681     dst->prev_poc_lsb           = src->prev_poc_lsb;
1682     dst->prev_frame_num_offset  = src->prev_frame_num_offset;
1683     dst->prev_frame_num         = src->prev_frame_num;
1684     dst->short_ref_count        = src->short_ref_count;
1685
1686     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
1687     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
1688     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1689     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
1690
1691     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
1692     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
1693 }
1694
1695 /**
1696  * decodes a slice header.
1697  * This will also call MPV_common_init() and frame_start() as needed.
1698  *
1699  * @param h h264context
1700  * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1701  *
1702  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1703  */
1704 static int decode_slice_header(H264Context *h, H264Context *h0){
1705     MpegEncContext * const s = &h->s;
1706     MpegEncContext * const s0 = &h0->s;
1707     unsigned int first_mb_in_slice;
1708     unsigned int pps_id;
1709     int num_ref_idx_active_override_flag;
1710     unsigned int slice_type, tmp, i, j;
1711     int default_ref_list_done = 0;
1712     int last_pic_structure;
1713
1714     s->dropable= h->nal_ref_idc == 0;
1715
1716     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1717         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1718         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1719     }else{
1720         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1721         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1722     }
1723
1724     first_mb_in_slice= get_ue_golomb(&s->gb);
1725
1726     if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1727         if(h0->current_slice && FIELD_PICTURE){
1728             field_end(h);
1729         }
1730
1731         h0->current_slice = 0;
1732         if (!s0->first_field)
1733             s->current_picture_ptr= NULL;
1734     }
1735
1736     slice_type= get_ue_golomb_31(&s->gb);
1737     if(slice_type > 9){
1738         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);
1739         return -1;
1740     }
1741     if(slice_type > 4){
1742         slice_type -= 5;
1743         h->slice_type_fixed=1;
1744     }else
1745         h->slice_type_fixed=0;
1746
1747     slice_type= golomb_to_pict_type[ slice_type ];
1748     if (slice_type == FF_I_TYPE
1749         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
1750         default_ref_list_done = 1;
1751     }
1752     h->slice_type= slice_type;
1753     h->slice_type_nos= slice_type & 3;
1754
1755     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
1756     if (s->pict_type == FF_B_TYPE && s0->last_picture_ptr == NULL) {
1757         av_log(h->s.avctx, AV_LOG_ERROR,
1758                "B picture before any references, skipping\n");
1759         return -1;
1760     }
1761
1762     pps_id= get_ue_golomb(&s->gb);
1763     if(pps_id>=MAX_PPS_COUNT){
1764         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
1765         return -1;
1766     }
1767     if(!h0->pps_buffers[pps_id]) {
1768         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
1769         return -1;
1770     }
1771     h->pps= *h0->pps_buffers[pps_id];
1772
1773     if(!h0->sps_buffers[h->pps.sps_id]) {
1774         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
1775         return -1;
1776     }
1777     h->sps = *h0->sps_buffers[h->pps.sps_id];
1778
1779     if(h == h0 && h->dequant_coeff_pps != pps_id){
1780         h->dequant_coeff_pps = pps_id;
1781         init_dequant_tables(h);
1782     }
1783
1784     s->mb_width= h->sps.mb_width;
1785     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1786
1787     h->b_stride=  s->mb_width*4;
1788     h->b8_stride= s->mb_width*2;
1789
1790     s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
1791     if(h->sps.frame_mbs_only_flag)
1792         s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
1793     else
1794         s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
1795
1796     if (s->context_initialized
1797         && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
1798         if(h != h0)
1799             return -1;   // width / height changed during parallelized decoding
1800         free_tables(h);
1801         flush_dpb(s->avctx);
1802         MPV_common_end(s);
1803     }
1804     if (!s->context_initialized) {
1805         if(h != h0)
1806             return -1;  // we cant (re-)initialize context during parallel decoding
1807
1808         avcodec_set_dimensions(s->avctx, s->width, s->height);
1809         s->avctx->sample_aspect_ratio= h->sps.sar;
1810         if(!s->avctx->sample_aspect_ratio.den)
1811             s->avctx->sample_aspect_ratio.den = 1;
1812
1813         if(h->sps.video_signal_type_present_flag){
1814             s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1815             if(h->sps.colour_description_present_flag){
1816                 s->avctx->color_primaries = h->sps.color_primaries;
1817                 s->avctx->color_trc       = h->sps.color_trc;
1818                 s->avctx->colorspace      = h->sps.colorspace;
1819             }
1820         }
1821
1822         if(h->sps.timing_info_present_flag){
1823             s->avctx->time_base= (AVRational){h->sps.num_units_in_tick, h->sps.time_scale};
1824             if(h->x264_build > 0 && h->x264_build < 44)
1825                 s->avctx->time_base.den *= 2;
1826             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
1827                       s->avctx->time_base.num, s->avctx->time_base.den, 1<<30);
1828         }
1829         s->avctx->pix_fmt = s->avctx->get_format(s->avctx, s->avctx->codec->pix_fmts);
1830         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1831
1832         if (MPV_common_init(s) < 0)
1833             return -1;
1834         s->first_field = 0;
1835         h->prev_interlaced_frame = 1;
1836
1837         init_scan_tables(h);
1838         ff_h264_alloc_tables(h);
1839
1840         for(i = 1; i < s->avctx->thread_count; i++) {
1841             H264Context *c;
1842             c = h->thread_context[i] = av_malloc(sizeof(H264Context));
1843             memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
1844             memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
1845             c->sps = h->sps;
1846             c->pps = h->pps;
1847             init_scan_tables(c);
1848             clone_tables(c, h);
1849         }
1850
1851         for(i = 0; i < s->avctx->thread_count; i++)
1852             if(context_init(h->thread_context[i]) < 0)
1853                 return -1;
1854     }
1855
1856     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1857
1858     h->mb_mbaff = 0;
1859     h->mb_aff_frame = 0;
1860     last_pic_structure = s0->picture_structure;
1861     if(h->sps.frame_mbs_only_flag){
1862         s->picture_structure= PICT_FRAME;
1863     }else{
1864         if(get_bits1(&s->gb)) { //field_pic_flag
1865             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
1866         } else {
1867             s->picture_structure= PICT_FRAME;
1868             h->mb_aff_frame = h->sps.mb_aff;
1869         }
1870     }
1871     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
1872
1873     if(h0->current_slice == 0){
1874         while(h->frame_num !=  h->prev_frame_num &&
1875               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
1876             av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
1877             if (ff_h264_frame_start(h) < 0)
1878                 return -1;
1879             h->prev_frame_num++;
1880             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
1881             s->current_picture_ptr->frame_num= h->prev_frame_num;
1882             ff_h264_execute_ref_pic_marking(h, NULL, 0);
1883         }
1884
1885         /* See if we have a decoded first field looking for a pair... */
1886         if (s0->first_field) {
1887             assert(s0->current_picture_ptr);
1888             assert(s0->current_picture_ptr->data[0]);
1889             assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
1890
1891             /* figure out if we have a complementary field pair */
1892             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
1893                 /*
1894                  * Previous field is unmatched. Don't display it, but let it
1895                  * remain for reference if marked as such.
1896                  */
1897                 s0->current_picture_ptr = NULL;
1898                 s0->first_field = FIELD_PICTURE;
1899
1900             } else {
1901                 if (h->nal_ref_idc &&
1902                         s0->current_picture_ptr->reference &&
1903                         s0->current_picture_ptr->frame_num != h->frame_num) {
1904                     /*
1905                      * This and previous field were reference, but had
1906                      * different frame_nums. Consider this field first in
1907                      * pair. Throw away previous field except for reference
1908                      * purposes.
1909                      */
1910                     s0->first_field = 1;
1911                     s0->current_picture_ptr = NULL;
1912
1913                 } else {
1914                     /* Second field in complementary pair */
1915                     s0->first_field = 0;
1916                 }
1917             }
1918
1919         } else {
1920             /* Frame or first field in a potentially complementary pair */
1921             assert(!s0->current_picture_ptr);
1922             s0->first_field = FIELD_PICTURE;
1923         }
1924
1925         if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
1926             s0->first_field = 0;
1927             return -1;
1928         }
1929     }
1930     if(h != h0)
1931         clone_slice(h, h0);
1932
1933     s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
1934
1935     assert(s->mb_num == s->mb_width * s->mb_height);
1936     if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
1937        first_mb_in_slice                    >= s->mb_num){
1938         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1939         return -1;
1940     }
1941     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
1942     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
1943     if (s->picture_structure == PICT_BOTTOM_FIELD)
1944         s->resync_mb_y = s->mb_y = s->mb_y + 1;
1945     assert(s->mb_y < s->mb_height);
1946
1947     if(s->picture_structure==PICT_FRAME){
1948         h->curr_pic_num=   h->frame_num;
1949         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
1950     }else{
1951         h->curr_pic_num= 2*h->frame_num + 1;
1952         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
1953     }
1954
1955     if(h->nal_unit_type == NAL_IDR_SLICE){
1956         get_ue_golomb(&s->gb); /* idr_pic_id */
1957     }
1958
1959     if(h->sps.poc_type==0){
1960         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
1961
1962         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
1963             h->delta_poc_bottom= get_se_golomb(&s->gb);
1964         }
1965     }
1966
1967     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
1968         h->delta_poc[0]= get_se_golomb(&s->gb);
1969
1970         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
1971             h->delta_poc[1]= get_se_golomb(&s->gb);
1972     }
1973
1974     init_poc(h);
1975
1976     if(h->pps.redundant_pic_cnt_present){
1977         h->redundant_pic_count= get_ue_golomb(&s->gb);
1978     }
1979
1980     //set defaults, might be overridden a few lines later
1981     h->ref_count[0]= h->pps.ref_count[0];
1982     h->ref_count[1]= h->pps.ref_count[1];
1983
1984     if(h->slice_type_nos != FF_I_TYPE){
1985         if(h->slice_type_nos == FF_B_TYPE){
1986             h->direct_spatial_mv_pred= get_bits1(&s->gb);
1987         }
1988         num_ref_idx_active_override_flag= get_bits1(&s->gb);
1989
1990         if(num_ref_idx_active_override_flag){
1991             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
1992             if(h->slice_type_nos==FF_B_TYPE)
1993                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
1994
1995             if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
1996                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
1997                 h->ref_count[0]= h->ref_count[1]= 1;
1998                 return -1;
1999             }
2000         }
2001         if(h->slice_type_nos == FF_B_TYPE)
2002             h->list_count= 2;
2003         else
2004             h->list_count= 1;
2005     }else
2006         h->list_count= 0;
2007
2008     if(!default_ref_list_done){
2009         ff_h264_fill_default_ref_list(h);
2010     }
2011
2012     if(h->slice_type_nos!=FF_I_TYPE && ff_h264_decode_ref_pic_list_reordering(h) < 0)
2013         return -1;
2014
2015     if(h->slice_type_nos!=FF_I_TYPE){
2016         s->last_picture_ptr= &h->ref_list[0][0];
2017         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
2018     }
2019     if(h->slice_type_nos==FF_B_TYPE){
2020         s->next_picture_ptr= &h->ref_list[1][0];
2021         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
2022     }
2023
2024     if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
2025        ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
2026         pred_weight_table(h);
2027     else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE)
2028         implicit_weight_table(h);
2029     else {
2030         h->use_weight = 0;
2031         for (i = 0; i < 2; i++) {
2032             h->luma_weight_flag[i]   = 0;
2033             h->chroma_weight_flag[i] = 0;
2034         }
2035     }
2036
2037     if(h->nal_ref_idc)
2038         ff_h264_decode_ref_pic_marking(h0, &s->gb);
2039
2040     if(FRAME_MBAFF)
2041         ff_h264_fill_mbaff_ref_list(h);
2042
2043     if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
2044         ff_h264_direct_dist_scale_factor(h);
2045     ff_h264_direct_ref_list_init(h);
2046
2047     if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
2048         tmp = get_ue_golomb_31(&s->gb);
2049         if(tmp > 2){
2050             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2051             return -1;
2052         }
2053         h->cabac_init_idc= tmp;
2054     }
2055
2056     h->last_qscale_diff = 0;
2057     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2058     if(tmp>51){
2059         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2060         return -1;
2061     }
2062     s->qscale= tmp;
2063     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2064     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2065     //FIXME qscale / qp ... stuff
2066     if(h->slice_type == FF_SP_TYPE){
2067         get_bits1(&s->gb); /* sp_for_switch_flag */
2068     }
2069     if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
2070         get_se_golomb(&s->gb); /* slice_qs_delta */
2071     }
2072
2073     h->deblocking_filter = 1;
2074     h->slice_alpha_c0_offset = 0;
2075     h->slice_beta_offset = 0;
2076     if( h->pps.deblocking_filter_parameters_present ) {
2077         tmp= get_ue_golomb_31(&s->gb);
2078         if(tmp > 2){
2079             av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2080             return -1;
2081         }
2082         h->deblocking_filter= tmp;
2083         if(h->deblocking_filter < 2)
2084             h->deblocking_filter^= 1; // 1<->0
2085
2086         if( h->deblocking_filter ) {
2087             h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
2088             h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
2089         }
2090     }
2091
2092     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
2093        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
2094        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
2095        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2096         h->deblocking_filter= 0;
2097
2098     if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2099         if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2100             /* Cheat slightly for speed:
2101                Do not bother to deblock across slices. */
2102             h->deblocking_filter = 2;
2103         } else {
2104             h0->max_contexts = 1;
2105             if(!h0->single_decode_warning) {
2106                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2107                 h0->single_decode_warning = 1;
2108             }
2109             if(h != h0)
2110                 return 1; // deblocking switched inside frame
2111         }
2112     }
2113     h->qp_thresh= 15 - 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]);
2114
2115 #if 0 //FMO
2116     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2117         slice_group_change_cycle= get_bits(&s->gb, ?);
2118 #endif
2119
2120     h0->last_slice_type = slice_type;
2121     h->slice_num = ++h0->current_slice;
2122     if(h->slice_num >= MAX_SLICES){
2123         av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2124     }
2125
2126     for(j=0; j<2; j++){
2127         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2128         ref2frm[0]=
2129         ref2frm[1]= -1;
2130         for(i=0; i<16; i++)
2131             ref2frm[i+2]= 4*h->ref_list[j][i].frame_num
2132                           +(h->ref_list[j][i].reference&3);
2133         ref2frm[18+0]=
2134         ref2frm[18+1]= -1;
2135         for(i=16; i<48; i++)
2136             ref2frm[i+4]= 4*h->ref_list[j][i].frame_num
2137                           +(h->ref_list[j][i].reference&3);
2138     }
2139
2140     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
2141     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2142
2143     s->avctx->refs= h->sps.ref_frame_count;
2144
2145     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2146         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",
2147                h->slice_num,
2148                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2149                first_mb_in_slice,
2150                av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2151                pps_id, h->frame_num,
2152                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2153                h->ref_count[0], h->ref_count[1],
2154                s->qscale,
2155                h->deblocking_filter, h->slice_alpha_c0_offset/2, h->slice_beta_offset/2,
2156                h->use_weight,
2157                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2158                h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2159                );
2160     }
2161
2162     return 0;
2163 }
2164
2165 int ff_h264_get_slice_type(H264Context *h)
2166 {
2167     switch (h->slice_type) {
2168     case FF_P_TYPE:  return 0;
2169     case FF_B_TYPE:  return 1;
2170     case FF_I_TYPE:  return 2;
2171     case FF_SP_TYPE: return 3;
2172     case FF_SI_TYPE: return 4;
2173     default:         return -1;
2174     }
2175 }
2176
2177 static int decode_slice(struct AVCodecContext *avctx, void *arg){
2178     H264Context *h = *(void**)arg;
2179     MpegEncContext * const s = &h->s;
2180     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2181
2182     s->mb_skip_run= -1;
2183
2184     h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
2185                     (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
2186
2187     if( h->pps.cabac ) {
2188         /* realign */
2189         align_get_bits( &s->gb );
2190
2191         /* init cabac */
2192         ff_init_cabac_states( &h->cabac);
2193         ff_init_cabac_decoder( &h->cabac,
2194                                s->gb.buffer + get_bits_count(&s->gb)/8,
2195                                (get_bits_left(&s->gb) + 7)/8);
2196
2197         ff_h264_init_cabac_states(h);
2198
2199         for(;;){
2200 //START_TIMER
2201             int ret = ff_h264_decode_mb_cabac(h);
2202             int eos;
2203 //STOP_TIMER("decode_mb_cabac")
2204
2205             if(ret>=0) ff_h264_hl_decode_mb(h);
2206
2207             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
2208                 s->mb_y++;
2209
2210                 ret = ff_h264_decode_mb_cabac(h);
2211
2212                 if(ret>=0) ff_h264_hl_decode_mb(h);
2213                 s->mb_y--;
2214             }
2215             eos = get_cabac_terminate( &h->cabac );
2216
2217             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2218                 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);
2219                 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);
2220                 return -1;
2221             }
2222
2223             if( ++s->mb_x >= s->mb_width ) {
2224                 s->mb_x = 0;
2225                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2226                 ++s->mb_y;
2227                 if(FIELD_OR_MBAFF_PICTURE) {
2228                     ++s->mb_y;
2229                 }
2230             }
2231
2232             if( eos || s->mb_y >= s->mb_height ) {
2233                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2234                 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);
2235                 return 0;
2236             }
2237         }
2238
2239     } else {
2240         for(;;){
2241             int ret = ff_h264_decode_mb_cavlc(h);
2242
2243             if(ret>=0) ff_h264_hl_decode_mb(h);
2244
2245             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
2246                 s->mb_y++;
2247                 ret = ff_h264_decode_mb_cavlc(h);
2248
2249                 if(ret>=0) ff_h264_hl_decode_mb(h);
2250                 s->mb_y--;
2251             }
2252
2253             if(ret<0){
2254                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2255                 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);
2256
2257                 return -1;
2258             }
2259
2260             if(++s->mb_x >= s->mb_width){
2261                 s->mb_x=0;
2262                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2263                 ++s->mb_y;
2264                 if(FIELD_OR_MBAFF_PICTURE) {
2265                     ++s->mb_y;
2266                 }
2267                 if(s->mb_y >= s->mb_height){
2268                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2269
2270                     if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
2271                         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);
2272
2273                         return 0;
2274                     }else{
2275                         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);
2276
2277                         return -1;
2278                     }
2279                 }
2280             }
2281
2282             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
2283                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2284                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2285                     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);
2286
2287                     return 0;
2288                 }else{
2289                     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);
2290
2291                     return -1;
2292                 }
2293             }
2294         }
2295     }
2296
2297 #if 0
2298     for(;s->mb_y < s->mb_height; s->mb_y++){
2299         for(;s->mb_x < s->mb_width; s->mb_x++){
2300             int ret= decode_mb(h);
2301
2302             ff_h264_hl_decode_mb(h);
2303
2304             if(ret<0){
2305                 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2306                 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);
2307
2308                 return -1;
2309             }
2310
2311             if(++s->mb_x >= s->mb_width){
2312                 s->mb_x=0;
2313                 if(++s->mb_y >= s->mb_height){
2314                     if(get_bits_count(s->gb) == s->gb.size_in_bits){
2315                         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);
2316
2317                         return 0;
2318                     }else{
2319                         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);
2320
2321                         return -1;
2322                     }
2323                 }
2324             }
2325
2326             if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2327                 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2328                     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);
2329
2330                     return 0;
2331                 }else{
2332                     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);
2333
2334                     return -1;
2335                 }
2336             }
2337         }
2338         s->mb_x=0;
2339         ff_draw_horiz_band(s, 16*s->mb_y, 16);
2340     }
2341 #endif
2342     return -1; //not reached
2343 }
2344
2345 /**
2346  * Call decode_slice() for each context.
2347  *
2348  * @param h h264 master context
2349  * @param context_count number of contexts to execute
2350  */
2351 static void execute_decode_slices(H264Context *h, int context_count){
2352     MpegEncContext * const s = &h->s;
2353     AVCodecContext * const avctx= s->avctx;
2354     H264Context *hx;
2355     int i;
2356
2357     if (s->avctx->hwaccel)
2358         return;
2359     if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2360         return;
2361     if(context_count == 1) {
2362         decode_slice(avctx, &h);
2363     } else {
2364         for(i = 1; i < context_count; i++) {
2365             hx = h->thread_context[i];
2366             hx->s.error_recognition = avctx->error_recognition;
2367             hx->s.error_count = 0;
2368         }
2369
2370         avctx->execute(avctx, (void *)decode_slice,
2371                        h->thread_context, NULL, context_count, sizeof(void*));
2372
2373         /* pull back stuff from slices to master context */
2374         hx = h->thread_context[context_count - 1];
2375         s->mb_x = hx->s.mb_x;
2376         s->mb_y = hx->s.mb_y;
2377         s->dropable = hx->s.dropable;
2378         s->picture_structure = hx->s.picture_structure;
2379         for(i = 1; i < context_count; i++)
2380             h->s.error_count += h->thread_context[i]->s.error_count;
2381     }
2382 }
2383
2384
2385 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
2386     MpegEncContext * const s = &h->s;
2387     AVCodecContext * const avctx= s->avctx;
2388     int buf_index=0;
2389     H264Context *hx; ///< thread context
2390     int context_count = 0;
2391     int next_avc= h->is_avc ? 0 : buf_size;
2392
2393     h->max_contexts = avctx->thread_count;
2394 #if 0
2395     int i;
2396     for(i=0; i<50; i++){
2397         av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
2398     }
2399 #endif
2400     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
2401         h->current_slice = 0;
2402         if (!s->first_field)
2403             s->current_picture_ptr= NULL;
2404         ff_h264_reset_sei(h);
2405     }
2406
2407     for(;;){
2408         int consumed;
2409         int dst_length;
2410         int bit_length;
2411         const uint8_t *ptr;
2412         int i, nalsize = 0;
2413         int err;
2414
2415         if(buf_index >= next_avc) {
2416             if(buf_index >= buf_size) break;
2417             nalsize = 0;
2418             for(i = 0; i < h->nal_length_size; i++)
2419                 nalsize = (nalsize << 8) | buf[buf_index++];
2420             if(nalsize <= 1 || nalsize > buf_size - buf_index){
2421                 if(nalsize == 1){
2422                     buf_index++;
2423                     continue;
2424                 }else{
2425                     av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2426                     break;
2427                 }
2428             }
2429             next_avc= buf_index + nalsize;
2430         } else {
2431             // start code prefix search
2432             for(; buf_index + 3 < next_avc; buf_index++){
2433                 // This should always succeed in the first iteration.
2434                 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2435                     break;
2436             }
2437
2438             if(buf_index+3 >= buf_size) break;
2439
2440             buf_index+=3;
2441             if(buf_index >= next_avc) continue;
2442         }
2443
2444         hx = h->thread_context[context_count];
2445
2446         ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
2447         if (ptr==NULL || dst_length < 0){
2448             return -1;
2449         }
2450         while(ptr[dst_length - 1] == 0 && dst_length > 0)
2451             dst_length--;
2452         bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
2453
2454         if(s->avctx->debug&FF_DEBUG_STARTCODE){
2455             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);
2456         }
2457
2458         if (h->is_avc && (nalsize != consumed) && nalsize){
2459             av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
2460         }
2461
2462         buf_index += consumed;
2463
2464         if(  (s->hurry_up == 1 && h->nal_ref_idc  == 0) //FIXME do not discard SEI id
2465            ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0))
2466             continue;
2467
2468       again:
2469         err = 0;
2470         switch(hx->nal_unit_type){
2471         case NAL_IDR_SLICE:
2472             if (h->nal_unit_type != NAL_IDR_SLICE) {
2473                 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2474                 return -1;
2475             }
2476             idr(h); //FIXME ensure we don't loose some frames if there is reordering
2477         case NAL_SLICE:
2478             init_get_bits(&hx->s.gb, ptr, bit_length);
2479             hx->intra_gb_ptr=
2480             hx->inter_gb_ptr= &hx->s.gb;
2481             hx->s.data_partitioning = 0;
2482
2483             if((err = decode_slice_header(hx, h)))
2484                break;
2485
2486             if (s->avctx->hwaccel && h->current_slice == 1) {
2487                 if (s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
2488                     return -1;
2489             }
2490
2491             s->current_picture_ptr->key_frame |=
2492                     (hx->nal_unit_type == NAL_IDR_SLICE) ||
2493                     (h->sei_recovery_frame_cnt >= 0);
2494             if(hx->redundant_pic_count==0 && hx->s.hurry_up < 5
2495                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2496                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2497                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2498                && avctx->skip_frame < AVDISCARD_ALL){
2499                 if(avctx->hwaccel) {
2500                     if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
2501                         return -1;
2502                 }else
2503                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
2504                     static const uint8_t start_code[] = {0x00, 0x00, 0x01};
2505                     ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
2506                     ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
2507                 }else
2508                     context_count++;
2509             }
2510             break;
2511         case NAL_DPA:
2512             init_get_bits(&hx->s.gb, ptr, bit_length);
2513             hx->intra_gb_ptr=
2514             hx->inter_gb_ptr= NULL;
2515
2516             if ((err = decode_slice_header(hx, h)) < 0)
2517                 break;
2518
2519             hx->s.data_partitioning = 1;
2520
2521             break;
2522         case NAL_DPB:
2523             init_get_bits(&hx->intra_gb, ptr, bit_length);
2524             hx->intra_gb_ptr= &hx->intra_gb;
2525             break;
2526         case NAL_DPC:
2527             init_get_bits(&hx->inter_gb, ptr, bit_length);
2528             hx->inter_gb_ptr= &hx->inter_gb;
2529
2530             if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
2531                && s->context_initialized
2532                && s->hurry_up < 5
2533                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2534                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2535                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2536                && avctx->skip_frame < AVDISCARD_ALL)
2537                 context_count++;
2538             break;
2539         case NAL_SEI:
2540             init_get_bits(&s->gb, ptr, bit_length);
2541             ff_h264_decode_sei(h);
2542             break;
2543         case NAL_SPS:
2544             init_get_bits(&s->gb, ptr, bit_length);
2545             ff_h264_decode_seq_parameter_set(h);
2546
2547             if(s->flags& CODEC_FLAG_LOW_DELAY)
2548                 s->low_delay=1;
2549
2550             if(avctx->has_b_frames < 2)
2551                 avctx->has_b_frames= !s->low_delay;
2552             break;
2553         case NAL_PPS:
2554             init_get_bits(&s->gb, ptr, bit_length);
2555
2556             ff_h264_decode_picture_parameter_set(h, bit_length);
2557
2558             break;
2559         case NAL_AUD:
2560         case NAL_END_SEQUENCE:
2561         case NAL_END_STREAM:
2562         case NAL_FILLER_DATA:
2563         case NAL_SPS_EXT:
2564         case NAL_AUXILIARY_SLICE:
2565             break;
2566         default:
2567             av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
2568         }
2569
2570         if(context_count == h->max_contexts) {
2571             execute_decode_slices(h, context_count);
2572             context_count = 0;
2573         }
2574
2575         if (err < 0)
2576             av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
2577         else if(err == 1) {
2578             /* Slice could not be decoded in parallel mode, copy down
2579              * NAL unit stuff to context 0 and restart. Note that
2580              * rbsp_buffer is not transferred, but since we no longer
2581              * run in parallel mode this should not be an issue. */
2582             h->nal_unit_type = hx->nal_unit_type;
2583             h->nal_ref_idc   = hx->nal_ref_idc;
2584             hx = h;
2585             goto again;
2586         }
2587     }
2588     if(context_count)
2589         execute_decode_slices(h, context_count);
2590     return buf_index;
2591 }
2592
2593 /**
2594  * returns the number of bytes consumed for building the current frame
2595  */
2596 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
2597         if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
2598         if(pos+10>buf_size) pos=buf_size; // oops ;)
2599
2600         return pos;
2601 }
2602
2603 static int decode_frame(AVCodecContext *avctx,
2604                              void *data, int *data_size,
2605                              AVPacket *avpkt)
2606 {
2607     const uint8_t *buf = avpkt->data;
2608     int buf_size = avpkt->size;
2609     H264Context *h = avctx->priv_data;
2610     MpegEncContext *s = &h->s;
2611     AVFrame *pict = data;
2612     int buf_index;
2613
2614     s->flags= avctx->flags;
2615     s->flags2= avctx->flags2;
2616
2617    /* end of stream, output what is still in the buffers */
2618     if (buf_size == 0) {
2619         Picture *out;
2620         int i, out_idx;
2621
2622 //FIXME factorize this with the output code below
2623         out = h->delayed_pic[0];
2624         out_idx = 0;
2625         for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2626             if(h->delayed_pic[i]->poc < out->poc){
2627                 out = h->delayed_pic[i];
2628                 out_idx = i;
2629             }
2630
2631         for(i=out_idx; h->delayed_pic[i]; i++)
2632             h->delayed_pic[i] = h->delayed_pic[i+1];
2633
2634         if(out){
2635             *data_size = sizeof(AVFrame);
2636             *pict= *(AVFrame*)out;
2637         }
2638
2639         return 0;
2640     }
2641
2642     if(h->is_avc && !h->got_avcC) {
2643         int i, cnt, nalsize;
2644         unsigned char *p = avctx->extradata;
2645         if(avctx->extradata_size < 7) {
2646             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
2647             return -1;
2648         }
2649         if(*p != 1) {
2650             av_log(avctx, AV_LOG_ERROR, "Unknown avcC version %d\n", *p);
2651             return -1;
2652         }
2653         /* sps and pps in the avcC always have length coded with 2 bytes,
2654            so put a fake nal_length_size = 2 while parsing them */
2655         h->nal_length_size = 2;
2656         // Decode sps from avcC
2657         cnt = *(p+5) & 0x1f; // Number of sps
2658         p += 6;
2659         for (i = 0; i < cnt; i++) {
2660             nalsize = AV_RB16(p) + 2;
2661             if(decode_nal_units(h, p, nalsize) < 0) {
2662                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
2663                 return -1;
2664             }
2665             p += nalsize;
2666         }
2667         // Decode pps from avcC
2668         cnt = *(p++); // Number of pps
2669         for (i = 0; i < cnt; i++) {
2670             nalsize = AV_RB16(p) + 2;
2671             if(decode_nal_units(h, p, nalsize)  != nalsize) {
2672                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
2673                 return -1;
2674             }
2675             p += nalsize;
2676         }
2677         // Now store right nal length size, that will be use to parse all other nals
2678         h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
2679         // Do not reparse avcC
2680         h->got_avcC = 1;
2681     }
2682
2683     if(!h->got_avcC && !h->is_avc && s->avctx->extradata_size){
2684         if(decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) < 0)
2685             return -1;
2686         h->got_avcC = 1;
2687     }
2688
2689     buf_index=decode_nal_units(h, buf, buf_size);
2690     if(buf_index < 0)
2691         return -1;
2692
2693     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
2694         if (avctx->skip_frame >= AVDISCARD_NONREF || s->hurry_up) return 0;
2695         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
2696         return -1;
2697     }
2698
2699     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
2700         Picture *out = s->current_picture_ptr;
2701         Picture *cur = s->current_picture_ptr;
2702         int i, pics, out_of_order, out_idx;
2703
2704         field_end(h);
2705
2706         if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
2707             /* Wait for second field. */
2708             *data_size = 0;
2709
2710         } else {
2711             cur->interlaced_frame = 0;
2712             cur->repeat_pict = 0;
2713
2714             /* Signal interlacing information externally. */
2715             /* Prioritize picture timing SEI information over used decoding process if it exists. */
2716
2717             if(h->sps.pic_struct_present_flag){
2718                 switch (h->sei_pic_struct)
2719                 {
2720                 case SEI_PIC_STRUCT_FRAME:
2721                     break;
2722                 case SEI_PIC_STRUCT_TOP_FIELD:
2723                 case SEI_PIC_STRUCT_BOTTOM_FIELD:
2724                     cur->interlaced_frame = 1;
2725                     break;
2726                 case SEI_PIC_STRUCT_TOP_BOTTOM:
2727                 case SEI_PIC_STRUCT_BOTTOM_TOP:
2728                     if (FIELD_OR_MBAFF_PICTURE)
2729                         cur->interlaced_frame = 1;
2730                     else
2731                         // try to flag soft telecine progressive
2732                         cur->interlaced_frame = h->prev_interlaced_frame;
2733                     break;
2734                 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
2735                 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
2736                     // Signal the possibility of telecined film externally (pic_struct 5,6)
2737                     // From these hints, let the applications decide if they apply deinterlacing.
2738                     cur->repeat_pict = 1;
2739                     break;
2740                 case SEI_PIC_STRUCT_FRAME_DOUBLING:
2741                     // Force progressive here, as doubling interlaced frame is a bad idea.
2742                     cur->repeat_pict = 2;
2743                     break;
2744                 case SEI_PIC_STRUCT_FRAME_TRIPLING:
2745                     cur->repeat_pict = 4;
2746                     break;
2747                 }
2748
2749                 if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
2750                     cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
2751             }else{
2752                 /* Derive interlacing flag from used decoding process. */
2753                 cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
2754             }
2755             h->prev_interlaced_frame = cur->interlaced_frame;
2756
2757             if (cur->field_poc[0] != cur->field_poc[1]){
2758                 /* Derive top_field_first from field pocs. */
2759                 cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
2760             }else{
2761                 if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
2762                     /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
2763                     if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
2764                       || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
2765                         cur->top_field_first = 1;
2766                     else
2767                         cur->top_field_first = 0;
2768                 }else{
2769                     /* Most likely progressive */
2770                     cur->top_field_first = 0;
2771                 }
2772             }
2773
2774         //FIXME do something with unavailable reference frames
2775
2776             /* Sort B-frames into display order */
2777
2778             if(h->sps.bitstream_restriction_flag
2779                && s->avctx->has_b_frames < h->sps.num_reorder_frames){
2780                 s->avctx->has_b_frames = h->sps.num_reorder_frames;
2781                 s->low_delay = 0;
2782             }
2783
2784             if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
2785                && !h->sps.bitstream_restriction_flag){
2786                 s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
2787                 s->low_delay= 0;
2788             }
2789
2790             pics = 0;
2791             while(h->delayed_pic[pics]) pics++;
2792
2793             assert(pics <= MAX_DELAYED_PIC_COUNT);
2794
2795             h->delayed_pic[pics++] = cur;
2796             if(cur->reference == 0)
2797                 cur->reference = DELAYED_PIC_REF;
2798
2799             out = h->delayed_pic[0];
2800             out_idx = 0;
2801             for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2802                 if(h->delayed_pic[i]->poc < out->poc){
2803                     out = h->delayed_pic[i];
2804                     out_idx = i;
2805                 }
2806             if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
2807                 h->outputed_poc= INT_MIN;
2808             out_of_order = out->poc < h->outputed_poc;
2809
2810             if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
2811                 { }
2812             else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
2813                || (s->low_delay &&
2814                 ((h->outputed_poc != INT_MIN && out->poc > h->outputed_poc + 2)
2815                  || cur->pict_type == FF_B_TYPE)))
2816             {
2817                 s->low_delay = 0;
2818                 s->avctx->has_b_frames++;
2819             }
2820
2821             if(out_of_order || pics > s->avctx->has_b_frames){
2822                 out->reference &= ~DELAYED_PIC_REF;
2823                 for(i=out_idx; h->delayed_pic[i]; i++)
2824                     h->delayed_pic[i] = h->delayed_pic[i+1];
2825             }
2826             if(!out_of_order && pics > s->avctx->has_b_frames){
2827                 *data_size = sizeof(AVFrame);
2828
2829                 if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
2830                     h->outputed_poc = INT_MIN;
2831                 } else
2832                     h->outputed_poc = out->poc;
2833                 *pict= *(AVFrame*)out;
2834             }else{
2835                 av_log(avctx, AV_LOG_DEBUG, "no picture\n");
2836             }
2837         }
2838     }
2839
2840     assert(pict->data[0] || !*data_size);
2841     ff_print_debug_info(s, pict);
2842 //printf("out %d\n", (int)pict->data[0]);
2843
2844     return get_consumed_bytes(s, buf_index, buf_size);
2845 }
2846 #if 0
2847 static inline void fill_mb_avail(H264Context *h){
2848     MpegEncContext * const s = &h->s;
2849     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2850
2851     if(s->mb_y){
2852         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
2853         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
2854         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
2855     }else{
2856         h->mb_avail[0]=
2857         h->mb_avail[1]=
2858         h->mb_avail[2]= 0;
2859     }
2860     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
2861     h->mb_avail[4]= 1; //FIXME move out
2862     h->mb_avail[5]= 0; //FIXME move out
2863 }
2864 #endif
2865
2866 #ifdef TEST
2867 #undef printf
2868 #undef random
2869 #define COUNT 8000
2870 #define SIZE (COUNT*40)
2871 int main(void){
2872     int i;
2873     uint8_t temp[SIZE];
2874     PutBitContext pb;
2875     GetBitContext gb;
2876 //    int int_temp[10000];
2877     DSPContext dsp;
2878     AVCodecContext avctx;
2879
2880     dsputil_init(&dsp, &avctx);
2881
2882     init_put_bits(&pb, temp, SIZE);
2883     printf("testing unsigned exp golomb\n");
2884     for(i=0; i<COUNT; i++){
2885         START_TIMER
2886         set_ue_golomb(&pb, i);
2887         STOP_TIMER("set_ue_golomb");
2888     }
2889     flush_put_bits(&pb);
2890
2891     init_get_bits(&gb, temp, 8*SIZE);
2892     for(i=0; i<COUNT; i++){
2893         int j, s;
2894
2895         s= show_bits(&gb, 24);
2896
2897         START_TIMER
2898         j= get_ue_golomb(&gb);
2899         if(j != i){
2900             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
2901 //            return -1;
2902         }
2903         STOP_TIMER("get_ue_golomb");
2904     }
2905
2906
2907     init_put_bits(&pb, temp, SIZE);
2908     printf("testing signed exp golomb\n");
2909     for(i=0; i<COUNT; i++){
2910         START_TIMER
2911         set_se_golomb(&pb, i - COUNT/2);
2912         STOP_TIMER("set_se_golomb");
2913     }
2914     flush_put_bits(&pb);
2915
2916     init_get_bits(&gb, temp, 8*SIZE);
2917     for(i=0; i<COUNT; i++){
2918         int j, s;
2919
2920         s= show_bits(&gb, 24);
2921
2922         START_TIMER
2923         j= get_se_golomb(&gb);
2924         if(j != i - COUNT/2){
2925             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
2926 //            return -1;
2927         }
2928         STOP_TIMER("get_se_golomb");
2929     }
2930
2931 #if 0
2932     printf("testing 4x4 (I)DCT\n");
2933
2934     DCTELEM block[16];
2935     uint8_t src[16], ref[16];
2936     uint64_t error= 0, max_error=0;
2937
2938     for(i=0; i<COUNT; i++){
2939         int j;
2940 //        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
2941         for(j=0; j<16; j++){
2942             ref[j]= random()%255;
2943             src[j]= random()%255;
2944         }
2945
2946         h264_diff_dct_c(block, src, ref, 4);
2947
2948         //normalize
2949         for(j=0; j<16; j++){
2950 //            printf("%d ", block[j]);
2951             block[j]= block[j]*4;
2952             if(j&1) block[j]= (block[j]*4 + 2)/5;
2953             if(j&4) block[j]= (block[j]*4 + 2)/5;
2954         }
2955 //        printf("\n");
2956
2957         s->dsp.h264_idct_add(ref, block, 4);
2958 /*        for(j=0; j<16; j++){
2959             printf("%d ", ref[j]);
2960         }
2961         printf("\n");*/
2962
2963         for(j=0; j<16; j++){
2964             int diff= FFABS(src[j] - ref[j]);
2965
2966             error+= diff*diff;
2967             max_error= FFMAX(max_error, diff);
2968         }
2969     }
2970     printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
2971     printf("testing quantizer\n");
2972     for(qp=0; qp<52; qp++){
2973         for(i=0; i<16; i++)
2974             src1_block[i]= src2_block[i]= random()%255;
2975
2976     }
2977     printf("Testing NAL layer\n");
2978
2979     uint8_t bitstream[COUNT];
2980     uint8_t nal[COUNT*2];
2981     H264Context h;
2982     memset(&h, 0, sizeof(H264Context));
2983
2984     for(i=0; i<COUNT; i++){
2985         int zeros= i;
2986         int nal_length;
2987         int consumed;
2988         int out_length;
2989         uint8_t *out;
2990         int j;
2991
2992         for(j=0; j<COUNT; j++){
2993             bitstream[j]= (random() % 255) + 1;
2994         }
2995
2996         for(j=0; j<zeros; j++){
2997             int pos= random() % COUNT;
2998             while(bitstream[pos] == 0){
2999                 pos++;
3000                 pos %= COUNT;
3001             }
3002             bitstream[pos]=0;
3003         }
3004
3005         START_TIMER
3006
3007         nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
3008         if(nal_length<0){
3009             printf("encoding failed\n");
3010             return -1;
3011         }
3012
3013         out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
3014
3015         STOP_TIMER("NAL")
3016
3017         if(out_length != COUNT){
3018             printf("incorrect length %d %d\n", out_length, COUNT);
3019             return -1;
3020         }
3021
3022         if(consumed != nal_length){
3023             printf("incorrect consumed length %d %d\n", nal_length, consumed);
3024             return -1;
3025         }
3026
3027         if(memcmp(bitstream, out, COUNT)){
3028             printf("mismatch\n");
3029             return -1;
3030         }
3031     }
3032 #endif
3033
3034     printf("Testing RBSP\n");
3035
3036
3037     return 0;
3038 }
3039 #endif /* TEST */
3040
3041
3042 av_cold void ff_h264_free_context(H264Context *h)
3043 {
3044     int i;
3045
3046     free_tables(h); //FIXME cleanup init stuff perhaps
3047
3048     for(i = 0; i < MAX_SPS_COUNT; i++)
3049         av_freep(h->sps_buffers + i);
3050
3051     for(i = 0; i < MAX_PPS_COUNT; i++)
3052         av_freep(h->pps_buffers + i);
3053 }
3054
3055 av_cold int ff_h264_decode_end(AVCodecContext *avctx)
3056 {
3057     H264Context *h = avctx->priv_data;
3058     MpegEncContext *s = &h->s;
3059
3060     ff_h264_free_context(h);
3061
3062     MPV_common_end(s);
3063
3064 //    memset(h, 0, sizeof(H264Context));
3065
3066     return 0;
3067 }
3068
3069
3070 AVCodec h264_decoder = {
3071     "h264",
3072     CODEC_TYPE_VIDEO,
3073     CODEC_ID_H264,
3074     sizeof(H264Context),
3075     ff_h264_decode_init,
3076     NULL,
3077     ff_h264_decode_end,
3078     decode_frame,
3079     /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3080     .flush= flush_dpb,
3081     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
3082     .pix_fmts= ff_hwaccel_pixfmt_list_420,
3083 };
3084
3085 #if CONFIG_H264_VDPAU_DECODER
3086 AVCodec h264_vdpau_decoder = {
3087     "h264_vdpau",
3088     CODEC_TYPE_VIDEO,
3089     CODEC_ID_H264,
3090     sizeof(H264Context),
3091     ff_h264_decode_init,
3092     NULL,
3093     ff_h264_decode_end,
3094     decode_frame,
3095     CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3096     .flush= flush_dpb,
3097     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
3098     .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
3099 };
3100 #endif