git.sesse.net Git - ffmpeg/blob - libavcodec/h264.c

   1 /*
   2  * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
   3  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
   4  *
   5  * This file is part of FFmpeg.
   6  *
   7  * FFmpeg is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2.1 of the License, or (at your option) any later version.
  11  *
  12  * FFmpeg is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with FFmpeg; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 /**
  23  * @file libavcodec/h264.c
  24  * H.264 / AVC / MPEG4 part10 codec.
  25  * @author Michael Niedermayer <michaelni@gmx.at>
  26  */
  27
  28 #include "internal.h"
  29 #include "dsputil.h"
  30 #include "avcodec.h"
  31 #include "mpegvideo.h"
  32 #include "h264.h"
  33 #include "h264data.h"
  34 #include "h264_mvpred.h"
  35 #include "h264_parser.h"
  36 #include "golomb.h"
  37 #include "mathops.h"
  38 #include "rectangle.h"
  39 #include "vdpau_internal.h"
  40
  41 #include "cabac.h"
  42
  43 //#undef NDEBUG
  44 #include <assert.h>
  45
  46 static const uint8_t rem6[52]={
  47 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
  48 };
  49
  50 static const uint8_t div6[52]={
  51 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8,
  52 };
  53
  54 void ff_h264_write_back_intra_pred_mode(H264Context *h){
  55     const int mb_xy= h->mb_xy;
  56
  57     h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
  58     h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
  59     h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
  60     h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
  61     h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
  62     h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
  63     h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
  64 }
  65
  66 /**
  67  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
  68  */
  69 int ff_h264_check_intra4x4_pred_mode(H264Context *h){
  70     MpegEncContext * const s = &h->s;
  71     static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
  72     static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
  73     int i;
  74
  75     if(!(h->top_samples_available&0x8000)){
  76         for(i=0; i<4; i++){
  77             int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
  78             if(status<0){
  79                 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
  80                 return -1;
  81             } else if(status){
  82                 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
  83             }
  84         }
  85     }
  86
  87     if((h->left_samples_available&0x8888)!=0x8888){
  88         static const int mask[4]={0x8000,0x2000,0x80,0x20};
  89         for(i=0; i<4; i++){
  90             if(!(h->left_samples_available&mask[i])){
  91                 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
  92                 if(status<0){
  93                     av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
  94                     return -1;
  95                 } else if(status){
  96                     h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
  97                 }
  98             }
  99         }
 100     }
 101
 102     return 0;
 103 } //FIXME cleanup like ff_h264_check_intra_pred_mode
 104
 105 /**
 106  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
 107  */
 108 int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
 109     MpegEncContext * const s = &h->s;
 110     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
 111     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
 112
 113     if(mode > 6U) {
 114         av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
 115         return -1;
 116     }
 117
 118     if(!(h->top_samples_available&0x8000)){
 119         mode= top[ mode ];
 120         if(mode<0){
 121             av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
 122             return -1;
 123         }
 124     }
 125
 126     if((h->left_samples_available&0x8080) != 0x8080){
 127         mode= left[ mode ];
 128         if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
 129             mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
 130         }
 131         if(mode<0){
 132             av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
 133             return -1;
 134         }
 135     }
 136
 137     return mode;
 138 }
 139
 140 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
 141     int i, si, di;
 142     uint8_t *dst;
 143     int bufidx;
 144
 145 //    src[0]&0x80;                //forbidden bit
 146     h->nal_ref_idc= src[0]>>5;
 147     h->nal_unit_type= src[0]&0x1F;
 148
 149     src++; length--;
 150 #if 0
 151     for(i=0; i<length; i++)
 152         printf("%2X ", src[i]);
 153 #endif
 154
 155 #if HAVE_FAST_UNALIGNED
 156 # if HAVE_FAST_64BIT
 157 #   define RS 7
 158     for(i=0; i+1<length; i+=9){
 159         if(!((~*(const uint64_t*)(src+i) & (*(const uint64_t*)(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
 160 # else
 161 #   define RS 3
 162     for(i=0; i+1<length; i+=5){
 163         if(!((~*(const uint32_t*)(src+i) & (*(const uint32_t*)(src+i) - 0x01000101U)) & 0x80008080U))
 164 # endif
 165             continue;
 166         if(i>0 && !src[i]) i--;
 167         while(src[i]) i++;
 168 #else
 169 #   define RS 0
 170     for(i=0; i+1<length; i+=2){
 171         if(src[i]) continue;
 172         if(i>0 && src[i-1]==0) i--;
 173 #endif
 174         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
 175             if(src[i+2]!=3){
 176                 /* startcode, so we must be past the end */
 177                 length=i;
 178             }
 179             break;
 180         }
 181         i-= RS;
 182     }
 183
 184     if(i>=length-1){ //no escaped 0
 185         *dst_length= length;
 186         *consumed= length+1; //+1 for the header
 187         return src;
 188     }
 189
 190     bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
 191     av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
 192     dst= h->rbsp_buffer[bufidx];
 193
 194     if (dst == NULL){
 195         return NULL;
 196     }
 197
 198 //printf("decoding esc\n");
 199     memcpy(dst, src, i);
 200     si=di=i;
 201     while(si+2<length){
 202         //remove escapes (very rare 1:2^22)
 203         if(src[si+2]>3){
 204             dst[di++]= src[si++];
 205             dst[di++]= src[si++];
 206         }else if(src[si]==0 && src[si+1]==0){
 207             if(src[si+2]==3){ //escape
 208                 dst[di++]= 0;
 209                 dst[di++]= 0;
 210                 si+=3;
 211                 continue;
 212             }else //next start code
 213                 goto nsc;
 214         }
 215
 216         dst[di++]= src[si++];
 217     }
 218     while(si<length)
 219         dst[di++]= src[si++];
 220 nsc:
 221
 222     memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
 223
 224     *dst_length= di;
 225     *consumed= si + 1;//+1 for the header
 226 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
 227     return dst;
 228 }
 229
 230 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
 231     int v= *src;
 232     int r;
 233
 234     tprintf(h->s.avctx, "rbsp trailing %X\n", v);
 235
 236     for(r=1; r<9; r++){
 237         if(v&1) return r;
 238         v>>=1;
 239     }
 240     return 0;
 241 }
 242
 243 /**
 244  * IDCT transforms the 16 dc values and dequantizes them.
 245  * @param qp quantization parameter
 246  */
 247 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
 248 #define stride 16
 249     int i;
 250     int temp[16]; //FIXME check if this is a good idea
 251     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
 252     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
 253
 254 //memset(block, 64, 2*256);
 255 //return;
 256     for(i=0; i<4; i++){
 257         const int offset= y_offset[i];
 258         const int z0= block[offset+stride*0] + block[offset+stride*4];
 259         const int z1= block[offset+stride*0] - block[offset+stride*4];
 260         const int z2= block[offset+stride*1] - block[offset+stride*5];
 261         const int z3= block[offset+stride*1] + block[offset+stride*5];
 262
 263         temp[4*i+0]= z0+z3;
 264         temp[4*i+1]= z1+z2;
 265         temp[4*i+2]= z1-z2;
 266         temp[4*i+3]= z0-z3;
 267     }
 268
 269     for(i=0; i<4; i++){
 270         const int offset= x_offset[i];
 271         const int z0= temp[4*0+i] + temp[4*2+i];
 272         const int z1= temp[4*0+i] - temp[4*2+i];
 273         const int z2= temp[4*1+i] - temp[4*3+i];
 274         const int z3= temp[4*1+i] + temp[4*3+i];
 275
 276         block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
 277         block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
 278         block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
 279         block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
 280     }
 281 }
 282
 283 #if 0
 284 /**
 285  * DCT transforms the 16 dc values.
 286  * @param qp quantization parameter ??? FIXME
 287  */
 288 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
 289 //    const int qmul= dequant_coeff[qp][0];
 290     int i;
 291     int temp[16]; //FIXME check if this is a good idea
 292     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
 293     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
 294
 295     for(i=0; i<4; i++){
 296         const int offset= y_offset[i];
 297         const int z0= block[offset+stride*0] + block[offset+stride*4];
 298         const int z1= block[offset+stride*0] - block[offset+stride*4];
 299         const int z2= block[offset+stride*1] - block[offset+stride*5];
 300         const int z3= block[offset+stride*1] + block[offset+stride*5];
 301
 302         temp[4*i+0]= z0+z3;
 303         temp[4*i+1]= z1+z2;
 304         temp[4*i+2]= z1-z2;
 305         temp[4*i+3]= z0-z3;
 306     }
 307
 308     for(i=0; i<4; i++){
 309         const int offset= x_offset[i];
 310         const int z0= temp[4*0+i] + temp[4*2+i];
 311         const int z1= temp[4*0+i] - temp[4*2+i];
 312         const int z2= temp[4*1+i] - temp[4*3+i];
 313         const int z3= temp[4*1+i] + temp[4*3+i];
 314
 315         block[stride*0 +offset]= (z0 + z3)>>1;
 316         block[stride*2 +offset]= (z1 + z2)>>1;
 317         block[stride*8 +offset]= (z1 - z2)>>1;
 318         block[stride*10+offset]= (z0 - z3)>>1;
 319     }
 320 }
 321 #endif
 322
 323 #undef xStride
 324 #undef stride
 325
 326 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
 327     const int stride= 16*2;
 328     const int xStride= 16;
 329     int a,b,c,d,e;
 330
 331     a= block[stride*0 + xStride*0];
 332     b= block[stride*0 + xStride*1];
 333     c= block[stride*1 + xStride*0];
 334     d= block[stride*1 + xStride*1];
 335
 336     e= a-b;
 337     a= a+b;
 338     b= c-d;
 339     c= c+d;
 340
 341     block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
 342     block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
 343     block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
 344     block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
 345 }
 346
 347 #if 0
 348 static void chroma_dc_dct_c(DCTELEM *block){
 349     const int stride= 16*2;
 350     const int xStride= 16;
 351     int a,b,c,d,e;
 352
 353     a= block[stride*0 + xStride*0];
 354     b= block[stride*0 + xStride*1];
 355     c= block[stride*1 + xStride*0];
 356     d= block[stride*1 + xStride*1];
 357
 358     e= a-b;
 359     a= a+b;
 360     b= c-d;
 361     c= c+d;
 362
 363     block[stride*0 + xStride*0]= (a+c);
 364     block[stride*0 + xStride*1]= (e+b);
 365     block[stride*1 + xStride*0]= (a-c);
 366     block[stride*1 + xStride*1]= (e-b);
 367 }
 368 #endif
 369
 370 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
 371                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 372                            int src_x_offset, int src_y_offset,
 373                            qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
 374     MpegEncContext * const s = &h->s;
 375     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
 376     int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
 377     const int luma_xy= (mx&3) + ((my&3)<<2);
 378     uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
 379     uint8_t * src_cb, * src_cr;
 380     int extra_width= h->emu_edge_width;
 381     int extra_height= h->emu_edge_height;
 382     int emu=0;
 383     const int full_mx= mx>>2;
 384     const int full_my= my>>2;
 385     const int pic_width  = 16*s->mb_width;
 386     const int pic_height = 16*s->mb_height >> MB_FIELD;
 387
 388     if(mx&7) extra_width -= 3;
 389     if(my&7) extra_height -= 3;
 390
 391     if(   full_mx < 0-extra_width
 392        || full_my < 0-extra_height
 393        || full_mx + 16/*FIXME*/ > pic_width + extra_width
 394        || full_my + 16/*FIXME*/ > pic_height + extra_height){
 395         ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
 396             src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
 397         emu=1;
 398     }
 399
 400     qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
 401     if(!square){
 402         qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
 403     }
 404
 405     if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
 406
 407     if(MB_FIELD){
 408         // chroma offset when predicting from a field of opposite parity
 409         my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
 410         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
 411     }
 412     src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
 413     src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
 414
 415     if(emu){
 416         ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
 417             src_cb= s->edge_emu_buffer;
 418     }
 419     chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
 420
 421     if(emu){
 422         ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
 423             src_cr= s->edge_emu_buffer;
 424     }
 425     chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
 426 }
 427
 428 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
 429                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 430                            int x_offset, int y_offset,
 431                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 432                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
 433                            int list0, int list1){
 434     MpegEncContext * const s = &h->s;
 435     qpel_mc_func *qpix_op=  qpix_put;
 436     h264_chroma_mc_func chroma_op= chroma_put;
 437
 438     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
 439     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
 440     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
 441     x_offset += 8*s->mb_x;
 442     y_offset += 8*(s->mb_y >> MB_FIELD);
 443
 444     if(list0){
 445         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
 446         mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
 447                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
 448                            qpix_op, chroma_op);
 449
 450         qpix_op=  qpix_avg;
 451         chroma_op= chroma_avg;
 452     }
 453
 454     if(list1){
 455         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
 456         mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
 457                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
 458                            qpix_op, chroma_op);
 459     }
 460 }
 461
 462 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
 463                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 464                            int x_offset, int y_offset,
 465                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 466                            h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
 467                            h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
 468                            int list0, int list1){
 469     MpegEncContext * const s = &h->s;
 470
 471     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
 472     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
 473     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
 474     x_offset += 8*s->mb_x;
 475     y_offset += 8*(s->mb_y >> MB_FIELD);
 476
 477     if(list0 && list1){
 478         /* don't optimize for luma-only case, since B-frames usually
 479          * use implicit weights => chroma too. */
 480         uint8_t *tmp_cb = s->obmc_scratchpad;
 481         uint8_t *tmp_cr = s->obmc_scratchpad + 8;
 482         uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
 483         int refn0 = h->ref_cache[0][ scan8[n] ];
 484         int refn1 = h->ref_cache[1][ scan8[n] ];
 485
 486         mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
 487                     dest_y, dest_cb, dest_cr,
 488                     x_offset, y_offset, qpix_put, chroma_put);
 489         mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
 490                     tmp_y, tmp_cb, tmp_cr,
 491                     x_offset, y_offset, qpix_put, chroma_put);
 492
 493         if(h->use_weight == 2){
 494             int weight0 = h->implicit_weight[refn0][refn1];
 495             int weight1 = 64 - weight0;
 496             luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
 497             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
 498             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
 499         }else{
 500             luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
 501                             h->luma_weight[0][refn0], h->luma_weight[1][refn1],
 502                             h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
 503             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 504                             h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
 505                             h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
 506             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 507                             h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
 508                             h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
 509         }
 510     }else{
 511         int list = list1 ? 1 : 0;
 512         int refn = h->ref_cache[list][ scan8[n] ];
 513         Picture *ref= &h->ref_list[list][refn];
 514         mc_dir_part(h, ref, n, square, chroma_height, delta, list,
 515                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
 516                     qpix_put, chroma_put);
 517
 518         luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
 519                        h->luma_weight[list][refn], h->luma_offset[list][refn]);
 520         if(h->use_weight_chroma){
 521             chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 522                              h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
 523             chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 524                              h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
 525         }
 526     }
 527 }
 528
 529 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
 530                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 531                            int x_offset, int y_offset,
 532                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 533                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
 534                            h264_weight_func *weight_op, h264_biweight_func *weight_avg,
 535                            int list0, int list1){
 536     if((h->use_weight==2 && list0 && list1
 537         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
 538        || h->use_weight==1)
 539         mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
 540                          x_offset, y_offset, qpix_put, chroma_put,
 541                          weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
 542     else
 543         mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
 544                     x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
 545 }
 546
 547 static inline void prefetch_motion(H264Context *h, int list){
 548     /* fetch pixels for estimated mv 4 macroblocks ahead
 549      * optimized for 64byte cache lines */
 550     MpegEncContext * const s = &h->s;
 551     const int refn = h->ref_cache[list][scan8[0]];
 552     if(refn >= 0){
 553         const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
 554         const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
 555         uint8_t **src= h->ref_list[list][refn].data;
 556         int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
 557         s->dsp.prefetch(src[0]+off, s->linesize, 4);
 558         off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
 559         s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
 560     }
 561 }
 562
 563 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 564                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
 565                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
 566                       h264_weight_func *weight_op, h264_biweight_func *weight_avg){
 567     MpegEncContext * const s = &h->s;
 568     const int mb_xy= h->mb_xy;
 569     const int mb_type= s->current_picture.mb_type[mb_xy];
 570
 571     assert(IS_INTER(mb_type));
 572
 573     prefetch_motion(h, 0);
 574
 575     if(IS_16X16(mb_type)){
 576         mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
 577                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
 578                 weight_op, weight_avg,
 579                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
 580     }else if(IS_16X8(mb_type)){
 581         mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
 582                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
 583                 &weight_op[1], &weight_avg[1],
 584                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
 585         mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
 586                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
 587                 &weight_op[1], &weight_avg[1],
 588                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
 589     }else if(IS_8X16(mb_type)){
 590         mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
 591                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 592                 &weight_op[2], &weight_avg[2],
 593                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
 594         mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
 595                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 596                 &weight_op[2], &weight_avg[2],
 597                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
 598     }else{
 599         int i;
 600
 601         assert(IS_8X8(mb_type));
 602
 603         for(i=0; i<4; i++){
 604             const int sub_mb_type= h->sub_mb_type[i];
 605             const int n= 4*i;
 606             int x_offset= (i&1)<<2;
 607             int y_offset= (i&2)<<1;
 608
 609             if(IS_SUB_8X8(sub_mb_type)){
 610                 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 611                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 612                     &weight_op[3], &weight_avg[3],
 613                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 614             }else if(IS_SUB_8X4(sub_mb_type)){
 615                 mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 616                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
 617                     &weight_op[4], &weight_avg[4],
 618                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 619                 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
 620                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
 621                     &weight_op[4], &weight_avg[4],
 622                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 623             }else if(IS_SUB_4X8(sub_mb_type)){
 624                 mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 625                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 626                     &weight_op[5], &weight_avg[5],
 627                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 628                 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
 629                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 630                     &weight_op[5], &weight_avg[5],
 631                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 632             }else{
 633                 int j;
 634                 assert(IS_SUB_4X4(sub_mb_type));
 635                 for(j=0; j<4; j++){
 636                     int sub_x_offset= x_offset + 2*(j&1);
 637                     int sub_y_offset= y_offset +   (j&2);
 638                     mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
 639                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 640                         &weight_op[6], &weight_avg[6],
 641                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 642                 }
 643             }
 644         }
 645     }
 646
 647     prefetch_motion(h, 1);
 648 }
 649
 650
 651 static void free_tables(H264Context *h){
 652     int i;
 653     H264Context *hx;
 654     av_freep(&h->intra4x4_pred_mode);
 655     av_freep(&h->chroma_pred_mode_table);
 656     av_freep(&h->cbp_table);
 657     av_freep(&h->mvd_table[0]);
 658     av_freep(&h->mvd_table[1]);
 659     av_freep(&h->direct_table);
 660     av_freep(&h->non_zero_count);
 661     av_freep(&h->slice_table_base);
 662     h->slice_table= NULL;
 663     av_freep(&h->list_counts);
 664
 665     av_freep(&h->mb2b_xy);
 666     av_freep(&h->mb2b8_xy);
 667
 668     for(i = 0; i < MAX_THREADS; i++) {
 669         hx = h->thread_context[i];
 670         if(!hx) continue;
 671         av_freep(&hx->top_borders[1]);
 672         av_freep(&hx->top_borders[0]);
 673         av_freep(&hx->s.obmc_scratchpad);
 674         av_freep(&hx->rbsp_buffer[1]);
 675         av_freep(&hx->rbsp_buffer[0]);
 676         hx->rbsp_buffer_size[0] = 0;
 677         hx->rbsp_buffer_size[1] = 0;
 678         if (i) av_freep(&h->thread_context[i]);
 679     }
 680 }
 681
 682 static void init_dequant8_coeff_table(H264Context *h){
 683     int i,q,x;
 684     const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
 685     h->dequant8_coeff[0] = h->dequant8_buffer[0];
 686     h->dequant8_coeff[1] = h->dequant8_buffer[1];
 687
 688     for(i=0; i<2; i++ ){
 689         if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
 690             h->dequant8_coeff[1] = h->dequant8_buffer[0];
 691             break;
 692         }
 693
 694         for(q=0; q<52; q++){
 695             int shift = div6[q];
 696             int idx = rem6[q];
 697             for(x=0; x<64; x++)
 698                 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
 699                     ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
 700                     h->pps.scaling_matrix8[i][x]) << shift;
 701         }
 702     }
 703 }
 704
 705 static void init_dequant4_coeff_table(H264Context *h){
 706     int i,j,q,x;
 707     const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
 708     for(i=0; i<6; i++ ){
 709         h->dequant4_coeff[i] = h->dequant4_buffer[i];
 710         for(j=0; j<i; j++){
 711             if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
 712                 h->dequant4_coeff[i] = h->dequant4_buffer[j];
 713                 break;
 714             }
 715         }
 716         if(j<i)
 717             continue;
 718
 719         for(q=0; q<52; q++){
 720             int shift = div6[q] + 2;
 721             int idx = rem6[q];
 722             for(x=0; x<16; x++)
 723                 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
 724                     ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
 725                     h->pps.scaling_matrix4[i][x]) << shift;
 726         }
 727     }
 728 }
 729
 730 static void init_dequant_tables(H264Context *h){
 731     int i,x;
 732     init_dequant4_coeff_table(h);
 733     if(h->pps.transform_8x8_mode)
 734         init_dequant8_coeff_table(h);
 735     if(h->sps.transform_bypass){
 736         for(i=0; i<6; i++)
 737             for(x=0; x<16; x++)
 738                 h->dequant4_coeff[i][0][x] = 1<<6;
 739         if(h->pps.transform_8x8_mode)
 740             for(i=0; i<2; i++)
 741                 for(x=0; x<64; x++)
 742                     h->dequant8_coeff[i][0][x] = 1<<6;
 743     }
 744 }
 745
 746
 747 int ff_h264_alloc_tables(H264Context *h){
 748     MpegEncContext * const s = &h->s;
 749     const int big_mb_num= s->mb_stride * (s->mb_height+1);
 750     int x,y;
 751
 752     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t), fail)
 753
 754     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 32 * sizeof(uint8_t), fail)
 755     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
 756     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
 757
 758     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
 759     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t), fail);
 760     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t), fail);
 761     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 32*big_mb_num * sizeof(uint8_t) , fail);
 762     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
 763
 764     memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
 765     h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
 766
 767     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy  , big_mb_num * sizeof(uint32_t), fail);
 768     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b8_xy , big_mb_num * sizeof(uint32_t), fail);
 769     for(y=0; y<s->mb_height; y++){
 770         for(x=0; x<s->mb_width; x++){
 771             const int mb_xy= x + y*s->mb_stride;
 772             const int b_xy = 4*x + 4*y*h->b_stride;
 773             const int b8_xy= 2*x + 2*y*h->b8_stride;
 774
 775             h->mb2b_xy [mb_xy]= b_xy;
 776             h->mb2b8_xy[mb_xy]= b8_xy;
 777         }
 778     }
 779
 780     s->obmc_scratchpad = NULL;
 781
 782     if(!h->dequant4_coeff[0])
 783         init_dequant_tables(h);
 784
 785     return 0;
 786 fail:
 787     free_tables(h);
 788     return -1;
 789 }
 790
 791 /**
 792  * Mimic alloc_tables(), but for every context thread.
 793  */
 794 static void clone_tables(H264Context *dst, H264Context *src){
 795     dst->intra4x4_pred_mode       = src->intra4x4_pred_mode;
 796     dst->non_zero_count           = src->non_zero_count;
 797     dst->slice_table              = src->slice_table;
 798     dst->cbp_table                = src->cbp_table;
 799     dst->mb2b_xy                  = src->mb2b_xy;
 800     dst->mb2b8_xy                 = src->mb2b8_xy;
 801     dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
 802     dst->mvd_table[0]             = src->mvd_table[0];
 803     dst->mvd_table[1]             = src->mvd_table[1];
 804     dst->direct_table             = src->direct_table;
 805     dst->list_counts              = src->list_counts;
 806
 807     dst->s.obmc_scratchpad = NULL;
 808     ff_h264_pred_init(&dst->hpc, src->s.codec_id);
 809 }
 810
 811 /**
 812  * Init context
 813  * Allocate buffers which are not shared amongst multiple threads.
 814  */
 815 static int context_init(H264Context *h){
 816     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
 817     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
 818
 819     return 0;
 820 fail:
 821     return -1; // free_tables will clean up for us
 822 }
 823
 824 static av_cold void common_init(H264Context *h){
 825     MpegEncContext * const s = &h->s;
 826
 827     s->width = s->avctx->width;
 828     s->height = s->avctx->height;
 829     s->codec_id= s->avctx->codec->id;
 830
 831     ff_h264_pred_init(&h->hpc, s->codec_id);
 832
 833     h->dequant_coeff_pps= -1;
 834     s->unrestricted_mv=1;
 835     s->decode=1; //FIXME
 836
 837     dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
 838
 839     memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
 840     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
 841 }
 842
 843 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
 844     H264Context *h= avctx->priv_data;
 845     MpegEncContext * const s = &h->s;
 846
 847     MPV_decode_defaults(s);
 848
 849     s->avctx = avctx;
 850     common_init(h);
 851
 852     s->out_format = FMT_H264;
 853     s->workaround_bugs= avctx->workaround_bugs;
 854
 855     // set defaults
 856 //    s->decode_mb= ff_h263_decode_mb;
 857     s->quarter_sample = 1;
 858     if(!avctx->has_b_frames)
 859     s->low_delay= 1;
 860
 861     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
 862
 863     ff_h264_decode_init_vlc();
 864
 865     if(avctx->extradata_size > 0 && avctx->extradata &&
 866        *(char *)avctx->extradata == 1){
 867         h->is_avc = 1;
 868         h->got_avcC = 0;
 869     } else {
 870         h->is_avc = 0;
 871     }
 872
 873     h->thread_context[0] = h;
 874     h->outputed_poc = INT_MIN;
 875     h->prev_poc_msb= 1<<16;
 876     h->x264_build = -1;
 877     ff_h264_reset_sei(h);
 878     if(avctx->codec_id == CODEC_ID_H264){
 879         if(avctx->ticks_per_frame == 1){
 880             s->avctx->time_base.den *=2;
 881         }
 882         avctx->ticks_per_frame = 2;
 883     }
 884     return 0;
 885 }
 886
 887 int ff_h264_frame_start(H264Context *h){
 888     MpegEncContext * const s = &h->s;
 889     int i;
 890
 891     if(MPV_frame_start(s, s->avctx) < 0)
 892         return -1;
 893     ff_er_frame_start(s);
 894     /*
 895      * MPV_frame_start uses pict_type to derive key_frame.
 896      * This is incorrect for H.264; IDR markings must be used.
 897      * Zero here; IDR markings per slice in frame or fields are ORed in later.
 898      * See decode_nal_units().
 899      */
 900     s->current_picture_ptr->key_frame= 0;
 901     s->current_picture_ptr->mmco_reset= 0;
 902
 903     assert(s->linesize && s->uvlinesize);
 904
 905     for(i=0; i<16; i++){
 906         h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
 907         h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
 908     }
 909     for(i=0; i<4; i++){
 910         h->block_offset[16+i]=
 911         h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
 912         h->block_offset[24+16+i]=
 913         h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
 914     }
 915
 916     /* can't be in alloc_tables because linesize isn't known there.
 917      * FIXME: redo bipred weight to not require extra buffer? */
 918     for(i = 0; i < s->avctx->thread_count; i++)
 919         if(!h->thread_context[i]->s.obmc_scratchpad)
 920             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
 921
 922     /* some macroblocks will be accessed before they're available */
 923     if(FRAME_MBAFF || s->avctx->thread_count > 1)
 924         memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
 925
 926 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
 927
 928     // We mark the current picture as non-reference after allocating it, so
 929     // that if we break out due to an error it can be released automatically
 930     // in the next MPV_frame_start().
 931     // SVQ3 as well as most other codecs have only last/next/current and thus
 932     // get released even with set reference, besides SVQ3 and others do not
 933     // mark frames as reference later "naturally".
 934     if(s->codec_id != CODEC_ID_SVQ3)
 935         s->current_picture_ptr->reference= 0;
 936
 937     s->current_picture_ptr->field_poc[0]=
 938     s->current_picture_ptr->field_poc[1]= INT_MAX;
 939     assert(s->current_picture_ptr->long_ref==0);
 940
 941     return 0;
 942 }
 943
 944 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int simple){
 945     MpegEncContext * const s = &h->s;
 946     uint8_t *top_border;
 947     int top_idx = 1;
 948
 949     src_y  -=   linesize;
 950     src_cb -= uvlinesize;
 951     src_cr -= uvlinesize;
 952
 953     if(!simple && FRAME_MBAFF){
 954         if(s->mb_y&1){
 955             if(!MB_MBAFF){
 956                 top_border = h->top_borders[0][s->mb_x];
 957                 AV_COPY128(top_border, src_y + 15*linesize);
 958                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
 959                     AV_COPY64(top_border+16, src_cb+7*uvlinesize);
 960                     AV_COPY64(top_border+24, src_cr+7*uvlinesize);
 961                 }
 962             }
 963         }else if(MB_MBAFF){
 964             top_idx = 0;
 965         }else
 966             return;
 967     }
 968
 969     top_border = h->top_borders[top_idx][s->mb_x];
 970     // There are two lines saved, the line above the the top macroblock of a pair,
 971     // and the line above the bottom macroblock
 972     AV_COPY128(top_border, src_y + 16*linesize);
 973
 974     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
 975         AV_COPY64(top_border+16, src_cb+8*uvlinesize);
 976         AV_COPY64(top_border+24, src_cr+8*uvlinesize);
 977     }
 978 }
 979
 980 static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg, int simple){
 981     MpegEncContext * const s = &h->s;
 982     int temp8, i;
 983     uint64_t temp64;
 984     int deblock_left;
 985     int deblock_top;
 986     int mb_xy;
 987     int top_idx = 1;
 988     uint8_t *top_border_m1;
 989     uint8_t *top_border;
 990
 991     if(!simple && FRAME_MBAFF){
 992         if(s->mb_y&1){
 993             if(!MB_MBAFF)
 994                 return;
 995         }else{
 996             top_idx = MB_MBAFF ? 0 : 1;
 997         }
 998     }
 999
1000     if(h->deblocking_filter == 2) {
1001         mb_xy = h->mb_xy;
1002         deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
1003         deblock_top  = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
1004     } else {
1005         deblock_left = (s->mb_x > 0);
1006         deblock_top =  (s->mb_y > !!MB_FIELD);
1007     }
1008
1009     src_y  -=   linesize + 1;
1010     src_cb -= uvlinesize + 1;
1011     src_cr -= uvlinesize + 1;
1012
1013     top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1014     top_border    = h->top_borders[top_idx][s->mb_x];
1015
1016 #define XCHG(a,b,xchg)\
1017 if (xchg) AV_SWAP64(b,a);\
1018 else      AV_COPY64(b,a);
1019
1020     if(deblock_top){
1021         if(deblock_left){
1022             XCHG(top_border_m1+8, src_y -7, 1);
1023         }
1024         XCHG(top_border+0, src_y +1, xchg);
1025         XCHG(top_border+8, src_y +9, 1);
1026         if(s->mb_x+1 < s->mb_width){
1027             XCHG(h->top_borders[top_idx][s->mb_x+1], src_y +17, 1);
1028         }
1029     }
1030
1031     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1032         if(deblock_top){
1033             if(deblock_left){
1034                 XCHG(top_border_m1+16, src_cb -7, 1);
1035                 XCHG(top_border_m1+24, src_cr -7, 1);
1036             }
1037             XCHG(top_border+16, src_cb+1, 1);
1038             XCHG(top_border+24, src_cr+1, 1);
1039         }
1040     }
1041 }
1042
1043 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
1044     MpegEncContext * const s = &h->s;
1045     const int mb_x= s->mb_x;
1046     const int mb_y= s->mb_y;
1047     const int mb_xy= h->mb_xy;
1048     const int mb_type= s->current_picture.mb_type[mb_xy];
1049     uint8_t  *dest_y, *dest_cb, *dest_cr;
1050     int linesize, uvlinesize /*dct_offset*/;
1051     int i;
1052     int *block_offset = &h->block_offset[0];
1053     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1054     /* is_h264 should always be true if SVQ3 is disabled. */
1055     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1056     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1057     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1058
1059     dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
1060     dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
1061     dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
1062
1063     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
1064     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
1065
1066     h->list_counts[mb_xy]= h->list_count;
1067
1068     if (!simple && MB_FIELD) {
1069         linesize   = h->mb_linesize   = s->linesize * 2;
1070         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1071         block_offset = &h->block_offset[24];
1072         if(mb_y&1){ //FIXME move out of this function?
1073             dest_y -= s->linesize*15;
1074             dest_cb-= s->uvlinesize*7;
1075             dest_cr-= s->uvlinesize*7;
1076         }
1077         if(FRAME_MBAFF) {
1078             int list;
1079             for(list=0; list<h->list_count; list++){
1080                 if(!USES_LIST(mb_type, list))
1081                     continue;
1082                 if(IS_16X16(mb_type)){
1083                     int8_t *ref = &h->ref_cache[list][scan8[0]];
1084                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1085                 }else{
1086                     for(i=0; i<16; i+=4){
1087                         int ref = h->ref_cache[list][scan8[i]];
1088                         if(ref >= 0)
1089                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1090                     }
1091                 }
1092             }
1093         }
1094     } else {
1095         linesize   = h->mb_linesize   = s->linesize;
1096         uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1097 //        dct_offset = s->linesize * 16;
1098     }
1099
1100     if (!simple && IS_INTRA_PCM(mb_type)) {
1101         for (i=0; i<16; i++) {
1102             memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1103         }
1104         for (i=0; i<8; i++) {
1105             memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
1106             memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
1107         }
1108     } else {
1109         if(IS_INTRA(mb_type)){
1110             if(h->deblocking_filter)
1111                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
1112
1113             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1114                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1115                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1116             }
1117
1118             if(IS_INTRA4x4(mb_type)){
1119                 if(simple || !s->encoding){
1120                     if(IS_8x8DCT(mb_type)){
1121                         if(transform_bypass){
1122                             idct_dc_add =
1123                             idct_add    = s->dsp.add_pixels8;
1124                         }else{
1125                             idct_dc_add = s->dsp.h264_idct8_dc_add;
1126                             idct_add    = s->dsp.h264_idct8_add;
1127                         }
1128                         for(i=0; i<16; i+=4){
1129                             uint8_t * const ptr= dest_y + block_offset[i];
1130                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1131                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1132                                 h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
1133                             }else{
1134                                 const int nnz = h->non_zero_count_cache[ scan8[i] ];
1135                                 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1136                                                             (h->topright_samples_available<<i)&0x4000, linesize);
1137                                 if(nnz){
1138                                     if(nnz == 1 && h->mb[i*16])
1139                                         idct_dc_add(ptr, h->mb + i*16, linesize);
1140                                     else
1141                                         idct_add   (ptr, h->mb + i*16, linesize);
1142                                 }
1143                             }
1144                         }
1145                     }else{
1146                         if(transform_bypass){
1147                             idct_dc_add =
1148                             idct_add    = s->dsp.add_pixels4;
1149                         }else{
1150                             idct_dc_add = s->dsp.h264_idct_dc_add;
1151                             idct_add    = s->dsp.h264_idct_add;
1152                         }
1153                         for(i=0; i<16; i++){
1154                             uint8_t * const ptr= dest_y + block_offset[i];
1155                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1156
1157                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1158                                 h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
1159                             }else{
1160                                 uint8_t *topright;
1161                                 int nnz, tr;
1162                                 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1163                                     const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1164                                     assert(mb_y || linesize <= block_offset[i]);
1165                                     if(!topright_avail){
1166                                         tr= ptr[3 - linesize]*0x01010101;
1167                                         topright= (uint8_t*) &tr;
1168                                     }else
1169                                         topright= ptr + 4 - linesize;
1170                                 }else
1171                                     topright= NULL;
1172
1173                                 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1174                                 nnz = h->non_zero_count_cache[ scan8[i] ];
1175                                 if(nnz){
1176                                     if(is_h264){
1177                                         if(nnz == 1 && h->mb[i*16])
1178                                             idct_dc_add(ptr, h->mb + i*16, linesize);
1179                                         else
1180                                             idct_add   (ptr, h->mb + i*16, linesize);
1181                                     }else
1182                                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1183                                 }
1184                             }
1185                         }
1186                     }
1187                 }
1188             }else{
1189                 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1190                 if(is_h264){
1191                     if(!transform_bypass)
1192                         h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
1193                 }else
1194                     ff_svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
1195             }
1196             if(h->deblocking_filter)
1197                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
1198         }else if(is_h264){
1199             hl_motion(h, dest_y, dest_cb, dest_cr,
1200                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1201                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1202                       s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
1203         }
1204
1205
1206         if(!IS_INTRA4x4(mb_type)){
1207             if(is_h264){
1208                 if(IS_INTRA16x16(mb_type)){
1209                     if(transform_bypass){
1210                         if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1211                             h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1212                         }else{
1213                             for(i=0; i<16; i++){
1214                                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1215                                     s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
1216                             }
1217                         }
1218                     }else{
1219                          s->dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1220                     }
1221                 }else if(h->cbp&15){
1222                     if(transform_bypass){
1223                         const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1224                         idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1225                         for(i=0; i<16; i+=di){
1226                             if(h->non_zero_count_cache[ scan8[i] ]){
1227                                 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
1228                             }
1229                         }
1230                     }else{
1231                         if(IS_8x8DCT(mb_type)){
1232                             s->dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1233                         }else{
1234                             s->dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1235                         }
1236                     }
1237                 }
1238             }else{
1239                 for(i=0; i<16; i++){
1240                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1241                         uint8_t * const ptr= dest_y + block_offset[i];
1242                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1243                     }
1244                 }
1245             }
1246         }
1247
1248         if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1249             uint8_t *dest[2] = {dest_cb, dest_cr};
1250             if(transform_bypass){
1251                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1252                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
1253                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
1254                 }else{
1255                     idct_add = s->dsp.add_pixels4;
1256                     for(i=16; i<16+8; i++){
1257                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1258                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1259                     }
1260                 }
1261             }else{
1262                 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]);
1263                 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]);
1264                 if(is_h264){
1265                     idct_add = s->dsp.h264_idct_add;
1266                     idct_dc_add = s->dsp.h264_idct_dc_add;
1267                     for(i=16; i<16+8; i++){
1268                         if(h->non_zero_count_cache[ scan8[i] ])
1269                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1270                         else if(h->mb[i*16])
1271                             idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1272                     }
1273                 }else{
1274                     for(i=16; i<16+8; i++){
1275                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1276                             uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1277                             ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
1278                         }
1279                     }
1280                 }
1281             }
1282         }
1283     }
1284     if(h->cbp || IS_INTRA(mb_type))
1285         s->dsp.clear_blocks(h->mb);
1286
1287     if(h->deblocking_filter && 0) {
1288         backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, simple);
1289         fill_filter_caches(h, mb_type); //FIXME don't fill stuff which isn't used by filter_mb
1290         h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
1291         h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
1292         if (!simple && FRAME_MBAFF) {
1293             ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
1294         } else {
1295             ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
1296         }
1297     }
1298 }
1299
1300 /**
1301  * Process a macroblock; this case avoids checks for expensive uncommon cases.
1302  */
1303 static void hl_decode_mb_simple(H264Context *h){
1304     hl_decode_mb_internal(h, 1);
1305 }
1306
1307 /**
1308  * Process a macroblock; this handles edge cases, such as interlacing.
1309  */
1310 static void av_noinline hl_decode_mb_complex(H264Context *h){
1311     hl_decode_mb_internal(h, 0);
1312 }
1313
1314 void ff_h264_hl_decode_mb(H264Context *h){
1315     MpegEncContext * const s = &h->s;
1316     const int mb_xy= h->mb_xy;
1317     const int mb_type= s->current_picture.mb_type[mb_xy];
1318     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1319
1320     if (is_complex)
1321         hl_decode_mb_complex(h);
1322     else hl_decode_mb_simple(h);
1323 }
1324
1325 static int pred_weight_table(H264Context *h){
1326     MpegEncContext * const s = &h->s;
1327     int list, i;
1328     int luma_def, chroma_def;
1329
1330     h->use_weight= 0;
1331     h->use_weight_chroma= 0;
1332     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1333     h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1334     luma_def = 1<<h->luma_log2_weight_denom;
1335     chroma_def = 1<<h->chroma_log2_weight_denom;
1336
1337     for(list=0; list<2; list++){
1338         h->luma_weight_flag[list]   = 0;
1339         h->chroma_weight_flag[list] = 0;
1340         for(i=0; i<h->ref_count[list]; i++){
1341             int luma_weight_flag, chroma_weight_flag;
1342
1343             luma_weight_flag= get_bits1(&s->gb);
1344             if(luma_weight_flag){
1345                 h->luma_weight[list][i]= get_se_golomb(&s->gb);
1346                 h->luma_offset[list][i]= get_se_golomb(&s->gb);
1347                 if(   h->luma_weight[list][i] != luma_def
1348                    || h->luma_offset[list][i] != 0) {
1349                     h->use_weight= 1;
1350                     h->luma_weight_flag[list]= 1;
1351                 }
1352             }else{
1353                 h->luma_weight[list][i]= luma_def;
1354                 h->luma_offset[list][i]= 0;
1355             }
1356
1357             if(CHROMA){
1358                 chroma_weight_flag= get_bits1(&s->gb);
1359                 if(chroma_weight_flag){
1360                     int j;
1361                     for(j=0; j<2; j++){
1362                         h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
1363                         h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
1364                         if(   h->chroma_weight[list][i][j] != chroma_def
1365                            || h->chroma_offset[list][i][j] != 0) {
1366                             h->use_weight_chroma= 1;
1367                             h->chroma_weight_flag[list]= 1;
1368                         }
1369                     }
1370                 }else{
1371                     int j;
1372                     for(j=0; j<2; j++){
1373                         h->chroma_weight[list][i][j]= chroma_def;
1374                         h->chroma_offset[list][i][j]= 0;
1375                     }
1376                 }
1377             }
1378         }
1379         if(h->slice_type_nos != FF_B_TYPE) break;
1380     }
1381     h->use_weight= h->use_weight || h->use_weight_chroma;
1382     return 0;
1383 }
1384
1385 static void implicit_weight_table(H264Context *h){
1386     MpegEncContext * const s = &h->s;
1387     int ref0, ref1, i;
1388     int cur_poc = s->current_picture_ptr->poc;
1389
1390     for (i = 0; i < 2; i++) {
1391         h->luma_weight_flag[i]   = 0;
1392         h->chroma_weight_flag[i] = 0;
1393     }
1394
1395     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
1396        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1397         h->use_weight= 0;
1398         h->use_weight_chroma= 0;
1399         return;
1400     }
1401
1402     h->use_weight= 2;
1403     h->use_weight_chroma= 2;
1404     h->luma_log2_weight_denom= 5;
1405     h->chroma_log2_weight_denom= 5;
1406
1407     for(ref0=0; ref0 < h->ref_count[0]; ref0++){
1408         int poc0 = h->ref_list[0][ref0].poc;
1409         for(ref1=0; ref1 < h->ref_count[1]; ref1++){
1410             int poc1 = h->ref_list[1][ref1].poc;
1411             int td = av_clip(poc1 - poc0, -128, 127);
1412             if(td){
1413                 int tb = av_clip(cur_poc - poc0, -128, 127);
1414                 int tx = (16384 + (FFABS(td) >> 1)) / td;
1415                 int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
1416                 if(dist_scale_factor < -64 || dist_scale_factor > 128)
1417                     h->implicit_weight[ref0][ref1] = 32;
1418                 else
1419                     h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
1420             }else
1421                 h->implicit_weight[ref0][ref1] = 32;
1422         }
1423     }
1424 }
1425
1426 /**
1427  * instantaneous decoder refresh.
1428  */
1429 static void idr(H264Context *h){
1430     ff_h264_remove_all_refs(h);
1431     h->prev_frame_num= 0;
1432     h->prev_frame_num_offset= 0;
1433     h->prev_poc_msb=
1434     h->prev_poc_lsb= 0;
1435 }
1436
1437 /* forget old pics after a seek */
1438 static void flush_dpb(AVCodecContext *avctx){
1439     H264Context *h= avctx->priv_data;
1440     int i;
1441     for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1442         if(h->delayed_pic[i])
1443             h->delayed_pic[i]->reference= 0;
1444         h->delayed_pic[i]= NULL;
1445     }
1446     h->outputed_poc= INT_MIN;
1447     h->prev_interlaced_frame = 1;
1448     idr(h);
1449     if(h->s.current_picture_ptr)
1450         h->s.current_picture_ptr->reference= 0;
1451     h->s.first_field= 0;
1452     ff_h264_reset_sei(h);
1453     ff_mpeg_flush(avctx);
1454 }
1455
1456 static int init_poc(H264Context *h){
1457     MpegEncContext * const s = &h->s;
1458     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1459     int field_poc[2];
1460     Picture *cur = s->current_picture_ptr;
1461
1462     h->frame_num_offset= h->prev_frame_num_offset;
1463     if(h->frame_num < h->prev_frame_num)
1464         h->frame_num_offset += max_frame_num;
1465
1466     if(h->sps.poc_type==0){
1467         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1468
1469         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1470             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1471         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1472             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1473         else
1474             h->poc_msb = h->prev_poc_msb;
1475 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1476         field_poc[0] =
1477         field_poc[1] = h->poc_msb + h->poc_lsb;
1478         if(s->picture_structure == PICT_FRAME)
1479             field_poc[1] += h->delta_poc_bottom;
1480     }else if(h->sps.poc_type==1){
1481         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1482         int i;
1483
1484         if(h->sps.poc_cycle_length != 0)
1485             abs_frame_num = h->frame_num_offset + h->frame_num;
1486         else
1487             abs_frame_num = 0;
1488
1489         if(h->nal_ref_idc==0 && abs_frame_num > 0)
1490             abs_frame_num--;
1491
1492         expected_delta_per_poc_cycle = 0;
1493         for(i=0; i < h->sps.poc_cycle_length; i++)
1494             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1495
1496         if(abs_frame_num > 0){
1497             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1498             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1499
1500             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1501             for(i = 0; i <= frame_num_in_poc_cycle; i++)
1502                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1503         } else
1504             expectedpoc = 0;
1505
1506         if(h->nal_ref_idc == 0)
1507             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1508
1509         field_poc[0] = expectedpoc + h->delta_poc[0];
1510         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1511
1512         if(s->picture_structure == PICT_FRAME)
1513             field_poc[1] += h->delta_poc[1];
1514     }else{
1515         int poc= 2*(h->frame_num_offset + h->frame_num);
1516
1517         if(!h->nal_ref_idc)
1518             poc--;
1519
1520         field_poc[0]= poc;
1521         field_poc[1]= poc;
1522     }
1523
1524     if(s->picture_structure != PICT_BOTTOM_FIELD)
1525         s->current_picture_ptr->field_poc[0]= field_poc[0];
1526     if(s->picture_structure != PICT_TOP_FIELD)
1527         s->current_picture_ptr->field_poc[1]= field_poc[1];
1528     cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1529
1530     return 0;
1531 }
1532
1533
1534 /**
1535  * initialize scan tables
1536  */
1537 static void init_scan_tables(H264Context *h){
1538     MpegEncContext * const s = &h->s;
1539     int i;
1540     if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
1541         memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
1542         memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
1543     }else{
1544         for(i=0; i<16; i++){
1545 #define T(x) (x>>2) | ((x<<2) & 0xF)
1546             h->zigzag_scan[i] = T(zigzag_scan[i]);
1547             h-> field_scan[i] = T( field_scan[i]);
1548 #undef T
1549         }
1550     }
1551     if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
1552         memcpy(h->zigzag_scan8x8,       ff_zigzag_direct,     64*sizeof(uint8_t));
1553         memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
1554         memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
1555         memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
1556     }else{
1557         for(i=0; i<64; i++){
1558 #define T(x) (x>>3) | ((x&7)<<3)
1559             h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
1560             h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1561             h->field_scan8x8[i]        = T(field_scan8x8[i]);
1562             h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
1563 #undef T
1564         }
1565     }
1566     if(h->sps.transform_bypass){ //FIXME same ugly
1567         h->zigzag_scan_q0          = zigzag_scan;
1568         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
1569         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1570         h->field_scan_q0           = field_scan;
1571         h->field_scan8x8_q0        = field_scan8x8;
1572         h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
1573     }else{
1574         h->zigzag_scan_q0          = h->zigzag_scan;
1575         h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
1576         h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1577         h->field_scan_q0           = h->field_scan;
1578         h->field_scan8x8_q0        = h->field_scan8x8;
1579         h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
1580     }
1581 }
1582
1583 static void field_end(H264Context *h){
1584     MpegEncContext * const s = &h->s;
1585     AVCodecContext * const avctx= s->avctx;
1586     s->mb_y= 0;
1587
1588     s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1589     s->current_picture_ptr->pict_type= s->pict_type;
1590
1591     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1592         ff_vdpau_h264_set_reference_frames(s);
1593
1594     if(!s->dropable) {
1595         ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1596         h->prev_poc_msb= h->poc_msb;
1597         h->prev_poc_lsb= h->poc_lsb;
1598     }
1599     h->prev_frame_num_offset= h->frame_num_offset;
1600     h->prev_frame_num= h->frame_num;
1601
1602     if (avctx->hwaccel) {
1603         if (avctx->hwaccel->end_frame(avctx) < 0)
1604             av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1605     }
1606
1607     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1608         ff_vdpau_h264_picture_complete(s);
1609
1610     /*
1611      * FIXME: Error handling code does not seem to support interlaced
1612      * when slices span multiple rows
1613      * The ff_er_add_slice calls don't work right for bottom
1614      * fields; they cause massive erroneous error concealing
1615      * Error marking covers both fields (top and bottom).
1616      * This causes a mismatched s->error_count
1617      * and a bad error table. Further, the error count goes to
1618      * INT_MAX when called for bottom field, because mb_y is
1619      * past end by one (callers fault) and resync_mb_y != 0
1620      * causes problems for the first MB line, too.
1621      */
1622     if (!FIELD_PICTURE)
1623         ff_er_frame_end(s);
1624
1625     MPV_frame_end(s);
1626
1627     h->current_slice=0;
1628 }
1629
1630 /**
1631  * Replicates H264 "master" context to thread contexts.
1632  */
1633 static void clone_slice(H264Context *dst, H264Context *src)
1634 {
1635     memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
1636     dst->s.current_picture_ptr  = src->s.current_picture_ptr;
1637     dst->s.current_picture      = src->s.current_picture;
1638     dst->s.linesize             = src->s.linesize;
1639     dst->s.uvlinesize           = src->s.uvlinesize;
1640     dst->s.first_field          = src->s.first_field;
1641
1642     dst->prev_poc_msb           = src->prev_poc_msb;
1643     dst->prev_poc_lsb           = src->prev_poc_lsb;
1644     dst->prev_frame_num_offset  = src->prev_frame_num_offset;
1645     dst->prev_frame_num         = src->prev_frame_num;
1646     dst->short_ref_count        = src->short_ref_count;
1647
1648     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
1649     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
1650     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1651     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
1652
1653     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
1654     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
1655 }
1656
1657 /**
1658  * decodes a slice header.
1659  * This will also call MPV_common_init() and frame_start() as needed.
1660  *
1661  * @param h h264context
1662  * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1663  *
1664  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1665  */
1666 static int decode_slice_header(H264Context *h, H264Context *h0){
1667     MpegEncContext * const s = &h->s;
1668     MpegEncContext * const s0 = &h0->s;
1669     unsigned int first_mb_in_slice;
1670     unsigned int pps_id;
1671     int num_ref_idx_active_override_flag;
1672     unsigned int slice_type, tmp, i, j;
1673     int default_ref_list_done = 0;
1674     int last_pic_structure;
1675
1676     s->dropable= h->nal_ref_idc == 0;
1677
1678     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1679         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1680         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1681     }else{
1682         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1683         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1684     }
1685
1686     first_mb_in_slice= get_ue_golomb(&s->gb);
1687
1688     if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1689         if(h0->current_slice && FIELD_PICTURE){
1690             field_end(h);
1691         }
1692
1693         h0->current_slice = 0;
1694         if (!s0->first_field)
1695             s->current_picture_ptr= NULL;
1696     }
1697
1698     slice_type= get_ue_golomb_31(&s->gb);
1699     if(slice_type > 9){
1700         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);
1701         return -1;
1702     }
1703     if(slice_type > 4){
1704         slice_type -= 5;
1705         h->slice_type_fixed=1;
1706     }else
1707         h->slice_type_fixed=0;
1708
1709     slice_type= golomb_to_pict_type[ slice_type ];
1710     if (slice_type == FF_I_TYPE
1711         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
1712         default_ref_list_done = 1;
1713     }
1714     h->slice_type= slice_type;
1715     h->slice_type_nos= slice_type & 3;
1716
1717     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
1718     if (s->pict_type == FF_B_TYPE && s0->last_picture_ptr == NULL) {
1719         av_log(h->s.avctx, AV_LOG_ERROR,
1720                "B picture before any references, skipping\n");
1721         return -1;
1722     }
1723
1724     pps_id= get_ue_golomb(&s->gb);
1725     if(pps_id>=MAX_PPS_COUNT){
1726         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
1727         return -1;
1728     }
1729     if(!h0->pps_buffers[pps_id]) {
1730         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
1731         return -1;
1732     }
1733     h->pps= *h0->pps_buffers[pps_id];
1734
1735     if(!h0->sps_buffers[h->pps.sps_id]) {
1736         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
1737         return -1;
1738     }
1739     h->sps = *h0->sps_buffers[h->pps.sps_id];
1740
1741     if(h == h0 && h->dequant_coeff_pps != pps_id){
1742         h->dequant_coeff_pps = pps_id;
1743         init_dequant_tables(h);
1744     }
1745
1746     s->mb_width= h->sps.mb_width;
1747     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1748
1749     h->b_stride=  s->mb_width*4;
1750     h->b8_stride= s->mb_width*2;
1751
1752     s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
1753     if(h->sps.frame_mbs_only_flag)
1754         s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
1755     else
1756         s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
1757
1758     if (s->context_initialized
1759         && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
1760         if(h != h0)
1761             return -1;   // width / height changed during parallelized decoding
1762         free_tables(h);
1763         flush_dpb(s->avctx);
1764         MPV_common_end(s);
1765     }
1766     if (!s->context_initialized) {
1767         if(h != h0)
1768             return -1;  // we cant (re-)initialize context during parallel decoding
1769
1770         avcodec_set_dimensions(s->avctx, s->width, s->height);
1771         s->avctx->sample_aspect_ratio= h->sps.sar;
1772         if(!s->avctx->sample_aspect_ratio.den)
1773             s->avctx->sample_aspect_ratio.den = 1;
1774
1775         if(h->sps.video_signal_type_present_flag){
1776             s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1777             if(h->sps.colour_description_present_flag){
1778                 s->avctx->color_primaries = h->sps.color_primaries;
1779                 s->avctx->color_trc       = h->sps.color_trc;
1780                 s->avctx->colorspace      = h->sps.colorspace;
1781             }
1782         }
1783
1784         if(h->sps.timing_info_present_flag){
1785             int64_t den= h->sps.time_scale;
1786             if(h->x264_build < 44U)
1787                 den *= 2;
1788             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
1789                       h->sps.num_units_in_tick, den, 1<<30);
1790         }
1791         s->avctx->pix_fmt = s->avctx->get_format(s->avctx, s->avctx->codec->pix_fmts);
1792         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1793
1794         if (MPV_common_init(s) < 0)
1795             return -1;
1796         s->first_field = 0;
1797         h->prev_interlaced_frame = 1;
1798
1799         init_scan_tables(h);
1800         ff_h264_alloc_tables(h);
1801
1802         for(i = 1; i < s->avctx->thread_count; i++) {
1803             H264Context *c;
1804             c = h->thread_context[i] = av_malloc(sizeof(H264Context));
1805             memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
1806             memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
1807             c->sps = h->sps;
1808             c->pps = h->pps;
1809             init_scan_tables(c);
1810             clone_tables(c, h);
1811         }
1812
1813         for(i = 0; i < s->avctx->thread_count; i++)
1814             if(context_init(h->thread_context[i]) < 0)
1815                 return -1;
1816     }
1817
1818     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1819
1820     h->mb_mbaff = 0;
1821     h->mb_aff_frame = 0;
1822     last_pic_structure = s0->picture_structure;
1823     if(h->sps.frame_mbs_only_flag){
1824         s->picture_structure= PICT_FRAME;
1825     }else{
1826         if(get_bits1(&s->gb)) { //field_pic_flag
1827             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
1828         } else {
1829             s->picture_structure= PICT_FRAME;
1830             h->mb_aff_frame = h->sps.mb_aff;
1831         }
1832     }
1833     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
1834
1835     if(h0->current_slice == 0){
1836         while(h->frame_num !=  h->prev_frame_num &&
1837               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
1838             av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
1839             if (ff_h264_frame_start(h) < 0)
1840                 return -1;
1841             h->prev_frame_num++;
1842             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
1843             s->current_picture_ptr->frame_num= h->prev_frame_num;
1844             ff_h264_execute_ref_pic_marking(h, NULL, 0);
1845         }
1846
1847         /* See if we have a decoded first field looking for a pair... */
1848         if (s0->first_field) {
1849             assert(s0->current_picture_ptr);
1850             assert(s0->current_picture_ptr->data[0]);
1851             assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
1852
1853             /* figure out if we have a complementary field pair */
1854             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
1855                 /*
1856                  * Previous field is unmatched. Don't display it, but let it
1857                  * remain for reference if marked as such.
1858                  */
1859                 s0->current_picture_ptr = NULL;
1860                 s0->first_field = FIELD_PICTURE;
1861
1862             } else {
1863                 if (h->nal_ref_idc &&
1864                         s0->current_picture_ptr->reference &&
1865                         s0->current_picture_ptr->frame_num != h->frame_num) {
1866                     /*
1867                      * This and previous field were reference, but had
1868                      * different frame_nums. Consider this field first in
1869                      * pair. Throw away previous field except for reference
1870                      * purposes.
1871                      */
1872                     s0->first_field = 1;
1873                     s0->current_picture_ptr = NULL;
1874
1875                 } else {
1876                     /* Second field in complementary pair */
1877                     s0->first_field = 0;
1878                 }
1879             }
1880
1881         } else {
1882             /* Frame or first field in a potentially complementary pair */
1883             assert(!s0->current_picture_ptr);
1884             s0->first_field = FIELD_PICTURE;
1885         }
1886
1887         if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
1888             s0->first_field = 0;
1889             return -1;
1890         }
1891     }
1892     if(h != h0)
1893         clone_slice(h, h0);
1894
1895     s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
1896
1897     assert(s->mb_num == s->mb_width * s->mb_height);
1898     if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
1899        first_mb_in_slice                    >= s->mb_num){
1900         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1901         return -1;
1902     }
1903     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
1904     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
1905     if (s->picture_structure == PICT_BOTTOM_FIELD)
1906         s->resync_mb_y = s->mb_y = s->mb_y + 1;
1907     assert(s->mb_y < s->mb_height);
1908
1909     if(s->picture_structure==PICT_FRAME){
1910         h->curr_pic_num=   h->frame_num;
1911         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
1912     }else{
1913         h->curr_pic_num= 2*h->frame_num + 1;
1914         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
1915     }
1916
1917     if(h->nal_unit_type == NAL_IDR_SLICE){
1918         get_ue_golomb(&s->gb); /* idr_pic_id */
1919     }
1920
1921     if(h->sps.poc_type==0){
1922         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
1923
1924         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
1925             h->delta_poc_bottom= get_se_golomb(&s->gb);
1926         }
1927     }
1928
1929     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
1930         h->delta_poc[0]= get_se_golomb(&s->gb);
1931
1932         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
1933             h->delta_poc[1]= get_se_golomb(&s->gb);
1934     }
1935
1936     init_poc(h);
1937
1938     if(h->pps.redundant_pic_cnt_present){
1939         h->redundant_pic_count= get_ue_golomb(&s->gb);
1940     }
1941
1942     //set defaults, might be overridden a few lines later
1943     h->ref_count[0]= h->pps.ref_count[0];
1944     h->ref_count[1]= h->pps.ref_count[1];
1945
1946     if(h->slice_type_nos != FF_I_TYPE){
1947         if(h->slice_type_nos == FF_B_TYPE){
1948             h->direct_spatial_mv_pred= get_bits1(&s->gb);
1949         }
1950         num_ref_idx_active_override_flag= get_bits1(&s->gb);
1951
1952         if(num_ref_idx_active_override_flag){
1953             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
1954             if(h->slice_type_nos==FF_B_TYPE)
1955                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
1956
1957             if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
1958                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
1959                 h->ref_count[0]= h->ref_count[1]= 1;
1960                 return -1;
1961             }
1962         }
1963         if(h->slice_type_nos == FF_B_TYPE)
1964             h->list_count= 2;
1965         else
1966             h->list_count= 1;
1967     }else
1968         h->list_count= 0;
1969
1970     if(!default_ref_list_done){
1971         ff_h264_fill_default_ref_list(h);
1972     }
1973
1974     if(h->slice_type_nos!=FF_I_TYPE && ff_h264_decode_ref_pic_list_reordering(h) < 0)
1975         return -1;
1976
1977     if(h->slice_type_nos!=FF_I_TYPE){
1978         s->last_picture_ptr= &h->ref_list[0][0];
1979         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
1980     }
1981     if(h->slice_type_nos==FF_B_TYPE){
1982         s->next_picture_ptr= &h->ref_list[1][0];
1983         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
1984     }
1985
1986     if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
1987        ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
1988         pred_weight_table(h);
1989     else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE)
1990         implicit_weight_table(h);
1991     else {
1992         h->use_weight = 0;
1993         for (i = 0; i < 2; i++) {
1994             h->luma_weight_flag[i]   = 0;
1995             h->chroma_weight_flag[i] = 0;
1996         }
1997     }
1998
1999     if(h->nal_ref_idc)
2000         ff_h264_decode_ref_pic_marking(h0, &s->gb);
2001
2002     if(FRAME_MBAFF)
2003         ff_h264_fill_mbaff_ref_list(h);
2004
2005     if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
2006         ff_h264_direct_dist_scale_factor(h);
2007     ff_h264_direct_ref_list_init(h);
2008
2009     if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
2010         tmp = get_ue_golomb_31(&s->gb);
2011         if(tmp > 2){
2012             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2013             return -1;
2014         }
2015         h->cabac_init_idc= tmp;
2016     }
2017
2018     h->last_qscale_diff = 0;
2019     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2020     if(tmp>51){
2021         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2022         return -1;
2023     }
2024     s->qscale= tmp;
2025     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2026     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2027     //FIXME qscale / qp ... stuff
2028     if(h->slice_type == FF_SP_TYPE){
2029         get_bits1(&s->gb); /* sp_for_switch_flag */
2030     }
2031     if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
2032         get_se_golomb(&s->gb); /* slice_qs_delta */
2033     }
2034
2035     h->deblocking_filter = 1;
2036     h->slice_alpha_c0_offset = 52;
2037     h->slice_beta_offset = 52;
2038     if( h->pps.deblocking_filter_parameters_present ) {
2039         tmp= get_ue_golomb_31(&s->gb);
2040         if(tmp > 2){
2041             av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2042             return -1;
2043         }
2044         h->deblocking_filter= tmp;
2045         if(h->deblocking_filter < 2)
2046             h->deblocking_filter^= 1; // 1<->0
2047
2048         if( h->deblocking_filter ) {
2049             h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2050             h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
2051             if(   h->slice_alpha_c0_offset > 104U
2052                || h->slice_beta_offset     > 104U){
2053                 av_log(s->avctx, AV_LOG_ERROR, "deblocking filter parameters %d %d out of range\n", h->slice_alpha_c0_offset, h->slice_beta_offset);
2054                 return -1;
2055             }
2056         }
2057     }
2058
2059     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
2060        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
2061        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
2062        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2063         h->deblocking_filter= 0;
2064
2065     if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2066         if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2067             /* Cheat slightly for speed:
2068                Do not bother to deblock across slices. */
2069             h->deblocking_filter = 2;
2070         } else {
2071             h0->max_contexts = 1;
2072             if(!h0->single_decode_warning) {
2073                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2074                 h0->single_decode_warning = 1;
2075             }
2076             if(h != h0)
2077                 return 1; // deblocking switched inside frame
2078         }
2079     }
2080     h->qp_thresh= 15 + 52 - FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1]);
2081
2082 #if 0 //FMO
2083     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2084         slice_group_change_cycle= get_bits(&s->gb, ?);
2085 #endif
2086
2087     h0->last_slice_type = slice_type;
2088     h->slice_num = ++h0->current_slice;
2089     if(h->slice_num >= MAX_SLICES){
2090         av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2091     }
2092
2093     for(j=0; j<2; j++){
2094         int id_list[16];
2095         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2096         for(i=0; i<16; i++){
2097             id_list[i]= 60;
2098             if(h->ref_list[j][i].data[0]){
2099                 int k;
2100                 uint8_t *base= h->ref_list[j][i].base[0];
2101                 for(k=0; k<h->short_ref_count; k++)
2102                     if(h->short_ref[k]->base[0] == base){
2103                         id_list[i]= k;
2104                         break;
2105                     }
2106                 for(k=0; k<h->long_ref_count; k++)
2107                     if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2108                         id_list[i]= h->short_ref_count + k;
2109                         break;
2110                     }
2111             }
2112         }
2113
2114         ref2frm[0]=
2115         ref2frm[1]= -1;
2116         for(i=0; i<16; i++)
2117             ref2frm[i+2]= 4*id_list[i]
2118                           +(h->ref_list[j][i].reference&3);
2119         ref2frm[18+0]=
2120         ref2frm[18+1]= -1;
2121         for(i=16; i<48; i++)
2122             ref2frm[i+4]= 4*id_list[(i-16)>>1]
2123                           +(h->ref_list[j][i].reference&3);
2124     }
2125
2126     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
2127     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2128
2129     s->avctx->refs= h->sps.ref_frame_count;
2130
2131     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2132         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",
2133                h->slice_num,
2134                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2135                first_mb_in_slice,
2136                av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2137                pps_id, h->frame_num,
2138                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2139                h->ref_count[0], h->ref_count[1],
2140                s->qscale,
2141                h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
2142                h->use_weight,
2143                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2144                h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2145                );
2146     }
2147
2148     return 0;
2149 }
2150
2151 int ff_h264_get_slice_type(const H264Context *h)
2152 {
2153     switch (h->slice_type) {
2154     case FF_P_TYPE:  return 0;
2155     case FF_B_TYPE:  return 1;
2156     case FF_I_TYPE:  return 2;
2157     case FF_SP_TYPE: return 3;
2158     case FF_SI_TYPE: return 4;
2159     default:         return -1;
2160     }
2161 }
2162
2163 static void loop_filter(H264Context *h){
2164     MpegEncContext * const s = &h->s;
2165     uint8_t  *dest_y, *dest_cb, *dest_cr;
2166     int linesize, uvlinesize, mb_x, mb_y;
2167     const int end_mb_y= s->mb_y + FRAME_MBAFF;
2168     const int old_slice_type= h->slice_type;
2169
2170     if(h->deblocking_filter) {
2171         for(mb_x= 0; mb_x<s->mb_width; mb_x++){
2172             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
2173                 int list, mb_xy, mb_type;
2174                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2175                 h->slice_num= h->slice_table[mb_xy];
2176                 mb_type= s->current_picture.mb_type[mb_xy];
2177                 h->list_count= h->list_counts[mb_xy];
2178
2179                 if(FRAME_MBAFF)
2180                     h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2181
2182                 s->mb_x= mb_x;
2183                 s->mb_y= mb_y;
2184                 dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2185                 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2186                 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2187                     //FIXME simplify above
2188
2189                 if (MB_FIELD) {
2190                     linesize   = h->mb_linesize   = s->linesize * 2;
2191                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2192                     if(mb_y&1){ //FIXME move out of this function?
2193                         dest_y -= s->linesize*15;
2194                         dest_cb-= s->uvlinesize*7;
2195                         dest_cr-= s->uvlinesize*7;
2196                     }
2197                 } else {
2198                     linesize   = h->mb_linesize   = s->linesize;
2199                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2200                 }
2201                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2202                 if(fill_filter_caches(h, mb_type))
2203                     continue;
2204                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2205                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2206
2207                 if (FRAME_MBAFF) {
2208                     ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2209                 } else {
2210                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2211                 }
2212             }
2213         }
2214     }
2215     h->slice_type= old_slice_type;
2216     s->mb_x= 0;
2217     s->mb_y= end_mb_y - FRAME_MBAFF;
2218 }
2219
2220 static int decode_slice(struct AVCodecContext *avctx, void *arg){
2221     H264Context *h = *(void**)arg;
2222     MpegEncContext * const s = &h->s;
2223     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2224
2225     s->mb_skip_run= -1;
2226
2227     h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
2228                     (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
2229
2230     if( h->pps.cabac ) {
2231         /* realign */
2232         align_get_bits( &s->gb );
2233
2234         /* init cabac */
2235         ff_init_cabac_states( &h->cabac);
2236         ff_init_cabac_decoder( &h->cabac,
2237                                s->gb.buffer + get_bits_count(&s->gb)/8,
2238                                (get_bits_left(&s->gb) + 7)/8);
2239
2240         ff_h264_init_cabac_states(h);
2241
2242         for(;;){
2243 //START_TIMER
2244             int ret = ff_h264_decode_mb_cabac(h);
2245             int eos;
2246 //STOP_TIMER("decode_mb_cabac")
2247
2248             if(ret>=0) ff_h264_hl_decode_mb(h);
2249
2250             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
2251                 s->mb_y++;
2252
2253                 ret = ff_h264_decode_mb_cabac(h);
2254
2255                 if(ret>=0) ff_h264_hl_decode_mb(h);
2256                 s->mb_y--;
2257             }
2258             eos = get_cabac_terminate( &h->cabac );
2259
2260             if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
2261                 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);
2262                 return 0;
2263             }
2264             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2265                 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);
2266                 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);
2267                 return -1;
2268             }
2269
2270             if( ++s->mb_x >= s->mb_width ) {
2271                 s->mb_x = 0;
2272                 loop_filter(h);
2273                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2274                 ++s->mb_y;
2275                 if(FIELD_OR_MBAFF_PICTURE) {
2276                     ++s->mb_y;
2277                 }
2278             }
2279
2280             if( eos || s->mb_y >= s->mb_height ) {
2281                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2282                 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);
2283                 return 0;
2284             }
2285         }
2286
2287     } else {
2288         for(;;){
2289             int ret = ff_h264_decode_mb_cavlc(h);
2290
2291             if(ret>=0) ff_h264_hl_decode_mb(h);
2292
2293             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
2294                 s->mb_y++;
2295                 ret = ff_h264_decode_mb_cavlc(h);
2296
2297                 if(ret>=0) ff_h264_hl_decode_mb(h);
2298                 s->mb_y--;
2299             }
2300
2301             if(ret<0){
2302                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2303                 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);
2304
2305                 return -1;
2306             }
2307
2308             if(++s->mb_x >= s->mb_width){
2309                 s->mb_x=0;
2310                 loop_filter(h);
2311                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2312                 ++s->mb_y;
2313                 if(FIELD_OR_MBAFF_PICTURE) {
2314                     ++s->mb_y;
2315                 }
2316                 if(s->mb_y >= s->mb_height){
2317                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2318
2319                     if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
2320                         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);
2321
2322                         return 0;
2323                     }else{
2324                         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);
2325
2326                         return -1;
2327                     }
2328                 }
2329             }
2330
2331             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
2332                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2333                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2334                     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);
2335
2336                     return 0;
2337                 }else{
2338                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
2339
2340                     return -1;
2341                 }
2342             }
2343         }
2344     }
2345
2346 #if 0
2347     for(;s->mb_y < s->mb_height; s->mb_y++){
2348         for(;s->mb_x < s->mb_width; s->mb_x++){
2349             int ret= decode_mb(h);
2350
2351             ff_h264_hl_decode_mb(h);
2352
2353             if(ret<0){
2354                 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2355                 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);
2356
2357                 return -1;
2358             }
2359
2360             if(++s->mb_x >= s->mb_width){
2361                 s->mb_x=0;
2362                 if(++s->mb_y >= s->mb_height){
2363                     if(get_bits_count(s->gb) == s->gb.size_in_bits){
2364                         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);
2365
2366                         return 0;
2367                     }else{
2368                         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);
2369
2370                         return -1;
2371                     }
2372                 }
2373             }
2374
2375             if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2376                 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2377                     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);
2378
2379                     return 0;
2380                 }else{
2381                     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);
2382
2383                     return -1;
2384                 }
2385             }
2386         }
2387         s->mb_x=0;
2388         ff_draw_horiz_band(s, 16*s->mb_y, 16);
2389     }
2390 #endif
2391     return -1; //not reached
2392 }
2393
2394 /**
2395  * Call decode_slice() for each context.
2396  *
2397  * @param h h264 master context
2398  * @param context_count number of contexts to execute
2399  */
2400 static void execute_decode_slices(H264Context *h, int context_count){
2401     MpegEncContext * const s = &h->s;
2402     AVCodecContext * const avctx= s->avctx;
2403     H264Context *hx;
2404     int i;
2405
2406     if (s->avctx->hwaccel)
2407         return;
2408     if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2409         return;
2410     if(context_count == 1) {
2411         decode_slice(avctx, &h);
2412     } else {
2413         for(i = 1; i < context_count; i++) {
2414             hx = h->thread_context[i];
2415             hx->s.error_recognition = avctx->error_recognition;
2416             hx->s.error_count = 0;
2417         }
2418
2419         avctx->execute(avctx, (void *)decode_slice,
2420                        h->thread_context, NULL, context_count, sizeof(void*));
2421
2422         /* pull back stuff from slices to master context */
2423         hx = h->thread_context[context_count - 1];
2424         s->mb_x = hx->s.mb_x;
2425         s->mb_y = hx->s.mb_y;
2426         s->dropable = hx->s.dropable;
2427         s->picture_structure = hx->s.picture_structure;
2428         for(i = 1; i < context_count; i++)
2429             h->s.error_count += h->thread_context[i]->s.error_count;
2430     }
2431 }
2432
2433
2434 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
2435     MpegEncContext * const s = &h->s;
2436     AVCodecContext * const avctx= s->avctx;
2437     int buf_index=0;
2438     H264Context *hx; ///< thread context
2439     int context_count = 0;
2440     int next_avc= h->is_avc ? 0 : buf_size;
2441
2442     h->max_contexts = avctx->thread_count;
2443 #if 0
2444     int i;
2445     for(i=0; i<50; i++){
2446         av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
2447     }
2448 #endif
2449     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
2450         h->current_slice = 0;
2451         if (!s->first_field)
2452             s->current_picture_ptr= NULL;
2453         ff_h264_reset_sei(h);
2454     }
2455
2456     for(;;){
2457         int consumed;
2458         int dst_length;
2459         int bit_length;
2460         const uint8_t *ptr;
2461         int i, nalsize = 0;
2462         int err;
2463
2464         if(buf_index >= next_avc) {
2465             if(buf_index >= buf_size) break;
2466             nalsize = 0;
2467             for(i = 0; i < h->nal_length_size; i++)
2468                 nalsize = (nalsize << 8) | buf[buf_index++];
2469             if(nalsize <= 1 || nalsize > buf_size - buf_index){
2470                 if(nalsize == 1){
2471                     buf_index++;
2472                     continue;
2473                 }else{
2474                     av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2475                     break;
2476                 }
2477             }
2478             next_avc= buf_index + nalsize;
2479         } else {
2480             // start code prefix search
2481             for(; buf_index + 3 < next_avc; buf_index++){
2482                 // This should always succeed in the first iteration.
2483                 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2484                     break;
2485             }
2486
2487             if(buf_index+3 >= buf_size) break;
2488
2489             buf_index+=3;
2490             if(buf_index >= next_avc) continue;
2491         }
2492
2493         hx = h->thread_context[context_count];
2494
2495         ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
2496         if (ptr==NULL || dst_length < 0){
2497             return -1;
2498         }
2499         i= buf_index + consumed;
2500         if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
2501            buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
2502             s->workaround_bugs |= FF_BUG_TRUNCATED;
2503
2504         if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
2505         while(ptr[dst_length - 1] == 0 && dst_length > 0)
2506             dst_length--;
2507         }
2508         bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
2509
2510         if(s->avctx->debug&FF_DEBUG_STARTCODE){
2511             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);
2512         }
2513
2514         if (h->is_avc && (nalsize != consumed) && nalsize){
2515             av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
2516         }
2517
2518         buf_index += consumed;
2519
2520         if(  (s->hurry_up == 1 && h->nal_ref_idc  == 0) //FIXME do not discard SEI id
2521            ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0))
2522             continue;
2523
2524       again:
2525         err = 0;
2526         switch(hx->nal_unit_type){
2527         case NAL_IDR_SLICE:
2528             if (h->nal_unit_type != NAL_IDR_SLICE) {
2529                 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2530                 return -1;
2531             }
2532             idr(h); //FIXME ensure we don't loose some frames if there is reordering
2533         case NAL_SLICE:
2534             init_get_bits(&hx->s.gb, ptr, bit_length);
2535             hx->intra_gb_ptr=
2536             hx->inter_gb_ptr= &hx->s.gb;
2537             hx->s.data_partitioning = 0;
2538
2539             if((err = decode_slice_header(hx, h)))
2540                break;
2541
2542             avctx->profile = hx->sps.profile_idc;
2543             avctx->level   = hx->sps.level_idc;
2544
2545             if (s->avctx->hwaccel && h->current_slice == 1) {
2546                 if (s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
2547                     return -1;
2548             }
2549
2550             s->current_picture_ptr->key_frame |=
2551                     (hx->nal_unit_type == NAL_IDR_SLICE) ||
2552                     (h->sei_recovery_frame_cnt >= 0);
2553             if(hx->redundant_pic_count==0 && hx->s.hurry_up < 5
2554                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2555                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2556                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2557                && avctx->skip_frame < AVDISCARD_ALL){
2558                 if(avctx->hwaccel) {
2559                     if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
2560                         return -1;
2561                 }else
2562                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
2563                     static const uint8_t start_code[] = {0x00, 0x00, 0x01};
2564                     ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
2565                     ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
2566                 }else
2567                     context_count++;
2568             }
2569             break;
2570         case NAL_DPA:
2571             init_get_bits(&hx->s.gb, ptr, bit_length);
2572             hx->intra_gb_ptr=
2573             hx->inter_gb_ptr= NULL;
2574
2575             if ((err = decode_slice_header(hx, h)) < 0)
2576                 break;
2577
2578             avctx->profile = hx->sps.profile_idc;
2579             avctx->level   = hx->sps.level_idc;
2580
2581             hx->s.data_partitioning = 1;
2582
2583             break;
2584         case NAL_DPB:
2585             init_get_bits(&hx->intra_gb, ptr, bit_length);
2586             hx->intra_gb_ptr= &hx->intra_gb;
2587             break;
2588         case NAL_DPC:
2589             init_get_bits(&hx->inter_gb, ptr, bit_length);
2590             hx->inter_gb_ptr= &hx->inter_gb;
2591
2592             if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
2593                && s->context_initialized
2594                && s->hurry_up < 5
2595                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2596                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2597                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2598                && avctx->skip_frame < AVDISCARD_ALL)
2599                 context_count++;
2600             break;
2601         case NAL_SEI:
2602             init_get_bits(&s->gb, ptr, bit_length);
2603             ff_h264_decode_sei(h);
2604             break;
2605         case NAL_SPS:
2606             init_get_bits(&s->gb, ptr, bit_length);
2607             ff_h264_decode_seq_parameter_set(h);
2608
2609             if(s->flags& CODEC_FLAG_LOW_DELAY)
2610                 s->low_delay=1;
2611
2612             if(avctx->has_b_frames < 2)
2613                 avctx->has_b_frames= !s->low_delay;
2614             break;
2615         case NAL_PPS:
2616             init_get_bits(&s->gb, ptr, bit_length);
2617
2618             ff_h264_decode_picture_parameter_set(h, bit_length);
2619
2620             break;
2621         case NAL_AUD:
2622         case NAL_END_SEQUENCE:
2623         case NAL_END_STREAM:
2624         case NAL_FILLER_DATA:
2625         case NAL_SPS_EXT:
2626         case NAL_AUXILIARY_SLICE:
2627             break;
2628         default:
2629             av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
2630         }
2631
2632         if(context_count == h->max_contexts) {
2633             execute_decode_slices(h, context_count);
2634             context_count = 0;
2635         }
2636
2637         if (err < 0)
2638             av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
2639         else if(err == 1) {
2640             /* Slice could not be decoded in parallel mode, copy down
2641              * NAL unit stuff to context 0 and restart. Note that
2642              * rbsp_buffer is not transferred, but since we no longer
2643              * run in parallel mode this should not be an issue. */
2644             h->nal_unit_type = hx->nal_unit_type;
2645             h->nal_ref_idc   = hx->nal_ref_idc;
2646             hx = h;
2647             goto again;
2648         }
2649     }
2650     if(context_count)
2651         execute_decode_slices(h, context_count);
2652     return buf_index;
2653 }
2654
2655 /**
2656  * returns the number of bytes consumed for building the current frame
2657  */
2658 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
2659         if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
2660         if(pos+10>buf_size) pos=buf_size; // oops ;)
2661
2662         return pos;
2663 }
2664
2665 static int decode_frame(AVCodecContext *avctx,
2666                              void *data, int *data_size,
2667                              AVPacket *avpkt)
2668 {
2669     const uint8_t *buf = avpkt->data;
2670     int buf_size = avpkt->size;
2671     H264Context *h = avctx->priv_data;
2672     MpegEncContext *s = &h->s;
2673     AVFrame *pict = data;
2674     int buf_index;
2675
2676     s->flags= avctx->flags;
2677     s->flags2= avctx->flags2;
2678
2679    /* end of stream, output what is still in the buffers */
2680     if (buf_size == 0) {
2681         Picture *out;
2682         int i, out_idx;
2683
2684 //FIXME factorize this with the output code below
2685         out = h->delayed_pic[0];
2686         out_idx = 0;
2687         for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2688             if(h->delayed_pic[i]->poc < out->poc){
2689                 out = h->delayed_pic[i];
2690                 out_idx = i;
2691             }
2692
2693         for(i=out_idx; h->delayed_pic[i]; i++)
2694             h->delayed_pic[i] = h->delayed_pic[i+1];
2695
2696         if(out){
2697             *data_size = sizeof(AVFrame);
2698             *pict= *(AVFrame*)out;
2699         }
2700
2701         return 0;
2702     }
2703
2704     if(h->is_avc && !h->got_avcC) {
2705         int i, cnt, nalsize;
2706         unsigned char *p = avctx->extradata;
2707         if(avctx->extradata_size < 7) {
2708             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
2709             return -1;
2710         }
2711         if(*p != 1) {
2712             av_log(avctx, AV_LOG_ERROR, "Unknown avcC version %d\n", *p);
2713             return -1;
2714         }
2715         /* sps and pps in the avcC always have length coded with 2 bytes,
2716            so put a fake nal_length_size = 2 while parsing them */
2717         h->nal_length_size = 2;
2718         // Decode sps from avcC
2719         cnt = *(p+5) & 0x1f; // Number of sps
2720         p += 6;
2721         for (i = 0; i < cnt; i++) {
2722             nalsize = AV_RB16(p) + 2;
2723             if(decode_nal_units(h, p, nalsize) < 0) {
2724                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
2725                 return -1;
2726             }
2727             p += nalsize;
2728         }
2729         // Decode pps from avcC
2730         cnt = *(p++); // Number of pps
2731         for (i = 0; i < cnt; i++) {
2732             nalsize = AV_RB16(p) + 2;
2733             if(decode_nal_units(h, p, nalsize)  != nalsize) {
2734                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
2735                 return -1;
2736             }
2737             p += nalsize;
2738         }
2739         // Now store right nal length size, that will be use to parse all other nals
2740         h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
2741         // Do not reparse avcC
2742         h->got_avcC = 1;
2743     }
2744
2745     if(!h->got_avcC && !h->is_avc && s->avctx->extradata_size){
2746         if(decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) < 0)
2747             return -1;
2748         h->got_avcC = 1;
2749     }
2750
2751     buf_index=decode_nal_units(h, buf, buf_size);
2752     if(buf_index < 0)
2753         return -1;
2754
2755     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
2756         if (avctx->skip_frame >= AVDISCARD_NONREF || s->hurry_up) return 0;
2757         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
2758         return -1;
2759     }
2760
2761     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
2762         Picture *out = s->current_picture_ptr;
2763         Picture *cur = s->current_picture_ptr;
2764         int i, pics, out_of_order, out_idx;
2765
2766         field_end(h);
2767
2768         if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
2769             /* Wait for second field. */
2770             *data_size = 0;
2771
2772         } else {
2773             cur->interlaced_frame = 0;
2774             cur->repeat_pict = 0;
2775
2776             /* Signal interlacing information externally. */
2777             /* Prioritize picture timing SEI information over used decoding process if it exists. */
2778
2779             if(h->sps.pic_struct_present_flag){
2780                 switch (h->sei_pic_struct)
2781                 {
2782                 case SEI_PIC_STRUCT_FRAME:
2783                     break;
2784                 case SEI_PIC_STRUCT_TOP_FIELD:
2785                 case SEI_PIC_STRUCT_BOTTOM_FIELD:
2786                     cur->interlaced_frame = 1;
2787                     break;
2788                 case SEI_PIC_STRUCT_TOP_BOTTOM:
2789                 case SEI_PIC_STRUCT_BOTTOM_TOP:
2790                     if (FIELD_OR_MBAFF_PICTURE)
2791                         cur->interlaced_frame = 1;
2792                     else
2793                         // try to flag soft telecine progressive
2794                         cur->interlaced_frame = h->prev_interlaced_frame;
2795                     break;
2796                 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
2797                 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
2798                     // Signal the possibility of telecined film externally (pic_struct 5,6)
2799                     // From these hints, let the applications decide if they apply deinterlacing.
2800                     cur->repeat_pict = 1;
2801                     break;
2802                 case SEI_PIC_STRUCT_FRAME_DOUBLING:
2803                     // Force progressive here, as doubling interlaced frame is a bad idea.
2804                     cur->repeat_pict = 2;
2805                     break;
2806                 case SEI_PIC_STRUCT_FRAME_TRIPLING:
2807                     cur->repeat_pict = 4;
2808                     break;
2809                 }
2810
2811                 if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
2812                     cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
2813             }else{
2814                 /* Derive interlacing flag from used decoding process. */
2815                 cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
2816             }
2817             h->prev_interlaced_frame = cur->interlaced_frame;
2818
2819             if (cur->field_poc[0] != cur->field_poc[1]){
2820                 /* Derive top_field_first from field pocs. */
2821                 cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
2822             }else{
2823                 if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
2824                     /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
2825                     if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
2826                       || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
2827                         cur->top_field_first = 1;
2828                     else
2829                         cur->top_field_first = 0;
2830                 }else{
2831                     /* Most likely progressive */
2832                     cur->top_field_first = 0;
2833                 }
2834             }
2835
2836         //FIXME do something with unavailable reference frames
2837
2838             /* Sort B-frames into display order */
2839
2840             if(h->sps.bitstream_restriction_flag
2841                && s->avctx->has_b_frames < h->sps.num_reorder_frames){
2842                 s->avctx->has_b_frames = h->sps.num_reorder_frames;
2843                 s->low_delay = 0;
2844             }
2845
2846             if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
2847                && !h->sps.bitstream_restriction_flag){
2848                 s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
2849                 s->low_delay= 0;
2850             }
2851
2852             pics = 0;
2853             while(h->delayed_pic[pics]) pics++;
2854
2855             assert(pics <= MAX_DELAYED_PIC_COUNT);
2856
2857             h->delayed_pic[pics++] = cur;
2858             if(cur->reference == 0)
2859                 cur->reference = DELAYED_PIC_REF;
2860
2861             out = h->delayed_pic[0];
2862             out_idx = 0;
2863             for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2864                 if(h->delayed_pic[i]->poc < out->poc){
2865                     out = h->delayed_pic[i];
2866                     out_idx = i;
2867                 }
2868             if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
2869                 h->outputed_poc= INT_MIN;
2870             out_of_order = out->poc < h->outputed_poc;
2871
2872             if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
2873                 { }
2874             else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
2875                || (s->low_delay &&
2876                 ((h->outputed_poc != INT_MIN && out->poc > h->outputed_poc + 2)
2877                  || cur->pict_type == FF_B_TYPE)))
2878             {
2879                 s->low_delay = 0;
2880                 s->avctx->has_b_frames++;
2881             }
2882
2883             if(out_of_order || pics > s->avctx->has_b_frames){
2884                 out->reference &= ~DELAYED_PIC_REF;
2885                 for(i=out_idx; h->delayed_pic[i]; i++)
2886                     h->delayed_pic[i] = h->delayed_pic[i+1];
2887             }
2888             if(!out_of_order && pics > s->avctx->has_b_frames){
2889                 *data_size = sizeof(AVFrame);
2890
2891                 if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
2892                     h->outputed_poc = INT_MIN;
2893                 } else
2894                     h->outputed_poc = out->poc;
2895                 *pict= *(AVFrame*)out;
2896             }else{
2897                 av_log(avctx, AV_LOG_DEBUG, "no picture\n");
2898             }
2899         }
2900     }
2901
2902     assert(pict->data[0] || !*data_size);
2903     ff_print_debug_info(s, pict);
2904 //printf("out %d\n", (int)pict->data[0]);
2905
2906     return get_consumed_bytes(s, buf_index, buf_size);
2907 }
2908 #if 0
2909 static inline void fill_mb_avail(H264Context *h){
2910     MpegEncContext * const s = &h->s;
2911     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2912
2913     if(s->mb_y){
2914         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
2915         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
2916         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
2917     }else{
2918         h->mb_avail[0]=
2919         h->mb_avail[1]=
2920         h->mb_avail[2]= 0;
2921     }
2922     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
2923     h->mb_avail[4]= 1; //FIXME move out
2924     h->mb_avail[5]= 0; //FIXME move out
2925 }
2926 #endif
2927
2928 #ifdef TEST
2929 #undef printf
2930 #undef random
2931 #define COUNT 8000
2932 #define SIZE (COUNT*40)
2933 int main(void){
2934     int i;
2935     uint8_t temp[SIZE];
2936     PutBitContext pb;
2937     GetBitContext gb;
2938 //    int int_temp[10000];
2939     DSPContext dsp;
2940     AVCodecContext avctx;
2941
2942     dsputil_init(&dsp, &avctx);
2943
2944     init_put_bits(&pb, temp, SIZE);
2945     printf("testing unsigned exp golomb\n");
2946     for(i=0; i<COUNT; i++){
2947         START_TIMER
2948         set_ue_golomb(&pb, i);
2949         STOP_TIMER("set_ue_golomb");
2950     }
2951     flush_put_bits(&pb);
2952
2953     init_get_bits(&gb, temp, 8*SIZE);
2954     for(i=0; i<COUNT; i++){
2955         int j, s;
2956
2957         s= show_bits(&gb, 24);
2958
2959         START_TIMER
2960         j= get_ue_golomb(&gb);
2961         if(j != i){
2962             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
2963 //            return -1;
2964         }
2965         STOP_TIMER("get_ue_golomb");
2966     }
2967
2968
2969     init_put_bits(&pb, temp, SIZE);
2970     printf("testing signed exp golomb\n");
2971     for(i=0; i<COUNT; i++){
2972         START_TIMER
2973         set_se_golomb(&pb, i - COUNT/2);
2974         STOP_TIMER("set_se_golomb");
2975     }
2976     flush_put_bits(&pb);
2977
2978     init_get_bits(&gb, temp, 8*SIZE);
2979     for(i=0; i<COUNT; i++){
2980         int j, s;
2981
2982         s= show_bits(&gb, 24);
2983
2984         START_TIMER
2985         j= get_se_golomb(&gb);
2986         if(j != i - COUNT/2){
2987             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
2988 //            return -1;
2989         }
2990         STOP_TIMER("get_se_golomb");
2991     }
2992
2993 #if 0
2994     printf("testing 4x4 (I)DCT\n");
2995
2996     DCTELEM block[16];
2997     uint8_t src[16], ref[16];
2998     uint64_t error= 0, max_error=0;
2999
3000     for(i=0; i<COUNT; i++){
3001         int j;
3002 //        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
3003         for(j=0; j<16; j++){
3004             ref[j]= random()%255;
3005             src[j]= random()%255;
3006         }
3007
3008         h264_diff_dct_c(block, src, ref, 4);
3009
3010         //normalize
3011         for(j=0; j<16; j++){
3012 //            printf("%d ", block[j]);
3013             block[j]= block[j]*4;
3014             if(j&1) block[j]= (block[j]*4 + 2)/5;
3015             if(j&4) block[j]= (block[j]*4 + 2)/5;
3016         }
3017 //        printf("\n");
3018
3019         s->dsp.h264_idct_add(ref, block, 4);
3020 /*        for(j=0; j<16; j++){
3021             printf("%d ", ref[j]);
3022         }
3023         printf("\n");*/
3024
3025         for(j=0; j<16; j++){
3026             int diff= FFABS(src[j] - ref[j]);
3027
3028             error+= diff*diff;
3029             max_error= FFMAX(max_error, diff);
3030         }
3031     }
3032     printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
3033     printf("testing quantizer\n");
3034     for(qp=0; qp<52; qp++){
3035         for(i=0; i<16; i++)
3036             src1_block[i]= src2_block[i]= random()%255;
3037
3038     }
3039     printf("Testing NAL layer\n");
3040
3041     uint8_t bitstream[COUNT];
3042     uint8_t nal[COUNT*2];
3043     H264Context h;
3044     memset(&h, 0, sizeof(H264Context));
3045
3046     for(i=0; i<COUNT; i++){
3047         int zeros= i;
3048         int nal_length;
3049         int consumed;
3050         int out_length;
3051         uint8_t *out;
3052         int j;
3053
3054         for(j=0; j<COUNT; j++){
3055             bitstream[j]= (random() % 255) + 1;
3056         }
3057
3058         for(j=0; j<zeros; j++){
3059             int pos= random() % COUNT;
3060             while(bitstream[pos] == 0){
3061                 pos++;
3062                 pos %= COUNT;
3063             }
3064             bitstream[pos]=0;
3065         }
3066
3067         START_TIMER
3068
3069         nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
3070         if(nal_length<0){
3071             printf("encoding failed\n");
3072             return -1;
3073         }
3074
3075         out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
3076
3077         STOP_TIMER("NAL")
3078
3079         if(out_length != COUNT){
3080             printf("incorrect length %d %d\n", out_length, COUNT);
3081             return -1;
3082         }
3083
3084         if(consumed != nal_length){
3085             printf("incorrect consumed length %d %d\n", nal_length, consumed);
3086             return -1;
3087         }
3088
3089         if(memcmp(bitstream, out, COUNT)){
3090             printf("mismatch\n");
3091             return -1;
3092         }
3093     }
3094 #endif
3095
3096     printf("Testing RBSP\n");
3097
3098
3099     return 0;
3100 }
3101 #endif /* TEST */
3102
3103
3104 av_cold void ff_h264_free_context(H264Context *h)
3105 {
3106     int i;
3107
3108     free_tables(h); //FIXME cleanup init stuff perhaps
3109
3110     for(i = 0; i < MAX_SPS_COUNT; i++)
3111         av_freep(h->sps_buffers + i);
3112
3113     for(i = 0; i < MAX_PPS_COUNT; i++)
3114         av_freep(h->pps_buffers + i);
3115 }
3116
3117 av_cold int ff_h264_decode_end(AVCodecContext *avctx)
3118 {
3119     H264Context *h = avctx->priv_data;
3120     MpegEncContext *s = &h->s;
3121
3122     ff_h264_free_context(h);
3123
3124     MPV_common_end(s);
3125
3126 //    memset(h, 0, sizeof(H264Context));
3127
3128     return 0;
3129 }
3130
3131
3132 AVCodec h264_decoder = {
3133     "h264",
3134     CODEC_TYPE_VIDEO,
3135     CODEC_ID_H264,
3136     sizeof(H264Context),
3137     ff_h264_decode_init,
3138     NULL,
3139     ff_h264_decode_end,
3140     decode_frame,
3141     /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3142     .flush= flush_dpb,
3143     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
3144     .pix_fmts= ff_hwaccel_pixfmt_list_420,
3145 };
3146
3147 #if CONFIG_H264_VDPAU_DECODER
3148 AVCodec h264_vdpau_decoder = {
3149     "h264_vdpau",
3150     CODEC_TYPE_VIDEO,
3151     CODEC_ID_H264,
3152     sizeof(H264Context),
3153     ff_h264_decode_init,
3154     NULL,
3155     ff_h264_decode_end,
3156     decode_frame,
3157     CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3158     .flush= flush_dpb,
3159     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
3160     .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
3161 };
3162 #endif