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     int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
  56
  57     AV_COPY32(mode, h->intra4x4_pred_mode_cache + 4 + 8*4);
  58     mode[4]= h->intra4x4_pred_mode_cache[7+8*3];
  59     mode[5]= h->intra4x4_pred_mode_cache[7+8*2];
  60     mode[6]= h->intra4x4_pred_mode_cache[7+8*1];
  61 }
  62
  63 /**
  64  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
  65  */
  66 int ff_h264_check_intra4x4_pred_mode(H264Context *h){
  67     MpegEncContext * const s = &h->s;
  68     static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
  69     static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
  70     int i;
  71
  72     if(!(h->top_samples_available&0x8000)){
  73         for(i=0; i<4; i++){
  74             int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
  75             if(status<0){
  76                 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
  77                 return -1;
  78             } else if(status){
  79                 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
  80             }
  81         }
  82     }
  83
  84     if((h->left_samples_available&0x8888)!=0x8888){
  85         static const int mask[4]={0x8000,0x2000,0x80,0x20};
  86         for(i=0; i<4; i++){
  87             if(!(h->left_samples_available&mask[i])){
  88                 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
  89                 if(status<0){
  90                     av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
  91                     return -1;
  92                 } else if(status){
  93                     h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
  94                 }
  95             }
  96         }
  97     }
  98
  99     return 0;
 100 } //FIXME cleanup like ff_h264_check_intra_pred_mode
 101
 102 /**
 103  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
 104  */
 105 int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
 106     MpegEncContext * const s = &h->s;
 107     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
 108     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
 109
 110     if(mode > 6U) {
 111         av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
 112         return -1;
 113     }
 114
 115     if(!(h->top_samples_available&0x8000)){
 116         mode= top[ mode ];
 117         if(mode<0){
 118             av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
 119             return -1;
 120         }
 121     }
 122
 123     if((h->left_samples_available&0x8080) != 0x8080){
 124         mode= left[ mode ];
 125         if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
 126             mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
 127         }
 128         if(mode<0){
 129             av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
 130             return -1;
 131         }
 132     }
 133
 134     return mode;
 135 }
 136
 137 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
 138     int i, si, di;
 139     uint8_t *dst;
 140     int bufidx;
 141
 142 //    src[0]&0x80;                //forbidden bit
 143     h->nal_ref_idc= src[0]>>5;
 144     h->nal_unit_type= src[0]&0x1F;
 145
 146     src++; length--;
 147 #if 0
 148     for(i=0; i<length; i++)
 149         printf("%2X ", src[i]);
 150 #endif
 151
 152 #if HAVE_FAST_UNALIGNED
 153 # if HAVE_FAST_64BIT
 154 #   define RS 7
 155     for(i=0; i+1<length; i+=9){
 156         if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
 157 # else
 158 #   define RS 3
 159     for(i=0; i+1<length; i+=5){
 160         if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
 161 # endif
 162             continue;
 163         if(i>0 && !src[i]) i--;
 164         while(src[i]) i++;
 165 #else
 166 #   define RS 0
 167     for(i=0; i+1<length; i+=2){
 168         if(src[i]) continue;
 169         if(i>0 && src[i-1]==0) i--;
 170 #endif
 171         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
 172             if(src[i+2]!=3){
 173                 /* startcode, so we must be past the end */
 174                 length=i;
 175             }
 176             break;
 177         }
 178         i-= RS;
 179     }
 180
 181     if(i>=length-1){ //no escaped 0
 182         *dst_length= length;
 183         *consumed= length+1; //+1 for the header
 184         return src;
 185     }
 186
 187     bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
 188     av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
 189     dst= h->rbsp_buffer[bufidx];
 190
 191     if (dst == NULL){
 192         return NULL;
 193     }
 194
 195 //printf("decoding esc\n");
 196     memcpy(dst, src, i);
 197     si=di=i;
 198     while(si+2<length){
 199         //remove escapes (very rare 1:2^22)
 200         if(src[si+2]>3){
 201             dst[di++]= src[si++];
 202             dst[di++]= src[si++];
 203         }else if(src[si]==0 && src[si+1]==0){
 204             if(src[si+2]==3){ //escape
 205                 dst[di++]= 0;
 206                 dst[di++]= 0;
 207                 si+=3;
 208                 continue;
 209             }else //next start code
 210                 goto nsc;
 211         }
 212
 213         dst[di++]= src[si++];
 214     }
 215     while(si<length)
 216         dst[di++]= src[si++];
 217 nsc:
 218
 219     memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
 220
 221     *dst_length= di;
 222     *consumed= si + 1;//+1 for the header
 223 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
 224     return dst;
 225 }
 226
 227 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
 228     int v= *src;
 229     int r;
 230
 231     tprintf(h->s.avctx, "rbsp trailing %X\n", v);
 232
 233     for(r=1; r<9; r++){
 234         if(v&1) return r;
 235         v>>=1;
 236     }
 237     return 0;
 238 }
 239
 240 /**
 241  * IDCT transforms the 16 dc values and dequantizes them.
 242  * @param qp quantization parameter
 243  */
 244 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
 245 #define stride 16
 246     int i;
 247     int temp[16]; //FIXME check if this is a good idea
 248     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
 249     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
 250
 251 //memset(block, 64, 2*256);
 252 //return;
 253     for(i=0; i<4; i++){
 254         const int offset= y_offset[i];
 255         const int z0= block[offset+stride*0] + block[offset+stride*4];
 256         const int z1= block[offset+stride*0] - block[offset+stride*4];
 257         const int z2= block[offset+stride*1] - block[offset+stride*5];
 258         const int z3= block[offset+stride*1] + block[offset+stride*5];
 259
 260         temp[4*i+0]= z0+z3;
 261         temp[4*i+1]= z1+z2;
 262         temp[4*i+2]= z1-z2;
 263         temp[4*i+3]= z0-z3;
 264     }
 265
 266     for(i=0; i<4; i++){
 267         const int offset= x_offset[i];
 268         const int z0= temp[4*0+i] + temp[4*2+i];
 269         const int z1= temp[4*0+i] - temp[4*2+i];
 270         const int z2= temp[4*1+i] - temp[4*3+i];
 271         const int z3= temp[4*1+i] + temp[4*3+i];
 272
 273         block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
 274         block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
 275         block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
 276         block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
 277     }
 278 }
 279
 280 #if 0
 281 /**
 282  * DCT transforms the 16 dc values.
 283  * @param qp quantization parameter ??? FIXME
 284  */
 285 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
 286 //    const int qmul= dequant_coeff[qp][0];
 287     int i;
 288     int temp[16]; //FIXME check if this is a good idea
 289     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
 290     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
 291
 292     for(i=0; i<4; i++){
 293         const int offset= y_offset[i];
 294         const int z0= block[offset+stride*0] + block[offset+stride*4];
 295         const int z1= block[offset+stride*0] - block[offset+stride*4];
 296         const int z2= block[offset+stride*1] - block[offset+stride*5];
 297         const int z3= block[offset+stride*1] + block[offset+stride*5];
 298
 299         temp[4*i+0]= z0+z3;
 300         temp[4*i+1]= z1+z2;
 301         temp[4*i+2]= z1-z2;
 302         temp[4*i+3]= z0-z3;
 303     }
 304
 305     for(i=0; i<4; i++){
 306         const int offset= x_offset[i];
 307         const int z0= temp[4*0+i] + temp[4*2+i];
 308         const int z1= temp[4*0+i] - temp[4*2+i];
 309         const int z2= temp[4*1+i] - temp[4*3+i];
 310         const int z3= temp[4*1+i] + temp[4*3+i];
 311
 312         block[stride*0 +offset]= (z0 + z3)>>1;
 313         block[stride*2 +offset]= (z1 + z2)>>1;
 314         block[stride*8 +offset]= (z1 - z2)>>1;
 315         block[stride*10+offset]= (z0 - z3)>>1;
 316     }
 317 }
 318 #endif
 319
 320 #undef xStride
 321 #undef stride
 322
 323 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
 324     const int stride= 16*2;
 325     const int xStride= 16;
 326     int a,b,c,d,e;
 327
 328     a= block[stride*0 + xStride*0];
 329     b= block[stride*0 + xStride*1];
 330     c= block[stride*1 + xStride*0];
 331     d= block[stride*1 + xStride*1];
 332
 333     e= a-b;
 334     a= a+b;
 335     b= c-d;
 336     c= c+d;
 337
 338     block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
 339     block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
 340     block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
 341     block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
 342 }
 343
 344 #if 0
 345 static void chroma_dc_dct_c(DCTELEM *block){
 346     const int stride= 16*2;
 347     const int xStride= 16;
 348     int a,b,c,d,e;
 349
 350     a= block[stride*0 + xStride*0];
 351     b= block[stride*0 + xStride*1];
 352     c= block[stride*1 + xStride*0];
 353     d= block[stride*1 + xStride*1];
 354
 355     e= a-b;
 356     a= a+b;
 357     b= c-d;
 358     c= c+d;
 359
 360     block[stride*0 + xStride*0]= (a+c);
 361     block[stride*0 + xStride*1]= (e+b);
 362     block[stride*1 + xStride*0]= (a-c);
 363     block[stride*1 + xStride*1]= (e-b);
 364 }
 365 #endif
 366
 367 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
 368                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 369                            int src_x_offset, int src_y_offset,
 370                            qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
 371     MpegEncContext * const s = &h->s;
 372     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
 373     int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
 374     const int luma_xy= (mx&3) + ((my&3)<<2);
 375     uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
 376     uint8_t * src_cb, * src_cr;
 377     int extra_width= h->emu_edge_width;
 378     int extra_height= h->emu_edge_height;
 379     int emu=0;
 380     const int full_mx= mx>>2;
 381     const int full_my= my>>2;
 382     const int pic_width  = 16*s->mb_width;
 383     const int pic_height = 16*s->mb_height >> MB_FIELD;
 384
 385     if(mx&7) extra_width -= 3;
 386     if(my&7) extra_height -= 3;
 387
 388     if(   full_mx < 0-extra_width
 389        || full_my < 0-extra_height
 390        || full_mx + 16/*FIXME*/ > pic_width + extra_width
 391        || full_my + 16/*FIXME*/ > pic_height + extra_height){
 392         ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
 393             src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
 394         emu=1;
 395     }
 396
 397     qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
 398     if(!square){
 399         qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
 400     }
 401
 402     if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
 403
 404     if(MB_FIELD){
 405         // chroma offset when predicting from a field of opposite parity
 406         my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
 407         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
 408     }
 409     src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
 410     src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
 411
 412     if(emu){
 413         ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
 414             src_cb= s->edge_emu_buffer;
 415     }
 416     chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
 417
 418     if(emu){
 419         ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
 420             src_cr= s->edge_emu_buffer;
 421     }
 422     chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
 423 }
 424
 425 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
 426                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 427                            int x_offset, int y_offset,
 428                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 429                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
 430                            int list0, int list1){
 431     MpegEncContext * const s = &h->s;
 432     qpel_mc_func *qpix_op=  qpix_put;
 433     h264_chroma_mc_func chroma_op= chroma_put;
 434
 435     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
 436     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
 437     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
 438     x_offset += 8*s->mb_x;
 439     y_offset += 8*(s->mb_y >> MB_FIELD);
 440
 441     if(list0){
 442         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
 443         mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
 444                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
 445                            qpix_op, chroma_op);
 446
 447         qpix_op=  qpix_avg;
 448         chroma_op= chroma_avg;
 449     }
 450
 451     if(list1){
 452         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
 453         mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
 454                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
 455                            qpix_op, chroma_op);
 456     }
 457 }
 458
 459 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
 460                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 461                            int x_offset, int y_offset,
 462                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 463                            h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
 464                            h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
 465                            int list0, int list1){
 466     MpegEncContext * const s = &h->s;
 467
 468     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
 469     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
 470     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
 471     x_offset += 8*s->mb_x;
 472     y_offset += 8*(s->mb_y >> MB_FIELD);
 473
 474     if(list0 && list1){
 475         /* don't optimize for luma-only case, since B-frames usually
 476          * use implicit weights => chroma too. */
 477         uint8_t *tmp_cb = s->obmc_scratchpad;
 478         uint8_t *tmp_cr = s->obmc_scratchpad + 8;
 479         uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
 480         int refn0 = h->ref_cache[0][ scan8[n] ];
 481         int refn1 = h->ref_cache[1][ scan8[n] ];
 482
 483         mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
 484                     dest_y, dest_cb, dest_cr,
 485                     x_offset, y_offset, qpix_put, chroma_put);
 486         mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
 487                     tmp_y, tmp_cb, tmp_cr,
 488                     x_offset, y_offset, qpix_put, chroma_put);
 489
 490         if(h->use_weight == 2){
 491             int weight0 = h->implicit_weight[refn0][refn1];
 492             int weight1 = 64 - weight0;
 493             luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
 494             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
 495             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
 496         }else{
 497             luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
 498                             h->luma_weight[0][refn0], h->luma_weight[1][refn1],
 499                             h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
 500             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 501                             h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
 502                             h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
 503             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 504                             h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
 505                             h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
 506         }
 507     }else{
 508         int list = list1 ? 1 : 0;
 509         int refn = h->ref_cache[list][ scan8[n] ];
 510         Picture *ref= &h->ref_list[list][refn];
 511         mc_dir_part(h, ref, n, square, chroma_height, delta, list,
 512                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
 513                     qpix_put, chroma_put);
 514
 515         luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
 516                        h->luma_weight[list][refn], h->luma_offset[list][refn]);
 517         if(h->use_weight_chroma){
 518             chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 519                              h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
 520             chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 521                              h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
 522         }
 523     }
 524 }
 525
 526 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
 527                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 528                            int x_offset, int y_offset,
 529                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 530                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
 531                            h264_weight_func *weight_op, h264_biweight_func *weight_avg,
 532                            int list0, int list1){
 533     if((h->use_weight==2 && list0 && list1
 534         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
 535        || h->use_weight==1)
 536         mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
 537                          x_offset, y_offset, qpix_put, chroma_put,
 538                          weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
 539     else
 540         mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
 541                     x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
 542 }
 543
 544 static inline void prefetch_motion(H264Context *h, int list){
 545     /* fetch pixels for estimated mv 4 macroblocks ahead
 546      * optimized for 64byte cache lines */
 547     MpegEncContext * const s = &h->s;
 548     const int refn = h->ref_cache[list][scan8[0]];
 549     if(refn >= 0){
 550         const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
 551         const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
 552         uint8_t **src= h->ref_list[list][refn].data;
 553         int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
 554         s->dsp.prefetch(src[0]+off, s->linesize, 4);
 555         off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
 556         s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
 557     }
 558 }
 559
 560 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 561                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
 562                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
 563                       h264_weight_func *weight_op, h264_biweight_func *weight_avg){
 564     MpegEncContext * const s = &h->s;
 565     const int mb_xy= h->mb_xy;
 566     const int mb_type= s->current_picture.mb_type[mb_xy];
 567
 568     assert(IS_INTER(mb_type));
 569
 570     prefetch_motion(h, 0);
 571
 572     if(IS_16X16(mb_type)){
 573         mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
 574                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
 575                 weight_op, weight_avg,
 576                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
 577     }else if(IS_16X8(mb_type)){
 578         mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
 579                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
 580                 &weight_op[1], &weight_avg[1],
 581                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
 582         mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
 583                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
 584                 &weight_op[1], &weight_avg[1],
 585                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
 586     }else if(IS_8X16(mb_type)){
 587         mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
 588                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 589                 &weight_op[2], &weight_avg[2],
 590                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
 591         mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
 592                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 593                 &weight_op[2], &weight_avg[2],
 594                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
 595     }else{
 596         int i;
 597
 598         assert(IS_8X8(mb_type));
 599
 600         for(i=0; i<4; i++){
 601             const int sub_mb_type= h->sub_mb_type[i];
 602             const int n= 4*i;
 603             int x_offset= (i&1)<<2;
 604             int y_offset= (i&2)<<1;
 605
 606             if(IS_SUB_8X8(sub_mb_type)){
 607                 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 608                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 609                     &weight_op[3], &weight_avg[3],
 610                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 611             }else if(IS_SUB_8X4(sub_mb_type)){
 612                 mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 613                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
 614                     &weight_op[4], &weight_avg[4],
 615                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 616                 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
 617                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
 618                     &weight_op[4], &weight_avg[4],
 619                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 620             }else if(IS_SUB_4X8(sub_mb_type)){
 621                 mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 622                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 623                     &weight_op[5], &weight_avg[5],
 624                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 625                 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
 626                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 627                     &weight_op[5], &weight_avg[5],
 628                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 629             }else{
 630                 int j;
 631                 assert(IS_SUB_4X4(sub_mb_type));
 632                 for(j=0; j<4; j++){
 633                     int sub_x_offset= x_offset + 2*(j&1);
 634                     int sub_y_offset= y_offset +   (j&2);
 635                     mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
 636                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 637                         &weight_op[6], &weight_avg[6],
 638                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 639                 }
 640             }
 641         }
 642     }
 643
 644     prefetch_motion(h, 1);
 645 }
 646
 647
 648 static void free_tables(H264Context *h){
 649     int i;
 650     H264Context *hx;
 651     av_freep(&h->intra4x4_pred_mode);
 652     av_freep(&h->chroma_pred_mode_table);
 653     av_freep(&h->cbp_table);
 654     av_freep(&h->mvd_table[0]);
 655     av_freep(&h->mvd_table[1]);
 656     av_freep(&h->direct_table);
 657     av_freep(&h->non_zero_count);
 658     av_freep(&h->slice_table_base);
 659     h->slice_table= NULL;
 660     av_freep(&h->list_counts);
 661
 662     av_freep(&h->mb2b_xy);
 663     av_freep(&h->mb2br_xy);
 664
 665     for(i = 0; i < MAX_THREADS; i++) {
 666         hx = h->thread_context[i];
 667         if(!hx) continue;
 668         av_freep(&hx->top_borders[1]);
 669         av_freep(&hx->top_borders[0]);
 670         av_freep(&hx->s.obmc_scratchpad);
 671         av_freep(&hx->rbsp_buffer[1]);
 672         av_freep(&hx->rbsp_buffer[0]);
 673         hx->rbsp_buffer_size[0] = 0;
 674         hx->rbsp_buffer_size[1] = 0;
 675         if (i) av_freep(&h->thread_context[i]);
 676     }
 677 }
 678
 679 static void init_dequant8_coeff_table(H264Context *h){
 680     int i,q,x;
 681     const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
 682     h->dequant8_coeff[0] = h->dequant8_buffer[0];
 683     h->dequant8_coeff[1] = h->dequant8_buffer[1];
 684
 685     for(i=0; i<2; i++ ){
 686         if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
 687             h->dequant8_coeff[1] = h->dequant8_buffer[0];
 688             break;
 689         }
 690
 691         for(q=0; q<52; q++){
 692             int shift = div6[q];
 693             int idx = rem6[q];
 694             for(x=0; x<64; x++)
 695                 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
 696                     ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
 697                     h->pps.scaling_matrix8[i][x]) << shift;
 698         }
 699     }
 700 }
 701
 702 static void init_dequant4_coeff_table(H264Context *h){
 703     int i,j,q,x;
 704     const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
 705     for(i=0; i<6; i++ ){
 706         h->dequant4_coeff[i] = h->dequant4_buffer[i];
 707         for(j=0; j<i; j++){
 708             if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
 709                 h->dequant4_coeff[i] = h->dequant4_buffer[j];
 710                 break;
 711             }
 712         }
 713         if(j<i)
 714             continue;
 715
 716         for(q=0; q<52; q++){
 717             int shift = div6[q] + 2;
 718             int idx = rem6[q];
 719             for(x=0; x<16; x++)
 720                 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
 721                     ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
 722                     h->pps.scaling_matrix4[i][x]) << shift;
 723         }
 724     }
 725 }
 726
 727 static void init_dequant_tables(H264Context *h){
 728     int i,x;
 729     init_dequant4_coeff_table(h);
 730     if(h->pps.transform_8x8_mode)
 731         init_dequant8_coeff_table(h);
 732     if(h->sps.transform_bypass){
 733         for(i=0; i<6; i++)
 734             for(x=0; x<16; x++)
 735                 h->dequant4_coeff[i][0][x] = 1<<6;
 736         if(h->pps.transform_8x8_mode)
 737             for(i=0; i<2; i++)
 738                 for(x=0; x<64; x++)
 739                     h->dequant8_coeff[i][0][x] = 1<<6;
 740     }
 741 }
 742
 743
 744 int ff_h264_alloc_tables(H264Context *h){
 745     MpegEncContext * const s = &h->s;
 746     const int big_mb_num= s->mb_stride * (s->mb_height+1);
 747     int x,y;
 748
 749     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t), fail)
 750
 751     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 32 * sizeof(uint8_t), fail)
 752     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
 753     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
 754
 755     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
 756     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*big_mb_num * sizeof(uint8_t), fail);
 757     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*big_mb_num * sizeof(uint8_t), fail);
 758     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
 759     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
 760
 761     memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
 762     h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
 763
 764     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy  , big_mb_num * sizeof(uint32_t), fail);
 765     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
 766     for(y=0; y<s->mb_height; y++){
 767         for(x=0; x<s->mb_width; x++){
 768             const int mb_xy= x + y*s->mb_stride;
 769             const int b_xy = 4*x + 4*y*h->b_stride;
 770
 771             h->mb2b_xy [mb_xy]= b_xy;
 772             h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
 773         }
 774     }
 775
 776     s->obmc_scratchpad = NULL;
 777
 778     if(!h->dequant4_coeff[0])
 779         init_dequant_tables(h);
 780
 781     return 0;
 782 fail:
 783     free_tables(h);
 784     return -1;
 785 }
 786
 787 /**
 788  * Mimic alloc_tables(), but for every context thread.
 789  */
 790 static void clone_tables(H264Context *dst, H264Context *src){
 791     dst->intra4x4_pred_mode       = src->intra4x4_pred_mode;
 792     dst->non_zero_count           = src->non_zero_count;
 793     dst->slice_table              = src->slice_table;
 794     dst->cbp_table                = src->cbp_table;
 795     dst->mb2b_xy                  = src->mb2b_xy;
 796     dst->mb2br_xy                 = src->mb2br_xy;
 797     dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
 798     dst->mvd_table[0]             = src->mvd_table[0];
 799     dst->mvd_table[1]             = src->mvd_table[1];
 800     dst->direct_table             = src->direct_table;
 801     dst->list_counts              = src->list_counts;
 802
 803     dst->s.obmc_scratchpad = NULL;
 804     ff_h264_pred_init(&dst->hpc, src->s.codec_id);
 805 }
 806
 807 /**
 808  * Init context
 809  * Allocate buffers which are not shared amongst multiple threads.
 810  */
 811 static int context_init(H264Context *h){
 812     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
 813     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
 814
 815     return 0;
 816 fail:
 817     return -1; // free_tables will clean up for us
 818 }
 819
 820 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
 821
 822 static av_cold void common_init(H264Context *h){
 823     MpegEncContext * const s = &h->s;
 824
 825     s->width = s->avctx->width;
 826     s->height = s->avctx->height;
 827     s->codec_id= s->avctx->codec->id;
 828
 829     ff_h264_pred_init(&h->hpc, s->codec_id);
 830
 831     h->dequant_coeff_pps= -1;
 832     s->unrestricted_mv=1;
 833     s->decode=1; //FIXME
 834
 835     dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
 836
 837     memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
 838     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
 839 }
 840
 841 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
 842     H264Context *h= avctx->priv_data;
 843     MpegEncContext * const s = &h->s;
 844
 845     MPV_decode_defaults(s);
 846
 847     s->avctx = avctx;
 848     common_init(h);
 849
 850     s->out_format = FMT_H264;
 851     s->workaround_bugs= avctx->workaround_bugs;
 852
 853     // set defaults
 854 //    s->decode_mb= ff_h263_decode_mb;
 855     s->quarter_sample = 1;
 856     if(!avctx->has_b_frames)
 857     s->low_delay= 1;
 858
 859     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
 860
 861     ff_h264_decode_init_vlc();
 862
 863     h->thread_context[0] = h;
 864     h->outputed_poc = INT_MIN;
 865     h->prev_poc_msb= 1<<16;
 866     h->x264_build = -1;
 867     ff_h264_reset_sei(h);
 868     if(avctx->codec_id == CODEC_ID_H264){
 869         if(avctx->ticks_per_frame == 1){
 870             s->avctx->time_base.den *=2;
 871         }
 872         avctx->ticks_per_frame = 2;
 873     }
 874
 875     if(avctx->extradata_size > 0 && avctx->extradata && *(char *)avctx->extradata == 1){
 876         int i, cnt, nalsize;
 877         unsigned char *p = avctx->extradata;
 878
 879         h->is_avc = 1;
 880
 881         if(avctx->extradata_size < 7) {
 882             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
 883             return -1;
 884         }
 885         /* sps and pps in the avcC always have length coded with 2 bytes,
 886            so put a fake nal_length_size = 2 while parsing them */
 887         h->nal_length_size = 2;
 888         // Decode sps from avcC
 889         cnt = *(p+5) & 0x1f; // Number of sps
 890         p += 6;
 891         for (i = 0; i < cnt; i++) {
 892             nalsize = AV_RB16(p) + 2;
 893             if(decode_nal_units(h, p, nalsize) < 0) {
 894                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
 895                 return -1;
 896             }
 897             p += nalsize;
 898         }
 899         // Decode pps from avcC
 900         cnt = *(p++); // Number of pps
 901         for (i = 0; i < cnt; i++) {
 902             nalsize = AV_RB16(p) + 2;
 903             if(decode_nal_units(h, p, nalsize)  != nalsize) {
 904                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
 905                 return -1;
 906             }
 907             p += nalsize;
 908         }
 909         // Now store right nal length size, that will be use to parse all other nals
 910         h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
 911     } else {
 912         h->is_avc = 0;
 913         if(decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) < 0)
 914             return -1;
 915     }
 916     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
 917         s->avctx->has_b_frames = h->sps.num_reorder_frames;
 918         s->low_delay = 0;
 919     }
 920
 921     return 0;
 922 }
 923
 924 int ff_h264_frame_start(H264Context *h){
 925     MpegEncContext * const s = &h->s;
 926     int i;
 927
 928     if(MPV_frame_start(s, s->avctx) < 0)
 929         return -1;
 930     ff_er_frame_start(s);
 931     /*
 932      * MPV_frame_start uses pict_type to derive key_frame.
 933      * This is incorrect for H.264; IDR markings must be used.
 934      * Zero here; IDR markings per slice in frame or fields are ORed in later.
 935      * See decode_nal_units().
 936      */
 937     s->current_picture_ptr->key_frame= 0;
 938     s->current_picture_ptr->mmco_reset= 0;
 939
 940     assert(s->linesize && s->uvlinesize);
 941
 942     for(i=0; i<16; i++){
 943         h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
 944         h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
 945     }
 946     for(i=0; i<4; i++){
 947         h->block_offset[16+i]=
 948         h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
 949         h->block_offset[24+16+i]=
 950         h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
 951     }
 952
 953     /* can't be in alloc_tables because linesize isn't known there.
 954      * FIXME: redo bipred weight to not require extra buffer? */
 955     for(i = 0; i < s->avctx->thread_count; i++)
 956         if(!h->thread_context[i]->s.obmc_scratchpad)
 957             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
 958
 959     /* some macroblocks will be accessed before they're available */
 960     if(FRAME_MBAFF || s->avctx->thread_count > 1)
 961         memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
 962
 963 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
 964
 965     // We mark the current picture as non-reference after allocating it, so
 966     // that if we break out due to an error it can be released automatically
 967     // in the next MPV_frame_start().
 968     // SVQ3 as well as most other codecs have only last/next/current and thus
 969     // get released even with set reference, besides SVQ3 and others do not
 970     // mark frames as reference later "naturally".
 971     if(s->codec_id != CODEC_ID_SVQ3)
 972         s->current_picture_ptr->reference= 0;
 973
 974     s->current_picture_ptr->field_poc[0]=
 975     s->current_picture_ptr->field_poc[1]= INT_MAX;
 976     assert(s->current_picture_ptr->long_ref==0);
 977
 978     return 0;
 979 }
 980
 981 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){
 982     MpegEncContext * const s = &h->s;
 983     uint8_t *top_border;
 984     int top_idx = 1;
 985
 986     src_y  -=   linesize;
 987     src_cb -= uvlinesize;
 988     src_cr -= uvlinesize;
 989
 990     if(!simple && FRAME_MBAFF){
 991         if(s->mb_y&1){
 992             if(!MB_MBAFF){
 993                 top_border = h->top_borders[0][s->mb_x];
 994                 AV_COPY128(top_border, src_y + 15*linesize);
 995                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
 996                     AV_COPY64(top_border+16, src_cb+7*uvlinesize);
 997                     AV_COPY64(top_border+24, src_cr+7*uvlinesize);
 998                 }
 999             }
1000         }else if(MB_MBAFF){
1001             top_idx = 0;
1002         }else
1003             return;
1004     }
1005
1006     top_border = h->top_borders[top_idx][s->mb_x];
1007     // There are two lines saved, the line above the the top macroblock of a pair,
1008     // and the line above the bottom macroblock
1009     AV_COPY128(top_border, src_y + 16*linesize);
1010
1011     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1012         AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1013         AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1014     }
1015 }
1016
1017 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){
1018     MpegEncContext * const s = &h->s;
1019     int deblock_left;
1020     int deblock_top;
1021     int top_idx = 1;
1022     uint8_t *top_border_m1;
1023     uint8_t *top_border;
1024
1025     if(!simple && FRAME_MBAFF){
1026         if(s->mb_y&1){
1027             if(!MB_MBAFF)
1028                 return;
1029         }else{
1030             top_idx = MB_MBAFF ? 0 : 1;
1031         }
1032     }
1033
1034     if(h->deblocking_filter == 2) {
1035         deblock_left = h->left_type[0];
1036         deblock_top  = h->top_type;
1037     } else {
1038         deblock_left = (s->mb_x > 0);
1039         deblock_top =  (s->mb_y > !!MB_FIELD);
1040     }
1041
1042     src_y  -=   linesize + 1;
1043     src_cb -= uvlinesize + 1;
1044     src_cr -= uvlinesize + 1;
1045
1046     top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1047     top_border    = h->top_borders[top_idx][s->mb_x];
1048
1049 #define XCHG(a,b,xchg)\
1050 if (xchg) AV_SWAP64(b,a);\
1051 else      AV_COPY64(b,a);
1052
1053     if(deblock_top){
1054         if(deblock_left){
1055             XCHG(top_border_m1+8, src_y -7, 1);
1056         }
1057         XCHG(top_border+0, src_y +1, xchg);
1058         XCHG(top_border+8, src_y +9, 1);
1059         if(s->mb_x+1 < s->mb_width){
1060             XCHG(h->top_borders[top_idx][s->mb_x+1], src_y +17, 1);
1061         }
1062     }
1063
1064     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1065         if(deblock_top){
1066             if(deblock_left){
1067                 XCHG(top_border_m1+16, src_cb -7, 1);
1068                 XCHG(top_border_m1+24, src_cr -7, 1);
1069             }
1070             XCHG(top_border+16, src_cb+1, 1);
1071             XCHG(top_border+24, src_cr+1, 1);
1072         }
1073     }
1074 }
1075
1076 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
1077     MpegEncContext * const s = &h->s;
1078     const int mb_x= s->mb_x;
1079     const int mb_y= s->mb_y;
1080     const int mb_xy= h->mb_xy;
1081     const int mb_type= s->current_picture.mb_type[mb_xy];
1082     uint8_t  *dest_y, *dest_cb, *dest_cr;
1083     int linesize, uvlinesize /*dct_offset*/;
1084     int i;
1085     int *block_offset = &h->block_offset[0];
1086     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1087     /* is_h264 should always be true if SVQ3 is disabled. */
1088     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1089     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1090     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1091
1092     dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
1093     dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
1094     dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
1095
1096     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
1097     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
1098
1099     h->list_counts[mb_xy]= h->list_count;
1100
1101     if (!simple && MB_FIELD) {
1102         linesize   = h->mb_linesize   = s->linesize * 2;
1103         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1104         block_offset = &h->block_offset[24];
1105         if(mb_y&1){ //FIXME move out of this function?
1106             dest_y -= s->linesize*15;
1107             dest_cb-= s->uvlinesize*7;
1108             dest_cr-= s->uvlinesize*7;
1109         }
1110         if(FRAME_MBAFF) {
1111             int list;
1112             for(list=0; list<h->list_count; list++){
1113                 if(!USES_LIST(mb_type, list))
1114                     continue;
1115                 if(IS_16X16(mb_type)){
1116                     int8_t *ref = &h->ref_cache[list][scan8[0]];
1117                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1118                 }else{
1119                     for(i=0; i<16; i+=4){
1120                         int ref = h->ref_cache[list][scan8[i]];
1121                         if(ref >= 0)
1122                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1123                     }
1124                 }
1125             }
1126         }
1127     } else {
1128         linesize   = h->mb_linesize   = s->linesize;
1129         uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1130 //        dct_offset = s->linesize * 16;
1131     }
1132
1133     if (!simple && IS_INTRA_PCM(mb_type)) {
1134         for (i=0; i<16; i++) {
1135             memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1136         }
1137         for (i=0; i<8; i++) {
1138             memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
1139             memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
1140         }
1141     } else {
1142         if(IS_INTRA(mb_type)){
1143             if(h->deblocking_filter)
1144                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
1145
1146             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1147                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1148                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1149             }
1150
1151             if(IS_INTRA4x4(mb_type)){
1152                 if(simple || !s->encoding){
1153                     if(IS_8x8DCT(mb_type)){
1154                         if(transform_bypass){
1155                             idct_dc_add =
1156                             idct_add    = s->dsp.add_pixels8;
1157                         }else{
1158                             idct_dc_add = s->dsp.h264_idct8_dc_add;
1159                             idct_add    = s->dsp.h264_idct8_add;
1160                         }
1161                         for(i=0; i<16; i+=4){
1162                             uint8_t * const ptr= dest_y + block_offset[i];
1163                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1164                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1165                                 h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
1166                             }else{
1167                                 const int nnz = h->non_zero_count_cache[ scan8[i] ];
1168                                 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1169                                                             (h->topright_samples_available<<i)&0x4000, linesize);
1170                                 if(nnz){
1171                                     if(nnz == 1 && h->mb[i*16])
1172                                         idct_dc_add(ptr, h->mb + i*16, linesize);
1173                                     else
1174                                         idct_add   (ptr, h->mb + i*16, linesize);
1175                                 }
1176                             }
1177                         }
1178                     }else{
1179                         if(transform_bypass){
1180                             idct_dc_add =
1181                             idct_add    = s->dsp.add_pixels4;
1182                         }else{
1183                             idct_dc_add = s->dsp.h264_idct_dc_add;
1184                             idct_add    = s->dsp.h264_idct_add;
1185                         }
1186                         for(i=0; i<16; i++){
1187                             uint8_t * const ptr= dest_y + block_offset[i];
1188                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1189
1190                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1191                                 h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
1192                             }else{
1193                                 uint8_t *topright;
1194                                 int nnz, tr;
1195                                 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1196                                     const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1197                                     assert(mb_y || linesize <= block_offset[i]);
1198                                     if(!topright_avail){
1199                                         tr= ptr[3 - linesize]*0x01010101;
1200                                         topright= (uint8_t*) &tr;
1201                                     }else
1202                                         topright= ptr + 4 - linesize;
1203                                 }else
1204                                     topright= NULL;
1205
1206                                 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1207                                 nnz = h->non_zero_count_cache[ scan8[i] ];
1208                                 if(nnz){
1209                                     if(is_h264){
1210                                         if(nnz == 1 && h->mb[i*16])
1211                                             idct_dc_add(ptr, h->mb + i*16, linesize);
1212                                         else
1213                                             idct_add   (ptr, h->mb + i*16, linesize);
1214                                     }else
1215                                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1216                                 }
1217                             }
1218                         }
1219                     }
1220                 }
1221             }else{
1222                 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1223                 if(is_h264){
1224                     if(!transform_bypass)
1225                         h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
1226                 }else
1227                     ff_svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
1228             }
1229             if(h->deblocking_filter)
1230                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
1231         }else if(is_h264){
1232             hl_motion(h, dest_y, dest_cb, dest_cr,
1233                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1234                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1235                       s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
1236         }
1237
1238
1239         if(!IS_INTRA4x4(mb_type)){
1240             if(is_h264){
1241                 if(IS_INTRA16x16(mb_type)){
1242                     if(transform_bypass){
1243                         if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1244                             h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1245                         }else{
1246                             for(i=0; i<16; i++){
1247                                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1248                                     s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
1249                             }
1250                         }
1251                     }else{
1252                          s->dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1253                     }
1254                 }else if(h->cbp&15){
1255                     if(transform_bypass){
1256                         const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1257                         idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1258                         for(i=0; i<16; i+=di){
1259                             if(h->non_zero_count_cache[ scan8[i] ]){
1260                                 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
1261                             }
1262                         }
1263                     }else{
1264                         if(IS_8x8DCT(mb_type)){
1265                             s->dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1266                         }else{
1267                             s->dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1268                         }
1269                     }
1270                 }
1271             }else{
1272                 for(i=0; i<16; i++){
1273                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1274                         uint8_t * const ptr= dest_y + block_offset[i];
1275                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1276                     }
1277                 }
1278             }
1279         }
1280
1281         if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1282             uint8_t *dest[2] = {dest_cb, dest_cr};
1283             if(transform_bypass){
1284                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1285                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
1286                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
1287                 }else{
1288                     idct_add = s->dsp.add_pixels4;
1289                     for(i=16; i<16+8; i++){
1290                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1291                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1292                     }
1293                 }
1294             }else{
1295                 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]);
1296                 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]);
1297                 if(is_h264){
1298                     idct_add = s->dsp.h264_idct_add;
1299                     idct_dc_add = s->dsp.h264_idct_dc_add;
1300                     for(i=16; i<16+8; i++){
1301                         if(h->non_zero_count_cache[ scan8[i] ])
1302                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1303                         else if(h->mb[i*16])
1304                             idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1305                     }
1306                 }else{
1307                     for(i=16; i<16+8; i++){
1308                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1309                             uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1310                             ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
1311                         }
1312                     }
1313                 }
1314             }
1315         }
1316     }
1317     if(h->cbp || IS_INTRA(mb_type))
1318         s->dsp.clear_blocks(h->mb);
1319 }
1320
1321 /**
1322  * Process a macroblock; this case avoids checks for expensive uncommon cases.
1323  */
1324 static void hl_decode_mb_simple(H264Context *h){
1325     hl_decode_mb_internal(h, 1);
1326 }
1327
1328 /**
1329  * Process a macroblock; this handles edge cases, such as interlacing.
1330  */
1331 static void av_noinline hl_decode_mb_complex(H264Context *h){
1332     hl_decode_mb_internal(h, 0);
1333 }
1334
1335 void ff_h264_hl_decode_mb(H264Context *h){
1336     MpegEncContext * const s = &h->s;
1337     const int mb_xy= h->mb_xy;
1338     const int mb_type= s->current_picture.mb_type[mb_xy];
1339     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1340
1341     if (is_complex)
1342         hl_decode_mb_complex(h);
1343     else hl_decode_mb_simple(h);
1344 }
1345
1346 static int pred_weight_table(H264Context *h){
1347     MpegEncContext * const s = &h->s;
1348     int list, i;
1349     int luma_def, chroma_def;
1350
1351     h->use_weight= 0;
1352     h->use_weight_chroma= 0;
1353     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1354     h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1355     luma_def = 1<<h->luma_log2_weight_denom;
1356     chroma_def = 1<<h->chroma_log2_weight_denom;
1357
1358     for(list=0; list<2; list++){
1359         h->luma_weight_flag[list]   = 0;
1360         h->chroma_weight_flag[list] = 0;
1361         for(i=0; i<h->ref_count[list]; i++){
1362             int luma_weight_flag, chroma_weight_flag;
1363
1364             luma_weight_flag= get_bits1(&s->gb);
1365             if(luma_weight_flag){
1366                 h->luma_weight[list][i]= get_se_golomb(&s->gb);
1367                 h->luma_offset[list][i]= get_se_golomb(&s->gb);
1368                 if(   h->luma_weight[list][i] != luma_def
1369                    || h->luma_offset[list][i] != 0) {
1370                     h->use_weight= 1;
1371                     h->luma_weight_flag[list]= 1;
1372                 }
1373             }else{
1374                 h->luma_weight[list][i]= luma_def;
1375                 h->luma_offset[list][i]= 0;
1376             }
1377
1378             if(CHROMA){
1379                 chroma_weight_flag= get_bits1(&s->gb);
1380                 if(chroma_weight_flag){
1381                     int j;
1382                     for(j=0; j<2; j++){
1383                         h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
1384                         h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
1385                         if(   h->chroma_weight[list][i][j] != chroma_def
1386                            || h->chroma_offset[list][i][j] != 0) {
1387                             h->use_weight_chroma= 1;
1388                             h->chroma_weight_flag[list]= 1;
1389                         }
1390                     }
1391                 }else{
1392                     int j;
1393                     for(j=0; j<2; j++){
1394                         h->chroma_weight[list][i][j]= chroma_def;
1395                         h->chroma_offset[list][i][j]= 0;
1396                     }
1397                 }
1398             }
1399         }
1400         if(h->slice_type_nos != FF_B_TYPE) break;
1401     }
1402     h->use_weight= h->use_weight || h->use_weight_chroma;
1403     return 0;
1404 }
1405
1406 static void implicit_weight_table(H264Context *h){
1407     MpegEncContext * const s = &h->s;
1408     int ref0, ref1, i;
1409     int cur_poc = s->current_picture_ptr->poc;
1410
1411     for (i = 0; i < 2; i++) {
1412         h->luma_weight_flag[i]   = 0;
1413         h->chroma_weight_flag[i] = 0;
1414     }
1415
1416     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
1417        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1418         h->use_weight= 0;
1419         h->use_weight_chroma= 0;
1420         return;
1421     }
1422
1423     h->use_weight= 2;
1424     h->use_weight_chroma= 2;
1425     h->luma_log2_weight_denom= 5;
1426     h->chroma_log2_weight_denom= 5;
1427
1428     for(ref0=0; ref0 < h->ref_count[0]; ref0++){
1429         int poc0 = h->ref_list[0][ref0].poc;
1430         for(ref1=0; ref1 < h->ref_count[1]; ref1++){
1431             int poc1 = h->ref_list[1][ref1].poc;
1432             int td = av_clip(poc1 - poc0, -128, 127);
1433             if(td){
1434                 int tb = av_clip(cur_poc - poc0, -128, 127);
1435                 int tx = (16384 + (FFABS(td) >> 1)) / td;
1436                 int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
1437                 if(dist_scale_factor < -64 || dist_scale_factor > 128)
1438                     h->implicit_weight[ref0][ref1] = 32;
1439                 else
1440                     h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
1441             }else
1442                 h->implicit_weight[ref0][ref1] = 32;
1443         }
1444     }
1445 }
1446
1447 /**
1448  * instantaneous decoder refresh.
1449  */
1450 static void idr(H264Context *h){
1451     ff_h264_remove_all_refs(h);
1452     h->prev_frame_num= 0;
1453     h->prev_frame_num_offset= 0;
1454     h->prev_poc_msb=
1455     h->prev_poc_lsb= 0;
1456 }
1457
1458 /* forget old pics after a seek */
1459 static void flush_dpb(AVCodecContext *avctx){
1460     H264Context *h= avctx->priv_data;
1461     int i;
1462     for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1463         if(h->delayed_pic[i])
1464             h->delayed_pic[i]->reference= 0;
1465         h->delayed_pic[i]= NULL;
1466     }
1467     h->outputed_poc= INT_MIN;
1468     h->prev_interlaced_frame = 1;
1469     idr(h);
1470     if(h->s.current_picture_ptr)
1471         h->s.current_picture_ptr->reference= 0;
1472     h->s.first_field= 0;
1473     ff_h264_reset_sei(h);
1474     ff_mpeg_flush(avctx);
1475 }
1476
1477 static int init_poc(H264Context *h){
1478     MpegEncContext * const s = &h->s;
1479     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1480     int field_poc[2];
1481     Picture *cur = s->current_picture_ptr;
1482
1483     h->frame_num_offset= h->prev_frame_num_offset;
1484     if(h->frame_num < h->prev_frame_num)
1485         h->frame_num_offset += max_frame_num;
1486
1487     if(h->sps.poc_type==0){
1488         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1489
1490         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1491             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1492         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1493             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1494         else
1495             h->poc_msb = h->prev_poc_msb;
1496 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1497         field_poc[0] =
1498         field_poc[1] = h->poc_msb + h->poc_lsb;
1499         if(s->picture_structure == PICT_FRAME)
1500             field_poc[1] += h->delta_poc_bottom;
1501     }else if(h->sps.poc_type==1){
1502         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1503         int i;
1504
1505         if(h->sps.poc_cycle_length != 0)
1506             abs_frame_num = h->frame_num_offset + h->frame_num;
1507         else
1508             abs_frame_num = 0;
1509
1510         if(h->nal_ref_idc==0 && abs_frame_num > 0)
1511             abs_frame_num--;
1512
1513         expected_delta_per_poc_cycle = 0;
1514         for(i=0; i < h->sps.poc_cycle_length; i++)
1515             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1516
1517         if(abs_frame_num > 0){
1518             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1519             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1520
1521             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1522             for(i = 0; i <= frame_num_in_poc_cycle; i++)
1523                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1524         } else
1525             expectedpoc = 0;
1526
1527         if(h->nal_ref_idc == 0)
1528             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1529
1530         field_poc[0] = expectedpoc + h->delta_poc[0];
1531         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1532
1533         if(s->picture_structure == PICT_FRAME)
1534             field_poc[1] += h->delta_poc[1];
1535     }else{
1536         int poc= 2*(h->frame_num_offset + h->frame_num);
1537
1538         if(!h->nal_ref_idc)
1539             poc--;
1540
1541         field_poc[0]= poc;
1542         field_poc[1]= poc;
1543     }
1544
1545     if(s->picture_structure != PICT_BOTTOM_FIELD)
1546         s->current_picture_ptr->field_poc[0]= field_poc[0];
1547     if(s->picture_structure != PICT_TOP_FIELD)
1548         s->current_picture_ptr->field_poc[1]= field_poc[1];
1549     cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1550
1551     return 0;
1552 }
1553
1554
1555 /**
1556  * initialize scan tables
1557  */
1558 static void init_scan_tables(H264Context *h){
1559     MpegEncContext * const s = &h->s;
1560     int i;
1561     if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
1562         memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
1563         memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
1564     }else{
1565         for(i=0; i<16; i++){
1566 #define T(x) (x>>2) | ((x<<2) & 0xF)
1567             h->zigzag_scan[i] = T(zigzag_scan[i]);
1568             h-> field_scan[i] = T( field_scan[i]);
1569 #undef T
1570         }
1571     }
1572     if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
1573         memcpy(h->zigzag_scan8x8,       ff_zigzag_direct,     64*sizeof(uint8_t));
1574         memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
1575         memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
1576         memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
1577     }else{
1578         for(i=0; i<64; i++){
1579 #define T(x) (x>>3) | ((x&7)<<3)
1580             h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
1581             h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1582             h->field_scan8x8[i]        = T(field_scan8x8[i]);
1583             h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
1584 #undef T
1585         }
1586     }
1587     if(h->sps.transform_bypass){ //FIXME same ugly
1588         h->zigzag_scan_q0          = zigzag_scan;
1589         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
1590         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1591         h->field_scan_q0           = field_scan;
1592         h->field_scan8x8_q0        = field_scan8x8;
1593         h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
1594     }else{
1595         h->zigzag_scan_q0          = h->zigzag_scan;
1596         h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
1597         h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1598         h->field_scan_q0           = h->field_scan;
1599         h->field_scan8x8_q0        = h->field_scan8x8;
1600         h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
1601     }
1602 }
1603
1604 static void field_end(H264Context *h){
1605     MpegEncContext * const s = &h->s;
1606     AVCodecContext * const avctx= s->avctx;
1607     s->mb_y= 0;
1608
1609     s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1610     s->current_picture_ptr->pict_type= s->pict_type;
1611
1612     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1613         ff_vdpau_h264_set_reference_frames(s);
1614
1615     if(!s->dropable) {
1616         ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1617         h->prev_poc_msb= h->poc_msb;
1618         h->prev_poc_lsb= h->poc_lsb;
1619     }
1620     h->prev_frame_num_offset= h->frame_num_offset;
1621     h->prev_frame_num= h->frame_num;
1622
1623     if (avctx->hwaccel) {
1624         if (avctx->hwaccel->end_frame(avctx) < 0)
1625             av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1626     }
1627
1628     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1629         ff_vdpau_h264_picture_complete(s);
1630
1631     /*
1632      * FIXME: Error handling code does not seem to support interlaced
1633      * when slices span multiple rows
1634      * The ff_er_add_slice calls don't work right for bottom
1635      * fields; they cause massive erroneous error concealing
1636      * Error marking covers both fields (top and bottom).
1637      * This causes a mismatched s->error_count
1638      * and a bad error table. Further, the error count goes to
1639      * INT_MAX when called for bottom field, because mb_y is
1640      * past end by one (callers fault) and resync_mb_y != 0
1641      * causes problems for the first MB line, too.
1642      */
1643     if (!FIELD_PICTURE)
1644         ff_er_frame_end(s);
1645
1646     MPV_frame_end(s);
1647
1648     h->current_slice=0;
1649 }
1650
1651 /**
1652  * Replicates H264 "master" context to thread contexts.
1653  */
1654 static void clone_slice(H264Context *dst, H264Context *src)
1655 {
1656     memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
1657     dst->s.current_picture_ptr  = src->s.current_picture_ptr;
1658     dst->s.current_picture      = src->s.current_picture;
1659     dst->s.linesize             = src->s.linesize;
1660     dst->s.uvlinesize           = src->s.uvlinesize;
1661     dst->s.first_field          = src->s.first_field;
1662
1663     dst->prev_poc_msb           = src->prev_poc_msb;
1664     dst->prev_poc_lsb           = src->prev_poc_lsb;
1665     dst->prev_frame_num_offset  = src->prev_frame_num_offset;
1666     dst->prev_frame_num         = src->prev_frame_num;
1667     dst->short_ref_count        = src->short_ref_count;
1668
1669     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
1670     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
1671     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1672     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
1673
1674     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
1675     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
1676 }
1677
1678 /**
1679  * decodes a slice header.
1680  * This will also call MPV_common_init() and frame_start() as needed.
1681  *
1682  * @param h h264context
1683  * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1684  *
1685  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1686  */
1687 static int decode_slice_header(H264Context *h, H264Context *h0){
1688     MpegEncContext * const s = &h->s;
1689     MpegEncContext * const s0 = &h0->s;
1690     unsigned int first_mb_in_slice;
1691     unsigned int pps_id;
1692     int num_ref_idx_active_override_flag;
1693     unsigned int slice_type, tmp, i, j;
1694     int default_ref_list_done = 0;
1695     int last_pic_structure;
1696
1697     s->dropable= h->nal_ref_idc == 0;
1698
1699     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1700         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1701         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1702     }else{
1703         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1704         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1705     }
1706
1707     first_mb_in_slice= get_ue_golomb(&s->gb);
1708
1709     if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1710         if(h0->current_slice && FIELD_PICTURE){
1711             field_end(h);
1712         }
1713
1714         h0->current_slice = 0;
1715         if (!s0->first_field)
1716             s->current_picture_ptr= NULL;
1717     }
1718
1719     slice_type= get_ue_golomb_31(&s->gb);
1720     if(slice_type > 9){
1721         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);
1722         return -1;
1723     }
1724     if(slice_type > 4){
1725         slice_type -= 5;
1726         h->slice_type_fixed=1;
1727     }else
1728         h->slice_type_fixed=0;
1729
1730     slice_type= golomb_to_pict_type[ slice_type ];
1731     if (slice_type == FF_I_TYPE
1732         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
1733         default_ref_list_done = 1;
1734     }
1735     h->slice_type= slice_type;
1736     h->slice_type_nos= slice_type & 3;
1737
1738     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
1739
1740     pps_id= get_ue_golomb(&s->gb);
1741     if(pps_id>=MAX_PPS_COUNT){
1742         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
1743         return -1;
1744     }
1745     if(!h0->pps_buffers[pps_id]) {
1746         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
1747         return -1;
1748     }
1749     h->pps= *h0->pps_buffers[pps_id];
1750
1751     if(!h0->sps_buffers[h->pps.sps_id]) {
1752         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
1753         return -1;
1754     }
1755     h->sps = *h0->sps_buffers[h->pps.sps_id];
1756
1757     if(h == h0 && h->dequant_coeff_pps != pps_id){
1758         h->dequant_coeff_pps = pps_id;
1759         init_dequant_tables(h);
1760     }
1761
1762     s->mb_width= h->sps.mb_width;
1763     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1764
1765     h->b_stride=  s->mb_width*4;
1766
1767     s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
1768     if(h->sps.frame_mbs_only_flag)
1769         s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
1770     else
1771         s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
1772
1773     if (s->context_initialized
1774         && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
1775         if(h != h0)
1776             return -1;   // width / height changed during parallelized decoding
1777         free_tables(h);
1778         flush_dpb(s->avctx);
1779         MPV_common_end(s);
1780     }
1781     if (!s->context_initialized) {
1782         if(h != h0)
1783             return -1;  // we cant (re-)initialize context during parallel decoding
1784
1785         avcodec_set_dimensions(s->avctx, s->width, s->height);
1786         s->avctx->sample_aspect_ratio= h->sps.sar;
1787         if(!s->avctx->sample_aspect_ratio.den)
1788             s->avctx->sample_aspect_ratio.den = 1;
1789
1790         if(h->sps.video_signal_type_present_flag){
1791             s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1792             if(h->sps.colour_description_present_flag){
1793                 s->avctx->color_primaries = h->sps.color_primaries;
1794                 s->avctx->color_trc       = h->sps.color_trc;
1795                 s->avctx->colorspace      = h->sps.colorspace;
1796             }
1797         }
1798
1799         if(h->sps.timing_info_present_flag){
1800             int64_t den= h->sps.time_scale;
1801             if(h->x264_build < 44U)
1802                 den *= 2;
1803             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
1804                       h->sps.num_units_in_tick, den, 1<<30);
1805         }
1806         s->avctx->pix_fmt = s->avctx->get_format(s->avctx, s->avctx->codec->pix_fmts);
1807         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1808
1809         if (MPV_common_init(s) < 0)
1810             return -1;
1811         s->first_field = 0;
1812         h->prev_interlaced_frame = 1;
1813
1814         init_scan_tables(h);
1815         ff_h264_alloc_tables(h);
1816
1817         for(i = 1; i < s->avctx->thread_count; i++) {
1818             H264Context *c;
1819             c = h->thread_context[i] = av_malloc(sizeof(H264Context));
1820             memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
1821             memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
1822             c->sps = h->sps;
1823             c->pps = h->pps;
1824             init_scan_tables(c);
1825             clone_tables(c, h);
1826         }
1827
1828         for(i = 0; i < s->avctx->thread_count; i++)
1829             if(context_init(h->thread_context[i]) < 0)
1830                 return -1;
1831     }
1832
1833     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1834
1835     h->mb_mbaff = 0;
1836     h->mb_aff_frame = 0;
1837     last_pic_structure = s0->picture_structure;
1838     if(h->sps.frame_mbs_only_flag){
1839         s->picture_structure= PICT_FRAME;
1840     }else{
1841         if(get_bits1(&s->gb)) { //field_pic_flag
1842             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
1843         } else {
1844             s->picture_structure= PICT_FRAME;
1845             h->mb_aff_frame = h->sps.mb_aff;
1846         }
1847     }
1848     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
1849
1850     if(h0->current_slice == 0){
1851         while(h->frame_num !=  h->prev_frame_num &&
1852               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
1853             av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
1854             if (ff_h264_frame_start(h) < 0)
1855                 return -1;
1856             h->prev_frame_num++;
1857             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
1858             s->current_picture_ptr->frame_num= h->prev_frame_num;
1859             ff_h264_execute_ref_pic_marking(h, NULL, 0);
1860         }
1861
1862         /* See if we have a decoded first field looking for a pair... */
1863         if (s0->first_field) {
1864             assert(s0->current_picture_ptr);
1865             assert(s0->current_picture_ptr->data[0]);
1866             assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
1867
1868             /* figure out if we have a complementary field pair */
1869             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
1870                 /*
1871                  * Previous field is unmatched. Don't display it, but let it
1872                  * remain for reference if marked as such.
1873                  */
1874                 s0->current_picture_ptr = NULL;
1875                 s0->first_field = FIELD_PICTURE;
1876
1877             } else {
1878                 if (h->nal_ref_idc &&
1879                         s0->current_picture_ptr->reference &&
1880                         s0->current_picture_ptr->frame_num != h->frame_num) {
1881                     /*
1882                      * This and previous field were reference, but had
1883                      * different frame_nums. Consider this field first in
1884                      * pair. Throw away previous field except for reference
1885                      * purposes.
1886                      */
1887                     s0->first_field = 1;
1888                     s0->current_picture_ptr = NULL;
1889
1890                 } else {
1891                     /* Second field in complementary pair */
1892                     s0->first_field = 0;
1893                 }
1894             }
1895
1896         } else {
1897             /* Frame or first field in a potentially complementary pair */
1898             assert(!s0->current_picture_ptr);
1899             s0->first_field = FIELD_PICTURE;
1900         }
1901
1902         if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
1903             s0->first_field = 0;
1904             return -1;
1905         }
1906     }
1907     if(h != h0)
1908         clone_slice(h, h0);
1909
1910     s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
1911
1912     assert(s->mb_num == s->mb_width * s->mb_height);
1913     if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
1914        first_mb_in_slice                    >= s->mb_num){
1915         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1916         return -1;
1917     }
1918     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
1919     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
1920     if (s->picture_structure == PICT_BOTTOM_FIELD)
1921         s->resync_mb_y = s->mb_y = s->mb_y + 1;
1922     assert(s->mb_y < s->mb_height);
1923
1924     if(s->picture_structure==PICT_FRAME){
1925         h->curr_pic_num=   h->frame_num;
1926         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
1927     }else{
1928         h->curr_pic_num= 2*h->frame_num + 1;
1929         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
1930     }
1931
1932     if(h->nal_unit_type == NAL_IDR_SLICE){
1933         get_ue_golomb(&s->gb); /* idr_pic_id */
1934     }
1935
1936     if(h->sps.poc_type==0){
1937         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
1938
1939         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
1940             h->delta_poc_bottom= get_se_golomb(&s->gb);
1941         }
1942     }
1943
1944     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
1945         h->delta_poc[0]= get_se_golomb(&s->gb);
1946
1947         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
1948             h->delta_poc[1]= get_se_golomb(&s->gb);
1949     }
1950
1951     init_poc(h);
1952
1953     if(h->pps.redundant_pic_cnt_present){
1954         h->redundant_pic_count= get_ue_golomb(&s->gb);
1955     }
1956
1957     //set defaults, might be overridden a few lines later
1958     h->ref_count[0]= h->pps.ref_count[0];
1959     h->ref_count[1]= h->pps.ref_count[1];
1960
1961     if(h->slice_type_nos != FF_I_TYPE){
1962         if(h->slice_type_nos == FF_B_TYPE){
1963             h->direct_spatial_mv_pred= get_bits1(&s->gb);
1964         }
1965         num_ref_idx_active_override_flag= get_bits1(&s->gb);
1966
1967         if(num_ref_idx_active_override_flag){
1968             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
1969             if(h->slice_type_nos==FF_B_TYPE)
1970                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
1971
1972             if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
1973                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
1974                 h->ref_count[0]= h->ref_count[1]= 1;
1975                 return -1;
1976             }
1977         }
1978         if(h->slice_type_nos == FF_B_TYPE)
1979             h->list_count= 2;
1980         else
1981             h->list_count= 1;
1982     }else
1983         h->list_count= 0;
1984
1985     if(!default_ref_list_done){
1986         ff_h264_fill_default_ref_list(h);
1987     }
1988
1989     if(h->slice_type_nos!=FF_I_TYPE && ff_h264_decode_ref_pic_list_reordering(h) < 0)
1990         return -1;
1991
1992     if(h->slice_type_nos!=FF_I_TYPE){
1993         s->last_picture_ptr= &h->ref_list[0][0];
1994         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
1995     }
1996     if(h->slice_type_nos==FF_B_TYPE){
1997         s->next_picture_ptr= &h->ref_list[1][0];
1998         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
1999     }
2000
2001     if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
2002        ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
2003         pred_weight_table(h);
2004     else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE)
2005         implicit_weight_table(h);
2006     else {
2007         h->use_weight = 0;
2008         for (i = 0; i < 2; i++) {
2009             h->luma_weight_flag[i]   = 0;
2010             h->chroma_weight_flag[i] = 0;
2011         }
2012     }
2013
2014     if(h->nal_ref_idc)
2015         ff_h264_decode_ref_pic_marking(h0, &s->gb);
2016
2017     if(FRAME_MBAFF)
2018         ff_h264_fill_mbaff_ref_list(h);
2019
2020     if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
2021         ff_h264_direct_dist_scale_factor(h);
2022     ff_h264_direct_ref_list_init(h);
2023
2024     if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
2025         tmp = get_ue_golomb_31(&s->gb);
2026         if(tmp > 2){
2027             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2028             return -1;
2029         }
2030         h->cabac_init_idc= tmp;
2031     }
2032
2033     h->last_qscale_diff = 0;
2034     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2035     if(tmp>51){
2036         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2037         return -1;
2038     }
2039     s->qscale= tmp;
2040     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2041     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2042     //FIXME qscale / qp ... stuff
2043     if(h->slice_type == FF_SP_TYPE){
2044         get_bits1(&s->gb); /* sp_for_switch_flag */
2045     }
2046     if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
2047         get_se_golomb(&s->gb); /* slice_qs_delta */
2048     }
2049
2050     h->deblocking_filter = 1;
2051     h->slice_alpha_c0_offset = 52;
2052     h->slice_beta_offset = 52;
2053     if( h->pps.deblocking_filter_parameters_present ) {
2054         tmp= get_ue_golomb_31(&s->gb);
2055         if(tmp > 2){
2056             av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2057             return -1;
2058         }
2059         h->deblocking_filter= tmp;
2060         if(h->deblocking_filter < 2)
2061             h->deblocking_filter^= 1; // 1<->0
2062
2063         if( h->deblocking_filter ) {
2064             h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2065             h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
2066             if(   h->slice_alpha_c0_offset > 104U
2067                || h->slice_beta_offset     > 104U){
2068                 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);
2069                 return -1;
2070             }
2071         }
2072     }
2073
2074     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
2075        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
2076        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
2077        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2078         h->deblocking_filter= 0;
2079
2080     if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2081         if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2082             /* Cheat slightly for speed:
2083                Do not bother to deblock across slices. */
2084             h->deblocking_filter = 2;
2085         } else {
2086             h0->max_contexts = 1;
2087             if(!h0->single_decode_warning) {
2088                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2089                 h0->single_decode_warning = 1;
2090             }
2091             if(h != h0)
2092                 return 1; // deblocking switched inside frame
2093         }
2094     }
2095     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]);
2096
2097 #if 0 //FMO
2098     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2099         slice_group_change_cycle= get_bits(&s->gb, ?);
2100 #endif
2101
2102     h0->last_slice_type = slice_type;
2103     h->slice_num = ++h0->current_slice;
2104     if(h->slice_num >= MAX_SLICES){
2105         av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2106     }
2107
2108     for(j=0; j<2; j++){
2109         int id_list[16];
2110         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2111         for(i=0; i<16; i++){
2112             id_list[i]= 60;
2113             if(h->ref_list[j][i].data[0]){
2114                 int k;
2115                 uint8_t *base= h->ref_list[j][i].base[0];
2116                 for(k=0; k<h->short_ref_count; k++)
2117                     if(h->short_ref[k]->base[0] == base){
2118                         id_list[i]= k;
2119                         break;
2120                     }
2121                 for(k=0; k<h->long_ref_count; k++)
2122                     if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2123                         id_list[i]= h->short_ref_count + k;
2124                         break;
2125                     }
2126             }
2127         }
2128
2129         ref2frm[0]=
2130         ref2frm[1]= -1;
2131         for(i=0; i<16; i++)
2132             ref2frm[i+2]= 4*id_list[i]
2133                           +(h->ref_list[j][i].reference&3);
2134         ref2frm[18+0]=
2135         ref2frm[18+1]= -1;
2136         for(i=16; i<48; i++)
2137             ref2frm[i+4]= 4*id_list[(i-16)>>1]
2138                           +(h->ref_list[j][i].reference&3);
2139     }
2140
2141     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
2142     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2143
2144     s->avctx->refs= h->sps.ref_frame_count;
2145
2146     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2147         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",
2148                h->slice_num,
2149                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2150                first_mb_in_slice,
2151                av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2152                pps_id, h->frame_num,
2153                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2154                h->ref_count[0], h->ref_count[1],
2155                s->qscale,
2156                h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
2157                h->use_weight,
2158                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2159                h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2160                );
2161     }
2162
2163     return 0;
2164 }
2165
2166 int ff_h264_get_slice_type(const H264Context *h)
2167 {
2168     switch (h->slice_type) {
2169     case FF_P_TYPE:  return 0;
2170     case FF_B_TYPE:  return 1;
2171     case FF_I_TYPE:  return 2;
2172     case FF_SP_TYPE: return 3;
2173     case FF_SI_TYPE: return 4;
2174     default:         return -1;
2175     }
2176 }
2177
2178 static void loop_filter(H264Context *h){
2179     MpegEncContext * const s = &h->s;
2180     uint8_t  *dest_y, *dest_cb, *dest_cr;
2181     int linesize, uvlinesize, mb_x, mb_y;
2182     const int end_mb_y= s->mb_y + FRAME_MBAFF;
2183     const int old_slice_type= h->slice_type;
2184
2185     if(h->deblocking_filter) {
2186         for(mb_x= 0; mb_x<s->mb_width; mb_x++){
2187             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
2188                 int mb_xy, mb_type;
2189                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2190                 h->slice_num= h->slice_table[mb_xy];
2191                 mb_type= s->current_picture.mb_type[mb_xy];
2192                 h->list_count= h->list_counts[mb_xy];
2193
2194                 if(FRAME_MBAFF)
2195                     h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2196
2197                 s->mb_x= mb_x;
2198                 s->mb_y= mb_y;
2199                 dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2200                 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2201                 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2202                     //FIXME simplify above
2203
2204                 if (MB_FIELD) {
2205                     linesize   = h->mb_linesize   = s->linesize * 2;
2206                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2207                     if(mb_y&1){ //FIXME move out of this function?
2208                         dest_y -= s->linesize*15;
2209                         dest_cb-= s->uvlinesize*7;
2210                         dest_cr-= s->uvlinesize*7;
2211                     }
2212                 } else {
2213                     linesize   = h->mb_linesize   = s->linesize;
2214                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2215                 }
2216                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2217                 if(fill_filter_caches(h, mb_type))
2218                     continue;
2219                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2220                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2221
2222                 if (FRAME_MBAFF) {
2223                     ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2224                 } else {
2225                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2226                 }
2227             }
2228         }
2229     }
2230     h->slice_type= old_slice_type;
2231     s->mb_x= 0;
2232     s->mb_y= end_mb_y - FRAME_MBAFF;
2233     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2234     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2235 }
2236
2237 static void predict_field_decoding_flag(H264Context *h){
2238     MpegEncContext * const s = &h->s;
2239     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2240     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
2241                 ? s->current_picture.mb_type[mb_xy-1]
2242                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
2243                 ? s->current_picture.mb_type[mb_xy-s->mb_stride]
2244                 : 0;
2245     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2246 }
2247
2248 static int decode_slice(struct AVCodecContext *avctx, void *arg){
2249     H264Context *h = *(void**)arg;
2250     MpegEncContext * const s = &h->s;
2251     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2252
2253     s->mb_skip_run= -1;
2254
2255     h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
2256                     (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
2257
2258     if( h->pps.cabac ) {
2259         /* realign */
2260         align_get_bits( &s->gb );
2261
2262         /* init cabac */
2263         ff_init_cabac_states( &h->cabac);
2264         ff_init_cabac_decoder( &h->cabac,
2265                                s->gb.buffer + get_bits_count(&s->gb)/8,
2266                                (get_bits_left(&s->gb) + 7)/8);
2267
2268         ff_h264_init_cabac_states(h);
2269
2270         for(;;){
2271 //START_TIMER
2272             int ret = ff_h264_decode_mb_cabac(h);
2273             int eos;
2274 //STOP_TIMER("decode_mb_cabac")
2275
2276             if(ret>=0) ff_h264_hl_decode_mb(h);
2277
2278             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
2279                 s->mb_y++;
2280
2281                 ret = ff_h264_decode_mb_cabac(h);
2282
2283                 if(ret>=0) ff_h264_hl_decode_mb(h);
2284                 s->mb_y--;
2285             }
2286             eos = get_cabac_terminate( &h->cabac );
2287
2288             if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
2289                 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);
2290                 return 0;
2291             }
2292             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2293                 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);
2294                 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);
2295                 return -1;
2296             }
2297
2298             if( ++s->mb_x >= s->mb_width ) {
2299                 s->mb_x = 0;
2300                 loop_filter(h);
2301                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2302                 ++s->mb_y;
2303                 if(FIELD_OR_MBAFF_PICTURE) {
2304                     ++s->mb_y;
2305                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2306                         predict_field_decoding_flag(h);
2307                 }
2308             }
2309
2310             if( eos || s->mb_y >= s->mb_height ) {
2311                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2312                 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);
2313                 return 0;
2314             }
2315         }
2316
2317     } else {
2318         for(;;){
2319             int ret = ff_h264_decode_mb_cavlc(h);
2320
2321             if(ret>=0) ff_h264_hl_decode_mb(h);
2322
2323             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
2324                 s->mb_y++;
2325                 ret = ff_h264_decode_mb_cavlc(h);
2326
2327                 if(ret>=0) ff_h264_hl_decode_mb(h);
2328                 s->mb_y--;
2329             }
2330
2331             if(ret<0){
2332                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2333                 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);
2334
2335                 return -1;
2336             }
2337
2338             if(++s->mb_x >= s->mb_width){
2339                 s->mb_x=0;
2340                 loop_filter(h);
2341                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2342                 ++s->mb_y;
2343                 if(FIELD_OR_MBAFF_PICTURE) {
2344                     ++s->mb_y;
2345                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2346                         predict_field_decoding_flag(h);
2347                 }
2348                 if(s->mb_y >= s->mb_height){
2349                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2350
2351                     if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
2352                         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);
2353
2354                         return 0;
2355                     }else{
2356                         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);
2357
2358                         return -1;
2359                     }
2360                 }
2361             }
2362
2363             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
2364                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2365                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2366                     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);
2367
2368                     return 0;
2369                 }else{
2370                     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);
2371
2372                     return -1;
2373                 }
2374             }
2375         }
2376     }
2377
2378 #if 0
2379     for(;s->mb_y < s->mb_height; s->mb_y++){
2380         for(;s->mb_x < s->mb_width; s->mb_x++){
2381             int ret= decode_mb(h);
2382
2383             ff_h264_hl_decode_mb(h);
2384
2385             if(ret<0){
2386                 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2387                 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);
2388
2389                 return -1;
2390             }
2391
2392             if(++s->mb_x >= s->mb_width){
2393                 s->mb_x=0;
2394                 if(++s->mb_y >= s->mb_height){
2395                     if(get_bits_count(s->gb) == s->gb.size_in_bits){
2396                         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);
2397
2398                         return 0;
2399                     }else{
2400                         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);
2401
2402                         return -1;
2403                     }
2404                 }
2405             }
2406
2407             if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2408                 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2409                     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);
2410
2411                     return 0;
2412                 }else{
2413                     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);
2414
2415                     return -1;
2416                 }
2417             }
2418         }
2419         s->mb_x=0;
2420         ff_draw_horiz_band(s, 16*s->mb_y, 16);
2421     }
2422 #endif
2423     return -1; //not reached
2424 }
2425
2426 /**
2427  * Call decode_slice() for each context.
2428  *
2429  * @param h h264 master context
2430  * @param context_count number of contexts to execute
2431  */
2432 static void execute_decode_slices(H264Context *h, int context_count){
2433     MpegEncContext * const s = &h->s;
2434     AVCodecContext * const avctx= s->avctx;
2435     H264Context *hx;
2436     int i;
2437
2438     if (s->avctx->hwaccel)
2439         return;
2440     if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2441         return;
2442     if(context_count == 1) {
2443         decode_slice(avctx, &h);
2444     } else {
2445         for(i = 1; i < context_count; i++) {
2446             hx = h->thread_context[i];
2447             hx->s.error_recognition = avctx->error_recognition;
2448             hx->s.error_count = 0;
2449         }
2450
2451         avctx->execute(avctx, (void *)decode_slice,
2452                        h->thread_context, NULL, context_count, sizeof(void*));
2453
2454         /* pull back stuff from slices to master context */
2455         hx = h->thread_context[context_count - 1];
2456         s->mb_x = hx->s.mb_x;
2457         s->mb_y = hx->s.mb_y;
2458         s->dropable = hx->s.dropable;
2459         s->picture_structure = hx->s.picture_structure;
2460         for(i = 1; i < context_count; i++)
2461             h->s.error_count += h->thread_context[i]->s.error_count;
2462     }
2463 }
2464
2465
2466 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
2467     MpegEncContext * const s = &h->s;
2468     AVCodecContext * const avctx= s->avctx;
2469     int buf_index=0;
2470     H264Context *hx; ///< thread context
2471     int context_count = 0;
2472     int next_avc= h->is_avc ? 0 : buf_size;
2473
2474     h->max_contexts = avctx->thread_count;
2475 #if 0
2476     int i;
2477     for(i=0; i<50; i++){
2478         av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
2479     }
2480 #endif
2481     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
2482         h->current_slice = 0;
2483         if (!s->first_field)
2484             s->current_picture_ptr= NULL;
2485         ff_h264_reset_sei(h);
2486     }
2487
2488     for(;;){
2489         int consumed;
2490         int dst_length;
2491         int bit_length;
2492         const uint8_t *ptr;
2493         int i, nalsize = 0;
2494         int err;
2495
2496         if(buf_index >= next_avc) {
2497             if(buf_index >= buf_size) break;
2498             nalsize = 0;
2499             for(i = 0; i < h->nal_length_size; i++)
2500                 nalsize = (nalsize << 8) | buf[buf_index++];
2501             if(nalsize <= 1 || nalsize > buf_size - buf_index){
2502                 if(nalsize == 1){
2503                     buf_index++;
2504                     continue;
2505                 }else{
2506                     av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2507                     break;
2508                 }
2509             }
2510             next_avc= buf_index + nalsize;
2511         } else {
2512             // start code prefix search
2513             for(; buf_index + 3 < next_avc; buf_index++){
2514                 // This should always succeed in the first iteration.
2515                 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2516                     break;
2517             }
2518
2519             if(buf_index+3 >= buf_size) break;
2520
2521             buf_index+=3;
2522             if(buf_index >= next_avc) continue;
2523         }
2524
2525         hx = h->thread_context[context_count];
2526
2527         ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
2528         if (ptr==NULL || dst_length < 0){
2529             return -1;
2530         }
2531         i= buf_index + consumed;
2532         if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
2533            buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
2534             s->workaround_bugs |= FF_BUG_TRUNCATED;
2535
2536         if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
2537         while(ptr[dst_length - 1] == 0 && dst_length > 0)
2538             dst_length--;
2539         }
2540         bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
2541
2542         if(s->avctx->debug&FF_DEBUG_STARTCODE){
2543             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);
2544         }
2545
2546         if (h->is_avc && (nalsize != consumed) && nalsize){
2547             av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
2548         }
2549
2550         buf_index += consumed;
2551
2552         if(  (s->hurry_up == 1 && h->nal_ref_idc  == 0) //FIXME do not discard SEI id
2553            ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0))
2554             continue;
2555
2556       again:
2557         err = 0;
2558         switch(hx->nal_unit_type){
2559         case NAL_IDR_SLICE:
2560             if (h->nal_unit_type != NAL_IDR_SLICE) {
2561                 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2562                 return -1;
2563             }
2564             idr(h); //FIXME ensure we don't loose some frames if there is reordering
2565         case NAL_SLICE:
2566             init_get_bits(&hx->s.gb, ptr, bit_length);
2567             hx->intra_gb_ptr=
2568             hx->inter_gb_ptr= &hx->s.gb;
2569             hx->s.data_partitioning = 0;
2570
2571             if((err = decode_slice_header(hx, h)))
2572                break;
2573
2574             avctx->profile = hx->sps.profile_idc;
2575             avctx->level   = hx->sps.level_idc;
2576
2577             if (s->avctx->hwaccel && h->current_slice == 1) {
2578                 if (s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
2579                     return -1;
2580             }
2581
2582             s->current_picture_ptr->key_frame |=
2583                     (hx->nal_unit_type == NAL_IDR_SLICE) ||
2584                     (h->sei_recovery_frame_cnt >= 0);
2585             if(hx->redundant_pic_count==0 && hx->s.hurry_up < 5
2586                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2587                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2588                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2589                && avctx->skip_frame < AVDISCARD_ALL){
2590                 if(avctx->hwaccel) {
2591                     if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
2592                         return -1;
2593                 }else
2594                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
2595                     static const uint8_t start_code[] = {0x00, 0x00, 0x01};
2596                     ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
2597                     ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
2598                 }else
2599                     context_count++;
2600             }
2601             break;
2602         case NAL_DPA:
2603             init_get_bits(&hx->s.gb, ptr, bit_length);
2604             hx->intra_gb_ptr=
2605             hx->inter_gb_ptr= NULL;
2606
2607             if ((err = decode_slice_header(hx, h)) < 0)
2608                 break;
2609
2610             avctx->profile = hx->sps.profile_idc;
2611             avctx->level   = hx->sps.level_idc;
2612
2613             hx->s.data_partitioning = 1;
2614
2615             break;
2616         case NAL_DPB:
2617             init_get_bits(&hx->intra_gb, ptr, bit_length);
2618             hx->intra_gb_ptr= &hx->intra_gb;
2619             break;
2620         case NAL_DPC:
2621             init_get_bits(&hx->inter_gb, ptr, bit_length);
2622             hx->inter_gb_ptr= &hx->inter_gb;
2623
2624             if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
2625                && s->context_initialized
2626                && s->hurry_up < 5
2627                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2628                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2629                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2630                && avctx->skip_frame < AVDISCARD_ALL)
2631                 context_count++;
2632             break;
2633         case NAL_SEI:
2634             init_get_bits(&s->gb, ptr, bit_length);
2635             ff_h264_decode_sei(h);
2636             break;
2637         case NAL_SPS:
2638             init_get_bits(&s->gb, ptr, bit_length);
2639             ff_h264_decode_seq_parameter_set(h);
2640
2641             if(s->flags& CODEC_FLAG_LOW_DELAY)
2642                 s->low_delay=1;
2643
2644             if(avctx->has_b_frames < 2)
2645                 avctx->has_b_frames= !s->low_delay;
2646             break;
2647         case NAL_PPS:
2648             init_get_bits(&s->gb, ptr, bit_length);
2649
2650             ff_h264_decode_picture_parameter_set(h, bit_length);
2651
2652             break;
2653         case NAL_AUD:
2654         case NAL_END_SEQUENCE:
2655         case NAL_END_STREAM:
2656         case NAL_FILLER_DATA:
2657         case NAL_SPS_EXT:
2658         case NAL_AUXILIARY_SLICE:
2659             break;
2660         default:
2661             av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
2662         }
2663
2664         if(context_count == h->max_contexts) {
2665             execute_decode_slices(h, context_count);
2666             context_count = 0;
2667         }
2668
2669         if (err < 0)
2670             av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
2671         else if(err == 1) {
2672             /* Slice could not be decoded in parallel mode, copy down
2673              * NAL unit stuff to context 0 and restart. Note that
2674              * rbsp_buffer is not transferred, but since we no longer
2675              * run in parallel mode this should not be an issue. */
2676             h->nal_unit_type = hx->nal_unit_type;
2677             h->nal_ref_idc   = hx->nal_ref_idc;
2678             hx = h;
2679             goto again;
2680         }
2681     }
2682     if(context_count)
2683         execute_decode_slices(h, context_count);
2684     return buf_index;
2685 }
2686
2687 /**
2688  * returns the number of bytes consumed for building the current frame
2689  */
2690 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
2691         if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
2692         if(pos+10>buf_size) pos=buf_size; // oops ;)
2693
2694         return pos;
2695 }
2696
2697 static int decode_frame(AVCodecContext *avctx,
2698                              void *data, int *data_size,
2699                              AVPacket *avpkt)
2700 {
2701     const uint8_t *buf = avpkt->data;
2702     int buf_size = avpkt->size;
2703     H264Context *h = avctx->priv_data;
2704     MpegEncContext *s = &h->s;
2705     AVFrame *pict = data;
2706     int buf_index;
2707
2708     s->flags= avctx->flags;
2709     s->flags2= avctx->flags2;
2710
2711    /* end of stream, output what is still in the buffers */
2712     if (buf_size == 0) {
2713         Picture *out;
2714         int i, out_idx;
2715
2716 //FIXME factorize this with the output code below
2717         out = h->delayed_pic[0];
2718         out_idx = 0;
2719         for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2720             if(h->delayed_pic[i]->poc < out->poc){
2721                 out = h->delayed_pic[i];
2722                 out_idx = i;
2723             }
2724
2725         for(i=out_idx; h->delayed_pic[i]; i++)
2726             h->delayed_pic[i] = h->delayed_pic[i+1];
2727
2728         if(out){
2729             *data_size = sizeof(AVFrame);
2730             *pict= *(AVFrame*)out;
2731         }
2732
2733         return 0;
2734     }
2735
2736     buf_index=decode_nal_units(h, buf, buf_size);
2737     if(buf_index < 0)
2738         return -1;
2739
2740     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
2741         if (avctx->skip_frame >= AVDISCARD_NONREF || s->hurry_up) return 0;
2742         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
2743         return -1;
2744     }
2745
2746     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
2747         Picture *out = s->current_picture_ptr;
2748         Picture *cur = s->current_picture_ptr;
2749         int i, pics, out_of_order, out_idx;
2750
2751         field_end(h);
2752
2753         if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
2754             /* Wait for second field. */
2755             *data_size = 0;
2756
2757         } else {
2758             cur->interlaced_frame = 0;
2759             cur->repeat_pict = 0;
2760
2761             /* Signal interlacing information externally. */
2762             /* Prioritize picture timing SEI information over used decoding process if it exists. */
2763
2764             if(h->sps.pic_struct_present_flag){
2765                 switch (h->sei_pic_struct)
2766                 {
2767                 case SEI_PIC_STRUCT_FRAME:
2768                     break;
2769                 case SEI_PIC_STRUCT_TOP_FIELD:
2770                 case SEI_PIC_STRUCT_BOTTOM_FIELD:
2771                     cur->interlaced_frame = 1;
2772                     break;
2773                 case SEI_PIC_STRUCT_TOP_BOTTOM:
2774                 case SEI_PIC_STRUCT_BOTTOM_TOP:
2775                     if (FIELD_OR_MBAFF_PICTURE)
2776                         cur->interlaced_frame = 1;
2777                     else
2778                         // try to flag soft telecine progressive
2779                         cur->interlaced_frame = h->prev_interlaced_frame;
2780                     break;
2781                 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
2782                 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
2783                     // Signal the possibility of telecined film externally (pic_struct 5,6)
2784                     // From these hints, let the applications decide if they apply deinterlacing.
2785                     cur->repeat_pict = 1;
2786                     break;
2787                 case SEI_PIC_STRUCT_FRAME_DOUBLING:
2788                     // Force progressive here, as doubling interlaced frame is a bad idea.
2789                     cur->repeat_pict = 2;
2790                     break;
2791                 case SEI_PIC_STRUCT_FRAME_TRIPLING:
2792                     cur->repeat_pict = 4;
2793                     break;
2794                 }
2795
2796                 if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
2797                     cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
2798             }else{
2799                 /* Derive interlacing flag from used decoding process. */
2800                 cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
2801             }
2802             h->prev_interlaced_frame = cur->interlaced_frame;
2803
2804             if (cur->field_poc[0] != cur->field_poc[1]){
2805                 /* Derive top_field_first from field pocs. */
2806                 cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
2807             }else{
2808                 if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
2809                     /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
2810                     if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
2811                       || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
2812                         cur->top_field_first = 1;
2813                     else
2814                         cur->top_field_first = 0;
2815                 }else{
2816                     /* Most likely progressive */
2817                     cur->top_field_first = 0;
2818                 }
2819             }
2820
2821         //FIXME do something with unavailable reference frames
2822
2823             /* Sort B-frames into display order */
2824
2825             if(h->sps.bitstream_restriction_flag
2826                && s->avctx->has_b_frames < h->sps.num_reorder_frames){
2827                 s->avctx->has_b_frames = h->sps.num_reorder_frames;
2828                 s->low_delay = 0;
2829             }
2830
2831             if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
2832                && !h->sps.bitstream_restriction_flag){
2833                 s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
2834                 s->low_delay= 0;
2835             }
2836
2837             pics = 0;
2838             while(h->delayed_pic[pics]) pics++;
2839
2840             assert(pics <= MAX_DELAYED_PIC_COUNT);
2841
2842             h->delayed_pic[pics++] = cur;
2843             if(cur->reference == 0)
2844                 cur->reference = DELAYED_PIC_REF;
2845
2846             out = h->delayed_pic[0];
2847             out_idx = 0;
2848             for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2849                 if(h->delayed_pic[i]->poc < out->poc){
2850                     out = h->delayed_pic[i];
2851                     out_idx = i;
2852                 }
2853             if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
2854                 h->outputed_poc= INT_MIN;
2855             out_of_order = out->poc < h->outputed_poc;
2856
2857             if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
2858                 { }
2859             else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
2860                || (s->low_delay &&
2861                 ((h->outputed_poc != INT_MIN && out->poc > h->outputed_poc + 2)
2862                  || cur->pict_type == FF_B_TYPE)))
2863             {
2864                 s->low_delay = 0;
2865                 s->avctx->has_b_frames++;
2866             }
2867
2868             if(out_of_order || pics > s->avctx->has_b_frames){
2869                 out->reference &= ~DELAYED_PIC_REF;
2870                 for(i=out_idx; h->delayed_pic[i]; i++)
2871                     h->delayed_pic[i] = h->delayed_pic[i+1];
2872             }
2873             if(!out_of_order && pics > s->avctx->has_b_frames){
2874                 *data_size = sizeof(AVFrame);
2875
2876                 if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
2877                     h->outputed_poc = INT_MIN;
2878                 } else
2879                     h->outputed_poc = out->poc;
2880                 *pict= *(AVFrame*)out;
2881             }else{
2882                 av_log(avctx, AV_LOG_DEBUG, "no picture\n");
2883             }
2884         }
2885     }
2886
2887     assert(pict->data[0] || !*data_size);
2888     ff_print_debug_info(s, pict);
2889 //printf("out %d\n", (int)pict->data[0]);
2890
2891     return get_consumed_bytes(s, buf_index, buf_size);
2892 }
2893 #if 0
2894 static inline void fill_mb_avail(H264Context *h){
2895     MpegEncContext * const s = &h->s;
2896     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2897
2898     if(s->mb_y){
2899         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
2900         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
2901         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
2902     }else{
2903         h->mb_avail[0]=
2904         h->mb_avail[1]=
2905         h->mb_avail[2]= 0;
2906     }
2907     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
2908     h->mb_avail[4]= 1; //FIXME move out
2909     h->mb_avail[5]= 0; //FIXME move out
2910 }
2911 #endif
2912
2913 #ifdef TEST
2914 #undef printf
2915 #undef random
2916 #define COUNT 8000
2917 #define SIZE (COUNT*40)
2918 int main(void){
2919     int i;
2920     uint8_t temp[SIZE];
2921     PutBitContext pb;
2922     GetBitContext gb;
2923 //    int int_temp[10000];
2924     DSPContext dsp;
2925     AVCodecContext avctx;
2926
2927     dsputil_init(&dsp, &avctx);
2928
2929     init_put_bits(&pb, temp, SIZE);
2930     printf("testing unsigned exp golomb\n");
2931     for(i=0; i<COUNT; i++){
2932         START_TIMER
2933         set_ue_golomb(&pb, i);
2934         STOP_TIMER("set_ue_golomb");
2935     }
2936     flush_put_bits(&pb);
2937
2938     init_get_bits(&gb, temp, 8*SIZE);
2939     for(i=0; i<COUNT; i++){
2940         int j, s;
2941
2942         s= show_bits(&gb, 24);
2943
2944         START_TIMER
2945         j= get_ue_golomb(&gb);
2946         if(j != i){
2947             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
2948 //            return -1;
2949         }
2950         STOP_TIMER("get_ue_golomb");
2951     }
2952
2953
2954     init_put_bits(&pb, temp, SIZE);
2955     printf("testing signed exp golomb\n");
2956     for(i=0; i<COUNT; i++){
2957         START_TIMER
2958         set_se_golomb(&pb, i - COUNT/2);
2959         STOP_TIMER("set_se_golomb");
2960     }
2961     flush_put_bits(&pb);
2962
2963     init_get_bits(&gb, temp, 8*SIZE);
2964     for(i=0; i<COUNT; i++){
2965         int j, s;
2966
2967         s= show_bits(&gb, 24);
2968
2969         START_TIMER
2970         j= get_se_golomb(&gb);
2971         if(j != i - COUNT/2){
2972             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
2973 //            return -1;
2974         }
2975         STOP_TIMER("get_se_golomb");
2976     }
2977
2978 #if 0
2979     printf("testing 4x4 (I)DCT\n");
2980
2981     DCTELEM block[16];
2982     uint8_t src[16], ref[16];
2983     uint64_t error= 0, max_error=0;
2984
2985     for(i=0; i<COUNT; i++){
2986         int j;
2987 //        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
2988         for(j=0; j<16; j++){
2989             ref[j]= random()%255;
2990             src[j]= random()%255;
2991         }
2992
2993         h264_diff_dct_c(block, src, ref, 4);
2994
2995         //normalize
2996         for(j=0; j<16; j++){
2997 //            printf("%d ", block[j]);
2998             block[j]= block[j]*4;
2999             if(j&1) block[j]= (block[j]*4 + 2)/5;
3000             if(j&4) block[j]= (block[j]*4 + 2)/5;
3001         }
3002 //        printf("\n");
3003
3004         s->dsp.h264_idct_add(ref, block, 4);
3005 /*        for(j=0; j<16; j++){
3006             printf("%d ", ref[j]);
3007         }
3008         printf("\n");*/
3009
3010         for(j=0; j<16; j++){
3011             int diff= FFABS(src[j] - ref[j]);
3012
3013             error+= diff*diff;
3014             max_error= FFMAX(max_error, diff);
3015         }
3016     }
3017     printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
3018     printf("testing quantizer\n");
3019     for(qp=0; qp<52; qp++){
3020         for(i=0; i<16; i++)
3021             src1_block[i]= src2_block[i]= random()%255;
3022
3023     }
3024     printf("Testing NAL layer\n");
3025
3026     uint8_t bitstream[COUNT];
3027     uint8_t nal[COUNT*2];
3028     H264Context h;
3029     memset(&h, 0, sizeof(H264Context));
3030
3031     for(i=0; i<COUNT; i++){
3032         int zeros= i;
3033         int nal_length;
3034         int consumed;
3035         int out_length;
3036         uint8_t *out;
3037         int j;
3038
3039         for(j=0; j<COUNT; j++){
3040             bitstream[j]= (random() % 255) + 1;
3041         }
3042
3043         for(j=0; j<zeros; j++){
3044             int pos= random() % COUNT;
3045             while(bitstream[pos] == 0){
3046                 pos++;
3047                 pos %= COUNT;
3048             }
3049             bitstream[pos]=0;
3050         }
3051
3052         START_TIMER
3053
3054         nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
3055         if(nal_length<0){
3056             printf("encoding failed\n");
3057             return -1;
3058         }
3059
3060         out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
3061
3062         STOP_TIMER("NAL")
3063
3064         if(out_length != COUNT){
3065             printf("incorrect length %d %d\n", out_length, COUNT);
3066             return -1;
3067         }
3068
3069         if(consumed != nal_length){
3070             printf("incorrect consumed length %d %d\n", nal_length, consumed);
3071             return -1;
3072         }
3073
3074         if(memcmp(bitstream, out, COUNT)){
3075             printf("mismatch\n");
3076             return -1;
3077         }
3078     }
3079 #endif
3080
3081     printf("Testing RBSP\n");
3082
3083
3084     return 0;
3085 }
3086 #endif /* TEST */
3087
3088
3089 av_cold void ff_h264_free_context(H264Context *h)
3090 {
3091     int i;
3092
3093     free_tables(h); //FIXME cleanup init stuff perhaps
3094
3095     for(i = 0; i < MAX_SPS_COUNT; i++)
3096         av_freep(h->sps_buffers + i);
3097
3098     for(i = 0; i < MAX_PPS_COUNT; i++)
3099         av_freep(h->pps_buffers + i);
3100 }
3101
3102 av_cold int ff_h264_decode_end(AVCodecContext *avctx)
3103 {
3104     H264Context *h = avctx->priv_data;
3105     MpegEncContext *s = &h->s;
3106
3107     ff_h264_free_context(h);
3108
3109     MPV_common_end(s);
3110
3111 //    memset(h, 0, sizeof(H264Context));
3112
3113     return 0;
3114 }
3115
3116
3117 AVCodec h264_decoder = {
3118     "h264",
3119     CODEC_TYPE_VIDEO,
3120     CODEC_ID_H264,
3121     sizeof(H264Context),
3122     ff_h264_decode_init,
3123     NULL,
3124     ff_h264_decode_end,
3125     decode_frame,
3126     /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3127     .flush= flush_dpb,
3128     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
3129     .pix_fmts= ff_hwaccel_pixfmt_list_420,
3130 };
3131
3132 #if CONFIG_H264_VDPAU_DECODER
3133 AVCodec h264_vdpau_decoder = {
3134     "h264_vdpau",
3135     CODEC_TYPE_VIDEO,
3136     CODEC_ID_H264,
3137     sizeof(H264Context),
3138     ff_h264_decode_init,
3139     NULL,
3140     ff_h264_decode_end,
3141     decode_frame,
3142     CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3143     .flush= flush_dpb,
3144     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
3145     .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
3146 };
3147 #endif