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

   1 /*
   2  * H.26L/H.264/AVC/JVT/14496-10/... 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
  24  * H.264 / AVC / MPEG4 part10 codec.
  25  * @author Michael Niedermayer <michaelni@gmx.at>
  26  */
  27
  28 #include "libavcore/imgutils.h"
  29 #include "internal.h"
  30 #include "dsputil.h"
  31 #include "avcodec.h"
  32 #include "mpegvideo.h"
  33 #include "h264.h"
  34 #include "h264data.h"
  35 #include "h264_mvpred.h"
  36 #include "h264_parser.h"
  37 #include "golomb.h"
  38 #include "mathops.h"
  39 #include "rectangle.h"
  40 #include "vdpau_internal.h"
  41 #include "libavutil/avassert.h"
  42
  43 #include "cabac.h"
  44
  45 //#undef NDEBUG
  46 #include <assert.h>
  47
  48 static const uint8_t rem6[52]={
  49 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,
  50 };
  51
  52 static const uint8_t div6[52]={
  53 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,
  54 };
  55
  56 static const enum PixelFormat hwaccel_pixfmt_list_h264_jpeg_420[] = {
  57     PIX_FMT_DXVA2_VLD,
  58     PIX_FMT_VAAPI_VLD,
  59     PIX_FMT_YUVJ420P,
  60     PIX_FMT_NONE
  61 };
  62
  63 void ff_h264_write_back_intra_pred_mode(H264Context *h){
  64     int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
  65
  66     AV_COPY32(mode, h->intra4x4_pred_mode_cache + 4 + 8*4);
  67     mode[4]= h->intra4x4_pred_mode_cache[7+8*3];
  68     mode[5]= h->intra4x4_pred_mode_cache[7+8*2];
  69     mode[6]= h->intra4x4_pred_mode_cache[7+8*1];
  70 }
  71
  72 /**
  73  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
  74  */
  75 int ff_h264_check_intra4x4_pred_mode(H264Context *h){
  76     MpegEncContext * const s = &h->s;
  77     static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
  78     static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
  79     int i;
  80
  81     if(!(h->top_samples_available&0x8000)){
  82         for(i=0; i<4; i++){
  83             int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
  84             if(status<0){
  85                 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
  86                 return -1;
  87             } else if(status){
  88                 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
  89             }
  90         }
  91     }
  92
  93     if((h->left_samples_available&0x8888)!=0x8888){
  94         static const int mask[4]={0x8000,0x2000,0x80,0x20};
  95         for(i=0; i<4; i++){
  96             if(!(h->left_samples_available&mask[i])){
  97                 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
  98                 if(status<0){
  99                     av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
 100                     return -1;
 101                 } else if(status){
 102                     h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
 103                 }
 104             }
 105         }
 106     }
 107
 108     return 0;
 109 } //FIXME cleanup like ff_h264_check_intra_pred_mode
 110
 111 /**
 112  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
 113  */
 114 int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
 115     MpegEncContext * const s = &h->s;
 116     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
 117     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
 118
 119     if(mode > 6U) {
 120         av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
 121         return -1;
 122     }
 123
 124     if(!(h->top_samples_available&0x8000)){
 125         mode= top[ mode ];
 126         if(mode<0){
 127             av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
 128             return -1;
 129         }
 130     }
 131
 132     if((h->left_samples_available&0x8080) != 0x8080){
 133         mode= left[ mode ];
 134         if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
 135             mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
 136         }
 137         if(mode<0){
 138             av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
 139             return -1;
 140         }
 141     }
 142
 143     return mode;
 144 }
 145
 146 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
 147     int i, si, di;
 148     uint8_t *dst;
 149     int bufidx;
 150
 151 //    src[0]&0x80;                //forbidden bit
 152     h->nal_ref_idc= src[0]>>5;
 153     h->nal_unit_type= src[0]&0x1F;
 154
 155     src++; length--;
 156 #if 0
 157     for(i=0; i<length; i++)
 158         printf("%2X ", src[i]);
 159 #endif
 160
 161 #if HAVE_FAST_UNALIGNED
 162 # if HAVE_FAST_64BIT
 163 #   define RS 7
 164     for(i=0; i+1<length; i+=9){
 165         if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
 166 # else
 167 #   define RS 3
 168     for(i=0; i+1<length; i+=5){
 169         if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
 170 # endif
 171             continue;
 172         if(i>0 && !src[i]) i--;
 173         while(src[i]) i++;
 174 #else
 175 #   define RS 0
 176     for(i=0; i+1<length; i+=2){
 177         if(src[i]) continue;
 178         if(i>0 && src[i-1]==0) i--;
 179 #endif
 180         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
 181             if(src[i+2]!=3){
 182                 /* startcode, so we must be past the end */
 183                 length=i;
 184             }
 185             break;
 186         }
 187         i-= RS;
 188     }
 189
 190     if(i>=length-1){ //no escaped 0
 191         *dst_length= length;
 192         *consumed= length+1; //+1 for the header
 193         return src;
 194     }
 195
 196     bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
 197     av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
 198     dst= h->rbsp_buffer[bufidx];
 199
 200     if (dst == NULL){
 201         return NULL;
 202     }
 203
 204 //printf("decoding esc\n");
 205     memcpy(dst, src, i);
 206     si=di=i;
 207     while(si+2<length){
 208         //remove escapes (very rare 1:2^22)
 209         if(src[si+2]>3){
 210             dst[di++]= src[si++];
 211             dst[di++]= src[si++];
 212         }else if(src[si]==0 && src[si+1]==0){
 213             if(src[si+2]==3){ //escape
 214                 dst[di++]= 0;
 215                 dst[di++]= 0;
 216                 si+=3;
 217                 continue;
 218             }else //next start code
 219                 goto nsc;
 220         }
 221
 222         dst[di++]= src[si++];
 223     }
 224     while(si<length)
 225         dst[di++]= src[si++];
 226 nsc:
 227
 228     memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
 229
 230     *dst_length= di;
 231     *consumed= si + 1;//+1 for the header
 232 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
 233     return dst;
 234 }
 235
 236 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
 237     int v= *src;
 238     int r;
 239
 240     tprintf(h->s.avctx, "rbsp trailing %X\n", v);
 241
 242     for(r=1; r<9; r++){
 243         if(v&1) return r;
 244         v>>=1;
 245     }
 246     return 0;
 247 }
 248
 249 #if 0
 250 /**
 251  * DCT transforms the 16 dc values.
 252  * @param qp quantization parameter ??? FIXME
 253  */
 254 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
 255 //    const int qmul= dequant_coeff[qp][0];
 256     int i;
 257     int temp[16]; //FIXME check if this is a good idea
 258     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
 259     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
 260
 261     for(i=0; i<4; i++){
 262         const int offset= y_offset[i];
 263         const int z0= block[offset+stride*0] + block[offset+stride*4];
 264         const int z1= block[offset+stride*0] - block[offset+stride*4];
 265         const int z2= block[offset+stride*1] - block[offset+stride*5];
 266         const int z3= block[offset+stride*1] + block[offset+stride*5];
 267
 268         temp[4*i+0]= z0+z3;
 269         temp[4*i+1]= z1+z2;
 270         temp[4*i+2]= z1-z2;
 271         temp[4*i+3]= z0-z3;
 272     }
 273
 274     for(i=0; i<4; i++){
 275         const int offset= x_offset[i];
 276         const int z0= temp[4*0+i] + temp[4*2+i];
 277         const int z1= temp[4*0+i] - temp[4*2+i];
 278         const int z2= temp[4*1+i] - temp[4*3+i];
 279         const int z3= temp[4*1+i] + temp[4*3+i];
 280
 281         block[stride*0 +offset]= (z0 + z3)>>1;
 282         block[stride*2 +offset]= (z1 + z2)>>1;
 283         block[stride*8 +offset]= (z1 - z2)>>1;
 284         block[stride*10+offset]= (z0 - z3)>>1;
 285     }
 286 }
 287 #endif
 288
 289 #undef xStride
 290 #undef stride
 291
 292 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qmul){
 293     const int stride= 16*2;
 294     const int xStride= 16;
 295     int a,b,c,d,e;
 296
 297     a= block[stride*0 + xStride*0];
 298     b= block[stride*0 + xStride*1];
 299     c= block[stride*1 + xStride*0];
 300     d= block[stride*1 + xStride*1];
 301
 302     e= a-b;
 303     a= a+b;
 304     b= c-d;
 305     c= c+d;
 306
 307     block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
 308     block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
 309     block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
 310     block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
 311 }
 312
 313 #if 0
 314 static void chroma_dc_dct_c(DCTELEM *block){
 315     const int stride= 16*2;
 316     const int xStride= 16;
 317     int a,b,c,d,e;
 318
 319     a= block[stride*0 + xStride*0];
 320     b= block[stride*0 + xStride*1];
 321     c= block[stride*1 + xStride*0];
 322     d= block[stride*1 + xStride*1];
 323
 324     e= a-b;
 325     a= a+b;
 326     b= c-d;
 327     c= c+d;
 328
 329     block[stride*0 + xStride*0]= (a+c);
 330     block[stride*0 + xStride*1]= (e+b);
 331     block[stride*1 + xStride*0]= (a-c);
 332     block[stride*1 + xStride*1]= (e-b);
 333 }
 334 #endif
 335
 336 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
 337                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 338                            int src_x_offset, int src_y_offset,
 339                            qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
 340     MpegEncContext * const s = &h->s;
 341     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
 342     int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
 343     const int luma_xy= (mx&3) + ((my&3)<<2);
 344     uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
 345     uint8_t * src_cb, * src_cr;
 346     int extra_width= h->emu_edge_width;
 347     int extra_height= h->emu_edge_height;
 348     int emu=0;
 349     const int full_mx= mx>>2;
 350     const int full_my= my>>2;
 351     const int pic_width  = 16*s->mb_width;
 352     const int pic_height = 16*s->mb_height >> MB_FIELD;
 353
 354     if(mx&7) extra_width -= 3;
 355     if(my&7) extra_height -= 3;
 356
 357     if(   full_mx < 0-extra_width
 358        || full_my < 0-extra_height
 359        || full_mx + 16/*FIXME*/ > pic_width + extra_width
 360        || full_my + 16/*FIXME*/ > pic_height + extra_height){
 361         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);
 362             src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
 363         emu=1;
 364     }
 365
 366     qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
 367     if(!square){
 368         qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
 369     }
 370
 371     if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
 372
 373     if(MB_FIELD){
 374         // chroma offset when predicting from a field of opposite parity
 375         my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
 376         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
 377     }
 378     src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
 379     src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
 380
 381     if(emu){
 382         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);
 383             src_cb= s->edge_emu_buffer;
 384     }
 385     chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
 386
 387     if(emu){
 388         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);
 389             src_cr= s->edge_emu_buffer;
 390     }
 391     chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
 392 }
 393
 394 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
 395                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 396                            int x_offset, int y_offset,
 397                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 398                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
 399                            int list0, int list1){
 400     MpegEncContext * const s = &h->s;
 401     qpel_mc_func *qpix_op=  qpix_put;
 402     h264_chroma_mc_func chroma_op= chroma_put;
 403
 404     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
 405     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
 406     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
 407     x_offset += 8*s->mb_x;
 408     y_offset += 8*(s->mb_y >> MB_FIELD);
 409
 410     if(list0){
 411         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
 412         mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
 413                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
 414                            qpix_op, chroma_op);
 415
 416         qpix_op=  qpix_avg;
 417         chroma_op= chroma_avg;
 418     }
 419
 420     if(list1){
 421         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
 422         mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
 423                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
 424                            qpix_op, chroma_op);
 425     }
 426 }
 427
 428 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
 429                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 430                            int x_offset, int y_offset,
 431                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 432                            h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
 433                            h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
 434                            int list0, int list1){
 435     MpegEncContext * const s = &h->s;
 436
 437     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
 438     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
 439     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
 440     x_offset += 8*s->mb_x;
 441     y_offset += 8*(s->mb_y >> MB_FIELD);
 442
 443     if(list0 && list1){
 444         /* don't optimize for luma-only case, since B-frames usually
 445          * use implicit weights => chroma too. */
 446         uint8_t *tmp_cb = s->obmc_scratchpad;
 447         uint8_t *tmp_cr = s->obmc_scratchpad + 8;
 448         uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
 449         int refn0 = h->ref_cache[0][ scan8[n] ];
 450         int refn1 = h->ref_cache[1][ scan8[n] ];
 451
 452         mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
 453                     dest_y, dest_cb, dest_cr,
 454                     x_offset, y_offset, qpix_put, chroma_put);
 455         mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
 456                     tmp_y, tmp_cb, tmp_cr,
 457                     x_offset, y_offset, qpix_put, chroma_put);
 458
 459         if(h->use_weight == 2){
 460             int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
 461             int weight1 = 64 - weight0;
 462             luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
 463             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
 464             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
 465         }else{
 466             luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
 467                             h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
 468                             h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
 469             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 470                             h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
 471                             h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
 472             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 473                             h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
 474                             h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
 475         }
 476     }else{
 477         int list = list1 ? 1 : 0;
 478         int refn = h->ref_cache[list][ scan8[n] ];
 479         Picture *ref= &h->ref_list[list][refn];
 480         mc_dir_part(h, ref, n, square, chroma_height, delta, list,
 481                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
 482                     qpix_put, chroma_put);
 483
 484         luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
 485                        h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
 486         if(h->use_weight_chroma){
 487             chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 488                              h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
 489             chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 490                              h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
 491         }
 492     }
 493 }
 494
 495 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
 496                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 497                            int x_offset, int y_offset,
 498                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 499                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
 500                            h264_weight_func *weight_op, h264_biweight_func *weight_avg,
 501                            int list0, int list1){
 502     if((h->use_weight==2 && list0 && list1
 503         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
 504        || h->use_weight==1)
 505         mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
 506                          x_offset, y_offset, qpix_put, chroma_put,
 507                          weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
 508     else
 509         mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
 510                     x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
 511 }
 512
 513 static inline void prefetch_motion(H264Context *h, int list){
 514     /* fetch pixels for estimated mv 4 macroblocks ahead
 515      * optimized for 64byte cache lines */
 516     MpegEncContext * const s = &h->s;
 517     const int refn = h->ref_cache[list][scan8[0]];
 518     if(refn >= 0){
 519         const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
 520         const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
 521         uint8_t **src= h->ref_list[list][refn].data;
 522         int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
 523         s->dsp.prefetch(src[0]+off, s->linesize, 4);
 524         off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
 525         s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
 526     }
 527 }
 528
 529 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 530                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
 531                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
 532                       h264_weight_func *weight_op, h264_biweight_func *weight_avg){
 533     MpegEncContext * const s = &h->s;
 534     const int mb_xy= h->mb_xy;
 535     const int mb_type= s->current_picture.mb_type[mb_xy];
 536
 537     assert(IS_INTER(mb_type));
 538
 539     prefetch_motion(h, 0);
 540
 541     if(IS_16X16(mb_type)){
 542         mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
 543                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
 544                 weight_op, weight_avg,
 545                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
 546     }else if(IS_16X8(mb_type)){
 547         mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
 548                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
 549                 &weight_op[1], &weight_avg[1],
 550                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
 551         mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
 552                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
 553                 &weight_op[1], &weight_avg[1],
 554                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
 555     }else if(IS_8X16(mb_type)){
 556         mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
 557                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 558                 &weight_op[2], &weight_avg[2],
 559                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
 560         mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
 561                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 562                 &weight_op[2], &weight_avg[2],
 563                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
 564     }else{
 565         int i;
 566
 567         assert(IS_8X8(mb_type));
 568
 569         for(i=0; i<4; i++){
 570             const int sub_mb_type= h->sub_mb_type[i];
 571             const int n= 4*i;
 572             int x_offset= (i&1)<<2;
 573             int y_offset= (i&2)<<1;
 574
 575             if(IS_SUB_8X8(sub_mb_type)){
 576                 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 577                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 578                     &weight_op[3], &weight_avg[3],
 579                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 580             }else if(IS_SUB_8X4(sub_mb_type)){
 581                 mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 582                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
 583                     &weight_op[4], &weight_avg[4],
 584                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 585                 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
 586                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
 587                     &weight_op[4], &weight_avg[4],
 588                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 589             }else if(IS_SUB_4X8(sub_mb_type)){
 590                 mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 591                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 592                     &weight_op[5], &weight_avg[5],
 593                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 594                 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
 595                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 596                     &weight_op[5], &weight_avg[5],
 597                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 598             }else{
 599                 int j;
 600                 assert(IS_SUB_4X4(sub_mb_type));
 601                 for(j=0; j<4; j++){
 602                     int sub_x_offset= x_offset + 2*(j&1);
 603                     int sub_y_offset= y_offset +   (j&2);
 604                     mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
 605                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 606                         &weight_op[6], &weight_avg[6],
 607                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 608                 }
 609             }
 610         }
 611     }
 612
 613     prefetch_motion(h, 1);
 614 }
 615
 616
 617 static void free_tables(H264Context *h, int free_rbsp){
 618     int i;
 619     H264Context *hx;
 620     av_freep(&h->intra4x4_pred_mode);
 621     av_freep(&h->chroma_pred_mode_table);
 622     av_freep(&h->cbp_table);
 623     av_freep(&h->mvd_table[0]);
 624     av_freep(&h->mvd_table[1]);
 625     av_freep(&h->direct_table);
 626     av_freep(&h->non_zero_count);
 627     av_freep(&h->slice_table_base);
 628     h->slice_table= NULL;
 629     av_freep(&h->list_counts);
 630
 631     av_freep(&h->mb2b_xy);
 632     av_freep(&h->mb2br_xy);
 633
 634     for(i = 0; i < MAX_THREADS; i++) {
 635         hx = h->thread_context[i];
 636         if(!hx) continue;
 637         av_freep(&hx->top_borders[1]);
 638         av_freep(&hx->top_borders[0]);
 639         av_freep(&hx->s.obmc_scratchpad);
 640         if (free_rbsp){
 641             av_freep(&hx->rbsp_buffer[1]);
 642             av_freep(&hx->rbsp_buffer[0]);
 643             hx->rbsp_buffer_size[0] = 0;
 644             hx->rbsp_buffer_size[1] = 0;
 645         }
 646         if (i) av_freep(&h->thread_context[i]);
 647     }
 648 }
 649
 650 static void init_dequant8_coeff_table(H264Context *h){
 651     int i,q,x;
 652     h->dequant8_coeff[0] = h->dequant8_buffer[0];
 653     h->dequant8_coeff[1] = h->dequant8_buffer[1];
 654
 655     for(i=0; i<2; i++ ){
 656         if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
 657             h->dequant8_coeff[1] = h->dequant8_buffer[0];
 658             break;
 659         }
 660
 661         for(q=0; q<52; q++){
 662             int shift = div6[q];
 663             int idx = rem6[q];
 664             for(x=0; x<64; x++)
 665                 h->dequant8_coeff[i][q][(x>>3)|((x&7)<<3)] =
 666                     ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
 667                     h->pps.scaling_matrix8[i][x]) << shift;
 668         }
 669     }
 670 }
 671
 672 static void init_dequant4_coeff_table(H264Context *h){
 673     int i,j,q,x;
 674     for(i=0; i<6; i++ ){
 675         h->dequant4_coeff[i] = h->dequant4_buffer[i];
 676         for(j=0; j<i; j++){
 677             if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
 678                 h->dequant4_coeff[i] = h->dequant4_buffer[j];
 679                 break;
 680             }
 681         }
 682         if(j<i)
 683             continue;
 684
 685         for(q=0; q<52; q++){
 686             int shift = div6[q] + 2;
 687             int idx = rem6[q];
 688             for(x=0; x<16; x++)
 689                 h->dequant4_coeff[i][q][(x>>2)|((x<<2)&0xF)] =
 690                     ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
 691                     h->pps.scaling_matrix4[i][x]) << shift;
 692         }
 693     }
 694 }
 695
 696 static void init_dequant_tables(H264Context *h){
 697     int i,x;
 698     init_dequant4_coeff_table(h);
 699     if(h->pps.transform_8x8_mode)
 700         init_dequant8_coeff_table(h);
 701     if(h->sps.transform_bypass){
 702         for(i=0; i<6; i++)
 703             for(x=0; x<16; x++)
 704                 h->dequant4_coeff[i][0][x] = 1<<6;
 705         if(h->pps.transform_8x8_mode)
 706             for(i=0; i<2; i++)
 707                 for(x=0; x<64; x++)
 708                     h->dequant8_coeff[i][0][x] = 1<<6;
 709     }
 710 }
 711
 712
 713 int ff_h264_alloc_tables(H264Context *h){
 714     MpegEncContext * const s = &h->s;
 715     const int big_mb_num= s->mb_stride * (s->mb_height+1);
 716     const int row_mb_num= 2*s->mb_stride*s->avctx->thread_count;
 717     int x,y;
 718
 719     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8  * sizeof(uint8_t), fail)
 720
 721     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 32 * sizeof(uint8_t), fail)
 722     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
 723     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
 724
 725     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
 726     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
 727     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
 728     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
 729     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
 730
 731     memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
 732     h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
 733
 734     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy  , big_mb_num * sizeof(uint32_t), fail);
 735     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
 736     for(y=0; y<s->mb_height; y++){
 737         for(x=0; x<s->mb_width; x++){
 738             const int mb_xy= x + y*s->mb_stride;
 739             const int b_xy = 4*x + 4*y*h->b_stride;
 740
 741             h->mb2b_xy [mb_xy]= b_xy;
 742             h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
 743         }
 744     }
 745
 746     s->obmc_scratchpad = NULL;
 747
 748     if(!h->dequant4_coeff[0])
 749         init_dequant_tables(h);
 750
 751     return 0;
 752 fail:
 753     free_tables(h, 1);
 754     return -1;
 755 }
 756
 757 /**
 758  * Mimic alloc_tables(), but for every context thread.
 759  */
 760 static void clone_tables(H264Context *dst, H264Context *src, int i){
 761     MpegEncContext * const s = &src->s;
 762     dst->intra4x4_pred_mode       = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
 763     dst->non_zero_count           = src->non_zero_count;
 764     dst->slice_table              = src->slice_table;
 765     dst->cbp_table                = src->cbp_table;
 766     dst->mb2b_xy                  = src->mb2b_xy;
 767     dst->mb2br_xy                 = src->mb2br_xy;
 768     dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
 769     dst->mvd_table[0]             = src->mvd_table[0] + i*8*2*s->mb_stride;
 770     dst->mvd_table[1]             = src->mvd_table[1] + i*8*2*s->mb_stride;
 771     dst->direct_table             = src->direct_table;
 772     dst->list_counts              = src->list_counts;
 773
 774     dst->s.obmc_scratchpad = NULL;
 775     ff_h264_pred_init(&dst->hpc, src->s.codec_id);
 776 }
 777
 778 /**
 779  * Init context
 780  * Allocate buffers which are not shared amongst multiple threads.
 781  */
 782 static int context_init(H264Context *h){
 783     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
 784     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
 785
 786     h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
 787     h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
 788
 789     return 0;
 790 fail:
 791     return -1; // free_tables will clean up for us
 792 }
 793
 794 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
 795
 796 static av_cold void common_init(H264Context *h){
 797     MpegEncContext * const s = &h->s;
 798
 799     s->width = s->avctx->width;
 800     s->height = s->avctx->height;
 801     s->codec_id= s->avctx->codec->id;
 802
 803     ff_h264dsp_init(&h->h264dsp);
 804     ff_h264_pred_init(&h->hpc, s->codec_id);
 805
 806     h->dequant_coeff_pps= -1;
 807     s->unrestricted_mv=1;
 808     s->decode=1; //FIXME
 809
 810     dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
 811
 812     memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
 813     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
 814 }
 815
 816 int ff_h264_decode_extradata(H264Context *h)
 817 {
 818     AVCodecContext *avctx = h->s.avctx;
 819
 820     if(*(char *)avctx->extradata == 1){
 821         int i, cnt, nalsize;
 822         unsigned char *p = avctx->extradata;
 823
 824         h->is_avc = 1;
 825
 826         if(avctx->extradata_size < 7) {
 827             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
 828             return -1;
 829         }
 830         /* sps and pps in the avcC always have length coded with 2 bytes,
 831            so put a fake nal_length_size = 2 while parsing them */
 832         h->nal_length_size = 2;
 833         // Decode sps from avcC
 834         cnt = *(p+5) & 0x1f; // Number of sps
 835         p += 6;
 836         for (i = 0; i < cnt; i++) {
 837             nalsize = AV_RB16(p) + 2;
 838             if(decode_nal_units(h, p, nalsize) < 0) {
 839                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
 840                 return -1;
 841             }
 842             p += nalsize;
 843         }
 844         // Decode pps from avcC
 845         cnt = *(p++); // Number of pps
 846         for (i = 0; i < cnt; i++) {
 847             nalsize = AV_RB16(p) + 2;
 848             if(decode_nal_units(h, p, nalsize)  != nalsize) {
 849                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
 850                 return -1;
 851             }
 852             p += nalsize;
 853         }
 854         // Now store right nal length size, that will be use to parse all other nals
 855         h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
 856     } else {
 857         h->is_avc = 0;
 858         if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
 859             return -1;
 860     }
 861     return 0;
 862 }
 863
 864 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
 865     H264Context *h= avctx->priv_data;
 866     MpegEncContext * const s = &h->s;
 867
 868     MPV_decode_defaults(s);
 869
 870     s->avctx = avctx;
 871     common_init(h);
 872
 873     s->out_format = FMT_H264;
 874     s->workaround_bugs= avctx->workaround_bugs;
 875
 876     // set defaults
 877 //    s->decode_mb= ff_h263_decode_mb;
 878     s->quarter_sample = 1;
 879     if(!avctx->has_b_frames)
 880     s->low_delay= 1;
 881
 882     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
 883
 884     ff_h264_decode_init_vlc();
 885
 886     h->thread_context[0] = h;
 887     h->outputed_poc = INT_MIN;
 888     h->prev_poc_msb= 1<<16;
 889     h->x264_build = -1;
 890     ff_h264_reset_sei(h);
 891     if(avctx->codec_id == CODEC_ID_H264){
 892         if(avctx->ticks_per_frame == 1){
 893             s->avctx->time_base.den *=2;
 894         }
 895         avctx->ticks_per_frame = 2;
 896     }
 897
 898     if(avctx->extradata_size > 0 && avctx->extradata &&
 899         ff_h264_decode_extradata(h))
 900         return -1;
 901
 902     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
 903         s->avctx->has_b_frames = h->sps.num_reorder_frames;
 904         s->low_delay = 0;
 905     }
 906
 907     return 0;
 908 }
 909
 910 int ff_h264_frame_start(H264Context *h){
 911     MpegEncContext * const s = &h->s;
 912     int i;
 913
 914     if(MPV_frame_start(s, s->avctx) < 0)
 915         return -1;
 916     ff_er_frame_start(s);
 917     /*
 918      * MPV_frame_start uses pict_type to derive key_frame.
 919      * This is incorrect for H.264; IDR markings must be used.
 920      * Zero here; IDR markings per slice in frame or fields are ORed in later.
 921      * See decode_nal_units().
 922      */
 923     s->current_picture_ptr->key_frame= 0;
 924     s->current_picture_ptr->mmco_reset= 0;
 925
 926     assert(s->linesize && s->uvlinesize);
 927
 928     for(i=0; i<16; i++){
 929         h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
 930         h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
 931     }
 932     for(i=0; i<4; i++){
 933         h->block_offset[16+i]=
 934         h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
 935         h->block_offset[24+16+i]=
 936         h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
 937     }
 938
 939     /* can't be in alloc_tables because linesize isn't known there.
 940      * FIXME: redo bipred weight to not require extra buffer? */
 941     for(i = 0; i < s->avctx->thread_count; i++)
 942         if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
 943             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
 944
 945     /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
 946     memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
 947
 948 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
 949
 950     // We mark the current picture as non-reference after allocating it, so
 951     // that if we break out due to an error it can be released automatically
 952     // in the next MPV_frame_start().
 953     // SVQ3 as well as most other codecs have only last/next/current and thus
 954     // get released even with set reference, besides SVQ3 and others do not
 955     // mark frames as reference later "naturally".
 956     if(s->codec_id != CODEC_ID_SVQ3)
 957         s->current_picture_ptr->reference= 0;
 958
 959     s->current_picture_ptr->field_poc[0]=
 960     s->current_picture_ptr->field_poc[1]= INT_MAX;
 961     assert(s->current_picture_ptr->long_ref==0);
 962
 963     return 0;
 964 }
 965
 966 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){
 967     MpegEncContext * const s = &h->s;
 968     uint8_t *top_border;
 969     int top_idx = 1;
 970
 971     src_y  -=   linesize;
 972     src_cb -= uvlinesize;
 973     src_cr -= uvlinesize;
 974
 975     if(!simple && FRAME_MBAFF){
 976         if(s->mb_y&1){
 977             if(!MB_MBAFF){
 978                 top_border = h->top_borders[0][s->mb_x];
 979                 AV_COPY128(top_border, src_y + 15*linesize);
 980                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
 981                     AV_COPY64(top_border+16, src_cb+7*uvlinesize);
 982                     AV_COPY64(top_border+24, src_cr+7*uvlinesize);
 983                 }
 984             }
 985         }else if(MB_MBAFF){
 986             top_idx = 0;
 987         }else
 988             return;
 989     }
 990
 991     top_border = h->top_borders[top_idx][s->mb_x];
 992     // There are two lines saved, the line above the the top macroblock of a pair,
 993     // and the line above the bottom macroblock
 994     AV_COPY128(top_border, src_y + 16*linesize);
 995
 996     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
 997         AV_COPY64(top_border+16, src_cb+8*uvlinesize);
 998         AV_COPY64(top_border+24, src_cr+8*uvlinesize);
 999     }
1000 }
1001
1002 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){
1003     MpegEncContext * const s = &h->s;
1004     int deblock_left;
1005     int deblock_top;
1006     int top_idx = 1;
1007     uint8_t *top_border_m1;
1008     uint8_t *top_border;
1009
1010     if(!simple && FRAME_MBAFF){
1011         if(s->mb_y&1){
1012             if(!MB_MBAFF)
1013                 return;
1014         }else{
1015             top_idx = MB_MBAFF ? 0 : 1;
1016         }
1017     }
1018
1019     if(h->deblocking_filter == 2) {
1020         deblock_left = h->left_type[0];
1021         deblock_top  = h->top_type;
1022     } else {
1023         deblock_left = (s->mb_x > 0);
1024         deblock_top =  (s->mb_y > !!MB_FIELD);
1025     }
1026
1027     src_y  -=   linesize + 1;
1028     src_cb -= uvlinesize + 1;
1029     src_cr -= uvlinesize + 1;
1030
1031     top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1032     top_border    = h->top_borders[top_idx][s->mb_x];
1033
1034 #define XCHG(a,b,xchg)\
1035 if (xchg) AV_SWAP64(b,a);\
1036 else      AV_COPY64(b,a);
1037
1038     if(deblock_top){
1039         if(deblock_left){
1040             XCHG(top_border_m1+8, src_y -7, 1);
1041         }
1042         XCHG(top_border+0, src_y +1, xchg);
1043         XCHG(top_border+8, src_y +9, 1);
1044         if(s->mb_x+1 < s->mb_width){
1045             XCHG(h->top_borders[top_idx][s->mb_x+1], src_y +17, 1);
1046         }
1047     }
1048
1049     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1050         if(deblock_top){
1051             if(deblock_left){
1052                 XCHG(top_border_m1+16, src_cb -7, 1);
1053                 XCHG(top_border_m1+24, src_cr -7, 1);
1054             }
1055             XCHG(top_border+16, src_cb+1, 1);
1056             XCHG(top_border+24, src_cr+1, 1);
1057         }
1058     }
1059 }
1060
1061 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
1062     MpegEncContext * const s = &h->s;
1063     const int mb_x= s->mb_x;
1064     const int mb_y= s->mb_y;
1065     const int mb_xy= h->mb_xy;
1066     const int mb_type= s->current_picture.mb_type[mb_xy];
1067     uint8_t  *dest_y, *dest_cb, *dest_cr;
1068     int linesize, uvlinesize /*dct_offset*/;
1069     int i;
1070     int *block_offset = &h->block_offset[0];
1071     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1072     /* is_h264 should always be true if SVQ3 is disabled. */
1073     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1074     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1075     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1076
1077     dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
1078     dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
1079     dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
1080
1081     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
1082     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
1083
1084     h->list_counts[mb_xy]= h->list_count;
1085
1086     if (!simple && MB_FIELD) {
1087         linesize   = h->mb_linesize   = s->linesize * 2;
1088         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1089         block_offset = &h->block_offset[24];
1090         if(mb_y&1){ //FIXME move out of this function?
1091             dest_y -= s->linesize*15;
1092             dest_cb-= s->uvlinesize*7;
1093             dest_cr-= s->uvlinesize*7;
1094         }
1095         if(FRAME_MBAFF) {
1096             int list;
1097             for(list=0; list<h->list_count; list++){
1098                 if(!USES_LIST(mb_type, list))
1099                     continue;
1100                 if(IS_16X16(mb_type)){
1101                     int8_t *ref = &h->ref_cache[list][scan8[0]];
1102                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1103                 }else{
1104                     for(i=0; i<16; i+=4){
1105                         int ref = h->ref_cache[list][scan8[i]];
1106                         if(ref >= 0)
1107                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1108                     }
1109                 }
1110             }
1111         }
1112     } else {
1113         linesize   = h->mb_linesize   = s->linesize;
1114         uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1115 //        dct_offset = s->linesize * 16;
1116     }
1117
1118     if (!simple && IS_INTRA_PCM(mb_type)) {
1119         for (i=0; i<16; i++) {
1120             memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1121         }
1122         for (i=0; i<8; i++) {
1123             memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
1124             memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
1125         }
1126     } else {
1127         if(IS_INTRA(mb_type)){
1128             if(h->deblocking_filter)
1129                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
1130
1131             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1132                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1133                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1134             }
1135
1136             if(IS_INTRA4x4(mb_type)){
1137                 if(simple || !s->encoding){
1138                     if(IS_8x8DCT(mb_type)){
1139                         if(transform_bypass){
1140                             idct_dc_add =
1141                             idct_add    = s->dsp.add_pixels8;
1142                         }else{
1143                             idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1144                             idct_add    = h->h264dsp.h264_idct8_add;
1145                         }
1146                         for(i=0; i<16; i+=4){
1147                             uint8_t * const ptr= dest_y + block_offset[i];
1148                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1149                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1150                                 h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
1151                             }else{
1152                                 const int nnz = h->non_zero_count_cache[ scan8[i] ];
1153                                 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1154                                                             (h->topright_samples_available<<i)&0x4000, linesize);
1155                                 if(nnz){
1156                                     if(nnz == 1 && h->mb[i*16])
1157                                         idct_dc_add(ptr, h->mb + i*16, linesize);
1158                                     else
1159                                         idct_add   (ptr, h->mb + i*16, linesize);
1160                                 }
1161                             }
1162                         }
1163                     }else{
1164                         if(transform_bypass){
1165                             idct_dc_add =
1166                             idct_add    = s->dsp.add_pixels4;
1167                         }else{
1168                             idct_dc_add = h->h264dsp.h264_idct_dc_add;
1169                             idct_add    = h->h264dsp.h264_idct_add;
1170                         }
1171                         for(i=0; i<16; i++){
1172                             uint8_t * const ptr= dest_y + block_offset[i];
1173                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1174
1175                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1176                                 h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
1177                             }else{
1178                                 uint8_t *topright;
1179                                 int nnz, tr;
1180                                 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1181                                     const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1182                                     assert(mb_y || linesize <= block_offset[i]);
1183                                     if(!topright_avail){
1184                                         tr= ptr[3 - linesize]*0x01010101;
1185                                         topright= (uint8_t*) &tr;
1186                                     }else
1187                                         topright= ptr + 4 - linesize;
1188                                 }else
1189                                     topright= NULL;
1190
1191                                 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1192                                 nnz = h->non_zero_count_cache[ scan8[i] ];
1193                                 if(nnz){
1194                                     if(is_h264){
1195                                         if(nnz == 1 && h->mb[i*16])
1196                                             idct_dc_add(ptr, h->mb + i*16, linesize);
1197                                         else
1198                                             idct_add   (ptr, h->mb + i*16, linesize);
1199                                     }else
1200                                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1201                                 }
1202                             }
1203                         }
1204                     }
1205                 }
1206             }else{
1207                 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1208                 if(is_h264){
1209                     if(h->non_zero_count_cache[ scan8[LUMA_DC_BLOCK_INDEX] ]){
1210                         if(!transform_bypass)
1211                             h->h264dsp.h264_luma_dc_dequant_idct(h->mb, h->mb_luma_dc, h->dequant4_coeff[0][s->qscale][0]);
1212                         else{
1213                             static const uint8_t dc_mapping[16] = { 0*16, 1*16, 4*16, 5*16, 2*16, 3*16, 6*16, 7*16,
1214                                                                     8*16, 9*16,12*16,13*16,10*16,11*16,14*16,15*16};
1215                             for(i = 0; i < 16; i++)
1216                                 h->mb[dc_mapping[i]] = h->mb_luma_dc[i];
1217                         }
1218                     }
1219                 }else
1220                     ff_svq3_luma_dc_dequant_idct_c(h->mb, h->mb_luma_dc, s->qscale);
1221             }
1222             if(h->deblocking_filter)
1223                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
1224         }else if(is_h264){
1225             hl_motion(h, dest_y, dest_cb, dest_cr,
1226                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1227                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1228                       h->h264dsp.weight_h264_pixels_tab, h->h264dsp.biweight_h264_pixels_tab);
1229         }
1230
1231
1232         if(!IS_INTRA4x4(mb_type)){
1233             if(is_h264){
1234                 if(IS_INTRA16x16(mb_type)){
1235                     if(transform_bypass){
1236                         if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1237                             h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1238                         }else{
1239                             for(i=0; i<16; i++){
1240                                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1241                                     s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
1242                             }
1243                         }
1244                     }else{
1245                          h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1246                     }
1247                 }else if(h->cbp&15){
1248                     if(transform_bypass){
1249                         const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1250                         idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1251                         for(i=0; i<16; i+=di){
1252                             if(h->non_zero_count_cache[ scan8[i] ]){
1253                                 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
1254                             }
1255                         }
1256                     }else{
1257                         if(IS_8x8DCT(mb_type)){
1258                             h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1259                         }else{
1260                             h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1261                         }
1262                     }
1263                 }
1264             }else{
1265                 for(i=0; i<16; i++){
1266                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1267                         uint8_t * const ptr= dest_y + block_offset[i];
1268                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1269                     }
1270                 }
1271             }
1272         }
1273
1274         if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1275             uint8_t *dest[2] = {dest_cb, dest_cr};
1276             if(transform_bypass){
1277                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1278                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
1279                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
1280                 }else{
1281                     idct_add = s->dsp.add_pixels4;
1282                     for(i=16; i<16+8; i++){
1283                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1284                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1285                     }
1286                 }
1287             }else{
1288                 if(is_h264){
1289                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+0] ])
1290                         chroma_dc_dequant_idct_c(h->mb + 16*16     , h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
1291                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+1] ])
1292                         chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
1293                     h->h264dsp.h264_idct_add8(dest, block_offset,
1294                                               h->mb, uvlinesize,
1295                                               h->non_zero_count_cache);
1296                 }else{
1297                     chroma_dc_dequant_idct_c(h->mb + 16*16     , h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
1298                     chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
1299                     for(i=16; i<16+8; i++){
1300                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1301                             uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1302                             ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
1303                         }
1304                     }
1305                 }
1306             }
1307         }
1308     }
1309     if(h->cbp || IS_INTRA(mb_type))
1310         s->dsp.clear_blocks(h->mb);
1311 }
1312
1313 /**
1314  * Process a macroblock; this case avoids checks for expensive uncommon cases.
1315  */
1316 static void hl_decode_mb_simple(H264Context *h){
1317     hl_decode_mb_internal(h, 1);
1318 }
1319
1320 /**
1321  * Process a macroblock; this handles edge cases, such as interlacing.
1322  */
1323 static void av_noinline hl_decode_mb_complex(H264Context *h){
1324     hl_decode_mb_internal(h, 0);
1325 }
1326
1327 void ff_h264_hl_decode_mb(H264Context *h){
1328     MpegEncContext * const s = &h->s;
1329     const int mb_xy= h->mb_xy;
1330     const int mb_type= s->current_picture.mb_type[mb_xy];
1331     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1332
1333     if (is_complex)
1334         hl_decode_mb_complex(h);
1335     else hl_decode_mb_simple(h);
1336 }
1337
1338 static int pred_weight_table(H264Context *h){
1339     MpegEncContext * const s = &h->s;
1340     int list, i;
1341     int luma_def, chroma_def;
1342
1343     h->use_weight= 0;
1344     h->use_weight_chroma= 0;
1345     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1346     if(CHROMA)
1347         h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1348     luma_def = 1<<h->luma_log2_weight_denom;
1349     chroma_def = 1<<h->chroma_log2_weight_denom;
1350
1351     for(list=0; list<2; list++){
1352         h->luma_weight_flag[list]   = 0;
1353         h->chroma_weight_flag[list] = 0;
1354         for(i=0; i<h->ref_count[list]; i++){
1355             int luma_weight_flag, chroma_weight_flag;
1356
1357             luma_weight_flag= get_bits1(&s->gb);
1358             if(luma_weight_flag){
1359                 h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
1360                 h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
1361                 if(   h->luma_weight[i][list][0] != luma_def
1362                    || h->luma_weight[i][list][1] != 0) {
1363                     h->use_weight= 1;
1364                     h->luma_weight_flag[list]= 1;
1365                 }
1366             }else{
1367                 h->luma_weight[i][list][0]= luma_def;
1368                 h->luma_weight[i][list][1]= 0;
1369             }
1370
1371             if(CHROMA){
1372                 chroma_weight_flag= get_bits1(&s->gb);
1373                 if(chroma_weight_flag){
1374                     int j;
1375                     for(j=0; j<2; j++){
1376                         h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
1377                         h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
1378                         if(   h->chroma_weight[i][list][j][0] != chroma_def
1379                            || h->chroma_weight[i][list][j][1] != 0) {
1380                             h->use_weight_chroma= 1;
1381                             h->chroma_weight_flag[list]= 1;
1382                         }
1383                     }
1384                 }else{
1385                     int j;
1386                     for(j=0; j<2; j++){
1387                         h->chroma_weight[i][list][j][0]= chroma_def;
1388                         h->chroma_weight[i][list][j][1]= 0;
1389                     }
1390                 }
1391             }
1392         }
1393         if(h->slice_type_nos != FF_B_TYPE) break;
1394     }
1395     h->use_weight= h->use_weight || h->use_weight_chroma;
1396     return 0;
1397 }
1398
1399 /**
1400  * Initialize implicit_weight table.
1401  * @param field  0/1 initialize the weight for interlaced MBAFF
1402  *                -1 initializes the rest
1403  */
1404 static void implicit_weight_table(H264Context *h, int field){
1405     MpegEncContext * const s = &h->s;
1406     int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
1407
1408     for (i = 0; i < 2; i++) {
1409         h->luma_weight_flag[i]   = 0;
1410         h->chroma_weight_flag[i] = 0;
1411     }
1412
1413     if(field < 0){
1414         cur_poc = s->current_picture_ptr->poc;
1415     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
1416        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1417         h->use_weight= 0;
1418         h->use_weight_chroma= 0;
1419         return;
1420     }
1421         ref_start= 0;
1422         ref_count0= h->ref_count[0];
1423         ref_count1= h->ref_count[1];
1424     }else{
1425         cur_poc = s->current_picture_ptr->field_poc[field];
1426         ref_start= 16;
1427         ref_count0= 16+2*h->ref_count[0];
1428         ref_count1= 16+2*h->ref_count[1];
1429     }
1430
1431     h->use_weight= 2;
1432     h->use_weight_chroma= 2;
1433     h->luma_log2_weight_denom= 5;
1434     h->chroma_log2_weight_denom= 5;
1435
1436     for(ref0=ref_start; ref0 < ref_count0; ref0++){
1437         int poc0 = h->ref_list[0][ref0].poc;
1438         for(ref1=ref_start; ref1 < ref_count1; ref1++){
1439             int poc1 = h->ref_list[1][ref1].poc;
1440             int td = av_clip(poc1 - poc0, -128, 127);
1441             int w= 32;
1442             if(td){
1443                 int tb = av_clip(cur_poc - poc0, -128, 127);
1444                 int tx = (16384 + (FFABS(td) >> 1)) / td;
1445                 int dist_scale_factor = (tb*tx + 32) >> 8;
1446                 if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
1447                     w = 64 - dist_scale_factor;
1448             }
1449             if(field<0){
1450                 h->implicit_weight[ref0][ref1][0]=
1451                 h->implicit_weight[ref0][ref1][1]= w;
1452             }else{
1453                 h->implicit_weight[ref0][ref1][field]=w;
1454             }
1455         }
1456     }
1457 }
1458
1459 /**
1460  * instantaneous decoder refresh.
1461  */
1462 static void idr(H264Context *h){
1463     ff_h264_remove_all_refs(h);
1464     h->prev_frame_num= 0;
1465     h->prev_frame_num_offset= 0;
1466     h->prev_poc_msb=
1467     h->prev_poc_lsb= 0;
1468 }
1469
1470 /* forget old pics after a seek */
1471 static void flush_dpb(AVCodecContext *avctx){
1472     H264Context *h= avctx->priv_data;
1473     int i;
1474     for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1475         if(h->delayed_pic[i])
1476             h->delayed_pic[i]->reference= 0;
1477         h->delayed_pic[i]= NULL;
1478     }
1479     h->outputed_poc= INT_MIN;
1480     h->prev_interlaced_frame = 1;
1481     idr(h);
1482     if(h->s.current_picture_ptr)
1483         h->s.current_picture_ptr->reference= 0;
1484     h->s.first_field= 0;
1485     ff_h264_reset_sei(h);
1486     ff_mpeg_flush(avctx);
1487 }
1488
1489 static int init_poc(H264Context *h){
1490     MpegEncContext * const s = &h->s;
1491     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1492     int field_poc[2];
1493     Picture *cur = s->current_picture_ptr;
1494
1495     h->frame_num_offset= h->prev_frame_num_offset;
1496     if(h->frame_num < h->prev_frame_num)
1497         h->frame_num_offset += max_frame_num;
1498
1499     if(h->sps.poc_type==0){
1500         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1501
1502         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1503             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1504         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1505             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1506         else
1507             h->poc_msb = h->prev_poc_msb;
1508 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1509         field_poc[0] =
1510         field_poc[1] = h->poc_msb + h->poc_lsb;
1511         if(s->picture_structure == PICT_FRAME)
1512             field_poc[1] += h->delta_poc_bottom;
1513     }else if(h->sps.poc_type==1){
1514         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1515         int i;
1516
1517         if(h->sps.poc_cycle_length != 0)
1518             abs_frame_num = h->frame_num_offset + h->frame_num;
1519         else
1520             abs_frame_num = 0;
1521
1522         if(h->nal_ref_idc==0 && abs_frame_num > 0)
1523             abs_frame_num--;
1524
1525         expected_delta_per_poc_cycle = 0;
1526         for(i=0; i < h->sps.poc_cycle_length; i++)
1527             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1528
1529         if(abs_frame_num > 0){
1530             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1531             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1532
1533             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1534             for(i = 0; i <= frame_num_in_poc_cycle; i++)
1535                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1536         } else
1537             expectedpoc = 0;
1538
1539         if(h->nal_ref_idc == 0)
1540             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1541
1542         field_poc[0] = expectedpoc + h->delta_poc[0];
1543         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1544
1545         if(s->picture_structure == PICT_FRAME)
1546             field_poc[1] += h->delta_poc[1];
1547     }else{
1548         int poc= 2*(h->frame_num_offset + h->frame_num);
1549
1550         if(!h->nal_ref_idc)
1551             poc--;
1552
1553         field_poc[0]= poc;
1554         field_poc[1]= poc;
1555     }
1556
1557     if(s->picture_structure != PICT_BOTTOM_FIELD)
1558         s->current_picture_ptr->field_poc[0]= field_poc[0];
1559     if(s->picture_structure != PICT_TOP_FIELD)
1560         s->current_picture_ptr->field_poc[1]= field_poc[1];
1561     cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1562
1563     return 0;
1564 }
1565
1566
1567 /**
1568  * initialize scan tables
1569  */
1570 static void init_scan_tables(H264Context *h){
1571     int i;
1572     for(i=0; i<16; i++){
1573 #define T(x) (x>>2) | ((x<<2) & 0xF)
1574         h->zigzag_scan[i] = T(zigzag_scan[i]);
1575         h-> field_scan[i] = T( field_scan[i]);
1576 #undef T
1577     }
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     if(h->sps.transform_bypass){ //FIXME same ugly
1587         h->zigzag_scan_q0          = zigzag_scan;
1588         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
1589         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1590         h->field_scan_q0           = field_scan;
1591         h->field_scan8x8_q0        = field_scan8x8;
1592         h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
1593     }else{
1594         h->zigzag_scan_q0          = h->zigzag_scan;
1595         h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
1596         h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1597         h->field_scan_q0           = h->field_scan;
1598         h->field_scan8x8_q0        = h->field_scan8x8;
1599         h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
1600     }
1601 }
1602
1603 static void field_end(H264Context *h){
1604     MpegEncContext * const s = &h->s;
1605     AVCodecContext * const avctx= s->avctx;
1606     s->mb_y= 0;
1607
1608     s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1609     s->current_picture_ptr->pict_type= s->pict_type;
1610
1611     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1612         ff_vdpau_h264_set_reference_frames(s);
1613
1614     if(!s->dropable) {
1615         ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1616         h->prev_poc_msb= h->poc_msb;
1617         h->prev_poc_lsb= h->poc_lsb;
1618     }
1619     h->prev_frame_num_offset= h->frame_num_offset;
1620     h->prev_frame_num= h->frame_num;
1621
1622     if (avctx->hwaccel) {
1623         if (avctx->hwaccel->end_frame(avctx) < 0)
1624             av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1625     }
1626
1627     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1628         ff_vdpau_h264_picture_complete(s);
1629
1630     /*
1631      * FIXME: Error handling code does not seem to support interlaced
1632      * when slices span multiple rows
1633      * The ff_er_add_slice calls don't work right for bottom
1634      * fields; they cause massive erroneous error concealing
1635      * Error marking covers both fields (top and bottom).
1636      * This causes a mismatched s->error_count
1637      * and a bad error table. Further, the error count goes to
1638      * INT_MAX when called for bottom field, because mb_y is
1639      * past end by one (callers fault) and resync_mb_y != 0
1640      * causes problems for the first MB line, too.
1641      */
1642     if (!FIELD_PICTURE)
1643         ff_er_frame_end(s);
1644
1645     MPV_frame_end(s);
1646
1647     h->current_slice=0;
1648 }
1649
1650 /**
1651  * Replicate H264 "master" context to thread contexts.
1652  */
1653 static void clone_slice(H264Context *dst, H264Context *src)
1654 {
1655     memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
1656     dst->s.current_picture_ptr  = src->s.current_picture_ptr;
1657     dst->s.current_picture      = src->s.current_picture;
1658     dst->s.linesize             = src->s.linesize;
1659     dst->s.uvlinesize           = src->s.uvlinesize;
1660     dst->s.first_field          = src->s.first_field;
1661
1662     dst->prev_poc_msb           = src->prev_poc_msb;
1663     dst->prev_poc_lsb           = src->prev_poc_lsb;
1664     dst->prev_frame_num_offset  = src->prev_frame_num_offset;
1665     dst->prev_frame_num         = src->prev_frame_num;
1666     dst->short_ref_count        = src->short_ref_count;
1667
1668     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
1669     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
1670     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1671     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
1672
1673     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
1674     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
1675 }
1676
1677 /**
1678  * decodes a slice header.
1679  * This will also call MPV_common_init() and frame_start() as needed.
1680  *
1681  * @param h h264context
1682  * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1683  *
1684  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1685  */
1686 static int decode_slice_header(H264Context *h, H264Context *h0){
1687     MpegEncContext * const s = &h->s;
1688     MpegEncContext * const s0 = &h0->s;
1689     unsigned int first_mb_in_slice;
1690     unsigned int pps_id;
1691     int num_ref_idx_active_override_flag;
1692     unsigned int slice_type, tmp, i, j;
1693     int default_ref_list_done = 0;
1694     int last_pic_structure;
1695
1696     s->dropable= h->nal_ref_idc == 0;
1697
1698     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1699         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1700         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1701     }else{
1702         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1703         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1704     }
1705
1706     first_mb_in_slice= get_ue_golomb(&s->gb);
1707
1708     if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1709         if(h0->current_slice && FIELD_PICTURE){
1710             field_end(h);
1711         }
1712
1713         h0->current_slice = 0;
1714         if (!s0->first_field)
1715             s->current_picture_ptr= NULL;
1716     }
1717
1718     slice_type= get_ue_golomb_31(&s->gb);
1719     if(slice_type > 9){
1720         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);
1721         return -1;
1722     }
1723     if(slice_type > 4){
1724         slice_type -= 5;
1725         h->slice_type_fixed=1;
1726     }else
1727         h->slice_type_fixed=0;
1728
1729     slice_type= golomb_to_pict_type[ slice_type ];
1730     if (slice_type == FF_I_TYPE
1731         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
1732         default_ref_list_done = 1;
1733     }
1734     h->slice_type= slice_type;
1735     h->slice_type_nos= slice_type & 3;
1736
1737     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
1738
1739     pps_id= get_ue_golomb(&s->gb);
1740     if(pps_id>=MAX_PPS_COUNT){
1741         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
1742         return -1;
1743     }
1744     if(!h0->pps_buffers[pps_id]) {
1745         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
1746         return -1;
1747     }
1748     h->pps= *h0->pps_buffers[pps_id];
1749
1750     if(!h0->sps_buffers[h->pps.sps_id]) {
1751         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
1752         return -1;
1753     }
1754     h->sps = *h0->sps_buffers[h->pps.sps_id];
1755
1756     s->avctx->profile = h->sps.profile_idc;
1757     s->avctx->level   = h->sps.level_idc;
1758     s->avctx->refs    = h->sps.ref_frame_count;
1759
1760     if(h == h0 && h->dequant_coeff_pps != pps_id){
1761         h->dequant_coeff_pps = pps_id;
1762         init_dequant_tables(h);
1763     }
1764
1765     s->mb_width= h->sps.mb_width;
1766     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1767
1768     h->b_stride=  s->mb_width*4;
1769
1770     s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
1771     if(h->sps.frame_mbs_only_flag)
1772         s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
1773     else
1774         s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 7);
1775
1776     if (s->context_initialized
1777         && (   s->width != s->avctx->width || s->height != s->avctx->height
1778             || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
1779         if(h != h0)
1780             return -1;   // width / height changed during parallelized decoding
1781         free_tables(h, 0);
1782         flush_dpb(s->avctx);
1783         MPV_common_end(s);
1784     }
1785     if (!s->context_initialized) {
1786         if(h != h0)
1787             return -1;  // we cant (re-)initialize context during parallel decoding
1788
1789         avcodec_set_dimensions(s->avctx, s->width, s->height);
1790         s->avctx->sample_aspect_ratio= h->sps.sar;
1791         av_assert0(s->avctx->sample_aspect_ratio.den);
1792
1793         if(h->sps.video_signal_type_present_flag){
1794             s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1795             if(h->sps.colour_description_present_flag){
1796                 s->avctx->color_primaries = h->sps.color_primaries;
1797                 s->avctx->color_trc       = h->sps.color_trc;
1798                 s->avctx->colorspace      = h->sps.colorspace;
1799             }
1800         }
1801
1802         if(h->sps.timing_info_present_flag){
1803             int64_t den= h->sps.time_scale;
1804             if(h->x264_build < 44U)
1805                 den *= 2;
1806             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
1807                       h->sps.num_units_in_tick, den, 1<<30);
1808         }
1809         s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
1810                                                  s->avctx->codec->pix_fmts ?
1811                                                  s->avctx->codec->pix_fmts :
1812                                                  s->avctx->color_range == AVCOL_RANGE_JPEG ?
1813                                                  hwaccel_pixfmt_list_h264_jpeg_420 :
1814                                                  ff_hwaccel_pixfmt_list_420);
1815         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1816
1817         if (MPV_common_init(s) < 0)
1818             return -1;
1819         s->first_field = 0;
1820         h->prev_interlaced_frame = 1;
1821
1822         init_scan_tables(h);
1823         ff_h264_alloc_tables(h);
1824
1825         for(i = 1; i < s->avctx->thread_count; i++) {
1826             H264Context *c;
1827             c = h->thread_context[i] = av_malloc(sizeof(H264Context));
1828             memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
1829             memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
1830             c->h264dsp = h->h264dsp;
1831             c->sps = h->sps;
1832             c->pps = h->pps;
1833             init_scan_tables(c);
1834             clone_tables(c, h, i);
1835         }
1836
1837         for(i = 0; i < s->avctx->thread_count; i++)
1838             if(context_init(h->thread_context[i]) < 0)
1839                 return -1;
1840     }
1841
1842     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1843
1844     h->mb_mbaff = 0;
1845     h->mb_aff_frame = 0;
1846     last_pic_structure = s0->picture_structure;
1847     if(h->sps.frame_mbs_only_flag){
1848         s->picture_structure= PICT_FRAME;
1849     }else{
1850         if(get_bits1(&s->gb)) { //field_pic_flag
1851             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
1852         } else {
1853             s->picture_structure= PICT_FRAME;
1854             h->mb_aff_frame = h->sps.mb_aff;
1855         }
1856     }
1857     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
1858
1859     if(h0->current_slice == 0){
1860         while(h->frame_num !=  h->prev_frame_num &&
1861               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
1862             Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1863             av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
1864             if (ff_h264_frame_start(h) < 0)
1865                 return -1;
1866             h->prev_frame_num++;
1867             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
1868             s->current_picture_ptr->frame_num= h->prev_frame_num;
1869             ff_generate_sliding_window_mmcos(h);
1870             ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1871             /* Error concealment: if a ref is missing, copy the previous ref in its place.
1872              * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
1873              * about there being no actual duplicates.
1874              * FIXME: this doesn't copy padding for out-of-frame motion vectors.  Given we're
1875              * concealing a lost frame, this probably isn't noticable by comparison, but it should
1876              * be fixed. */
1877             if (h->short_ref_count) {
1878                 if (prev) {
1879                     av_image_copy(h->short_ref[0]->data, h->short_ref[0]->linesize,
1880                                   (const uint8_t**)prev->data, prev->linesize,
1881                                   s->avctx->pix_fmt, s->mb_width*16, s->mb_height*16);
1882                     h->short_ref[0]->poc = prev->poc+2;
1883                 }
1884                 h->short_ref[0]->frame_num = h->prev_frame_num;
1885             }
1886         }
1887
1888         /* See if we have a decoded first field looking for a pair... */
1889         if (s0->first_field) {
1890             assert(s0->current_picture_ptr);
1891             assert(s0->current_picture_ptr->data[0]);
1892             assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
1893
1894             /* figure out if we have a complementary field pair */
1895             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
1896                 /*
1897                  * Previous field is unmatched. Don't display it, but let it
1898                  * remain for reference if marked as such.
1899                  */
1900                 s0->current_picture_ptr = NULL;
1901                 s0->first_field = FIELD_PICTURE;
1902
1903             } else {
1904                 if (h->nal_ref_idc &&
1905                         s0->current_picture_ptr->reference &&
1906                         s0->current_picture_ptr->frame_num != h->frame_num) {
1907                     /*
1908                      * This and previous field were reference, but had
1909                      * different frame_nums. Consider this field first in
1910                      * pair. Throw away previous field except for reference
1911                      * purposes.
1912                      */
1913                     s0->first_field = 1;
1914                     s0->current_picture_ptr = NULL;
1915
1916                 } else {
1917                     /* Second field in complementary pair */
1918                     s0->first_field = 0;
1919                 }
1920             }
1921
1922         } else {
1923             /* Frame or first field in a potentially complementary pair */
1924             assert(!s0->current_picture_ptr);
1925             s0->first_field = FIELD_PICTURE;
1926         }
1927
1928         if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
1929             s0->first_field = 0;
1930             return -1;
1931         }
1932     }
1933     if(h != h0)
1934         clone_slice(h, h0);
1935
1936     s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
1937
1938     assert(s->mb_num == s->mb_width * s->mb_height);
1939     if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
1940        first_mb_in_slice                    >= s->mb_num){
1941         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1942         return -1;
1943     }
1944     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
1945     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
1946     if (s->picture_structure == PICT_BOTTOM_FIELD)
1947         s->resync_mb_y = s->mb_y = s->mb_y + 1;
1948     assert(s->mb_y < s->mb_height);
1949
1950     if(s->picture_structure==PICT_FRAME){
1951         h->curr_pic_num=   h->frame_num;
1952         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
1953     }else{
1954         h->curr_pic_num= 2*h->frame_num + 1;
1955         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
1956     }
1957
1958     if(h->nal_unit_type == NAL_IDR_SLICE){
1959         get_ue_golomb(&s->gb); /* idr_pic_id */
1960     }
1961
1962     if(h->sps.poc_type==0){
1963         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
1964
1965         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
1966             h->delta_poc_bottom= get_se_golomb(&s->gb);
1967         }
1968     }
1969
1970     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
1971         h->delta_poc[0]= get_se_golomb(&s->gb);
1972
1973         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
1974             h->delta_poc[1]= get_se_golomb(&s->gb);
1975     }
1976
1977     init_poc(h);
1978
1979     if(h->pps.redundant_pic_cnt_present){
1980         h->redundant_pic_count= get_ue_golomb(&s->gb);
1981     }
1982
1983     //set defaults, might be overridden a few lines later
1984     h->ref_count[0]= h->pps.ref_count[0];
1985     h->ref_count[1]= h->pps.ref_count[1];
1986
1987     if(h->slice_type_nos != FF_I_TYPE){
1988         if(h->slice_type_nos == FF_B_TYPE){
1989             h->direct_spatial_mv_pred= get_bits1(&s->gb);
1990         }
1991         num_ref_idx_active_override_flag= get_bits1(&s->gb);
1992
1993         if(num_ref_idx_active_override_flag){
1994             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
1995             if(h->slice_type_nos==FF_B_TYPE)
1996                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
1997
1998             if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
1999                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
2000                 h->ref_count[0]= h->ref_count[1]= 1;
2001                 return -1;
2002             }
2003         }
2004         if(h->slice_type_nos == FF_B_TYPE)
2005             h->list_count= 2;
2006         else
2007             h->list_count= 1;
2008     }else
2009         h->list_count= 0;
2010
2011     if(!default_ref_list_done){
2012         ff_h264_fill_default_ref_list(h);
2013     }
2014
2015     if(h->slice_type_nos!=FF_I_TYPE && ff_h264_decode_ref_pic_list_reordering(h) < 0)
2016         return -1;
2017
2018     if(h->slice_type_nos!=FF_I_TYPE){
2019         s->last_picture_ptr= &h->ref_list[0][0];
2020         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
2021     }
2022     if(h->slice_type_nos==FF_B_TYPE){
2023         s->next_picture_ptr= &h->ref_list[1][0];
2024         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
2025     }
2026
2027     if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
2028        ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
2029         pred_weight_table(h);
2030     else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE){
2031         implicit_weight_table(h, -1);
2032     }else {
2033         h->use_weight = 0;
2034         for (i = 0; i < 2; i++) {
2035             h->luma_weight_flag[i]   = 0;
2036             h->chroma_weight_flag[i] = 0;
2037         }
2038     }
2039
2040     if(h->nal_ref_idc)
2041         ff_h264_decode_ref_pic_marking(h0, &s->gb);
2042
2043     if(FRAME_MBAFF){
2044         ff_h264_fill_mbaff_ref_list(h);
2045
2046         if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE){
2047             implicit_weight_table(h, 0);
2048             implicit_weight_table(h, 1);
2049         }
2050     }
2051
2052     if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
2053         ff_h264_direct_dist_scale_factor(h);
2054     ff_h264_direct_ref_list_init(h);
2055
2056     if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
2057         tmp = get_ue_golomb_31(&s->gb);
2058         if(tmp > 2){
2059             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2060             return -1;
2061         }
2062         h->cabac_init_idc= tmp;
2063     }
2064
2065     h->last_qscale_diff = 0;
2066     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2067     if(tmp>51){
2068         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2069         return -1;
2070     }
2071     s->qscale= tmp;
2072     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2073     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2074     //FIXME qscale / qp ... stuff
2075     if(h->slice_type == FF_SP_TYPE){
2076         get_bits1(&s->gb); /* sp_for_switch_flag */
2077     }
2078     if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
2079         get_se_golomb(&s->gb); /* slice_qs_delta */
2080     }
2081
2082     h->deblocking_filter = 1;
2083     h->slice_alpha_c0_offset = 52;
2084     h->slice_beta_offset = 52;
2085     if( h->pps.deblocking_filter_parameters_present ) {
2086         tmp= get_ue_golomb_31(&s->gb);
2087         if(tmp > 2){
2088             av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2089             return -1;
2090         }
2091         h->deblocking_filter= tmp;
2092         if(h->deblocking_filter < 2)
2093             h->deblocking_filter^= 1; // 1<->0
2094
2095         if( h->deblocking_filter ) {
2096             h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2097             h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
2098             if(   h->slice_alpha_c0_offset > 104U
2099                || h->slice_beta_offset     > 104U){
2100                 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);
2101                 return -1;
2102             }
2103         }
2104     }
2105
2106     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
2107        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
2108        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
2109        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2110         h->deblocking_filter= 0;
2111
2112     if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2113         if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2114             /* Cheat slightly for speed:
2115                Do not bother to deblock across slices. */
2116             h->deblocking_filter = 2;
2117         } else {
2118             h0->max_contexts = 1;
2119             if(!h0->single_decode_warning) {
2120                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2121                 h0->single_decode_warning = 1;
2122             }
2123             if(h != h0)
2124                 return 1; // deblocking switched inside frame
2125         }
2126     }
2127     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]);
2128
2129 #if 0 //FMO
2130     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2131         slice_group_change_cycle= get_bits(&s->gb, ?);
2132 #endif
2133
2134     h0->last_slice_type = slice_type;
2135     h->slice_num = ++h0->current_slice;
2136     if(h->slice_num >= MAX_SLICES){
2137         av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2138     }
2139
2140     for(j=0; j<2; j++){
2141         int id_list[16];
2142         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2143         for(i=0; i<16; i++){
2144             id_list[i]= 60;
2145             if(h->ref_list[j][i].data[0]){
2146                 int k;
2147                 uint8_t *base= h->ref_list[j][i].base[0];
2148                 for(k=0; k<h->short_ref_count; k++)
2149                     if(h->short_ref[k]->base[0] == base){
2150                         id_list[i]= k;
2151                         break;
2152                     }
2153                 for(k=0; k<h->long_ref_count; k++)
2154                     if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2155                         id_list[i]= h->short_ref_count + k;
2156                         break;
2157                     }
2158             }
2159         }
2160
2161         ref2frm[0]=
2162         ref2frm[1]= -1;
2163         for(i=0; i<16; i++)
2164             ref2frm[i+2]= 4*id_list[i]
2165                           +(h->ref_list[j][i].reference&3);
2166         ref2frm[18+0]=
2167         ref2frm[18+1]= -1;
2168         for(i=16; i<48; i++)
2169             ref2frm[i+4]= 4*id_list[(i-16)>>1]
2170                           +(h->ref_list[j][i].reference&3);
2171     }
2172
2173     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
2174     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2175
2176     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2177         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",
2178                h->slice_num,
2179                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2180                first_mb_in_slice,
2181                av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2182                pps_id, h->frame_num,
2183                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2184                h->ref_count[0], h->ref_count[1],
2185                s->qscale,
2186                h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
2187                h->use_weight,
2188                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2189                h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2190                );
2191     }
2192
2193     return 0;
2194 }
2195
2196 int ff_h264_get_slice_type(const H264Context *h)
2197 {
2198     switch (h->slice_type) {
2199     case FF_P_TYPE:  return 0;
2200     case FF_B_TYPE:  return 1;
2201     case FF_I_TYPE:  return 2;
2202     case FF_SP_TYPE: return 3;
2203     case FF_SI_TYPE: return 4;
2204     default:         return -1;
2205     }
2206 }
2207
2208 /**
2209  *
2210  * @return non zero if the loop filter can be skiped
2211  */
2212 static int fill_filter_caches(H264Context *h, int mb_type){
2213     MpegEncContext * const s = &h->s;
2214     const int mb_xy= h->mb_xy;
2215     int top_xy, left_xy[2];
2216     int top_type, left_type[2];
2217
2218     top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
2219
2220     //FIXME deblocking could skip the intra and nnz parts.
2221
2222     /* Wow, what a mess, why didn't they simplify the interlacing & intra
2223      * stuff, I can't imagine that these complex rules are worth it. */
2224
2225     left_xy[1] = left_xy[0] = mb_xy-1;
2226     if(FRAME_MBAFF){
2227         const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
2228         const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
2229         if(s->mb_y&1){
2230             if (left_mb_field_flag != curr_mb_field_flag) {
2231                 left_xy[0] -= s->mb_stride;
2232             }
2233         }else{
2234             if(curr_mb_field_flag){
2235                 top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
2236             }
2237             if (left_mb_field_flag != curr_mb_field_flag) {
2238                 left_xy[1] += s->mb_stride;
2239             }
2240         }
2241     }
2242
2243     h->top_mb_xy = top_xy;
2244     h->left_mb_xy[0] = left_xy[0];
2245     h->left_mb_xy[1] = left_xy[1];
2246     {
2247         //for sufficiently low qp, filtering wouldn't do anything
2248         //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
2249         int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
2250         int qp = s->current_picture.qscale_table[mb_xy];
2251         if(qp <= qp_thresh
2252            && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
2253            && (top_xy   < 0 || ((qp + s->current_picture.qscale_table[top_xy    ] + 1)>>1) <= qp_thresh)){
2254             if(!FRAME_MBAFF)
2255                 return 1;
2256             if(   (left_xy[0]< 0            || ((qp + s->current_picture.qscale_table[left_xy[1]             ] + 1)>>1) <= qp_thresh)
2257                && (top_xy    < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy    -s->mb_stride] + 1)>>1) <= qp_thresh))
2258                 return 1;
2259         }
2260     }
2261
2262     top_type     = s->current_picture.mb_type[top_xy]    ;
2263     left_type[0] = s->current_picture.mb_type[left_xy[0]];
2264     left_type[1] = s->current_picture.mb_type[left_xy[1]];
2265     if(h->deblocking_filter == 2){
2266         if(h->slice_table[top_xy     ] != h->slice_num) top_type= 0;
2267         if(h->slice_table[left_xy[0] ] != h->slice_num) left_type[0]= left_type[1]= 0;
2268     }else{
2269         if(h->slice_table[top_xy     ] == 0xFFFF) top_type= 0;
2270         if(h->slice_table[left_xy[0] ] == 0xFFFF) left_type[0]= left_type[1] =0;
2271     }
2272     h->top_type    = top_type    ;
2273     h->left_type[0]= left_type[0];
2274     h->left_type[1]= left_type[1];
2275
2276     if(IS_INTRA(mb_type))
2277         return 0;
2278
2279     AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
2280     AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
2281     AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]);
2282     AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]);
2283     AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
2284
2285     h->cbp= h->cbp_table[mb_xy];
2286
2287     {
2288         int list;
2289         for(list=0; list<h->list_count; list++){
2290             int8_t *ref;
2291             int y, b_stride;
2292             int16_t (*mv_dst)[2];
2293             int16_t (*mv_src)[2];
2294
2295             if(!USES_LIST(mb_type, list)){
2296                 fill_rectangle(  h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
2297                 AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2298                 AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2299                 AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2300                 AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2301                 continue;
2302             }
2303
2304             ref = &s->current_picture.ref_index[list][4*mb_xy];
2305             {
2306                 int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2307                 AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2308                 AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2309                 ref += 2;
2310                 AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2311                 AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2312             }
2313
2314             b_stride = h->b_stride;
2315             mv_dst   = &h->mv_cache[list][scan8[0]];
2316             mv_src   = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
2317             for(y=0; y<4; y++){
2318                 AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
2319             }
2320
2321         }
2322     }
2323
2324
2325 /*
2326 0 . T T. T T T T
2327 1 L . .L . . . .
2328 2 L . .L . . . .
2329 3 . T TL . . . .
2330 4 L . .L . . . .
2331 5 L . .. . . . .
2332 */
2333 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
2334     if(top_type){
2335         AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
2336     }
2337
2338     if(left_type[0]){
2339         h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
2340         h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
2341         h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
2342         h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
2343     }
2344
2345     // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
2346     if(!CABAC && h->pps.transform_8x8_mode){
2347         if(IS_8x8DCT(top_type)){
2348             h->non_zero_count_cache[4+8*0]=
2349             h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
2350             h->non_zero_count_cache[6+8*0]=
2351             h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
2352         }
2353         if(IS_8x8DCT(left_type[0])){
2354             h->non_zero_count_cache[3+8*1]=
2355             h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
2356         }
2357         if(IS_8x8DCT(left_type[1])){
2358             h->non_zero_count_cache[3+8*3]=
2359             h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
2360         }
2361
2362         if(IS_8x8DCT(mb_type)){
2363             h->non_zero_count_cache[scan8[0   ]]= h->non_zero_count_cache[scan8[1   ]]=
2364             h->non_zero_count_cache[scan8[2   ]]= h->non_zero_count_cache[scan8[3   ]]= h->cbp & 1;
2365
2366             h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
2367             h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
2368
2369             h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
2370             h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
2371
2372             h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
2373             h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
2374         }
2375     }
2376
2377     if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
2378         int list;
2379         for(list=0; list<h->list_count; list++){
2380             if(USES_LIST(top_type, list)){
2381                 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
2382                 const int b8_xy= 4*top_xy + 2;
2383                 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2384                 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
2385                 h->ref_cache[list][scan8[0] + 0 - 1*8]=
2386                 h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
2387                 h->ref_cache[list][scan8[0] + 2 - 1*8]=
2388                 h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
2389             }else{
2390                 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
2391                 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2392             }
2393
2394             if(!IS_INTERLACED(mb_type^left_type[0])){
2395                 if(USES_LIST(left_type[0], list)){
2396                     const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
2397                     const int b8_xy= 4*left_xy[0] + 1;
2398                     int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2399                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
2400                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
2401                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
2402                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
2403                     h->ref_cache[list][scan8[0] - 1 + 0 ]=
2404                     h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]];
2405                     h->ref_cache[list][scan8[0] - 1 +16 ]=
2406                     h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]];
2407                 }else{
2408                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
2409                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
2410                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
2411                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
2412                     h->ref_cache[list][scan8[0] - 1 + 0  ]=
2413                     h->ref_cache[list][scan8[0] - 1 + 8  ]=
2414                     h->ref_cache[list][scan8[0] - 1 + 16 ]=
2415                     h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
2416                 }
2417             }
2418         }
2419     }
2420
2421     return 0;
2422 }
2423
2424 static void loop_filter(H264Context *h){
2425     MpegEncContext * const s = &h->s;
2426     uint8_t  *dest_y, *dest_cb, *dest_cr;
2427     int linesize, uvlinesize, mb_x, mb_y;
2428     const int end_mb_y= s->mb_y + FRAME_MBAFF;
2429     const int old_slice_type= h->slice_type;
2430
2431     if(h->deblocking_filter) {
2432         for(mb_x= 0; mb_x<s->mb_width; mb_x++){
2433             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
2434                 int mb_xy, mb_type;
2435                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2436                 h->slice_num= h->slice_table[mb_xy];
2437                 mb_type= s->current_picture.mb_type[mb_xy];
2438                 h->list_count= h->list_counts[mb_xy];
2439
2440                 if(FRAME_MBAFF)
2441                     h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2442
2443                 s->mb_x= mb_x;
2444                 s->mb_y= mb_y;
2445                 dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2446                 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2447                 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2448                     //FIXME simplify above
2449
2450                 if (MB_FIELD) {
2451                     linesize   = h->mb_linesize   = s->linesize * 2;
2452                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2453                     if(mb_y&1){ //FIXME move out of this function?
2454                         dest_y -= s->linesize*15;
2455                         dest_cb-= s->uvlinesize*7;
2456                         dest_cr-= s->uvlinesize*7;
2457                     }
2458                 } else {
2459                     linesize   = h->mb_linesize   = s->linesize;
2460                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2461                 }
2462                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2463                 if(fill_filter_caches(h, mb_type))
2464                     continue;
2465                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2466                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2467
2468                 if (FRAME_MBAFF) {
2469                     ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2470                 } else {
2471                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2472                 }
2473             }
2474         }
2475     }
2476     h->slice_type= old_slice_type;
2477     s->mb_x= 0;
2478     s->mb_y= end_mb_y - FRAME_MBAFF;
2479     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2480     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2481 }
2482
2483 static void predict_field_decoding_flag(H264Context *h){
2484     MpegEncContext * const s = &h->s;
2485     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2486     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
2487                 ? s->current_picture.mb_type[mb_xy-1]
2488                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
2489                 ? s->current_picture.mb_type[mb_xy-s->mb_stride]
2490                 : 0;
2491     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2492 }
2493
2494 static int decode_slice(struct AVCodecContext *avctx, void *arg){
2495     H264Context *h = *(void**)arg;
2496     MpegEncContext * const s = &h->s;
2497     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2498
2499     s->mb_skip_run= -1;
2500
2501     h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
2502                     (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
2503
2504     if( h->pps.cabac ) {
2505         /* realign */
2506         align_get_bits( &s->gb );
2507
2508         /* init cabac */
2509         ff_init_cabac_states( &h->cabac);
2510         ff_init_cabac_decoder( &h->cabac,
2511                                s->gb.buffer + get_bits_count(&s->gb)/8,
2512                                (get_bits_left(&s->gb) + 7)/8);
2513
2514         ff_h264_init_cabac_states(h);
2515
2516         for(;;){
2517 //START_TIMER
2518             int ret = ff_h264_decode_mb_cabac(h);
2519             int eos;
2520 //STOP_TIMER("decode_mb_cabac")
2521
2522             if(ret>=0) ff_h264_hl_decode_mb(h);
2523
2524             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
2525                 s->mb_y++;
2526
2527                 ret = ff_h264_decode_mb_cabac(h);
2528
2529                 if(ret>=0) ff_h264_hl_decode_mb(h);
2530                 s->mb_y--;
2531             }
2532             eos = get_cabac_terminate( &h->cabac );
2533
2534             if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
2535                 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);
2536                 return 0;
2537             }
2538             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2539                 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);
2540                 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);
2541                 return -1;
2542             }
2543
2544             if( ++s->mb_x >= s->mb_width ) {
2545                 s->mb_x = 0;
2546                 loop_filter(h);
2547                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2548                 ++s->mb_y;
2549                 if(FIELD_OR_MBAFF_PICTURE) {
2550                     ++s->mb_y;
2551                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2552                         predict_field_decoding_flag(h);
2553                 }
2554             }
2555
2556             if( eos || s->mb_y >= s->mb_height ) {
2557                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2558                 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);
2559                 return 0;
2560             }
2561         }
2562
2563     } else {
2564         for(;;){
2565             int ret = ff_h264_decode_mb_cavlc(h);
2566
2567             if(ret>=0) ff_h264_hl_decode_mb(h);
2568
2569             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
2570                 s->mb_y++;
2571                 ret = ff_h264_decode_mb_cavlc(h);
2572
2573                 if(ret>=0) ff_h264_hl_decode_mb(h);
2574                 s->mb_y--;
2575             }
2576
2577             if(ret<0){
2578                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2579                 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);
2580
2581                 return -1;
2582             }
2583
2584             if(++s->mb_x >= s->mb_width){
2585                 s->mb_x=0;
2586                 loop_filter(h);
2587                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2588                 ++s->mb_y;
2589                 if(FIELD_OR_MBAFF_PICTURE) {
2590                     ++s->mb_y;
2591                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2592                         predict_field_decoding_flag(h);
2593                 }
2594                 if(s->mb_y >= s->mb_height){
2595                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2596
2597                     if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
2598                         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);
2599
2600                         return 0;
2601                     }else{
2602                         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);
2603
2604                         return -1;
2605                     }
2606                 }
2607             }
2608
2609             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
2610                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2611                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2612                     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);
2613
2614                     return 0;
2615                 }else{
2616                     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);
2617
2618                     return -1;
2619                 }
2620             }
2621         }
2622     }
2623
2624 #if 0
2625     for(;s->mb_y < s->mb_height; s->mb_y++){
2626         for(;s->mb_x < s->mb_width; s->mb_x++){
2627             int ret= decode_mb(h);
2628
2629             ff_h264_hl_decode_mb(h);
2630
2631             if(ret<0){
2632                 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2633                 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);
2634
2635                 return -1;
2636             }
2637
2638             if(++s->mb_x >= s->mb_width){
2639                 s->mb_x=0;
2640                 if(++s->mb_y >= s->mb_height){
2641                     if(get_bits_count(s->gb) == s->gb.size_in_bits){
2642                         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);
2643
2644                         return 0;
2645                     }else{
2646                         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);
2647
2648                         return -1;
2649                     }
2650                 }
2651             }
2652
2653             if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2654                 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2655                     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);
2656
2657                     return 0;
2658                 }else{
2659                     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);
2660
2661                     return -1;
2662                 }
2663             }
2664         }
2665         s->mb_x=0;
2666         ff_draw_horiz_band(s, 16*s->mb_y, 16);
2667     }
2668 #endif
2669     return -1; //not reached
2670 }
2671
2672 /**
2673  * Call decode_slice() for each context.
2674  *
2675  * @param h h264 master context
2676  * @param context_count number of contexts to execute
2677  */
2678 static void execute_decode_slices(H264Context *h, int context_count){
2679     MpegEncContext * const s = &h->s;
2680     AVCodecContext * const avctx= s->avctx;
2681     H264Context *hx;
2682     int i;
2683
2684     if (s->avctx->hwaccel)
2685         return;
2686     if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2687         return;
2688     if(context_count == 1) {
2689         decode_slice(avctx, &h);
2690     } else {
2691         for(i = 1; i < context_count; i++) {
2692             hx = h->thread_context[i];
2693             hx->s.error_recognition = avctx->error_recognition;
2694             hx->s.error_count = 0;
2695         }
2696
2697         avctx->execute(avctx, (void *)decode_slice,
2698                        h->thread_context, NULL, context_count, sizeof(void*));
2699
2700         /* pull back stuff from slices to master context */
2701         hx = h->thread_context[context_count - 1];
2702         s->mb_x = hx->s.mb_x;
2703         s->mb_y = hx->s.mb_y;
2704         s->dropable = hx->s.dropable;
2705         s->picture_structure = hx->s.picture_structure;
2706         for(i = 1; i < context_count; i++)
2707             h->s.error_count += h->thread_context[i]->s.error_count;
2708     }
2709 }
2710
2711
2712 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
2713     MpegEncContext * const s = &h->s;
2714     AVCodecContext * const avctx= s->avctx;
2715     int buf_index=0;
2716     H264Context *hx; ///< thread context
2717     int context_count = 0;
2718     int next_avc= h->is_avc ? 0 : buf_size;
2719
2720     h->max_contexts = avctx->thread_count;
2721 #if 0
2722     int i;
2723     for(i=0; i<50; i++){
2724         av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
2725     }
2726 #endif
2727     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
2728         h->current_slice = 0;
2729         if (!s->first_field)
2730             s->current_picture_ptr= NULL;
2731         ff_h264_reset_sei(h);
2732     }
2733
2734     for(;;){
2735         int consumed;
2736         int dst_length;
2737         int bit_length;
2738         const uint8_t *ptr;
2739         int i, nalsize = 0;
2740         int err;
2741
2742         if(buf_index >= next_avc) {
2743             if(buf_index >= buf_size) break;
2744             nalsize = 0;
2745             for(i = 0; i < h->nal_length_size; i++)
2746                 nalsize = (nalsize << 8) | buf[buf_index++];
2747             if(nalsize <= 0 || nalsize > buf_size - buf_index){
2748                 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2749                 break;
2750             }
2751             next_avc= buf_index + nalsize;
2752         } else {
2753             // start code prefix search
2754             for(; buf_index + 3 < next_avc; buf_index++){
2755                 // This should always succeed in the first iteration.
2756                 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2757                     break;
2758             }
2759
2760             if(buf_index+3 >= buf_size) break;
2761
2762             buf_index+=3;
2763             if(buf_index >= next_avc) continue;
2764         }
2765
2766         hx = h->thread_context[context_count];
2767
2768         ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
2769         if (ptr==NULL || dst_length < 0){
2770             return -1;
2771         }
2772         i= buf_index + consumed;
2773         if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
2774            buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
2775             s->workaround_bugs |= FF_BUG_TRUNCATED;
2776
2777         if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
2778         while(ptr[dst_length - 1] == 0 && dst_length > 0)
2779             dst_length--;
2780         }
2781         bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
2782
2783         if(s->avctx->debug&FF_DEBUG_STARTCODE){
2784             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);
2785         }
2786
2787         if (h->is_avc && (nalsize != consumed) && nalsize){
2788             av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
2789         }
2790
2791         buf_index += consumed;
2792
2793         if(  (s->hurry_up == 1 && h->nal_ref_idc  == 0) //FIXME do not discard SEI id
2794            ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0))
2795             continue;
2796
2797       again:
2798         err = 0;
2799         switch(hx->nal_unit_type){
2800         case NAL_IDR_SLICE:
2801             if (h->nal_unit_type != NAL_IDR_SLICE) {
2802                 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2803                 return -1;
2804             }
2805             idr(h); //FIXME ensure we don't loose some frames if there is reordering
2806         case NAL_SLICE:
2807             init_get_bits(&hx->s.gb, ptr, bit_length);
2808             hx->intra_gb_ptr=
2809             hx->inter_gb_ptr= &hx->s.gb;
2810             hx->s.data_partitioning = 0;
2811
2812             if((err = decode_slice_header(hx, h)))
2813                break;
2814
2815             if (h->current_slice == 1) {
2816                 if (s->avctx->hwaccel && s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
2817                     return -1;
2818                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2819                     ff_vdpau_h264_picture_start(s);
2820             }
2821
2822             s->current_picture_ptr->key_frame |=
2823                     (hx->nal_unit_type == NAL_IDR_SLICE) ||
2824                     (h->sei_recovery_frame_cnt >= 0);
2825             if(hx->redundant_pic_count==0 && hx->s.hurry_up < 5
2826                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2827                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2828                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2829                && avctx->skip_frame < AVDISCARD_ALL){
2830                 if(avctx->hwaccel) {
2831                     if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
2832                         return -1;
2833                 }else
2834                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
2835                     static const uint8_t start_code[] = {0x00, 0x00, 0x01};
2836                     ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
2837                     ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
2838                 }else
2839                     context_count++;
2840             }
2841             break;
2842         case NAL_DPA:
2843             init_get_bits(&hx->s.gb, ptr, bit_length);
2844             hx->intra_gb_ptr=
2845             hx->inter_gb_ptr= NULL;
2846
2847             if ((err = decode_slice_header(hx, h)) < 0)
2848                 break;
2849
2850             hx->s.data_partitioning = 1;
2851
2852             break;
2853         case NAL_DPB:
2854             init_get_bits(&hx->intra_gb, ptr, bit_length);
2855             hx->intra_gb_ptr= &hx->intra_gb;
2856             break;
2857         case NAL_DPC:
2858             init_get_bits(&hx->inter_gb, ptr, bit_length);
2859             hx->inter_gb_ptr= &hx->inter_gb;
2860
2861             if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
2862                && s->context_initialized
2863                && s->hurry_up < 5
2864                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2865                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2866                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2867                && avctx->skip_frame < AVDISCARD_ALL)
2868                 context_count++;
2869             break;
2870         case NAL_SEI:
2871             init_get_bits(&s->gb, ptr, bit_length);
2872             ff_h264_decode_sei(h);
2873             break;
2874         case NAL_SPS:
2875             init_get_bits(&s->gb, ptr, bit_length);
2876             ff_h264_decode_seq_parameter_set(h);
2877
2878             if(s->flags& CODEC_FLAG_LOW_DELAY)
2879                 s->low_delay=1;
2880
2881             if(avctx->has_b_frames < 2)
2882                 avctx->has_b_frames= !s->low_delay;
2883             break;
2884         case NAL_PPS:
2885             init_get_bits(&s->gb, ptr, bit_length);
2886
2887             ff_h264_decode_picture_parameter_set(h, bit_length);
2888
2889             break;
2890         case NAL_AUD:
2891         case NAL_END_SEQUENCE:
2892         case NAL_END_STREAM:
2893         case NAL_FILLER_DATA:
2894         case NAL_SPS_EXT:
2895         case NAL_AUXILIARY_SLICE:
2896             break;
2897         default:
2898             av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
2899         }
2900
2901         if(context_count == h->max_contexts) {
2902             execute_decode_slices(h, context_count);
2903             context_count = 0;
2904         }
2905
2906         if (err < 0)
2907             av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
2908         else if(err == 1) {
2909             /* Slice could not be decoded in parallel mode, copy down
2910              * NAL unit stuff to context 0 and restart. Note that
2911              * rbsp_buffer is not transferred, but since we no longer
2912              * run in parallel mode this should not be an issue. */
2913             h->nal_unit_type = hx->nal_unit_type;
2914             h->nal_ref_idc   = hx->nal_ref_idc;
2915             hx = h;
2916             goto again;
2917         }
2918     }
2919     if(context_count)
2920         execute_decode_slices(h, context_count);
2921     return buf_index;
2922 }
2923
2924 /**
2925  * returns the number of bytes consumed for building the current frame
2926  */
2927 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
2928         if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
2929         if(pos+10>buf_size) pos=buf_size; // oops ;)
2930
2931         return pos;
2932 }
2933
2934 static int decode_frame(AVCodecContext *avctx,
2935                              void *data, int *data_size,
2936                              AVPacket *avpkt)
2937 {
2938     const uint8_t *buf = avpkt->data;
2939     int buf_size = avpkt->size;
2940     H264Context *h = avctx->priv_data;
2941     MpegEncContext *s = &h->s;
2942     AVFrame *pict = data;
2943     int buf_index;
2944
2945     s->flags= avctx->flags;
2946     s->flags2= avctx->flags2;
2947
2948    /* end of stream, output what is still in the buffers */
2949  out:
2950     if (buf_size == 0) {
2951         Picture *out;
2952         int i, out_idx;
2953
2954 //FIXME factorize this with the output code below
2955         out = h->delayed_pic[0];
2956         out_idx = 0;
2957         for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2958             if(h->delayed_pic[i]->poc < out->poc){
2959                 out = h->delayed_pic[i];
2960                 out_idx = i;
2961             }
2962
2963         for(i=out_idx; h->delayed_pic[i]; i++)
2964             h->delayed_pic[i] = h->delayed_pic[i+1];
2965
2966         if(out){
2967             *data_size = sizeof(AVFrame);
2968             *pict= *(AVFrame*)out;
2969         }
2970
2971         return 0;
2972     }
2973
2974     buf_index=decode_nal_units(h, buf, buf_size);
2975     if(buf_index < 0)
2976         return -1;
2977
2978     if (!s->current_picture_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
2979         buf_size = 0;
2980         goto out;
2981     }
2982
2983     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
2984         if (avctx->skip_frame >= AVDISCARD_NONREF || s->hurry_up) return 0;
2985         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
2986         return -1;
2987     }
2988
2989     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
2990         Picture *out = s->current_picture_ptr;
2991         Picture *cur = s->current_picture_ptr;
2992         int i, pics, out_of_order, out_idx;
2993
2994         field_end(h);
2995
2996         if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
2997             /* Wait for second field. */
2998             *data_size = 0;
2999
3000         } else {
3001             cur->interlaced_frame = 0;
3002             cur->repeat_pict = 0;
3003
3004             /* Signal interlacing information externally. */
3005             /* Prioritize picture timing SEI information over used decoding process if it exists. */
3006
3007             if(h->sps.pic_struct_present_flag){
3008                 switch (h->sei_pic_struct)
3009                 {
3010                 case SEI_PIC_STRUCT_FRAME:
3011                     break;
3012                 case SEI_PIC_STRUCT_TOP_FIELD:
3013                 case SEI_PIC_STRUCT_BOTTOM_FIELD:
3014                     cur->interlaced_frame = 1;
3015                     break;
3016                 case SEI_PIC_STRUCT_TOP_BOTTOM:
3017                 case SEI_PIC_STRUCT_BOTTOM_TOP:
3018                     if (FIELD_OR_MBAFF_PICTURE)
3019                         cur->interlaced_frame = 1;
3020                     else
3021                         // try to flag soft telecine progressive
3022                         cur->interlaced_frame = h->prev_interlaced_frame;
3023                     break;
3024                 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
3025                 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
3026                     // Signal the possibility of telecined film externally (pic_struct 5,6)
3027                     // From these hints, let the applications decide if they apply deinterlacing.
3028                     cur->repeat_pict = 1;
3029                     break;
3030                 case SEI_PIC_STRUCT_FRAME_DOUBLING:
3031                     // Force progressive here, as doubling interlaced frame is a bad idea.
3032                     cur->repeat_pict = 2;
3033                     break;
3034                 case SEI_PIC_STRUCT_FRAME_TRIPLING:
3035                     cur->repeat_pict = 4;
3036                     break;
3037                 }
3038
3039                 if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
3040                     cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
3041             }else{
3042                 /* Derive interlacing flag from used decoding process. */
3043                 cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
3044             }
3045             h->prev_interlaced_frame = cur->interlaced_frame;
3046
3047             if (cur->field_poc[0] != cur->field_poc[1]){
3048                 /* Derive top_field_first from field pocs. */
3049                 cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
3050             }else{
3051                 if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
3052                     /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
3053                     if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
3054                       || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
3055                         cur->top_field_first = 1;
3056                     else
3057                         cur->top_field_first = 0;
3058                 }else{
3059                     /* Most likely progressive */
3060                     cur->top_field_first = 0;
3061                 }
3062             }
3063
3064         //FIXME do something with unavailable reference frames
3065
3066             /* Sort B-frames into display order */
3067
3068             if(h->sps.bitstream_restriction_flag
3069                && s->avctx->has_b_frames < h->sps.num_reorder_frames){
3070                 s->avctx->has_b_frames = h->sps.num_reorder_frames;
3071                 s->low_delay = 0;
3072             }
3073
3074             if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
3075                && !h->sps.bitstream_restriction_flag){
3076                 s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
3077                 s->low_delay= 0;
3078             }
3079
3080             pics = 0;
3081             while(h->delayed_pic[pics]) pics++;
3082
3083             assert(pics <= MAX_DELAYED_PIC_COUNT);
3084
3085             h->delayed_pic[pics++] = cur;
3086             if(cur->reference == 0)
3087                 cur->reference = DELAYED_PIC_REF;
3088
3089             out = h->delayed_pic[0];
3090             out_idx = 0;
3091             for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
3092                 if(h->delayed_pic[i]->poc < out->poc){
3093                     out = h->delayed_pic[i];
3094                     out_idx = i;
3095                 }
3096             if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
3097                 h->outputed_poc= INT_MIN;
3098             out_of_order = out->poc < h->outputed_poc;
3099
3100             if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
3101                 { }
3102             else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
3103                || (s->low_delay &&
3104                 ((h->outputed_poc != INT_MIN && out->poc > h->outputed_poc + 2)
3105                  || cur->pict_type == FF_B_TYPE)))
3106             {
3107                 s->low_delay = 0;
3108                 s->avctx->has_b_frames++;
3109             }
3110
3111             if(out_of_order || pics > s->avctx->has_b_frames){
3112                 out->reference &= ~DELAYED_PIC_REF;
3113                 for(i=out_idx; h->delayed_pic[i]; i++)
3114                     h->delayed_pic[i] = h->delayed_pic[i+1];
3115             }
3116             if(!out_of_order && pics > s->avctx->has_b_frames){
3117                 *data_size = sizeof(AVFrame);
3118
3119                 if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
3120                     h->outputed_poc = INT_MIN;
3121                 } else
3122                     h->outputed_poc = out->poc;
3123                 *pict= *(AVFrame*)out;
3124             }else{
3125                 av_log(avctx, AV_LOG_DEBUG, "no picture\n");
3126             }
3127         }
3128     }
3129
3130     assert(pict->data[0] || !*data_size);
3131     ff_print_debug_info(s, pict);
3132 //printf("out %d\n", (int)pict->data[0]);
3133
3134     return get_consumed_bytes(s, buf_index, buf_size);
3135 }
3136 #if 0
3137 static inline void fill_mb_avail(H264Context *h){
3138     MpegEncContext * const s = &h->s;
3139     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3140
3141     if(s->mb_y){
3142         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
3143         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
3144         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
3145     }else{
3146         h->mb_avail[0]=
3147         h->mb_avail[1]=
3148         h->mb_avail[2]= 0;
3149     }
3150     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
3151     h->mb_avail[4]= 1; //FIXME move out
3152     h->mb_avail[5]= 0; //FIXME move out
3153 }
3154 #endif
3155
3156 #ifdef TEST
3157 #undef printf
3158 #undef random
3159 #define COUNT 8000
3160 #define SIZE (COUNT*40)
3161 int main(void){
3162     int i;
3163     uint8_t temp[SIZE];
3164     PutBitContext pb;
3165     GetBitContext gb;
3166 //    int int_temp[10000];
3167     DSPContext dsp;
3168     AVCodecContext avctx;
3169
3170     dsputil_init(&dsp, &avctx);
3171
3172     init_put_bits(&pb, temp, SIZE);
3173     printf("testing unsigned exp golomb\n");
3174     for(i=0; i<COUNT; i++){
3175         START_TIMER
3176         set_ue_golomb(&pb, i);
3177         STOP_TIMER("set_ue_golomb");
3178     }
3179     flush_put_bits(&pb);
3180
3181     init_get_bits(&gb, temp, 8*SIZE);
3182     for(i=0; i<COUNT; i++){
3183         int j, s;
3184
3185         s= show_bits(&gb, 24);
3186
3187         START_TIMER
3188         j= get_ue_golomb(&gb);
3189         if(j != i){
3190             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
3191 //            return -1;
3192         }
3193         STOP_TIMER("get_ue_golomb");
3194     }
3195
3196
3197     init_put_bits(&pb, temp, SIZE);
3198     printf("testing signed exp golomb\n");
3199     for(i=0; i<COUNT; i++){
3200         START_TIMER
3201         set_se_golomb(&pb, i - COUNT/2);
3202         STOP_TIMER("set_se_golomb");
3203     }
3204     flush_put_bits(&pb);
3205
3206     init_get_bits(&gb, temp, 8*SIZE);
3207     for(i=0; i<COUNT; i++){
3208         int j, s;
3209
3210         s= show_bits(&gb, 24);
3211
3212         START_TIMER
3213         j= get_se_golomb(&gb);
3214         if(j != i - COUNT/2){
3215             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
3216 //            return -1;
3217         }
3218         STOP_TIMER("get_se_golomb");
3219     }
3220
3221 #if 0
3222     printf("testing 4x4 (I)DCT\n");
3223
3224     DCTELEM block[16];
3225     uint8_t src[16], ref[16];
3226     uint64_t error= 0, max_error=0;
3227
3228     for(i=0; i<COUNT; i++){
3229         int j;
3230 //        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
3231         for(j=0; j<16; j++){
3232             ref[j]= random()%255;
3233             src[j]= random()%255;
3234         }
3235
3236         h264_diff_dct_c(block, src, ref, 4);
3237
3238         //normalize
3239         for(j=0; j<16; j++){
3240 //            printf("%d ", block[j]);
3241             block[j]= block[j]*4;
3242             if(j&1) block[j]= (block[j]*4 + 2)/5;
3243             if(j&4) block[j]= (block[j]*4 + 2)/5;
3244         }
3245 //        printf("\n");
3246
3247         h->h264dsp.h264_idct_add(ref, block, 4);
3248 /*        for(j=0; j<16; j++){
3249             printf("%d ", ref[j]);
3250         }
3251         printf("\n");*/
3252
3253         for(j=0; j<16; j++){
3254             int diff= FFABS(src[j] - ref[j]);
3255
3256             error+= diff*diff;
3257             max_error= FFMAX(max_error, diff);
3258         }
3259     }
3260     printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
3261     printf("testing quantizer\n");
3262     for(qp=0; qp<52; qp++){
3263         for(i=0; i<16; i++)
3264             src1_block[i]= src2_block[i]= random()%255;
3265
3266     }
3267     printf("Testing NAL layer\n");
3268
3269     uint8_t bitstream[COUNT];
3270     uint8_t nal[COUNT*2];
3271     H264Context h;
3272     memset(&h, 0, sizeof(H264Context));
3273
3274     for(i=0; i<COUNT; i++){
3275         int zeros= i;
3276         int nal_length;
3277         int consumed;
3278         int out_length;
3279         uint8_t *out;
3280         int j;
3281
3282         for(j=0; j<COUNT; j++){
3283             bitstream[j]= (random() % 255) + 1;
3284         }
3285
3286         for(j=0; j<zeros; j++){
3287             int pos= random() % COUNT;
3288             while(bitstream[pos] == 0){
3289                 pos++;
3290                 pos %= COUNT;
3291             }
3292             bitstream[pos]=0;
3293         }
3294
3295         START_TIMER
3296
3297         nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
3298         if(nal_length<0){
3299             printf("encoding failed\n");
3300             return -1;
3301         }
3302
3303         out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
3304
3305         STOP_TIMER("NAL")
3306
3307         if(out_length != COUNT){
3308             printf("incorrect length %d %d\n", out_length, COUNT);
3309             return -1;
3310         }
3311
3312         if(consumed != nal_length){
3313             printf("incorrect consumed length %d %d\n", nal_length, consumed);
3314             return -1;
3315         }
3316
3317         if(memcmp(bitstream, out, COUNT)){
3318             printf("mismatch\n");
3319             return -1;
3320         }
3321     }
3322 #endif
3323
3324     printf("Testing RBSP\n");
3325
3326
3327     return 0;
3328 }
3329 #endif /* TEST */
3330
3331
3332 av_cold void ff_h264_free_context(H264Context *h)
3333 {
3334     int i;
3335
3336     free_tables(h, 1); //FIXME cleanup init stuff perhaps
3337
3338     for(i = 0; i < MAX_SPS_COUNT; i++)
3339         av_freep(h->sps_buffers + i);
3340
3341     for(i = 0; i < MAX_PPS_COUNT; i++)
3342         av_freep(h->pps_buffers + i);
3343 }
3344
3345 av_cold int ff_h264_decode_end(AVCodecContext *avctx)
3346 {
3347     H264Context *h = avctx->priv_data;
3348     MpegEncContext *s = &h->s;
3349
3350     ff_h264_free_context(h);
3351
3352     MPV_common_end(s);
3353
3354 //    memset(h, 0, sizeof(H264Context));
3355
3356     return 0;
3357 }
3358
3359
3360 AVCodec h264_decoder = {
3361     "h264",
3362     AVMEDIA_TYPE_VIDEO,
3363     CODEC_ID_H264,
3364     sizeof(H264Context),
3365     ff_h264_decode_init,
3366     NULL,
3367     ff_h264_decode_end,
3368     decode_frame,
3369     /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3370     .flush= flush_dpb,
3371     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
3372 };
3373
3374 #if CONFIG_H264_VDPAU_DECODER
3375 AVCodec h264_vdpau_decoder = {
3376     "h264_vdpau",
3377     AVMEDIA_TYPE_VIDEO,
3378     CODEC_ID_H264,
3379     sizeof(H264Context),
3380     ff_h264_decode_init,
3381     NULL,
3382     ff_h264_decode_end,
3383     decode_frame,
3384     CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3385     .flush= flush_dpb,
3386     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
3387     .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
3388 };
3389 #endif