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 Libav.
   6  *
   7  * Libav 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  * Libav 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 Libav; 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 "libavutil/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 "golomb.h"
  37 #include "mathops.h"
  38 #include "rectangle.h"
  39 #include "vdpau_internal.h"
  40 #include "libavutil/avassert.h"
  41
  42 #include "cabac.h"
  43
  44 //#undef NDEBUG
  45 #include <assert.h>
  46
  47 static const uint8_t rem6[52]={
  48 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,
  49 };
  50
  51 static const uint8_t div6[52]={
  52 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,
  53 };
  54
  55 static const enum PixelFormat hwaccel_pixfmt_list_h264_jpeg_420[] = {
  56     PIX_FMT_DXVA2_VLD,
  57     PIX_FMT_VAAPI_VLD,
  58     PIX_FMT_YUVJ420P,
  59     PIX_FMT_NONE
  60 };
  61
  62 void ff_h264_write_back_intra_pred_mode(H264Context *h){
  63     int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
  64
  65     AV_COPY32(mode, h->intra4x4_pred_mode_cache + 4 + 8*4);
  66     mode[4]= h->intra4x4_pred_mode_cache[7+8*3];
  67     mode[5]= h->intra4x4_pred_mode_cache[7+8*2];
  68     mode[6]= h->intra4x4_pred_mode_cache[7+8*1];
  69 }
  70
  71 /**
  72  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
  73  */
  74 int ff_h264_check_intra4x4_pred_mode(H264Context *h){
  75     MpegEncContext * const s = &h->s;
  76     static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
  77     static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
  78     int i;
  79
  80     if(!(h->top_samples_available&0x8000)){
  81         for(i=0; i<4; i++){
  82             int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
  83             if(status<0){
  84                 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);
  85                 return -1;
  86             } else if(status){
  87                 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
  88             }
  89         }
  90     }
  91
  92     if((h->left_samples_available&0x8888)!=0x8888){
  93         static const int mask[4]={0x8000,0x2000,0x80,0x20};
  94         for(i=0; i<4; i++){
  95             if(!(h->left_samples_available&mask[i])){
  96                 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
  97                 if(status<0){
  98                     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);
  99                     return -1;
 100                 } else if(status){
 101                     h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
 102                 }
 103             }
 104         }
 105     }
 106
 107     return 0;
 108 } //FIXME cleanup like ff_h264_check_intra_pred_mode
 109
 110 /**
 111  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
 112  */
 113 int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
 114     MpegEncContext * const s = &h->s;
 115     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
 116     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
 117
 118     if(mode > 6U) {
 119         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);
 120         return -1;
 121     }
 122
 123     if(!(h->top_samples_available&0x8000)){
 124         mode= top[ mode ];
 125         if(mode<0){
 126             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);
 127             return -1;
 128         }
 129     }
 130
 131     if((h->left_samples_available&0x8080) != 0x8080){
 132         mode= left[ mode ];
 133         if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
 134             mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
 135         }
 136         if(mode<0){
 137             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);
 138             return -1;
 139         }
 140     }
 141
 142     return mode;
 143 }
 144
 145 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
 146     int i, si, di;
 147     uint8_t *dst;
 148     int bufidx;
 149
 150 //    src[0]&0x80;                //forbidden bit
 151     h->nal_ref_idc= src[0]>>5;
 152     h->nal_unit_type= src[0]&0x1F;
 153
 154     src++; length--;
 155
 156 #if HAVE_FAST_UNALIGNED
 157 # if HAVE_FAST_64BIT
 158 #   define RS 7
 159     for(i=0; i+1<length; i+=9){
 160         if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
 161 # else
 162 #   define RS 3
 163     for(i=0; i+1<length; i+=5){
 164         if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
 165 # endif
 166             continue;
 167         if(i>0 && !src[i]) i--;
 168         while(src[i]) i++;
 169 #else
 170 #   define RS 0
 171     for(i=0; i+1<length; i+=2){
 172         if(src[i]) continue;
 173         if(i>0 && src[i-1]==0) i--;
 174 #endif
 175         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
 176             if(src[i+2]!=3){
 177                 /* startcode, so we must be past the end */
 178                 length=i;
 179             }
 180             break;
 181         }
 182         i-= RS;
 183     }
 184
 185     if(i>=length-1){ //no escaped 0
 186         *dst_length= length;
 187         *consumed= length+1; //+1 for the header
 188         return src;
 189     }
 190
 191     bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
 192     av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
 193     dst= h->rbsp_buffer[bufidx];
 194
 195     if (dst == NULL){
 196         return NULL;
 197     }
 198
 199 //printf("decoding esc\n");
 200     memcpy(dst, src, i);
 201     si=di=i;
 202     while(si+2<length){
 203         //remove escapes (very rare 1:2^22)
 204         if(src[si+2]>3){
 205             dst[di++]= src[si++];
 206             dst[di++]= src[si++];
 207         }else if(src[si]==0 && src[si+1]==0){
 208             if(src[si+2]==3){ //escape
 209                 dst[di++]= 0;
 210                 dst[di++]= 0;
 211                 si+=3;
 212                 continue;
 213             }else //next start code
 214                 goto nsc;
 215         }
 216
 217         dst[di++]= src[si++];
 218     }
 219     while(si<length)
 220         dst[di++]= src[si++];
 221 nsc:
 222
 223     memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
 224
 225     *dst_length= di;
 226     *consumed= si + 1;//+1 for the header
 227 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
 228     return dst;
 229 }
 230
 231 /**
 232  * Identify the exact end of the bitstream
 233  * @return the length of the trailing, or 0 if damaged
 234  */
 235 static int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
 236     int v= *src;
 237     int r;
 238
 239     tprintf(h->s.avctx, "rbsp trailing %X\n", v);
 240
 241     for(r=1; r<9; r++){
 242         if(v&1) return r;
 243         v>>=1;
 244     }
 245     return 0;
 246 }
 247
 248 #if 0
 249 /**
 250  * DCT transforms the 16 dc values.
 251  * @param qp quantization parameter ??? FIXME
 252  */
 253 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
 254 //    const int qmul= dequant_coeff[qp][0];
 255     int i;
 256     int temp[16]; //FIXME check if this is a good idea
 257     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
 258     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
 259
 260     for(i=0; i<4; i++){
 261         const int offset= y_offset[i];
 262         const int z0= block[offset+stride*0] + block[offset+stride*4];
 263         const int z1= block[offset+stride*0] - block[offset+stride*4];
 264         const int z2= block[offset+stride*1] - block[offset+stride*5];
 265         const int z3= block[offset+stride*1] + block[offset+stride*5];
 266
 267         temp[4*i+0]= z0+z3;
 268         temp[4*i+1]= z1+z2;
 269         temp[4*i+2]= z1-z2;
 270         temp[4*i+3]= z0-z3;
 271     }
 272
 273     for(i=0; i<4; i++){
 274         const int offset= x_offset[i];
 275         const int z0= temp[4*0+i] + temp[4*2+i];
 276         const int z1= temp[4*0+i] - temp[4*2+i];
 277         const int z2= temp[4*1+i] - temp[4*3+i];
 278         const int z3= temp[4*1+i] + temp[4*3+i];
 279
 280         block[stride*0 +offset]= (z0 + z3)>>1;
 281         block[stride*2 +offset]= (z1 + z2)>>1;
 282         block[stride*8 +offset]= (z1 - z2)>>1;
 283         block[stride*10+offset]= (z0 - z3)>>1;
 284     }
 285 }
 286 #endif
 287
 288 #undef xStride
 289 #undef stride
 290
 291 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qmul){
 292     const int stride= 16*2;
 293     const int xStride= 16;
 294     int a,b,c,d,e;
 295
 296     a= block[stride*0 + xStride*0];
 297     b= block[stride*0 + xStride*1];
 298     c= block[stride*1 + xStride*0];
 299     d= block[stride*1 + xStride*1];
 300
 301     e= a-b;
 302     a= a+b;
 303     b= c-d;
 304     c= c+d;
 305
 306     block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
 307     block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
 308     block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
 309     block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
 310 }
 311
 312 #if 0
 313 static void chroma_dc_dct_c(DCTELEM *block){
 314     const int stride= 16*2;
 315     const int xStride= 16;
 316     int a,b,c,d,e;
 317
 318     a= block[stride*0 + xStride*0];
 319     b= block[stride*0 + xStride*1];
 320     c= block[stride*1 + xStride*0];
 321     d= block[stride*1 + xStride*1];
 322
 323     e= a-b;
 324     a= a+b;
 325     b= c-d;
 326     c= c+d;
 327
 328     block[stride*0 + xStride*0]= (a+c);
 329     block[stride*0 + xStride*1]= (e+b);
 330     block[stride*1 + xStride*0]= (a-c);
 331     block[stride*1 + xStride*1]= (e-b);
 332 }
 333 #endif
 334
 335 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
 336                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 337                            int src_x_offset, int src_y_offset,
 338                            qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
 339     MpegEncContext * const s = &h->s;
 340     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
 341     int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
 342     const int luma_xy= (mx&3) + ((my&3)<<2);
 343     uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
 344     uint8_t * src_cb, * src_cr;
 345     int extra_width= h->emu_edge_width;
 346     int extra_height= h->emu_edge_height;
 347     int emu=0;
 348     const int full_mx= mx>>2;
 349     const int full_my= my>>2;
 350     const int pic_width  = 16*s->mb_width;
 351     const int pic_height = 16*s->mb_height >> MB_FIELD;
 352
 353     if(mx&7) extra_width -= 3;
 354     if(my&7) extra_height -= 3;
 355
 356     if(   full_mx < 0-extra_width
 357        || full_my < 0-extra_height
 358        || full_mx + 16/*FIXME*/ > pic_width + extra_width
 359        || full_my + 16/*FIXME*/ > pic_height + extra_height){
 360         s->dsp.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);
 361             src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
 362         emu=1;
 363     }
 364
 365     qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
 366     if(!square){
 367         qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
 368     }
 369
 370     if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
 371
 372     if(MB_FIELD){
 373         // chroma offset when predicting from a field of opposite parity
 374         my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
 375         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
 376     }
 377     src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
 378     src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
 379
 380     if(emu){
 381         s->dsp.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);
 382             src_cb= s->edge_emu_buffer;
 383     }
 384     chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
 385
 386     if(emu){
 387         s->dsp.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);
 388             src_cr= s->edge_emu_buffer;
 389     }
 390     chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
 391 }
 392
 393 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
 394                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 395                            int x_offset, int y_offset,
 396                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 397                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
 398                            int list0, int list1){
 399     MpegEncContext * const s = &h->s;
 400     qpel_mc_func *qpix_op=  qpix_put;
 401     h264_chroma_mc_func chroma_op= chroma_put;
 402
 403     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
 404     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
 405     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
 406     x_offset += 8*s->mb_x;
 407     y_offset += 8*(s->mb_y >> MB_FIELD);
 408
 409     if(list0){
 410         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
 411         mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
 412                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
 413                            qpix_op, chroma_op);
 414
 415         qpix_op=  qpix_avg;
 416         chroma_op= chroma_avg;
 417     }
 418
 419     if(list1){
 420         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
 421         mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
 422                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
 423                            qpix_op, chroma_op);
 424     }
 425 }
 426
 427 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
 428                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 429                            int x_offset, int y_offset,
 430                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 431                            h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
 432                            h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
 433                            int list0, int list1){
 434     MpegEncContext * const s = &h->s;
 435
 436     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
 437     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
 438     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
 439     x_offset += 8*s->mb_x;
 440     y_offset += 8*(s->mb_y >> MB_FIELD);
 441
 442     if(list0 && list1){
 443         /* don't optimize for luma-only case, since B-frames usually
 444          * use implicit weights => chroma too. */
 445         uint8_t *tmp_cb = s->obmc_scratchpad;
 446         uint8_t *tmp_cr = s->obmc_scratchpad + 8;
 447         uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
 448         int refn0 = h->ref_cache[0][ scan8[n] ];
 449         int refn1 = h->ref_cache[1][ scan8[n] ];
 450
 451         mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
 452                     dest_y, dest_cb, dest_cr,
 453                     x_offset, y_offset, qpix_put, chroma_put);
 454         mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
 455                     tmp_y, tmp_cb, tmp_cr,
 456                     x_offset, y_offset, qpix_put, chroma_put);
 457
 458         if(h->use_weight == 2){
 459             int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
 460             int weight1 = 64 - weight0;
 461             luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
 462             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
 463             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
 464         }else{
 465             luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
 466                             h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
 467                             h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
 468             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 469                             h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
 470                             h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
 471             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 472                             h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
 473                             h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
 474         }
 475     }else{
 476         int list = list1 ? 1 : 0;
 477         int refn = h->ref_cache[list][ scan8[n] ];
 478         Picture *ref= &h->ref_list[list][refn];
 479         mc_dir_part(h, ref, n, square, chroma_height, delta, list,
 480                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
 481                     qpix_put, chroma_put);
 482
 483         luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
 484                        h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
 485         if(h->use_weight_chroma){
 486             chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 487                              h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
 488             chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 489                              h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
 490         }
 491     }
 492 }
 493
 494 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
 495                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 496                            int x_offset, int y_offset,
 497                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 498                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
 499                            h264_weight_func *weight_op, h264_biweight_func *weight_avg,
 500                            int list0, int list1){
 501     if((h->use_weight==2 && list0 && list1
 502         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
 503        || h->use_weight==1)
 504         mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
 505                          x_offset, y_offset, qpix_put, chroma_put,
 506                          weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
 507     else
 508         mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
 509                     x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
 510 }
 511
 512 static inline void prefetch_motion(H264Context *h, int list){
 513     /* fetch pixels for estimated mv 4 macroblocks ahead
 514      * optimized for 64byte cache lines */
 515     MpegEncContext * const s = &h->s;
 516     const int refn = h->ref_cache[list][scan8[0]];
 517     if(refn >= 0){
 518         const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
 519         const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
 520         uint8_t **src= h->ref_list[list][refn].data;
 521         int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
 522         s->dsp.prefetch(src[0]+off, s->linesize, 4);
 523         off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
 524         s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
 525     }
 526 }
 527
 528 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 529                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
 530                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
 531                       h264_weight_func *weight_op, h264_biweight_func *weight_avg){
 532     MpegEncContext * const s = &h->s;
 533     const int mb_xy= h->mb_xy;
 534     const int mb_type= s->current_picture.mb_type[mb_xy];
 535
 536     assert(IS_INTER(mb_type));
 537
 538     prefetch_motion(h, 0);
 539
 540     if(IS_16X16(mb_type)){
 541         mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
 542                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
 543                 weight_op, weight_avg,
 544                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
 545     }else if(IS_16X8(mb_type)){
 546         mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
 547                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
 548                 &weight_op[1], &weight_avg[1],
 549                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
 550         mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
 551                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
 552                 &weight_op[1], &weight_avg[1],
 553                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
 554     }else if(IS_8X16(mb_type)){
 555         mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
 556                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 557                 &weight_op[2], &weight_avg[2],
 558                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
 559         mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
 560                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 561                 &weight_op[2], &weight_avg[2],
 562                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
 563     }else{
 564         int i;
 565
 566         assert(IS_8X8(mb_type));
 567
 568         for(i=0; i<4; i++){
 569             const int sub_mb_type= h->sub_mb_type[i];
 570             const int n= 4*i;
 571             int x_offset= (i&1)<<2;
 572             int y_offset= (i&2)<<1;
 573
 574             if(IS_SUB_8X8(sub_mb_type)){
 575                 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 576                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 577                     &weight_op[3], &weight_avg[3],
 578                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 579             }else if(IS_SUB_8X4(sub_mb_type)){
 580                 mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 581                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
 582                     &weight_op[4], &weight_avg[4],
 583                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 584                 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
 585                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
 586                     &weight_op[4], &weight_avg[4],
 587                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 588             }else if(IS_SUB_4X8(sub_mb_type)){
 589                 mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 590                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 591                     &weight_op[5], &weight_avg[5],
 592                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 593                 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
 594                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 595                     &weight_op[5], &weight_avg[5],
 596                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 597             }else{
 598                 int j;
 599                 assert(IS_SUB_4X4(sub_mb_type));
 600                 for(j=0; j<4; j++){
 601                     int sub_x_offset= x_offset + 2*(j&1);
 602                     int sub_y_offset= y_offset +   (j&2);
 603                     mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
 604                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 605                         &weight_op[6], &weight_avg[6],
 606                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 607                 }
 608             }
 609         }
 610     }
 611
 612     prefetch_motion(h, 1);
 613 }
 614
 615
 616 static void free_tables(H264Context *h, int free_rbsp){
 617     int i;
 618     H264Context *hx;
 619     av_freep(&h->intra4x4_pred_mode);
 620     av_freep(&h->chroma_pred_mode_table);
 621     av_freep(&h->cbp_table);
 622     av_freep(&h->mvd_table[0]);
 623     av_freep(&h->mvd_table[1]);
 624     av_freep(&h->direct_table);
 625     av_freep(&h->non_zero_count);
 626     av_freep(&h->slice_table_base);
 627     h->slice_table= NULL;
 628     av_freep(&h->list_counts);
 629
 630     av_freep(&h->mb2b_xy);
 631     av_freep(&h->mb2br_xy);
 632
 633     for(i = 0; i < MAX_THREADS; i++) {
 634         hx = h->thread_context[i];
 635         if(!hx) continue;
 636         av_freep(&hx->top_borders[1]);
 637         av_freep(&hx->top_borders[0]);
 638         av_freep(&hx->s.obmc_scratchpad);
 639         if (free_rbsp){
 640             av_freep(&hx->rbsp_buffer[1]);
 641             av_freep(&hx->rbsp_buffer[0]);
 642             hx->rbsp_buffer_size[0] = 0;
 643             hx->rbsp_buffer_size[1] = 0;
 644         }
 645         if (i) av_freep(&h->thread_context[i]);
 646     }
 647 }
 648
 649 static void init_dequant8_coeff_table(H264Context *h){
 650     int i,q,x;
 651     h->dequant8_coeff[0] = h->dequant8_buffer[0];
 652     h->dequant8_coeff[1] = h->dequant8_buffer[1];
 653
 654     for(i=0; i<2; i++ ){
 655         if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
 656             h->dequant8_coeff[1] = h->dequant8_buffer[0];
 657             break;
 658         }
 659
 660         for(q=0; q<52; q++){
 661             int shift = div6[q];
 662             int idx = rem6[q];
 663             for(x=0; x<64; x++)
 664                 h->dequant8_coeff[i][q][(x>>3)|((x&7)<<3)] =
 665                     ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
 666                     h->pps.scaling_matrix8[i][x]) << shift;
 667         }
 668     }
 669 }
 670
 671 static void init_dequant4_coeff_table(H264Context *h){
 672     int i,j,q,x;
 673     for(i=0; i<6; i++ ){
 674         h->dequant4_coeff[i] = h->dequant4_buffer[i];
 675         for(j=0; j<i; j++){
 676             if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
 677                 h->dequant4_coeff[i] = h->dequant4_buffer[j];
 678                 break;
 679             }
 680         }
 681         if(j<i)
 682             continue;
 683
 684         for(q=0; q<52; q++){
 685             int shift = div6[q] + 2;
 686             int idx = rem6[q];
 687             for(x=0; x<16; x++)
 688                 h->dequant4_coeff[i][q][(x>>2)|((x<<2)&0xF)] =
 689                     ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
 690                     h->pps.scaling_matrix4[i][x]) << shift;
 691         }
 692     }
 693 }
 694
 695 static void init_dequant_tables(H264Context *h){
 696     int i,x;
 697     init_dequant4_coeff_table(h);
 698     if(h->pps.transform_8x8_mode)
 699         init_dequant8_coeff_table(h);
 700     if(h->sps.transform_bypass){
 701         for(i=0; i<6; i++)
 702             for(x=0; x<16; x++)
 703                 h->dequant4_coeff[i][0][x] = 1<<6;
 704         if(h->pps.transform_8x8_mode)
 705             for(i=0; i<2; i++)
 706                 for(x=0; x<64; x++)
 707                     h->dequant8_coeff[i][0][x] = 1<<6;
 708     }
 709 }
 710
 711
 712 int ff_h264_alloc_tables(H264Context *h){
 713     MpegEncContext * const s = &h->s;
 714     const int big_mb_num= s->mb_stride * (s->mb_height+1);
 715     const int row_mb_num= 2*s->mb_stride*s->avctx->thread_count;
 716     int x,y;
 717
 718     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8  * sizeof(uint8_t), fail)
 719
 720     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 32 * sizeof(uint8_t), fail)
 721     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
 722     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
 723
 724     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
 725     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
 726     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
 727     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
 728     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
 729
 730     memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
 731     h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
 732
 733     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy  , big_mb_num * sizeof(uint32_t), fail);
 734     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
 735     for(y=0; y<s->mb_height; y++){
 736         for(x=0; x<s->mb_width; x++){
 737             const int mb_xy= x + y*s->mb_stride;
 738             const int b_xy = 4*x + 4*y*h->b_stride;
 739
 740             h->mb2b_xy [mb_xy]= b_xy;
 741             h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
 742         }
 743     }
 744
 745     s->obmc_scratchpad = NULL;
 746
 747     if(!h->dequant4_coeff[0])
 748         init_dequant_tables(h);
 749
 750     return 0;
 751 fail:
 752     free_tables(h, 1);
 753     return -1;
 754 }
 755
 756 /**
 757  * Mimic alloc_tables(), but for every context thread.
 758  */
 759 static void clone_tables(H264Context *dst, H264Context *src, int i){
 760     MpegEncContext * const s = &src->s;
 761     dst->intra4x4_pred_mode       = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
 762     dst->non_zero_count           = src->non_zero_count;
 763     dst->slice_table              = src->slice_table;
 764     dst->cbp_table                = src->cbp_table;
 765     dst->mb2b_xy                  = src->mb2b_xy;
 766     dst->mb2br_xy                 = src->mb2br_xy;
 767     dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
 768     dst->mvd_table[0]             = src->mvd_table[0] + i*8*2*s->mb_stride;
 769     dst->mvd_table[1]             = src->mvd_table[1] + i*8*2*s->mb_stride;
 770     dst->direct_table             = src->direct_table;
 771     dst->list_counts              = src->list_counts;
 772
 773     dst->s.obmc_scratchpad = NULL;
 774     ff_h264_pred_init(&dst->hpc, src->s.codec_id);
 775 }
 776
 777 /**
 778  * Init context
 779  * Allocate buffers which are not shared amongst multiple threads.
 780  */
 781 static int context_init(H264Context *h){
 782     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
 783     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
 784
 785     h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
 786     h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
 787
 788     return 0;
 789 fail:
 790     return -1; // free_tables will clean up for us
 791 }
 792
 793 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
 794
 795 static av_cold void common_init(H264Context *h){
 796     MpegEncContext * const s = &h->s;
 797
 798     s->width = s->avctx->width;
 799     s->height = s->avctx->height;
 800     s->codec_id= s->avctx->codec->id;
 801
 802     ff_h264dsp_init(&h->h264dsp);
 803     ff_h264_pred_init(&h->hpc, s->codec_id);
 804
 805     h->dequant_coeff_pps= -1;
 806     s->unrestricted_mv=1;
 807     s->decode=1; //FIXME
 808
 809     dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
 810
 811     memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
 812     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
 813 }
 814
 815 int ff_h264_decode_extradata(H264Context *h)
 816 {
 817     AVCodecContext *avctx = h->s.avctx;
 818
 819     if(*(char *)avctx->extradata == 1){
 820         int i, cnt, nalsize;
 821         unsigned char *p = avctx->extradata;
 822
 823         h->is_avc = 1;
 824
 825         if(avctx->extradata_size < 7) {
 826             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
 827             return -1;
 828         }
 829         /* sps and pps in the avcC always have length coded with 2 bytes,
 830            so put a fake nal_length_size = 2 while parsing them */
 831         h->nal_length_size = 2;
 832         // Decode sps from avcC
 833         cnt = *(p+5) & 0x1f; // Number of sps
 834         p += 6;
 835         for (i = 0; i < cnt; i++) {
 836             nalsize = AV_RB16(p) + 2;
 837             if(decode_nal_units(h, p, nalsize) < 0) {
 838                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
 839                 return -1;
 840             }
 841             p += nalsize;
 842         }
 843         // Decode pps from avcC
 844         cnt = *(p++); // Number of pps
 845         for (i = 0; i < cnt; i++) {
 846             nalsize = AV_RB16(p) + 2;
 847             if(decode_nal_units(h, p, nalsize)  != nalsize) {
 848                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
 849                 return -1;
 850             }
 851             p += nalsize;
 852         }
 853         // Now store right nal length size, that will be use to parse all other nals
 854         h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
 855     } else {
 856         h->is_avc = 0;
 857         if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
 858             return -1;
 859     }
 860     return 0;
 861 }
 862
 863 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
 864     H264Context *h= avctx->priv_data;
 865     MpegEncContext * const s = &h->s;
 866
 867     MPV_decode_defaults(s);
 868
 869     s->avctx = avctx;
 870     common_init(h);
 871
 872     s->out_format = FMT_H264;
 873     s->workaround_bugs= avctx->workaround_bugs;
 874
 875     // set defaults
 876 //    s->decode_mb= ff_h263_decode_mb;
 877     s->quarter_sample = 1;
 878     if(!avctx->has_b_frames)
 879     s->low_delay= 1;
 880
 881     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
 882
 883     ff_h264_decode_init_vlc();
 884
 885     h->thread_context[0] = h;
 886     h->outputed_poc = INT_MIN;
 887     h->prev_poc_msb= 1<<16;
 888     h->x264_build = -1;
 889     ff_h264_reset_sei(h);
 890     if(avctx->codec_id == CODEC_ID_H264){
 891         if(avctx->ticks_per_frame == 1){
 892             s->avctx->time_base.den *=2;
 893         }
 894         avctx->ticks_per_frame = 2;
 895     }
 896
 897     if(avctx->extradata_size > 0 && avctx->extradata &&
 898         ff_h264_decode_extradata(h))
 899         return -1;
 900
 901     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
 902         s->avctx->has_b_frames = h->sps.num_reorder_frames;
 903         s->low_delay = 0;
 904     }
 905
 906     return 0;
 907 }
 908
 909 int ff_h264_frame_start(H264Context *h){
 910     MpegEncContext * const s = &h->s;
 911     int i;
 912
 913     if(MPV_frame_start(s, s->avctx) < 0)
 914         return -1;
 915     ff_er_frame_start(s);
 916     /*
 917      * MPV_frame_start uses pict_type to derive key_frame.
 918      * This is incorrect for H.264; IDR markings must be used.
 919      * Zero here; IDR markings per slice in frame or fields are ORed in later.
 920      * See decode_nal_units().
 921      */
 922     s->current_picture_ptr->key_frame= 0;
 923     s->current_picture_ptr->mmco_reset= 0;
 924
 925     assert(s->linesize && s->uvlinesize);
 926
 927     for(i=0; i<16; i++){
 928         h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
 929         h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
 930     }
 931     for(i=0; i<4; i++){
 932         h->block_offset[16+i]=
 933         h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
 934         h->block_offset[24+16+i]=
 935         h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
 936     }
 937
 938     /* can't be in alloc_tables because linesize isn't known there.
 939      * FIXME: redo bipred weight to not require extra buffer? */
 940     for(i = 0; i < s->avctx->thread_count; i++)
 941         if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
 942             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
 943
 944     /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
 945     memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
 946
 947 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
 948
 949     // We mark the current picture as non-reference after allocating it, so
 950     // that if we break out due to an error it can be released automatically
 951     // in the next MPV_frame_start().
 952     // SVQ3 as well as most other codecs have only last/next/current and thus
 953     // get released even with set reference, besides SVQ3 and others do not
 954     // mark frames as reference later "naturally".
 955     if(s->codec_id != CODEC_ID_SVQ3)
 956         s->current_picture_ptr->reference= 0;
 957
 958     s->current_picture_ptr->field_poc[0]=
 959     s->current_picture_ptr->field_poc[1]= INT_MAX;
 960     assert(s->current_picture_ptr->long_ref==0);
 961
 962     return 0;
 963 }
 964
 965 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){
 966     MpegEncContext * const s = &h->s;
 967     uint8_t *top_border;
 968     int top_idx = 1;
 969
 970     src_y  -=   linesize;
 971     src_cb -= uvlinesize;
 972     src_cr -= uvlinesize;
 973
 974     if(!simple && FRAME_MBAFF){
 975         if(s->mb_y&1){
 976             if(!MB_MBAFF){
 977                 top_border = h->top_borders[0][s->mb_x];
 978                 AV_COPY128(top_border, src_y + 15*linesize);
 979                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
 980                     AV_COPY64(top_border+16, src_cb+7*uvlinesize);
 981                     AV_COPY64(top_border+24, src_cr+7*uvlinesize);
 982                 }
 983             }
 984         }else if(MB_MBAFF){
 985             top_idx = 0;
 986         }else
 987             return;
 988     }
 989
 990     top_border = h->top_borders[top_idx][s->mb_x];
 991     // There are two lines saved, the line above the the top macroblock of a pair,
 992     // and the line above the bottom macroblock
 993     AV_COPY128(top_border, src_y + 16*linesize);
 994
 995     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
 996         AV_COPY64(top_border+16, src_cb+8*uvlinesize);
 997         AV_COPY64(top_border+24, src_cr+8*uvlinesize);
 998     }
 999 }
1000
1001 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){
1002     MpegEncContext * const s = &h->s;
1003     int deblock_left;
1004     int deblock_top;
1005     int top_idx = 1;
1006     uint8_t *top_border_m1;
1007     uint8_t *top_border;
1008
1009     if(!simple && FRAME_MBAFF){
1010         if(s->mb_y&1){
1011             if(!MB_MBAFF)
1012                 return;
1013         }else{
1014             top_idx = MB_MBAFF ? 0 : 1;
1015         }
1016     }
1017
1018     if(h->deblocking_filter == 2) {
1019         deblock_left = h->left_type[0];
1020         deblock_top  = h->top_type;
1021     } else {
1022         deblock_left = (s->mb_x > 0);
1023         deblock_top =  (s->mb_y > !!MB_FIELD);
1024     }
1025
1026     src_y  -=   linesize + 1;
1027     src_cb -= uvlinesize + 1;
1028     src_cr -= uvlinesize + 1;
1029
1030     top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1031     top_border    = h->top_borders[top_idx][s->mb_x];
1032
1033 #define XCHG(a,b,xchg)\
1034 if (xchg) AV_SWAP64(b,a);\
1035 else      AV_COPY64(b,a);
1036
1037     if(deblock_top){
1038         if(deblock_left){
1039             XCHG(top_border_m1+8, src_y -7, 1);
1040         }
1041         XCHG(top_border+0, src_y +1, xchg);
1042         XCHG(top_border+8, src_y +9, 1);
1043         if(s->mb_x+1 < s->mb_width){
1044             XCHG(h->top_borders[top_idx][s->mb_x+1], src_y +17, 1);
1045         }
1046     }
1047
1048     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1049         if(deblock_top){
1050             if(deblock_left){
1051                 XCHG(top_border_m1+16, src_cb -7, 1);
1052                 XCHG(top_border_m1+24, src_cr -7, 1);
1053             }
1054             XCHG(top_border+16, src_cb+1, 1);
1055             XCHG(top_border+24, src_cr+1, 1);
1056         }
1057     }
1058 }
1059
1060 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
1061     MpegEncContext * const s = &h->s;
1062     const int mb_x= s->mb_x;
1063     const int mb_y= s->mb_y;
1064     const int mb_xy= h->mb_xy;
1065     const int mb_type= s->current_picture.mb_type[mb_xy];
1066     uint8_t  *dest_y, *dest_cb, *dest_cr;
1067     int linesize, uvlinesize /*dct_offset*/;
1068     int i;
1069     int *block_offset = &h->block_offset[0];
1070     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1071     /* is_h264 should always be true if SVQ3 is disabled. */
1072     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1073     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1074     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1075
1076     dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
1077     dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
1078     dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
1079
1080     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
1081     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
1082
1083     h->list_counts[mb_xy]= h->list_count;
1084
1085     if (!simple && MB_FIELD) {
1086         linesize   = h->mb_linesize   = s->linesize * 2;
1087         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1088         block_offset = &h->block_offset[24];
1089         if(mb_y&1){ //FIXME move out of this function?
1090             dest_y -= s->linesize*15;
1091             dest_cb-= s->uvlinesize*7;
1092             dest_cr-= s->uvlinesize*7;
1093         }
1094         if(FRAME_MBAFF) {
1095             int list;
1096             for(list=0; list<h->list_count; list++){
1097                 if(!USES_LIST(mb_type, list))
1098                     continue;
1099                 if(IS_16X16(mb_type)){
1100                     int8_t *ref = &h->ref_cache[list][scan8[0]];
1101                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1102                 }else{
1103                     for(i=0; i<16; i+=4){
1104                         int ref = h->ref_cache[list][scan8[i]];
1105                         if(ref >= 0)
1106                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1107                     }
1108                 }
1109             }
1110         }
1111     } else {
1112         linesize   = h->mb_linesize   = s->linesize;
1113         uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1114 //        dct_offset = s->linesize * 16;
1115     }
1116
1117     if (!simple && IS_INTRA_PCM(mb_type)) {
1118         for (i=0; i<16; i++) {
1119             memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1120         }
1121         for (i=0; i<8; i++) {
1122             memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
1123             memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
1124         }
1125     } else {
1126         if(IS_INTRA(mb_type)){
1127             if(h->deblocking_filter)
1128                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
1129
1130             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1131                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1132                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1133             }
1134
1135             if(IS_INTRA4x4(mb_type)){
1136                 if(simple || !s->encoding){
1137                     if(IS_8x8DCT(mb_type)){
1138                         if(transform_bypass){
1139                             idct_dc_add =
1140                             idct_add    = s->dsp.add_pixels8;
1141                         }else{
1142                             idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1143                             idct_add    = h->h264dsp.h264_idct8_add;
1144                         }
1145                         for(i=0; i<16; i+=4){
1146                             uint8_t * const ptr= dest_y + block_offset[i];
1147                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1148                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1149                                 h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
1150                             }else{
1151                                 const int nnz = h->non_zero_count_cache[ scan8[i] ];
1152                                 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1153                                                             (h->topright_samples_available<<i)&0x4000, linesize);
1154                                 if(nnz){
1155                                     if(nnz == 1 && h->mb[i*16])
1156                                         idct_dc_add(ptr, h->mb + i*16, linesize);
1157                                     else
1158                                         idct_add   (ptr, h->mb + i*16, linesize);
1159                                 }
1160                             }
1161                         }
1162                     }else{
1163                         if(transform_bypass){
1164                             idct_dc_add =
1165                             idct_add    = s->dsp.add_pixels4;
1166                         }else{
1167                             idct_dc_add = h->h264dsp.h264_idct_dc_add;
1168                             idct_add    = h->h264dsp.h264_idct_add;
1169                         }
1170                         for(i=0; i<16; i++){
1171                             uint8_t * const ptr= dest_y + block_offset[i];
1172                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1173
1174                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1175                                 h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
1176                             }else{
1177                                 uint8_t *topright;
1178                                 int nnz, tr;
1179                                 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1180                                     const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1181                                     assert(mb_y || linesize <= block_offset[i]);
1182                                     if(!topright_avail){
1183                                         tr= ptr[3 - linesize]*0x01010101;
1184                                         topright= (uint8_t*) &tr;
1185                                     }else
1186                                         topright= ptr + 4 - linesize;
1187                                 }else
1188                                     topright= NULL;
1189
1190                                 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1191                                 nnz = h->non_zero_count_cache[ scan8[i] ];
1192                                 if(nnz){
1193                                     if(is_h264){
1194                                         if(nnz == 1 && h->mb[i*16])
1195                                             idct_dc_add(ptr, h->mb + i*16, linesize);
1196                                         else
1197                                             idct_add   (ptr, h->mb + i*16, linesize);
1198                                     }else
1199                                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1200                                 }
1201                             }
1202                         }
1203                     }
1204                 }
1205             }else{
1206                 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1207                 if(is_h264){
1208                     if(h->non_zero_count_cache[ scan8[LUMA_DC_BLOCK_INDEX] ]){
1209                         if(!transform_bypass)
1210                             h->h264dsp.h264_luma_dc_dequant_idct(h->mb, h->mb_luma_dc, h->dequant4_coeff[0][s->qscale][0]);
1211                         else{
1212                             static const uint8_t dc_mapping[16] = { 0*16, 1*16, 4*16, 5*16, 2*16, 3*16, 6*16, 7*16,
1213                                                                     8*16, 9*16,12*16,13*16,10*16,11*16,14*16,15*16};
1214                             for(i = 0; i < 16; i++)
1215                                 h->mb[dc_mapping[i]] = h->mb_luma_dc[i];
1216                         }
1217                     }
1218                 }else
1219                     ff_svq3_luma_dc_dequant_idct_c(h->mb, h->mb_luma_dc, s->qscale);
1220             }
1221             if(h->deblocking_filter)
1222                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
1223         }else if(is_h264){
1224             hl_motion(h, dest_y, dest_cb, dest_cr,
1225                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1226                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1227                       h->h264dsp.weight_h264_pixels_tab, h->h264dsp.biweight_h264_pixels_tab);
1228         }
1229
1230
1231         if(!IS_INTRA4x4(mb_type)){
1232             if(is_h264){
1233                 if(IS_INTRA16x16(mb_type)){
1234                     if(transform_bypass){
1235                         if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1236                             h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1237                         }else{
1238                             for(i=0; i<16; i++){
1239                                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1240                                     s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
1241                             }
1242                         }
1243                     }else{
1244                          h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1245                     }
1246                 }else if(h->cbp&15){
1247                     if(transform_bypass){
1248                         const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1249                         idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1250                         for(i=0; i<16; i+=di){
1251                             if(h->non_zero_count_cache[ scan8[i] ]){
1252                                 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
1253                             }
1254                         }
1255                     }else{
1256                         if(IS_8x8DCT(mb_type)){
1257                             h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1258                         }else{
1259                             h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1260                         }
1261                     }
1262                 }
1263             }else{
1264                 for(i=0; i<16; i++){
1265                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1266                         uint8_t * const ptr= dest_y + block_offset[i];
1267                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1268                     }
1269                 }
1270             }
1271         }
1272
1273         if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1274             uint8_t *dest[2] = {dest_cb, dest_cr};
1275             if(transform_bypass){
1276                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1277                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
1278                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
1279                 }else{
1280                     idct_add = s->dsp.add_pixels4;
1281                     for(i=16; i<16+8; i++){
1282                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1283                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1284                     }
1285                 }
1286             }else{
1287                 if(is_h264){
1288                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+0] ])
1289                         chroma_dc_dequant_idct_c(h->mb + 16*16     , h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
1290                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+1] ])
1291                         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]);
1292                     h->h264dsp.h264_idct_add8(dest, block_offset,
1293                                               h->mb, uvlinesize,
1294                                               h->non_zero_count_cache);
1295                 }else{
1296                     chroma_dc_dequant_idct_c(h->mb + 16*16     , h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
1297                     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]);
1298                     for(i=16; i<16+8; i++){
1299                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1300                             uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1301                             ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
1302                         }
1303                     }
1304                 }
1305             }
1306         }
1307     }
1308     if(h->cbp || IS_INTRA(mb_type))
1309         s->dsp.clear_blocks(h->mb);
1310 }
1311
1312 /**
1313  * Process a macroblock; this case avoids checks for expensive uncommon cases.
1314  */
1315 static void hl_decode_mb_simple(H264Context *h){
1316     hl_decode_mb_internal(h, 1);
1317 }
1318
1319 /**
1320  * Process a macroblock; this handles edge cases, such as interlacing.
1321  */
1322 static void av_noinline hl_decode_mb_complex(H264Context *h){
1323     hl_decode_mb_internal(h, 0);
1324 }
1325
1326 void ff_h264_hl_decode_mb(H264Context *h){
1327     MpegEncContext * const s = &h->s;
1328     const int mb_xy= h->mb_xy;
1329     const int mb_type= s->current_picture.mb_type[mb_xy];
1330     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1331
1332     if (is_complex)
1333         hl_decode_mb_complex(h);
1334     else hl_decode_mb_simple(h);
1335 }
1336
1337 static int pred_weight_table(H264Context *h){
1338     MpegEncContext * const s = &h->s;
1339     int list, i;
1340     int luma_def, chroma_def;
1341
1342     h->use_weight= 0;
1343     h->use_weight_chroma= 0;
1344     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1345     if(CHROMA)
1346         h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1347     luma_def = 1<<h->luma_log2_weight_denom;
1348     chroma_def = 1<<h->chroma_log2_weight_denom;
1349
1350     for(list=0; list<2; list++){
1351         h->luma_weight_flag[list]   = 0;
1352         h->chroma_weight_flag[list] = 0;
1353         for(i=0; i<h->ref_count[list]; i++){
1354             int luma_weight_flag, chroma_weight_flag;
1355
1356             luma_weight_flag= get_bits1(&s->gb);
1357             if(luma_weight_flag){
1358                 h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
1359                 h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
1360                 if(   h->luma_weight[i][list][0] != luma_def
1361                    || h->luma_weight[i][list][1] != 0) {
1362                     h->use_weight= 1;
1363                     h->luma_weight_flag[list]= 1;
1364                 }
1365             }else{
1366                 h->luma_weight[i][list][0]= luma_def;
1367                 h->luma_weight[i][list][1]= 0;
1368             }
1369
1370             if(CHROMA){
1371                 chroma_weight_flag= get_bits1(&s->gb);
1372                 if(chroma_weight_flag){
1373                     int j;
1374                     for(j=0; j<2; j++){
1375                         h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
1376                         h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
1377                         if(   h->chroma_weight[i][list][j][0] != chroma_def
1378                            || h->chroma_weight[i][list][j][1] != 0) {
1379                             h->use_weight_chroma= 1;
1380                             h->chroma_weight_flag[list]= 1;
1381                         }
1382                     }
1383                 }else{
1384                     int j;
1385                     for(j=0; j<2; j++){
1386                         h->chroma_weight[i][list][j][0]= chroma_def;
1387                         h->chroma_weight[i][list][j][1]= 0;
1388                     }
1389                 }
1390             }
1391         }
1392         if(h->slice_type_nos != AV_PICTURE_TYPE_B) break;
1393     }
1394     h->use_weight= h->use_weight || h->use_weight_chroma;
1395     return 0;
1396 }
1397
1398 /**
1399  * Initialize implicit_weight table.
1400  * @param field  0/1 initialize the weight for interlaced MBAFF
1401  *                -1 initializes the rest
1402  */
1403 static void implicit_weight_table(H264Context *h, int field){
1404     MpegEncContext * const s = &h->s;
1405     int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
1406
1407     for (i = 0; i < 2; i++) {
1408         h->luma_weight_flag[i]   = 0;
1409         h->chroma_weight_flag[i] = 0;
1410     }
1411
1412     if(field < 0){
1413         cur_poc = s->current_picture_ptr->poc;
1414     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
1415        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1416         h->use_weight= 0;
1417         h->use_weight_chroma= 0;
1418         return;
1419     }
1420         ref_start= 0;
1421         ref_count0= h->ref_count[0];
1422         ref_count1= h->ref_count[1];
1423     }else{
1424         cur_poc = s->current_picture_ptr->field_poc[field];
1425         ref_start= 16;
1426         ref_count0= 16+2*h->ref_count[0];
1427         ref_count1= 16+2*h->ref_count[1];
1428     }
1429
1430     h->use_weight= 2;
1431     h->use_weight_chroma= 2;
1432     h->luma_log2_weight_denom= 5;
1433     h->chroma_log2_weight_denom= 5;
1434
1435     for(ref0=ref_start; ref0 < ref_count0; ref0++){
1436         int poc0 = h->ref_list[0][ref0].poc;
1437         for(ref1=ref_start; ref1 < ref_count1; ref1++){
1438             int poc1 = h->ref_list[1][ref1].poc;
1439             int td = av_clip(poc1 - poc0, -128, 127);
1440             int w= 32;
1441             if(td){
1442                 int tb = av_clip(cur_poc - poc0, -128, 127);
1443                 int tx = (16384 + (FFABS(td) >> 1)) / td;
1444                 int dist_scale_factor = (tb*tx + 32) >> 8;
1445                 if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
1446                     w = 64 - dist_scale_factor;
1447             }
1448             if(field<0){
1449                 h->implicit_weight[ref0][ref1][0]=
1450                 h->implicit_weight[ref0][ref1][1]= w;
1451             }else{
1452                 h->implicit_weight[ref0][ref1][field]=w;
1453             }
1454         }
1455     }
1456 }
1457
1458 /**
1459  * instantaneous decoder refresh.
1460  */
1461 static void idr(H264Context *h){
1462     ff_h264_remove_all_refs(h);
1463     h->prev_frame_num= 0;
1464     h->prev_frame_num_offset= 0;
1465     h->prev_poc_msb=
1466     h->prev_poc_lsb= 0;
1467 }
1468
1469 /* forget old pics after a seek */
1470 static void flush_dpb(AVCodecContext *avctx){
1471     H264Context *h= avctx->priv_data;
1472     int i;
1473     for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1474         if(h->delayed_pic[i])
1475             h->delayed_pic[i]->reference= 0;
1476         h->delayed_pic[i]= NULL;
1477     }
1478     h->outputed_poc= INT_MIN;
1479     h->prev_interlaced_frame = 1;
1480     idr(h);
1481     if(h->s.current_picture_ptr)
1482         h->s.current_picture_ptr->reference= 0;
1483     h->s.first_field= 0;
1484     ff_h264_reset_sei(h);
1485     ff_mpeg_flush(avctx);
1486 }
1487
1488 static int init_poc(H264Context *h){
1489     MpegEncContext * const s = &h->s;
1490     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1491     int field_poc[2];
1492     Picture *cur = s->current_picture_ptr;
1493
1494     h->frame_num_offset= h->prev_frame_num_offset;
1495     if(h->frame_num < h->prev_frame_num)
1496         h->frame_num_offset += max_frame_num;
1497
1498     if(h->sps.poc_type==0){
1499         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1500
1501         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1502             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1503         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1504             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1505         else
1506             h->poc_msb = h->prev_poc_msb;
1507 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1508         field_poc[0] =
1509         field_poc[1] = h->poc_msb + h->poc_lsb;
1510         if(s->picture_structure == PICT_FRAME)
1511             field_poc[1] += h->delta_poc_bottom;
1512     }else if(h->sps.poc_type==1){
1513         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1514         int i;
1515
1516         if(h->sps.poc_cycle_length != 0)
1517             abs_frame_num = h->frame_num_offset + h->frame_num;
1518         else
1519             abs_frame_num = 0;
1520
1521         if(h->nal_ref_idc==0 && abs_frame_num > 0)
1522             abs_frame_num--;
1523
1524         expected_delta_per_poc_cycle = 0;
1525         for(i=0; i < h->sps.poc_cycle_length; i++)
1526             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1527
1528         if(abs_frame_num > 0){
1529             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1530             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1531
1532             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1533             for(i = 0; i <= frame_num_in_poc_cycle; i++)
1534                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1535         } else
1536             expectedpoc = 0;
1537
1538         if(h->nal_ref_idc == 0)
1539             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1540
1541         field_poc[0] = expectedpoc + h->delta_poc[0];
1542         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1543
1544         if(s->picture_structure == PICT_FRAME)
1545             field_poc[1] += h->delta_poc[1];
1546     }else{
1547         int poc= 2*(h->frame_num_offset + h->frame_num);
1548
1549         if(!h->nal_ref_idc)
1550             poc--;
1551
1552         field_poc[0]= poc;
1553         field_poc[1]= poc;
1554     }
1555
1556     if(s->picture_structure != PICT_BOTTOM_FIELD)
1557         s->current_picture_ptr->field_poc[0]= field_poc[0];
1558     if(s->picture_structure != PICT_TOP_FIELD)
1559         s->current_picture_ptr->field_poc[1]= field_poc[1];
1560     cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1561
1562     return 0;
1563 }
1564
1565
1566 /**
1567  * initialize scan tables
1568  */
1569 static void init_scan_tables(H264Context *h){
1570     int i;
1571     for(i=0; i<16; i++){
1572 #define T(x) (x>>2) | ((x<<2) & 0xF)
1573         h->zigzag_scan[i] = T(zigzag_scan[i]);
1574         h-> field_scan[i] = T( field_scan[i]);
1575 #undef T
1576     }
1577     for(i=0; i<64; i++){
1578 #define T(x) (x>>3) | ((x&7)<<3)
1579         h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
1580         h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1581         h->field_scan8x8[i]        = T(field_scan8x8[i]);
1582         h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
1583 #undef T
1584     }
1585     if(h->sps.transform_bypass){ //FIXME same ugly
1586         h->zigzag_scan_q0          = zigzag_scan;
1587         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
1588         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1589         h->field_scan_q0           = field_scan;
1590         h->field_scan8x8_q0        = field_scan8x8;
1591         h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
1592     }else{
1593         h->zigzag_scan_q0          = h->zigzag_scan;
1594         h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
1595         h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1596         h->field_scan_q0           = h->field_scan;
1597         h->field_scan8x8_q0        = h->field_scan8x8;
1598         h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
1599     }
1600 }
1601
1602 static void field_end(H264Context *h){
1603     MpegEncContext * const s = &h->s;
1604     AVCodecContext * const avctx= s->avctx;
1605     s->mb_y= 0;
1606
1607     s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1608     s->current_picture_ptr->pict_type= s->pict_type;
1609
1610     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1611         ff_vdpau_h264_set_reference_frames(s);
1612
1613     if(!s->dropable) {
1614         ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1615         h->prev_poc_msb= h->poc_msb;
1616         h->prev_poc_lsb= h->poc_lsb;
1617     }
1618     h->prev_frame_num_offset= h->frame_num_offset;
1619     h->prev_frame_num= h->frame_num;
1620
1621     if (avctx->hwaccel) {
1622         if (avctx->hwaccel->end_frame(avctx) < 0)
1623             av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1624     }
1625
1626     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1627         ff_vdpau_h264_picture_complete(s);
1628
1629     /*
1630      * FIXME: Error handling code does not seem to support interlaced
1631      * when slices span multiple rows
1632      * The ff_er_add_slice calls don't work right for bottom
1633      * fields; they cause massive erroneous error concealing
1634      * Error marking covers both fields (top and bottom).
1635      * This causes a mismatched s->error_count
1636      * and a bad error table. Further, the error count goes to
1637      * INT_MAX when called for bottom field, because mb_y is
1638      * past end by one (callers fault) and resync_mb_y != 0
1639      * causes problems for the first MB line, too.
1640      */
1641     if (!FIELD_PICTURE)
1642         ff_er_frame_end(s);
1643
1644     MPV_frame_end(s);
1645
1646     h->current_slice=0;
1647 }
1648
1649 /**
1650  * Replicate H264 "master" context to thread contexts.
1651  */
1652 static void clone_slice(H264Context *dst, H264Context *src)
1653 {
1654     memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
1655     dst->s.current_picture_ptr  = src->s.current_picture_ptr;
1656     dst->s.current_picture      = src->s.current_picture;
1657     dst->s.linesize             = src->s.linesize;
1658     dst->s.uvlinesize           = src->s.uvlinesize;
1659     dst->s.first_field          = src->s.first_field;
1660
1661     dst->prev_poc_msb           = src->prev_poc_msb;
1662     dst->prev_poc_lsb           = src->prev_poc_lsb;
1663     dst->prev_frame_num_offset  = src->prev_frame_num_offset;
1664     dst->prev_frame_num         = src->prev_frame_num;
1665     dst->short_ref_count        = src->short_ref_count;
1666
1667     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
1668     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
1669     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1670     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
1671
1672     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
1673     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
1674 }
1675
1676 /**
1677  * computes profile from profile_idc and constraint_set?_flags
1678  *
1679  * @param sps SPS
1680  *
1681  * @return profile as defined by FF_PROFILE_H264_*
1682  */
1683 int ff_h264_get_profile(SPS *sps)
1684 {
1685     int profile = sps->profile_idc;
1686
1687     switch(sps->profile_idc) {
1688     case FF_PROFILE_H264_BASELINE:
1689         // constraint_set1_flag set to 1
1690         profile |= (sps->constraint_set_flags & 1<<1) ? FF_PROFILE_H264_CONSTRAINED : 0;
1691         break;
1692     case FF_PROFILE_H264_HIGH_10:
1693     case FF_PROFILE_H264_HIGH_422:
1694     case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
1695         // constraint_set3_flag set to 1
1696         profile |= (sps->constraint_set_flags & 1<<3) ? FF_PROFILE_H264_INTRA : 0;
1697         break;
1698     }
1699
1700     return profile;
1701 }
1702
1703 /**
1704  * decodes a slice header.
1705  * This will also call MPV_common_init() and frame_start() as needed.
1706  *
1707  * @param h h264context
1708  * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1709  *
1710  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1711  */
1712 static int decode_slice_header(H264Context *h, H264Context *h0){
1713     MpegEncContext * const s = &h->s;
1714     MpegEncContext * const s0 = &h0->s;
1715     unsigned int first_mb_in_slice;
1716     unsigned int pps_id;
1717     int num_ref_idx_active_override_flag;
1718     unsigned int slice_type, tmp, i, j;
1719     int default_ref_list_done = 0;
1720     int last_pic_structure;
1721
1722     s->dropable= h->nal_ref_idc == 0;
1723
1724     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1725         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1726         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1727     }else{
1728         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1729         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1730     }
1731
1732     first_mb_in_slice= get_ue_golomb(&s->gb);
1733
1734     if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1735         if(h0->current_slice && FIELD_PICTURE){
1736             field_end(h);
1737         }
1738
1739         h0->current_slice = 0;
1740         if (!s0->first_field)
1741             s->current_picture_ptr= NULL;
1742     }
1743
1744     slice_type= get_ue_golomb_31(&s->gb);
1745     if(slice_type > 9){
1746         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);
1747         return -1;
1748     }
1749     if(slice_type > 4){
1750         slice_type -= 5;
1751         h->slice_type_fixed=1;
1752     }else
1753         h->slice_type_fixed=0;
1754
1755     slice_type= golomb_to_pict_type[ slice_type ];
1756     if (slice_type == AV_PICTURE_TYPE_I
1757         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
1758         default_ref_list_done = 1;
1759     }
1760     h->slice_type= slice_type;
1761     h->slice_type_nos= slice_type & 3;
1762
1763     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
1764
1765     pps_id= get_ue_golomb(&s->gb);
1766     if(pps_id>=MAX_PPS_COUNT){
1767         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
1768         return -1;
1769     }
1770     if(!h0->pps_buffers[pps_id]) {
1771         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
1772         return -1;
1773     }
1774     h->pps= *h0->pps_buffers[pps_id];
1775
1776     if(!h0->sps_buffers[h->pps.sps_id]) {
1777         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
1778         return -1;
1779     }
1780     h->sps = *h0->sps_buffers[h->pps.sps_id];
1781
1782     s->avctx->profile = ff_h264_get_profile(&h->sps);
1783     s->avctx->level   = h->sps.level_idc;
1784     s->avctx->refs    = h->sps.ref_frame_count;
1785
1786     if(h == h0 && h->dequant_coeff_pps != pps_id){
1787         h->dequant_coeff_pps = pps_id;
1788         init_dequant_tables(h);
1789     }
1790
1791     s->mb_width= h->sps.mb_width;
1792     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1793
1794     h->b_stride=  s->mb_width*4;
1795
1796     s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
1797     if(h->sps.frame_mbs_only_flag)
1798         s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
1799     else
1800         s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 7);
1801
1802     if (s->context_initialized
1803         && (   s->width != s->avctx->width || s->height != s->avctx->height
1804             || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
1805         if(h != h0)
1806             return -1;   // width / height changed during parallelized decoding
1807         free_tables(h, 0);
1808         flush_dpb(s->avctx);
1809         MPV_common_end(s);
1810     }
1811     if (!s->context_initialized) {
1812         if(h != h0)
1813             return -1;  // we cant (re-)initialize context during parallel decoding
1814
1815         avcodec_set_dimensions(s->avctx, s->width, s->height);
1816         s->avctx->sample_aspect_ratio= h->sps.sar;
1817         av_assert0(s->avctx->sample_aspect_ratio.den);
1818
1819         if(h->sps.video_signal_type_present_flag){
1820             s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1821             if(h->sps.colour_description_present_flag){
1822                 s->avctx->color_primaries = h->sps.color_primaries;
1823                 s->avctx->color_trc       = h->sps.color_trc;
1824                 s->avctx->colorspace      = h->sps.colorspace;
1825             }
1826         }
1827
1828         if(h->sps.timing_info_present_flag){
1829             int64_t den= h->sps.time_scale;
1830             if(h->x264_build < 44U)
1831                 den *= 2;
1832             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
1833                       h->sps.num_units_in_tick, den, 1<<30);
1834         }
1835         s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
1836                                                  s->avctx->codec->pix_fmts ?
1837                                                  s->avctx->codec->pix_fmts :
1838                                                  s->avctx->color_range == AVCOL_RANGE_JPEG ?
1839                                                  hwaccel_pixfmt_list_h264_jpeg_420 :
1840                                                  ff_hwaccel_pixfmt_list_420);
1841         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1842
1843         if (MPV_common_init(s) < 0)
1844             return -1;
1845         s->first_field = 0;
1846         h->prev_interlaced_frame = 1;
1847
1848         init_scan_tables(h);
1849         ff_h264_alloc_tables(h);
1850
1851         for(i = 1; i < s->avctx->thread_count; i++) {
1852             H264Context *c;
1853             c = h->thread_context[i] = av_malloc(sizeof(H264Context));
1854             memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
1855             memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
1856             c->h264dsp = h->h264dsp;
1857             c->sps = h->sps;
1858             c->pps = h->pps;
1859             init_scan_tables(c);
1860             clone_tables(c, h, i);
1861         }
1862
1863         for(i = 0; i < s->avctx->thread_count; i++)
1864             if(context_init(h->thread_context[i]) < 0)
1865                 return -1;
1866     }
1867
1868     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1869
1870     h->mb_mbaff = 0;
1871     h->mb_aff_frame = 0;
1872     last_pic_structure = s0->picture_structure;
1873     if(h->sps.frame_mbs_only_flag){
1874         s->picture_structure= PICT_FRAME;
1875     }else{
1876         if(get_bits1(&s->gb)) { //field_pic_flag
1877             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
1878         } else {
1879             s->picture_structure= PICT_FRAME;
1880             h->mb_aff_frame = h->sps.mb_aff;
1881         }
1882     }
1883     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
1884
1885     if(h0->current_slice == 0){
1886         while(h->frame_num !=  h->prev_frame_num &&
1887               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
1888             Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1889             av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
1890             if (ff_h264_frame_start(h) < 0)
1891                 return -1;
1892             h->prev_frame_num++;
1893             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
1894             s->current_picture_ptr->frame_num= h->prev_frame_num;
1895             ff_generate_sliding_window_mmcos(h);
1896             ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1897             /* Error concealment: if a ref is missing, copy the previous ref in its place.
1898              * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
1899              * about there being no actual duplicates.
1900              * FIXME: this doesn't copy padding for out-of-frame motion vectors.  Given we're
1901              * concealing a lost frame, this probably isn't noticable by comparison, but it should
1902              * be fixed. */
1903             if (h->short_ref_count) {
1904                 if (prev) {
1905                     av_image_copy(h->short_ref[0]->data, h->short_ref[0]->linesize,
1906                                   (const uint8_t**)prev->data, prev->linesize,
1907                                   s->avctx->pix_fmt, s->mb_width*16, s->mb_height*16);
1908                     h->short_ref[0]->poc = prev->poc+2;
1909                 }
1910                 h->short_ref[0]->frame_num = h->prev_frame_num;
1911             }
1912         }
1913
1914         /* See if we have a decoded first field looking for a pair... */
1915         if (s0->first_field) {
1916             assert(s0->current_picture_ptr);
1917             assert(s0->current_picture_ptr->data[0]);
1918             assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
1919
1920             /* figure out if we have a complementary field pair */
1921             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
1922                 /*
1923                  * Previous field is unmatched. Don't display it, but let it
1924                  * remain for reference if marked as such.
1925                  */
1926                 s0->current_picture_ptr = NULL;
1927                 s0->first_field = FIELD_PICTURE;
1928
1929             } else {
1930                 if (h->nal_ref_idc &&
1931                         s0->current_picture_ptr->reference &&
1932                         s0->current_picture_ptr->frame_num != h->frame_num) {
1933                     /*
1934                      * This and previous field were reference, but had
1935                      * different frame_nums. Consider this field first in
1936                      * pair. Throw away previous field except for reference
1937                      * purposes.
1938                      */
1939                     s0->first_field = 1;
1940                     s0->current_picture_ptr = NULL;
1941
1942                 } else {
1943                     /* Second field in complementary pair */
1944                     s0->first_field = 0;
1945                 }
1946             }
1947
1948         } else {
1949             /* Frame or first field in a potentially complementary pair */
1950             assert(!s0->current_picture_ptr);
1951             s0->first_field = FIELD_PICTURE;
1952         }
1953
1954         if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
1955             s0->first_field = 0;
1956             return -1;
1957         }
1958     }
1959     if(h != h0)
1960         clone_slice(h, h0);
1961
1962     s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
1963
1964     assert(s->mb_num == s->mb_width * s->mb_height);
1965     if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
1966        first_mb_in_slice                    >= s->mb_num){
1967         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1968         return -1;
1969     }
1970     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
1971     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
1972     if (s->picture_structure == PICT_BOTTOM_FIELD)
1973         s->resync_mb_y = s->mb_y = s->mb_y + 1;
1974     assert(s->mb_y < s->mb_height);
1975
1976     if(s->picture_structure==PICT_FRAME){
1977         h->curr_pic_num=   h->frame_num;
1978         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
1979     }else{
1980         h->curr_pic_num= 2*h->frame_num + 1;
1981         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
1982     }
1983
1984     if(h->nal_unit_type == NAL_IDR_SLICE){
1985         get_ue_golomb(&s->gb); /* idr_pic_id */
1986     }
1987
1988     if(h->sps.poc_type==0){
1989         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
1990
1991         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
1992             h->delta_poc_bottom= get_se_golomb(&s->gb);
1993         }
1994     }
1995
1996     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
1997         h->delta_poc[0]= get_se_golomb(&s->gb);
1998
1999         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2000             h->delta_poc[1]= get_se_golomb(&s->gb);
2001     }
2002
2003     init_poc(h);
2004
2005     if(h->pps.redundant_pic_cnt_present){
2006         h->redundant_pic_count= get_ue_golomb(&s->gb);
2007     }
2008
2009     //set defaults, might be overridden a few lines later
2010     h->ref_count[0]= h->pps.ref_count[0];
2011     h->ref_count[1]= h->pps.ref_count[1];
2012
2013     if(h->slice_type_nos != AV_PICTURE_TYPE_I){
2014         if(h->slice_type_nos == AV_PICTURE_TYPE_B){
2015             h->direct_spatial_mv_pred= get_bits1(&s->gb);
2016         }
2017         num_ref_idx_active_override_flag= get_bits1(&s->gb);
2018
2019         if(num_ref_idx_active_override_flag){
2020             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2021             if(h->slice_type_nos==AV_PICTURE_TYPE_B)
2022                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2023
2024             if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
2025                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
2026                 h->ref_count[0]= h->ref_count[1]= 1;
2027                 return -1;
2028             }
2029         }
2030         if(h->slice_type_nos == AV_PICTURE_TYPE_B)
2031             h->list_count= 2;
2032         else
2033             h->list_count= 1;
2034     }else
2035         h->list_count= 0;
2036
2037     if(!default_ref_list_done){
2038         ff_h264_fill_default_ref_list(h);
2039     }
2040
2041     if(h->slice_type_nos!=AV_PICTURE_TYPE_I && ff_h264_decode_ref_pic_list_reordering(h) < 0)
2042         return -1;
2043
2044     if(h->slice_type_nos!=AV_PICTURE_TYPE_I){
2045         s->last_picture_ptr= &h->ref_list[0][0];
2046         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
2047     }
2048     if(h->slice_type_nos==AV_PICTURE_TYPE_B){
2049         s->next_picture_ptr= &h->ref_list[1][0];
2050         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
2051     }
2052
2053     if(   (h->pps.weighted_pred          && h->slice_type_nos == AV_PICTURE_TYPE_P )
2054        ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== AV_PICTURE_TYPE_B ) )
2055         pred_weight_table(h);
2056     else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
2057         implicit_weight_table(h, -1);
2058     }else {
2059         h->use_weight = 0;
2060         for (i = 0; i < 2; i++) {
2061             h->luma_weight_flag[i]   = 0;
2062             h->chroma_weight_flag[i] = 0;
2063         }
2064     }
2065
2066     if(h->nal_ref_idc)
2067         ff_h264_decode_ref_pic_marking(h0, &s->gb);
2068
2069     if(FRAME_MBAFF){
2070         ff_h264_fill_mbaff_ref_list(h);
2071
2072         if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
2073             implicit_weight_table(h, 0);
2074             implicit_weight_table(h, 1);
2075         }
2076     }
2077
2078     if(h->slice_type_nos==AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
2079         ff_h264_direct_dist_scale_factor(h);
2080     ff_h264_direct_ref_list_init(h);
2081
2082     if( h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac ){
2083         tmp = get_ue_golomb_31(&s->gb);
2084         if(tmp > 2){
2085             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2086             return -1;
2087         }
2088         h->cabac_init_idc= tmp;
2089     }
2090
2091     h->last_qscale_diff = 0;
2092     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2093     if(tmp>51){
2094         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2095         return -1;
2096     }
2097     s->qscale= tmp;
2098     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2099     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2100     //FIXME qscale / qp ... stuff
2101     if(h->slice_type == AV_PICTURE_TYPE_SP){
2102         get_bits1(&s->gb); /* sp_for_switch_flag */
2103     }
2104     if(h->slice_type==AV_PICTURE_TYPE_SP || h->slice_type == AV_PICTURE_TYPE_SI){
2105         get_se_golomb(&s->gb); /* slice_qs_delta */
2106     }
2107
2108     h->deblocking_filter = 1;
2109     h->slice_alpha_c0_offset = 52;
2110     h->slice_beta_offset = 52;
2111     if( h->pps.deblocking_filter_parameters_present ) {
2112         tmp= get_ue_golomb_31(&s->gb);
2113         if(tmp > 2){
2114             av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2115             return -1;
2116         }
2117         h->deblocking_filter= tmp;
2118         if(h->deblocking_filter < 2)
2119             h->deblocking_filter^= 1; // 1<->0
2120
2121         if( h->deblocking_filter ) {
2122             h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2123             h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
2124             if(   h->slice_alpha_c0_offset > 104U
2125                || h->slice_beta_offset     > 104U){
2126                 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);
2127                 return -1;
2128             }
2129         }
2130     }
2131
2132     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
2133        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != AV_PICTURE_TYPE_I)
2134        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == AV_PICTURE_TYPE_B)
2135        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2136         h->deblocking_filter= 0;
2137
2138     if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2139         if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2140             /* Cheat slightly for speed:
2141                Do not bother to deblock across slices. */
2142             h->deblocking_filter = 2;
2143         } else {
2144             h0->max_contexts = 1;
2145             if(!h0->single_decode_warning) {
2146                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2147                 h0->single_decode_warning = 1;
2148             }
2149             if(h != h0)
2150                 return 1; // deblocking switched inside frame
2151         }
2152     }
2153     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]);
2154
2155 #if 0 //FMO
2156     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2157         slice_group_change_cycle= get_bits(&s->gb, ?);
2158 #endif
2159
2160     h0->last_slice_type = slice_type;
2161     h->slice_num = ++h0->current_slice;
2162     if(h->slice_num >= MAX_SLICES){
2163         av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2164     }
2165
2166     for(j=0; j<2; j++){
2167         int id_list[16];
2168         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2169         for(i=0; i<16; i++){
2170             id_list[i]= 60;
2171             if(h->ref_list[j][i].data[0]){
2172                 int k;
2173                 uint8_t *base= h->ref_list[j][i].base[0];
2174                 for(k=0; k<h->short_ref_count; k++)
2175                     if(h->short_ref[k]->base[0] == base){
2176                         id_list[i]= k;
2177                         break;
2178                     }
2179                 for(k=0; k<h->long_ref_count; k++)
2180                     if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2181                         id_list[i]= h->short_ref_count + k;
2182                         break;
2183                     }
2184             }
2185         }
2186
2187         ref2frm[0]=
2188         ref2frm[1]= -1;
2189         for(i=0; i<16; i++)
2190             ref2frm[i+2]= 4*id_list[i]
2191                           +(h->ref_list[j][i].reference&3);
2192         ref2frm[18+0]=
2193         ref2frm[18+1]= -1;
2194         for(i=16; i<48; i++)
2195             ref2frm[i+4]= 4*id_list[(i-16)>>1]
2196                           +(h->ref_list[j][i].reference&3);
2197     }
2198
2199     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
2200     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2201
2202     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2203         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",
2204                h->slice_num,
2205                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2206                first_mb_in_slice,
2207                av_get_picture_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2208                pps_id, h->frame_num,
2209                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2210                h->ref_count[0], h->ref_count[1],
2211                s->qscale,
2212                h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
2213                h->use_weight,
2214                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2215                h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2216                );
2217     }
2218
2219     return 0;
2220 }
2221
2222 int ff_h264_get_slice_type(const H264Context *h)
2223 {
2224     switch (h->slice_type) {
2225     case AV_PICTURE_TYPE_P:  return 0;
2226     case AV_PICTURE_TYPE_B:  return 1;
2227     case AV_PICTURE_TYPE_I:  return 2;
2228     case AV_PICTURE_TYPE_SP: return 3;
2229     case AV_PICTURE_TYPE_SI: return 4;
2230     default:         return -1;
2231     }
2232 }
2233
2234 /**
2235  *
2236  * @return non zero if the loop filter can be skiped
2237  */
2238 static int fill_filter_caches(H264Context *h, int mb_type){
2239     MpegEncContext * const s = &h->s;
2240     const int mb_xy= h->mb_xy;
2241     int top_xy, left_xy[2];
2242     int top_type, left_type[2];
2243
2244     top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
2245
2246     //FIXME deblocking could skip the intra and nnz parts.
2247
2248     /* Wow, what a mess, why didn't they simplify the interlacing & intra
2249      * stuff, I can't imagine that these complex rules are worth it. */
2250
2251     left_xy[1] = left_xy[0] = mb_xy-1;
2252     if(FRAME_MBAFF){
2253         const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
2254         const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
2255         if(s->mb_y&1){
2256             if (left_mb_field_flag != curr_mb_field_flag) {
2257                 left_xy[0] -= s->mb_stride;
2258             }
2259         }else{
2260             if(curr_mb_field_flag){
2261                 top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
2262             }
2263             if (left_mb_field_flag != curr_mb_field_flag) {
2264                 left_xy[1] += s->mb_stride;
2265             }
2266         }
2267     }
2268
2269     h->top_mb_xy = top_xy;
2270     h->left_mb_xy[0] = left_xy[0];
2271     h->left_mb_xy[1] = left_xy[1];
2272     {
2273         //for sufficiently low qp, filtering wouldn't do anything
2274         //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
2275         int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
2276         int qp = s->current_picture.qscale_table[mb_xy];
2277         if(qp <= qp_thresh
2278            && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
2279            && (top_xy   < 0 || ((qp + s->current_picture.qscale_table[top_xy    ] + 1)>>1) <= qp_thresh)){
2280             if(!FRAME_MBAFF)
2281                 return 1;
2282             if(   (left_xy[0]< 0            || ((qp + s->current_picture.qscale_table[left_xy[1]             ] + 1)>>1) <= qp_thresh)
2283                && (top_xy    < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy    -s->mb_stride] + 1)>>1) <= qp_thresh))
2284                 return 1;
2285         }
2286     }
2287
2288     top_type     = s->current_picture.mb_type[top_xy]    ;
2289     left_type[0] = s->current_picture.mb_type[left_xy[0]];
2290     left_type[1] = s->current_picture.mb_type[left_xy[1]];
2291     if(h->deblocking_filter == 2){
2292         if(h->slice_table[top_xy     ] != h->slice_num) top_type= 0;
2293         if(h->slice_table[left_xy[0] ] != h->slice_num) left_type[0]= left_type[1]= 0;
2294     }else{
2295         if(h->slice_table[top_xy     ] == 0xFFFF) top_type= 0;
2296         if(h->slice_table[left_xy[0] ] == 0xFFFF) left_type[0]= left_type[1] =0;
2297     }
2298     h->top_type    = top_type    ;
2299     h->left_type[0]= left_type[0];
2300     h->left_type[1]= left_type[1];
2301
2302     if(IS_INTRA(mb_type))
2303         return 0;
2304
2305     AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
2306     AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
2307     AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]);
2308     AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]);
2309     AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
2310
2311     h->cbp= h->cbp_table[mb_xy];
2312
2313     {
2314         int list;
2315         for(list=0; list<h->list_count; list++){
2316             int8_t *ref;
2317             int y, b_stride;
2318             int16_t (*mv_dst)[2];
2319             int16_t (*mv_src)[2];
2320
2321             if(!USES_LIST(mb_type, list)){
2322                 fill_rectangle(  h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
2323                 AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2324                 AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2325                 AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2326                 AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2327                 continue;
2328             }
2329
2330             ref = &s->current_picture.ref_index[list][4*mb_xy];
2331             {
2332                 int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2333                 AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2334                 AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2335                 ref += 2;
2336                 AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2337                 AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2338             }
2339
2340             b_stride = h->b_stride;
2341             mv_dst   = &h->mv_cache[list][scan8[0]];
2342             mv_src   = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
2343             for(y=0; y<4; y++){
2344                 AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
2345             }
2346
2347         }
2348     }
2349
2350
2351 /*
2352 0 . T T. T T T T
2353 1 L . .L . . . .
2354 2 L . .L . . . .
2355 3 . T TL . . . .
2356 4 L . .L . . . .
2357 5 L . .. . . . .
2358 */
2359 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
2360     if(top_type){
2361         AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
2362     }
2363
2364     if(left_type[0]){
2365         h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
2366         h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
2367         h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
2368         h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
2369     }
2370
2371     // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
2372     if(!CABAC && h->pps.transform_8x8_mode){
2373         if(IS_8x8DCT(top_type)){
2374             h->non_zero_count_cache[4+8*0]=
2375             h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
2376             h->non_zero_count_cache[6+8*0]=
2377             h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
2378         }
2379         if(IS_8x8DCT(left_type[0])){
2380             h->non_zero_count_cache[3+8*1]=
2381             h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
2382         }
2383         if(IS_8x8DCT(left_type[1])){
2384             h->non_zero_count_cache[3+8*3]=
2385             h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
2386         }
2387
2388         if(IS_8x8DCT(mb_type)){
2389             h->non_zero_count_cache[scan8[0   ]]= h->non_zero_count_cache[scan8[1   ]]=
2390             h->non_zero_count_cache[scan8[2   ]]= h->non_zero_count_cache[scan8[3   ]]= h->cbp & 1;
2391
2392             h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
2393             h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
2394
2395             h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
2396             h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
2397
2398             h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
2399             h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
2400         }
2401     }
2402
2403     if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
2404         int list;
2405         for(list=0; list<h->list_count; list++){
2406             if(USES_LIST(top_type, list)){
2407                 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
2408                 const int b8_xy= 4*top_xy + 2;
2409                 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2410                 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
2411                 h->ref_cache[list][scan8[0] + 0 - 1*8]=
2412                 h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
2413                 h->ref_cache[list][scan8[0] + 2 - 1*8]=
2414                 h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
2415             }else{
2416                 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
2417                 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2418             }
2419
2420             if(!IS_INTERLACED(mb_type^left_type[0])){
2421                 if(USES_LIST(left_type[0], list)){
2422                     const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
2423                     const int b8_xy= 4*left_xy[0] + 1;
2424                     int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2425                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
2426                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
2427                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
2428                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
2429                     h->ref_cache[list][scan8[0] - 1 + 0 ]=
2430                     h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]];
2431                     h->ref_cache[list][scan8[0] - 1 +16 ]=
2432                     h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]];
2433                 }else{
2434                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
2435                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
2436                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
2437                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
2438                     h->ref_cache[list][scan8[0] - 1 + 0  ]=
2439                     h->ref_cache[list][scan8[0] - 1 + 8  ]=
2440                     h->ref_cache[list][scan8[0] - 1 + 16 ]=
2441                     h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
2442                 }
2443             }
2444         }
2445     }
2446
2447     return 0;
2448 }
2449
2450 static void loop_filter(H264Context *h){
2451     MpegEncContext * const s = &h->s;
2452     uint8_t  *dest_y, *dest_cb, *dest_cr;
2453     int linesize, uvlinesize, mb_x, mb_y;
2454     const int end_mb_y= s->mb_y + FRAME_MBAFF;
2455     const int old_slice_type= h->slice_type;
2456
2457     if(h->deblocking_filter) {
2458         for(mb_x= 0; mb_x<s->mb_width; mb_x++){
2459             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
2460                 int mb_xy, mb_type;
2461                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2462                 h->slice_num= h->slice_table[mb_xy];
2463                 mb_type= s->current_picture.mb_type[mb_xy];
2464                 h->list_count= h->list_counts[mb_xy];
2465
2466                 if(FRAME_MBAFF)
2467                     h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2468
2469                 s->mb_x= mb_x;
2470                 s->mb_y= mb_y;
2471                 dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2472                 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2473                 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2474                     //FIXME simplify above
2475
2476                 if (MB_FIELD) {
2477                     linesize   = h->mb_linesize   = s->linesize * 2;
2478                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2479                     if(mb_y&1){ //FIXME move out of this function?
2480                         dest_y -= s->linesize*15;
2481                         dest_cb-= s->uvlinesize*7;
2482                         dest_cr-= s->uvlinesize*7;
2483                     }
2484                 } else {
2485                     linesize   = h->mb_linesize   = s->linesize;
2486                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2487                 }
2488                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2489                 if(fill_filter_caches(h, mb_type))
2490                     continue;
2491                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2492                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2493
2494                 if (FRAME_MBAFF) {
2495                     ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2496                 } else {
2497                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2498                 }
2499             }
2500         }
2501     }
2502     h->slice_type= old_slice_type;
2503     s->mb_x= 0;
2504     s->mb_y= end_mb_y - FRAME_MBAFF;
2505     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2506     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2507 }
2508
2509 static void predict_field_decoding_flag(H264Context *h){
2510     MpegEncContext * const s = &h->s;
2511     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2512     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
2513                 ? s->current_picture.mb_type[mb_xy-1]
2514                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
2515                 ? s->current_picture.mb_type[mb_xy-s->mb_stride]
2516                 : 0;
2517     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2518 }
2519
2520 static int decode_slice(struct AVCodecContext *avctx, void *arg){
2521     H264Context *h = *(void**)arg;
2522     MpegEncContext * const s = &h->s;
2523     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2524
2525     s->mb_skip_run= -1;
2526
2527     h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
2528                     (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
2529
2530     if( h->pps.cabac ) {
2531         /* realign */
2532         align_get_bits( &s->gb );
2533
2534         /* init cabac */
2535         ff_init_cabac_states( &h->cabac);
2536         ff_init_cabac_decoder( &h->cabac,
2537                                s->gb.buffer + get_bits_count(&s->gb)/8,
2538                                (get_bits_left(&s->gb) + 7)/8);
2539
2540         ff_h264_init_cabac_states(h);
2541
2542         for(;;){
2543 //START_TIMER
2544             int ret = ff_h264_decode_mb_cabac(h);
2545             int eos;
2546 //STOP_TIMER("decode_mb_cabac")
2547
2548             if(ret>=0) ff_h264_hl_decode_mb(h);
2549
2550             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
2551                 s->mb_y++;
2552
2553                 ret = ff_h264_decode_mb_cabac(h);
2554
2555                 if(ret>=0) ff_h264_hl_decode_mb(h);
2556                 s->mb_y--;
2557             }
2558             eos = get_cabac_terminate( &h->cabac );
2559
2560             if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
2561                 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);
2562                 return 0;
2563             }
2564             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2565                 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);
2566                 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);
2567                 return -1;
2568             }
2569
2570             if( ++s->mb_x >= s->mb_width ) {
2571                 s->mb_x = 0;
2572                 loop_filter(h);
2573                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2574                 ++s->mb_y;
2575                 if(FIELD_OR_MBAFF_PICTURE) {
2576                     ++s->mb_y;
2577                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2578                         predict_field_decoding_flag(h);
2579                 }
2580             }
2581
2582             if( eos || s->mb_y >= s->mb_height ) {
2583                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2584                 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);
2585                 return 0;
2586             }
2587         }
2588
2589     } else {
2590         for(;;){
2591             int ret = ff_h264_decode_mb_cavlc(h);
2592
2593             if(ret>=0) ff_h264_hl_decode_mb(h);
2594
2595             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
2596                 s->mb_y++;
2597                 ret = ff_h264_decode_mb_cavlc(h);
2598
2599                 if(ret>=0) ff_h264_hl_decode_mb(h);
2600                 s->mb_y--;
2601             }
2602
2603             if(ret<0){
2604                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2605                 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);
2606
2607                 return -1;
2608             }
2609
2610             if(++s->mb_x >= s->mb_width){
2611                 s->mb_x=0;
2612                 loop_filter(h);
2613                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2614                 ++s->mb_y;
2615                 if(FIELD_OR_MBAFF_PICTURE) {
2616                     ++s->mb_y;
2617                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2618                         predict_field_decoding_flag(h);
2619                 }
2620                 if(s->mb_y >= s->mb_height){
2621                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2622
2623                     if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
2624                         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);
2625
2626                         return 0;
2627                     }else{
2628                         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);
2629
2630                         return -1;
2631                     }
2632                 }
2633             }
2634
2635             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
2636                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2637                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2638                     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);
2639
2640                     return 0;
2641                 }else{
2642                     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);
2643
2644                     return -1;
2645                 }
2646             }
2647         }
2648     }
2649
2650 #if 0
2651     for(;s->mb_y < s->mb_height; s->mb_y++){
2652         for(;s->mb_x < s->mb_width; s->mb_x++){
2653             int ret= decode_mb(h);
2654
2655             ff_h264_hl_decode_mb(h);
2656
2657             if(ret<0){
2658                 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
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             if(++s->mb_x >= s->mb_width){
2665                 s->mb_x=0;
2666                 if(++s->mb_y >= s->mb_height){
2667                     if(get_bits_count(s->gb) == s->gb.size_in_bits){
2668                         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);
2669
2670                         return 0;
2671                     }else{
2672                         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);
2673
2674                         return -1;
2675                     }
2676                 }
2677             }
2678
2679             if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2680                 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2681                     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);
2682
2683                     return 0;
2684                 }else{
2685                     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);
2686
2687                     return -1;
2688                 }
2689             }
2690         }
2691         s->mb_x=0;
2692         ff_draw_horiz_band(s, 16*s->mb_y, 16);
2693     }
2694 #endif
2695     return -1; //not reached
2696 }
2697
2698 /**
2699  * Call decode_slice() for each context.
2700  *
2701  * @param h h264 master context
2702  * @param context_count number of contexts to execute
2703  */
2704 static void execute_decode_slices(H264Context *h, int context_count){
2705     MpegEncContext * const s = &h->s;
2706     AVCodecContext * const avctx= s->avctx;
2707     H264Context *hx;
2708     int i;
2709
2710     if (s->avctx->hwaccel)
2711         return;
2712     if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2713         return;
2714     if(context_count == 1) {
2715         decode_slice(avctx, &h);
2716     } else {
2717         for(i = 1; i < context_count; i++) {
2718             hx = h->thread_context[i];
2719             hx->s.error_recognition = avctx->error_recognition;
2720             hx->s.error_count = 0;
2721         }
2722
2723         avctx->execute(avctx, (void *)decode_slice,
2724                        h->thread_context, NULL, context_count, sizeof(void*));
2725
2726         /* pull back stuff from slices to master context */
2727         hx = h->thread_context[context_count - 1];
2728         s->mb_x = hx->s.mb_x;
2729         s->mb_y = hx->s.mb_y;
2730         s->dropable = hx->s.dropable;
2731         s->picture_structure = hx->s.picture_structure;
2732         for(i = 1; i < context_count; i++)
2733             h->s.error_count += h->thread_context[i]->s.error_count;
2734     }
2735 }
2736
2737
2738 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
2739     MpegEncContext * const s = &h->s;
2740     AVCodecContext * const avctx= s->avctx;
2741     int buf_index=0;
2742     H264Context *hx; ///< thread context
2743     int context_count = 0;
2744     int next_avc= h->is_avc ? 0 : buf_size;
2745
2746     h->max_contexts = avctx->thread_count;
2747     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
2748         h->current_slice = 0;
2749         if (!s->first_field)
2750             s->current_picture_ptr= NULL;
2751         ff_h264_reset_sei(h);
2752     }
2753
2754     for(;;){
2755         int consumed;
2756         int dst_length;
2757         int bit_length;
2758         const uint8_t *ptr;
2759         int i, nalsize = 0;
2760         int err;
2761
2762         if(buf_index >= next_avc) {
2763             if(buf_index >= buf_size) break;
2764             nalsize = 0;
2765             for(i = 0; i < h->nal_length_size; i++)
2766                 nalsize = (nalsize << 8) | buf[buf_index++];
2767             if(nalsize <= 0 || nalsize > buf_size - buf_index){
2768                 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2769                 break;
2770             }
2771             next_avc= buf_index + nalsize;
2772         } else {
2773             // start code prefix search
2774             for(; buf_index + 3 < next_avc; buf_index++){
2775                 // This should always succeed in the first iteration.
2776                 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2777                     break;
2778             }
2779
2780             if(buf_index+3 >= buf_size) break;
2781
2782             buf_index+=3;
2783             if(buf_index >= next_avc) continue;
2784         }
2785
2786         hx = h->thread_context[context_count];
2787
2788         ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
2789         if (ptr==NULL || dst_length < 0){
2790             return -1;
2791         }
2792         i= buf_index + consumed;
2793         if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
2794            buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
2795             s->workaround_bugs |= FF_BUG_TRUNCATED;
2796
2797         if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
2798         while(ptr[dst_length - 1] == 0 && dst_length > 0)
2799             dst_length--;
2800         }
2801         bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
2802
2803         if(s->avctx->debug&FF_DEBUG_STARTCODE){
2804             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);
2805         }
2806
2807         if (h->is_avc && (nalsize != consumed) && nalsize){
2808             av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
2809         }
2810
2811         buf_index += consumed;
2812
2813         //FIXME do not discard SEI id
2814         if(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0)
2815             continue;
2816
2817       again:
2818         err = 0;
2819         switch(hx->nal_unit_type){
2820         case NAL_IDR_SLICE:
2821             if (h->nal_unit_type != NAL_IDR_SLICE) {
2822                 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2823                 return -1;
2824             }
2825             idr(h); //FIXME ensure we don't loose some frames if there is reordering
2826         case NAL_SLICE:
2827             init_get_bits(&hx->s.gb, ptr, bit_length);
2828             hx->intra_gb_ptr=
2829             hx->inter_gb_ptr= &hx->s.gb;
2830             hx->s.data_partitioning = 0;
2831
2832             if((err = decode_slice_header(hx, h)))
2833                break;
2834
2835             if (h->current_slice == 1) {
2836                 if (s->avctx->hwaccel && s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
2837                     return -1;
2838                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2839                     ff_vdpau_h264_picture_start(s);
2840             }
2841
2842             s->current_picture_ptr->key_frame |=
2843                     (hx->nal_unit_type == NAL_IDR_SLICE) ||
2844                     (h->sei_recovery_frame_cnt >= 0);
2845             if(hx->redundant_pic_count==0
2846                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2847                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
2848                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
2849                && avctx->skip_frame < AVDISCARD_ALL){
2850                 if(avctx->hwaccel) {
2851                     if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
2852                         return -1;
2853                 }else
2854                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
2855                     static const uint8_t start_code[] = {0x00, 0x00, 0x01};
2856                     ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
2857                     ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
2858                 }else
2859                     context_count++;
2860             }
2861             break;
2862         case NAL_DPA:
2863             init_get_bits(&hx->s.gb, ptr, bit_length);
2864             hx->intra_gb_ptr=
2865             hx->inter_gb_ptr= NULL;
2866
2867             if ((err = decode_slice_header(hx, h)) < 0)
2868                 break;
2869
2870             hx->s.data_partitioning = 1;
2871
2872             break;
2873         case NAL_DPB:
2874             init_get_bits(&hx->intra_gb, ptr, bit_length);
2875             hx->intra_gb_ptr= &hx->intra_gb;
2876             break;
2877         case NAL_DPC:
2878             init_get_bits(&hx->inter_gb, ptr, bit_length);
2879             hx->inter_gb_ptr= &hx->inter_gb;
2880
2881             if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
2882                && s->context_initialized
2883                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2884                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
2885                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
2886                && avctx->skip_frame < AVDISCARD_ALL)
2887                 context_count++;
2888             break;
2889         case NAL_SEI:
2890             init_get_bits(&s->gb, ptr, bit_length);
2891             ff_h264_decode_sei(h);
2892             break;
2893         case NAL_SPS:
2894             init_get_bits(&s->gb, ptr, bit_length);
2895             ff_h264_decode_seq_parameter_set(h);
2896
2897             if(s->flags& CODEC_FLAG_LOW_DELAY)
2898                 s->low_delay=1;
2899
2900             if(avctx->has_b_frames < 2)
2901                 avctx->has_b_frames= !s->low_delay;
2902             break;
2903         case NAL_PPS:
2904             init_get_bits(&s->gb, ptr, bit_length);
2905
2906             ff_h264_decode_picture_parameter_set(h, bit_length);
2907
2908             break;
2909         case NAL_AUD:
2910         case NAL_END_SEQUENCE:
2911         case NAL_END_STREAM:
2912         case NAL_FILLER_DATA:
2913         case NAL_SPS_EXT:
2914         case NAL_AUXILIARY_SLICE:
2915             break;
2916         default:
2917             av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
2918         }
2919
2920         if(context_count == h->max_contexts) {
2921             execute_decode_slices(h, context_count);
2922             context_count = 0;
2923         }
2924
2925         if (err < 0)
2926             av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
2927         else if(err == 1) {
2928             /* Slice could not be decoded in parallel mode, copy down
2929              * NAL unit stuff to context 0 and restart. Note that
2930              * rbsp_buffer is not transferred, but since we no longer
2931              * run in parallel mode this should not be an issue. */
2932             h->nal_unit_type = hx->nal_unit_type;
2933             h->nal_ref_idc   = hx->nal_ref_idc;
2934             hx = h;
2935             goto again;
2936         }
2937     }
2938     if(context_count)
2939         execute_decode_slices(h, context_count);
2940     return buf_index;
2941 }
2942
2943 /**
2944  * returns the number of bytes consumed for building the current frame
2945  */
2946 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
2947         if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
2948         if(pos+10>buf_size) pos=buf_size; // oops ;)
2949
2950         return pos;
2951 }
2952
2953 static int decode_frame(AVCodecContext *avctx,
2954                              void *data, int *data_size,
2955                              AVPacket *avpkt)
2956 {
2957     const uint8_t *buf = avpkt->data;
2958     int buf_size = avpkt->size;
2959     H264Context *h = avctx->priv_data;
2960     MpegEncContext *s = &h->s;
2961     AVFrame *pict = data;
2962     int buf_index;
2963
2964     s->flags= avctx->flags;
2965     s->flags2= avctx->flags2;
2966
2967    /* end of stream, output what is still in the buffers */
2968  out:
2969     if (buf_size == 0) {
2970         Picture *out;
2971         int i, out_idx;
2972
2973 //FIXME factorize this with the output code below
2974         out = h->delayed_pic[0];
2975         out_idx = 0;
2976         for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2977             if(h->delayed_pic[i]->poc < out->poc){
2978                 out = h->delayed_pic[i];
2979                 out_idx = i;
2980             }
2981
2982         for(i=out_idx; h->delayed_pic[i]; i++)
2983             h->delayed_pic[i] = h->delayed_pic[i+1];
2984
2985         if(out){
2986             *data_size = sizeof(AVFrame);
2987             *pict= *(AVFrame*)out;
2988         }
2989
2990         return 0;
2991     }
2992
2993     buf_index=decode_nal_units(h, buf, buf_size);
2994     if(buf_index < 0)
2995         return -1;
2996
2997     if (!s->current_picture_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
2998         buf_size = 0;
2999         goto out;
3000     }
3001
3002     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
3003         if (avctx->skip_frame >= AVDISCARD_NONREF)
3004             return 0;
3005         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
3006         return -1;
3007     }
3008
3009     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
3010         Picture *out = s->current_picture_ptr;
3011         Picture *cur = s->current_picture_ptr;
3012         int i, pics, out_of_order, out_idx;
3013
3014         field_end(h);
3015
3016         if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
3017             /* Wait for second field. */
3018             *data_size = 0;
3019
3020         } else {
3021             cur->interlaced_frame = 0;
3022             cur->repeat_pict = 0;
3023
3024             /* Signal interlacing information externally. */
3025             /* Prioritize picture timing SEI information over used decoding process if it exists. */
3026
3027             if(h->sps.pic_struct_present_flag){
3028                 switch (h->sei_pic_struct)
3029                 {
3030                 case SEI_PIC_STRUCT_FRAME:
3031                     break;
3032                 case SEI_PIC_STRUCT_TOP_FIELD:
3033                 case SEI_PIC_STRUCT_BOTTOM_FIELD:
3034                     cur->interlaced_frame = 1;
3035                     break;
3036                 case SEI_PIC_STRUCT_TOP_BOTTOM:
3037                 case SEI_PIC_STRUCT_BOTTOM_TOP:
3038                     if (FIELD_OR_MBAFF_PICTURE)
3039                         cur->interlaced_frame = 1;
3040                     else
3041                         // try to flag soft telecine progressive
3042                         cur->interlaced_frame = h->prev_interlaced_frame;
3043                     break;
3044                 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
3045                 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
3046                     // Signal the possibility of telecined film externally (pic_struct 5,6)
3047                     // From these hints, let the applications decide if they apply deinterlacing.
3048                     cur->repeat_pict = 1;
3049                     break;
3050                 case SEI_PIC_STRUCT_FRAME_DOUBLING:
3051                     // Force progressive here, as doubling interlaced frame is a bad idea.
3052                     cur->repeat_pict = 2;
3053                     break;
3054                 case SEI_PIC_STRUCT_FRAME_TRIPLING:
3055                     cur->repeat_pict = 4;
3056                     break;
3057                 }
3058
3059                 if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
3060                     cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
3061             }else{
3062                 /* Derive interlacing flag from used decoding process. */
3063                 cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
3064             }
3065             h->prev_interlaced_frame = cur->interlaced_frame;
3066
3067             if (cur->field_poc[0] != cur->field_poc[1]){
3068                 /* Derive top_field_first from field pocs. */
3069                 cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
3070             }else{
3071                 if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
3072                     /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
3073                     if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
3074                       || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
3075                         cur->top_field_first = 1;
3076                     else
3077                         cur->top_field_first = 0;
3078                 }else{
3079                     /* Most likely progressive */
3080                     cur->top_field_first = 0;
3081                 }
3082             }
3083
3084         //FIXME do something with unavailable reference frames
3085
3086             /* Sort B-frames into display order */
3087
3088             if(h->sps.bitstream_restriction_flag
3089                && s->avctx->has_b_frames < h->sps.num_reorder_frames){
3090                 s->avctx->has_b_frames = h->sps.num_reorder_frames;
3091                 s->low_delay = 0;
3092             }
3093
3094             if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
3095                && !h->sps.bitstream_restriction_flag){
3096                 s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
3097                 s->low_delay= 0;
3098             }
3099
3100             pics = 0;
3101             while(h->delayed_pic[pics]) pics++;
3102
3103             assert(pics <= MAX_DELAYED_PIC_COUNT);
3104
3105             h->delayed_pic[pics++] = cur;
3106             if(cur->reference == 0)
3107                 cur->reference = DELAYED_PIC_REF;
3108
3109             out = h->delayed_pic[0];
3110             out_idx = 0;
3111             for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
3112                 if(h->delayed_pic[i]->poc < out->poc){
3113                     out = h->delayed_pic[i];
3114                     out_idx = i;
3115                 }
3116             if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
3117                 h->outputed_poc= INT_MIN;
3118             out_of_order = out->poc < h->outputed_poc;
3119
3120             if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
3121                 { }
3122             else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
3123                || (s->low_delay &&
3124                 ((h->outputed_poc != INT_MIN && out->poc > h->outputed_poc + 2)
3125                  || cur->pict_type == AV_PICTURE_TYPE_B)))
3126             {
3127                 s->low_delay = 0;
3128                 s->avctx->has_b_frames++;
3129             }
3130
3131             if(out_of_order || pics > s->avctx->has_b_frames){
3132                 out->reference &= ~DELAYED_PIC_REF;
3133                 for(i=out_idx; h->delayed_pic[i]; i++)
3134                     h->delayed_pic[i] = h->delayed_pic[i+1];
3135             }
3136             if(!out_of_order && pics > s->avctx->has_b_frames){
3137                 *data_size = sizeof(AVFrame);
3138
3139                 if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
3140                     h->outputed_poc = INT_MIN;
3141                 } else
3142                     h->outputed_poc = out->poc;
3143                 *pict= *(AVFrame*)out;
3144             }else{
3145                 av_log(avctx, AV_LOG_DEBUG, "no picture\n");
3146             }
3147         }
3148     }
3149
3150     assert(pict->data[0] || !*data_size);
3151     ff_print_debug_info(s, pict);
3152 //printf("out %d\n", (int)pict->data[0]);
3153
3154     return get_consumed_bytes(s, buf_index, buf_size);
3155 }
3156 #if 0
3157 static inline void fill_mb_avail(H264Context *h){
3158     MpegEncContext * const s = &h->s;
3159     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3160
3161     if(s->mb_y){
3162         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
3163         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
3164         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
3165     }else{
3166         h->mb_avail[0]=
3167         h->mb_avail[1]=
3168         h->mb_avail[2]= 0;
3169     }
3170     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
3171     h->mb_avail[4]= 1; //FIXME move out
3172     h->mb_avail[5]= 0; //FIXME move out
3173 }
3174 #endif
3175
3176 #ifdef TEST
3177 #undef printf
3178 #undef random
3179 #define COUNT 8000
3180 #define SIZE (COUNT*40)
3181 int main(void){
3182     int i;
3183     uint8_t temp[SIZE];
3184     PutBitContext pb;
3185     GetBitContext gb;
3186 //    int int_temp[10000];
3187     DSPContext dsp;
3188     AVCodecContext avctx;
3189
3190     dsputil_init(&dsp, &avctx);
3191
3192     init_put_bits(&pb, temp, SIZE);
3193     printf("testing unsigned exp golomb\n");
3194     for(i=0; i<COUNT; i++){
3195         START_TIMER
3196         set_ue_golomb(&pb, i);
3197         STOP_TIMER("set_ue_golomb");
3198     }
3199     flush_put_bits(&pb);
3200
3201     init_get_bits(&gb, temp, 8*SIZE);
3202     for(i=0; i<COUNT; i++){
3203         int j, s;
3204
3205         s= show_bits(&gb, 24);
3206
3207         START_TIMER
3208         j= get_ue_golomb(&gb);
3209         if(j != i){
3210             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
3211 //            return -1;
3212         }
3213         STOP_TIMER("get_ue_golomb");
3214     }
3215
3216
3217     init_put_bits(&pb, temp, SIZE);
3218     printf("testing signed exp golomb\n");
3219     for(i=0; i<COUNT; i++){
3220         START_TIMER
3221         set_se_golomb(&pb, i - COUNT/2);
3222         STOP_TIMER("set_se_golomb");
3223     }
3224     flush_put_bits(&pb);
3225
3226     init_get_bits(&gb, temp, 8*SIZE);
3227     for(i=0; i<COUNT; i++){
3228         int j, s;
3229
3230         s= show_bits(&gb, 24);
3231
3232         START_TIMER
3233         j= get_se_golomb(&gb);
3234         if(j != i - COUNT/2){
3235             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
3236 //            return -1;
3237         }
3238         STOP_TIMER("get_se_golomb");
3239     }
3240
3241 #if 0
3242     printf("testing 4x4 (I)DCT\n");
3243
3244     DCTELEM block[16];
3245     uint8_t src[16], ref[16];
3246     uint64_t error= 0, max_error=0;
3247
3248     for(i=0; i<COUNT; i++){
3249         int j;
3250 //        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
3251         for(j=0; j<16; j++){
3252             ref[j]= random()%255;
3253             src[j]= random()%255;
3254         }
3255
3256         h264_diff_dct_c(block, src, ref, 4);
3257
3258         //normalize
3259         for(j=0; j<16; j++){
3260 //            printf("%d ", block[j]);
3261             block[j]= block[j]*4;
3262             if(j&1) block[j]= (block[j]*4 + 2)/5;
3263             if(j&4) block[j]= (block[j]*4 + 2)/5;
3264         }
3265 //        printf("\n");
3266
3267         h->h264dsp.h264_idct_add(ref, block, 4);
3268 /*        for(j=0; j<16; j++){
3269             printf("%d ", ref[j]);
3270         }
3271         printf("\n");*/
3272
3273         for(j=0; j<16; j++){
3274             int diff= FFABS(src[j] - ref[j]);
3275
3276             error+= diff*diff;
3277             max_error= FFMAX(max_error, diff);
3278         }
3279     }
3280     printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
3281     printf("testing quantizer\n");
3282     for(qp=0; qp<52; qp++){
3283         for(i=0; i<16; i++)
3284             src1_block[i]= src2_block[i]= random()%255;
3285
3286     }
3287     printf("Testing NAL layer\n");
3288
3289     uint8_t bitstream[COUNT];
3290     uint8_t nal[COUNT*2];
3291     H264Context h;
3292     memset(&h, 0, sizeof(H264Context));
3293
3294     for(i=0; i<COUNT; i++){
3295         int zeros= i;
3296         int nal_length;
3297         int consumed;
3298         int out_length;
3299         uint8_t *out;
3300         int j;
3301
3302         for(j=0; j<COUNT; j++){
3303             bitstream[j]= (random() % 255) + 1;
3304         }
3305
3306         for(j=0; j<zeros; j++){
3307             int pos= random() % COUNT;
3308             while(bitstream[pos] == 0){
3309                 pos++;
3310                 pos %= COUNT;
3311             }
3312             bitstream[pos]=0;
3313         }
3314
3315         START_TIMER
3316
3317         nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
3318         if(nal_length<0){
3319             printf("encoding failed\n");
3320             return -1;
3321         }
3322
3323         out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
3324
3325         STOP_TIMER("NAL")
3326
3327         if(out_length != COUNT){
3328             printf("incorrect length %d %d\n", out_length, COUNT);
3329             return -1;
3330         }
3331
3332         if(consumed != nal_length){
3333             printf("incorrect consumed length %d %d\n", nal_length, consumed);
3334             return -1;
3335         }
3336
3337         if(memcmp(bitstream, out, COUNT)){
3338             printf("mismatch\n");
3339             return -1;
3340         }
3341     }
3342 #endif
3343
3344     printf("Testing RBSP\n");
3345
3346
3347     return 0;
3348 }
3349 #endif /* TEST */
3350
3351
3352 av_cold void ff_h264_free_context(H264Context *h)
3353 {
3354     int i;
3355
3356     free_tables(h, 1); //FIXME cleanup init stuff perhaps
3357
3358     for(i = 0; i < MAX_SPS_COUNT; i++)
3359         av_freep(h->sps_buffers + i);
3360
3361     for(i = 0; i < MAX_PPS_COUNT; i++)
3362         av_freep(h->pps_buffers + i);
3363 }
3364
3365 av_cold int ff_h264_decode_end(AVCodecContext *avctx)
3366 {
3367     H264Context *h = avctx->priv_data;
3368     MpegEncContext *s = &h->s;
3369
3370     ff_h264_free_context(h);
3371
3372     MPV_common_end(s);
3373
3374 //    memset(h, 0, sizeof(H264Context));
3375
3376     return 0;
3377 }
3378
3379 static const AVProfile profiles[] = {
3380     { FF_PROFILE_H264_BASELINE,             "Baseline"              },
3381     { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline"  },
3382     { FF_PROFILE_H264_MAIN,                 "Main"                  },
3383     { FF_PROFILE_H264_EXTENDED,             "Extended"              },
3384     { FF_PROFILE_H264_HIGH,                 "High"                  },
3385     { FF_PROFILE_H264_HIGH_10,              "High 10"               },
3386     { FF_PROFILE_H264_HIGH_10_INTRA,        "High 10 Intra"         },
3387     { FF_PROFILE_H264_HIGH_422,             "High 4:2:2"            },
3388     { FF_PROFILE_H264_HIGH_422_INTRA,       "High 4:2:2 Intra"      },
3389     { FF_PROFILE_H264_HIGH_444,             "High 4:4:4"            },
3390     { FF_PROFILE_H264_HIGH_444_PREDICTIVE,  "High 4:4:4 Predictive" },
3391     { FF_PROFILE_H264_HIGH_444_INTRA,       "High 4:4:4 Intra"      },
3392     { FF_PROFILE_H264_CAVLC_444,            "CAVLC 4:4:4"           },
3393     { FF_PROFILE_UNKNOWN },
3394 };
3395
3396 AVCodec ff_h264_decoder = {
3397     "h264",
3398     AVMEDIA_TYPE_VIDEO,
3399     CODEC_ID_H264,
3400     sizeof(H264Context),
3401     ff_h264_decode_init,
3402     NULL,
3403     ff_h264_decode_end,
3404     decode_frame,
3405     /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY |
3406         CODEC_CAP_SLICE_THREADS,
3407     .flush= flush_dpb,
3408     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
3409     .profiles = NULL_IF_CONFIG_SMALL(profiles),
3410 };
3411
3412 #if CONFIG_H264_VDPAU_DECODER
3413 AVCodec ff_h264_vdpau_decoder = {
3414     "h264_vdpau",
3415     AVMEDIA_TYPE_VIDEO,
3416     CODEC_ID_H264,
3417     sizeof(H264Context),
3418     ff_h264_decode_init,
3419     NULL,
3420     ff_h264_decode_end,
3421     decode_frame,
3422     CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3423     .flush= flush_dpb,
3424     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
3425     .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
3426     .profiles = NULL_IF_CONFIG_SMALL(profiles),
3427 };
3428 #endif