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

   1 /*
   2  * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
   3  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
   4  *
   5  * This file is part of FFmpeg.
   6  *
   7  * FFmpeg is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2.1 of the License, or (at your option) any later version.
  11  *
  12  * FFmpeg is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with FFmpeg; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 /**
  23  * @file libavcodec/h264.c
  24  * H.264 / AVC / MPEG4 part10 codec.
  25  * @author Michael Niedermayer <michaelni@gmx.at>
  26  */
  27
  28 #include "internal.h"
  29 #include "dsputil.h"
  30 #include "avcodec.h"
  31 #include "mpegvideo.h"
  32 #include "h264.h"
  33 #include "h264data.h"
  34 #include "h264_mvpred.h"
  35 #include "h264_parser.h"
  36 #include "golomb.h"
  37 #include "mathops.h"
  38 #include "rectangle.h"
  39 #include "vdpau_internal.h"
  40
  41 #include "cabac.h"
  42
  43 //#undef NDEBUG
  44 #include <assert.h>
  45
  46 static const uint8_t rem6[52]={
  47 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
  48 };
  49
  50 static const uint8_t div6[52]={
  51 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8,
  52 };
  53
  54 void ff_h264_write_back_intra_pred_mode(H264Context *h){
  55     int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
  56
  57     AV_COPY32(mode, h->intra4x4_pred_mode_cache + 4 + 8*4);
  58     mode[4]= h->intra4x4_pred_mode_cache[7+8*3];
  59     mode[5]= h->intra4x4_pred_mode_cache[7+8*2];
  60     mode[6]= h->intra4x4_pred_mode_cache[7+8*1];
  61 }
  62
  63 /**
  64  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
  65  */
  66 int ff_h264_check_intra4x4_pred_mode(H264Context *h){
  67     MpegEncContext * const s = &h->s;
  68     static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
  69     static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
  70     int i;
  71
  72     if(!(h->top_samples_available&0x8000)){
  73         for(i=0; i<4; i++){
  74             int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
  75             if(status<0){
  76                 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
  77                 return -1;
  78             } else if(status){
  79                 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
  80             }
  81         }
  82     }
  83
  84     if((h->left_samples_available&0x8888)!=0x8888){
  85         static const int mask[4]={0x8000,0x2000,0x80,0x20};
  86         for(i=0; i<4; i++){
  87             if(!(h->left_samples_available&mask[i])){
  88                 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
  89                 if(status<0){
  90                     av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
  91                     return -1;
  92                 } else if(status){
  93                     h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
  94                 }
  95             }
  96         }
  97     }
  98
  99     return 0;
 100 } //FIXME cleanup like ff_h264_check_intra_pred_mode
 101
 102 /**
 103  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
 104  */
 105 int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
 106     MpegEncContext * const s = &h->s;
 107     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
 108     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
 109
 110     if(mode > 6U) {
 111         av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
 112         return -1;
 113     }
 114
 115     if(!(h->top_samples_available&0x8000)){
 116         mode= top[ mode ];
 117         if(mode<0){
 118             av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
 119             return -1;
 120         }
 121     }
 122
 123     if((h->left_samples_available&0x8080) != 0x8080){
 124         mode= left[ mode ];
 125         if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
 126             mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
 127         }
 128         if(mode<0){
 129             av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
 130             return -1;
 131         }
 132     }
 133
 134     return mode;
 135 }
 136
 137 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
 138     int i, si, di;
 139     uint8_t *dst;
 140     int bufidx;
 141
 142 //    src[0]&0x80;                //forbidden bit
 143     h->nal_ref_idc= src[0]>>5;
 144     h->nal_unit_type= src[0]&0x1F;
 145
 146     src++; length--;
 147 #if 0
 148     for(i=0; i<length; i++)
 149         printf("%2X ", src[i]);
 150 #endif
 151
 152 #if HAVE_FAST_UNALIGNED
 153 # if HAVE_FAST_64BIT
 154 #   define RS 7
 155     for(i=0; i+1<length; i+=9){
 156         if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
 157 # else
 158 #   define RS 3
 159     for(i=0; i+1<length; i+=5){
 160         if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
 161 # endif
 162             continue;
 163         if(i>0 && !src[i]) i--;
 164         while(src[i]) i++;
 165 #else
 166 #   define RS 0
 167     for(i=0; i+1<length; i+=2){
 168         if(src[i]) continue;
 169         if(i>0 && src[i-1]==0) i--;
 170 #endif
 171         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
 172             if(src[i+2]!=3){
 173                 /* startcode, so we must be past the end */
 174                 length=i;
 175             }
 176             break;
 177         }
 178         i-= RS;
 179     }
 180
 181     if(i>=length-1){ //no escaped 0
 182         *dst_length= length;
 183         *consumed= length+1; //+1 for the header
 184         return src;
 185     }
 186
 187     bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
 188     av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
 189     dst= h->rbsp_buffer[bufidx];
 190
 191     if (dst == NULL){
 192         return NULL;
 193     }
 194
 195 //printf("decoding esc\n");
 196     memcpy(dst, src, i);
 197     si=di=i;
 198     while(si+2<length){
 199         //remove escapes (very rare 1:2^22)
 200         if(src[si+2]>3){
 201             dst[di++]= src[si++];
 202             dst[di++]= src[si++];
 203         }else if(src[si]==0 && src[si+1]==0){
 204             if(src[si+2]==3){ //escape
 205                 dst[di++]= 0;
 206                 dst[di++]= 0;
 207                 si+=3;
 208                 continue;
 209             }else //next start code
 210                 goto nsc;
 211         }
 212
 213         dst[di++]= src[si++];
 214     }
 215     while(si<length)
 216         dst[di++]= src[si++];
 217 nsc:
 218
 219     memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
 220
 221     *dst_length= di;
 222     *consumed= si + 1;//+1 for the header
 223 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
 224     return dst;
 225 }
 226
 227 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
 228     int v= *src;
 229     int r;
 230
 231     tprintf(h->s.avctx, "rbsp trailing %X\n", v);
 232
 233     for(r=1; r<9; r++){
 234         if(v&1) return r;
 235         v>>=1;
 236     }
 237     return 0;
 238 }
 239
 240 /**
 241  * IDCT transforms the 16 dc values and dequantizes them.
 242  * @param qp quantization parameter
 243  */
 244 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
 245 #define stride 16
 246     int i;
 247     int temp[16]; //FIXME check if this is a good idea
 248     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
 249     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
 250
 251 //memset(block, 64, 2*256);
 252 //return;
 253     for(i=0; i<4; i++){
 254         const int offset= y_offset[i];
 255         const int z0= block[offset+stride*0] + block[offset+stride*4];
 256         const int z1= block[offset+stride*0] - block[offset+stride*4];
 257         const int z2= block[offset+stride*1] - block[offset+stride*5];
 258         const int z3= block[offset+stride*1] + block[offset+stride*5];
 259
 260         temp[4*i+0]= z0+z3;
 261         temp[4*i+1]= z1+z2;
 262         temp[4*i+2]= z1-z2;
 263         temp[4*i+3]= z0-z3;
 264     }
 265
 266     for(i=0; i<4; i++){
 267         const int offset= x_offset[i];
 268         const int z0= temp[4*0+i] + temp[4*2+i];
 269         const int z1= temp[4*0+i] - temp[4*2+i];
 270         const int z2= temp[4*1+i] - temp[4*3+i];
 271         const int z3= temp[4*1+i] + temp[4*3+i];
 272
 273         block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
 274         block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
 275         block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
 276         block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
 277     }
 278 }
 279
 280 #if 0
 281 /**
 282  * DCT transforms the 16 dc values.
 283  * @param qp quantization parameter ??? FIXME
 284  */
 285 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
 286 //    const int qmul= dequant_coeff[qp][0];
 287     int i;
 288     int temp[16]; //FIXME check if this is a good idea
 289     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
 290     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
 291
 292     for(i=0; i<4; i++){
 293         const int offset= y_offset[i];
 294         const int z0= block[offset+stride*0] + block[offset+stride*4];
 295         const int z1= block[offset+stride*0] - block[offset+stride*4];
 296         const int z2= block[offset+stride*1] - block[offset+stride*5];
 297         const int z3= block[offset+stride*1] + block[offset+stride*5];
 298
 299         temp[4*i+0]= z0+z3;
 300         temp[4*i+1]= z1+z2;
 301         temp[4*i+2]= z1-z2;
 302         temp[4*i+3]= z0-z3;
 303     }
 304
 305     for(i=0; i<4; i++){
 306         const int offset= x_offset[i];
 307         const int z0= temp[4*0+i] + temp[4*2+i];
 308         const int z1= temp[4*0+i] - temp[4*2+i];
 309         const int z2= temp[4*1+i] - temp[4*3+i];
 310         const int z3= temp[4*1+i] + temp[4*3+i];
 311
 312         block[stride*0 +offset]= (z0 + z3)>>1;
 313         block[stride*2 +offset]= (z1 + z2)>>1;
 314         block[stride*8 +offset]= (z1 - z2)>>1;
 315         block[stride*10+offset]= (z0 - z3)>>1;
 316     }
 317 }
 318 #endif
 319
 320 #undef xStride
 321 #undef stride
 322
 323 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
 324     const int stride= 16*2;
 325     const int xStride= 16;
 326     int a,b,c,d,e;
 327
 328     a= block[stride*0 + xStride*0];
 329     b= block[stride*0 + xStride*1];
 330     c= block[stride*1 + xStride*0];
 331     d= block[stride*1 + xStride*1];
 332
 333     e= a-b;
 334     a= a+b;
 335     b= c-d;
 336     c= c+d;
 337
 338     block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
 339     block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
 340     block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
 341     block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
 342 }
 343
 344 #if 0
 345 static void chroma_dc_dct_c(DCTELEM *block){
 346     const int stride= 16*2;
 347     const int xStride= 16;
 348     int a,b,c,d,e;
 349
 350     a= block[stride*0 + xStride*0];
 351     b= block[stride*0 + xStride*1];
 352     c= block[stride*1 + xStride*0];
 353     d= block[stride*1 + xStride*1];
 354
 355     e= a-b;
 356     a= a+b;
 357     b= c-d;
 358     c= c+d;
 359
 360     block[stride*0 + xStride*0]= (a+c);
 361     block[stride*0 + xStride*1]= (e+b);
 362     block[stride*1 + xStride*0]= (a-c);
 363     block[stride*1 + xStride*1]= (e-b);
 364 }
 365 #endif
 366
 367 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
 368                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 369                            int src_x_offset, int src_y_offset,
 370                            qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
 371     MpegEncContext * const s = &h->s;
 372     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
 373     int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
 374     const int luma_xy= (mx&3) + ((my&3)<<2);
 375     uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
 376     uint8_t * src_cb, * src_cr;
 377     int extra_width= h->emu_edge_width;
 378     int extra_height= h->emu_edge_height;
 379     int emu=0;
 380     const int full_mx= mx>>2;
 381     const int full_my= my>>2;
 382     const int pic_width  = 16*s->mb_width;
 383     const int pic_height = 16*s->mb_height >> MB_FIELD;
 384
 385     if(mx&7) extra_width -= 3;
 386     if(my&7) extra_height -= 3;
 387
 388     if(   full_mx < 0-extra_width
 389        || full_my < 0-extra_height
 390        || full_mx + 16/*FIXME*/ > pic_width + extra_width
 391        || full_my + 16/*FIXME*/ > pic_height + extra_height){
 392         ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
 393             src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
 394         emu=1;
 395     }
 396
 397     qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
 398     if(!square){
 399         qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
 400     }
 401
 402     if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
 403
 404     if(MB_FIELD){
 405         // chroma offset when predicting from a field of opposite parity
 406         my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
 407         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
 408     }
 409     src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
 410     src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
 411
 412     if(emu){
 413         ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
 414             src_cb= s->edge_emu_buffer;
 415     }
 416     chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
 417
 418     if(emu){
 419         ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
 420             src_cr= s->edge_emu_buffer;
 421     }
 422     chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
 423 }
 424
 425 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
 426                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 427                            int x_offset, int y_offset,
 428                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 429                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
 430                            int list0, int list1){
 431     MpegEncContext * const s = &h->s;
 432     qpel_mc_func *qpix_op=  qpix_put;
 433     h264_chroma_mc_func chroma_op= chroma_put;
 434
 435     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
 436     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
 437     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
 438     x_offset += 8*s->mb_x;
 439     y_offset += 8*(s->mb_y >> MB_FIELD);
 440
 441     if(list0){
 442         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
 443         mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
 444                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
 445                            qpix_op, chroma_op);
 446
 447         qpix_op=  qpix_avg;
 448         chroma_op= chroma_avg;
 449     }
 450
 451     if(list1){
 452         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
 453         mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
 454                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
 455                            qpix_op, chroma_op);
 456     }
 457 }
 458
 459 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
 460                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 461                            int x_offset, int y_offset,
 462                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 463                            h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
 464                            h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
 465                            int list0, int list1){
 466     MpegEncContext * const s = &h->s;
 467
 468     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
 469     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
 470     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
 471     x_offset += 8*s->mb_x;
 472     y_offset += 8*(s->mb_y >> MB_FIELD);
 473
 474     if(list0 && list1){
 475         /* don't optimize for luma-only case, since B-frames usually
 476          * use implicit weights => chroma too. */
 477         uint8_t *tmp_cb = s->obmc_scratchpad;
 478         uint8_t *tmp_cr = s->obmc_scratchpad + 8;
 479         uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
 480         int refn0 = h->ref_cache[0][ scan8[n] ];
 481         int refn1 = h->ref_cache[1][ scan8[n] ];
 482
 483         mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
 484                     dest_y, dest_cb, dest_cr,
 485                     x_offset, y_offset, qpix_put, chroma_put);
 486         mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
 487                     tmp_y, tmp_cb, tmp_cr,
 488                     x_offset, y_offset, qpix_put, chroma_put);
 489
 490         if(h->use_weight == 2){
 491             int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
 492             int weight1 = 64 - weight0;
 493             luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
 494             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
 495             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
 496         }else{
 497             luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
 498                             h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
 499                             h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
 500             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 501                             h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
 502                             h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
 503             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 504                             h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
 505                             h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
 506         }
 507     }else{
 508         int list = list1 ? 1 : 0;
 509         int refn = h->ref_cache[list][ scan8[n] ];
 510         Picture *ref= &h->ref_list[list][refn];
 511         mc_dir_part(h, ref, n, square, chroma_height, delta, list,
 512                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
 513                     qpix_put, chroma_put);
 514
 515         luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
 516                        h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
 517         if(h->use_weight_chroma){
 518             chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 519                              h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
 520             chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 521                              h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
 522         }
 523     }
 524 }
 525
 526 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
 527                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 528                            int x_offset, int y_offset,
 529                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 530                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
 531                            h264_weight_func *weight_op, h264_biweight_func *weight_avg,
 532                            int list0, int list1){
 533     if((h->use_weight==2 && list0 && list1
 534         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
 535        || h->use_weight==1)
 536         mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
 537                          x_offset, y_offset, qpix_put, chroma_put,
 538                          weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
 539     else
 540         mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
 541                     x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
 542 }
 543
 544 static inline void prefetch_motion(H264Context *h, int list){
 545     /* fetch pixels for estimated mv 4 macroblocks ahead
 546      * optimized for 64byte cache lines */
 547     MpegEncContext * const s = &h->s;
 548     const int refn = h->ref_cache[list][scan8[0]];
 549     if(refn >= 0){
 550         const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
 551         const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
 552         uint8_t **src= h->ref_list[list][refn].data;
 553         int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
 554         s->dsp.prefetch(src[0]+off, s->linesize, 4);
 555         off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
 556         s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
 557     }
 558 }
 559
 560 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 561                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
 562                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
 563                       h264_weight_func *weight_op, h264_biweight_func *weight_avg){
 564     MpegEncContext * const s = &h->s;
 565     const int mb_xy= h->mb_xy;
 566     const int mb_type= s->current_picture.mb_type[mb_xy];
 567
 568     assert(IS_INTER(mb_type));
 569
 570     prefetch_motion(h, 0);
 571
 572     if(IS_16X16(mb_type)){
 573         mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
 574                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
 575                 weight_op, weight_avg,
 576                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
 577     }else if(IS_16X8(mb_type)){
 578         mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
 579                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
 580                 &weight_op[1], &weight_avg[1],
 581                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
 582         mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
 583                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
 584                 &weight_op[1], &weight_avg[1],
 585                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
 586     }else if(IS_8X16(mb_type)){
 587         mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
 588                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 589                 &weight_op[2], &weight_avg[2],
 590                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
 591         mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
 592                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 593                 &weight_op[2], &weight_avg[2],
 594                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
 595     }else{
 596         int i;
 597
 598         assert(IS_8X8(mb_type));
 599
 600         for(i=0; i<4; i++){
 601             const int sub_mb_type= h->sub_mb_type[i];
 602             const int n= 4*i;
 603             int x_offset= (i&1)<<2;
 604             int y_offset= (i&2)<<1;
 605
 606             if(IS_SUB_8X8(sub_mb_type)){
 607                 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 608                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 609                     &weight_op[3], &weight_avg[3],
 610                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 611             }else if(IS_SUB_8X4(sub_mb_type)){
 612                 mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 613                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
 614                     &weight_op[4], &weight_avg[4],
 615                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 616                 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
 617                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
 618                     &weight_op[4], &weight_avg[4],
 619                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 620             }else if(IS_SUB_4X8(sub_mb_type)){
 621                 mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 622                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 623                     &weight_op[5], &weight_avg[5],
 624                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 625                 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
 626                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 627                     &weight_op[5], &weight_avg[5],
 628                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 629             }else{
 630                 int j;
 631                 assert(IS_SUB_4X4(sub_mb_type));
 632                 for(j=0; j<4; j++){
 633                     int sub_x_offset= x_offset + 2*(j&1);
 634                     int sub_y_offset= y_offset +   (j&2);
 635                     mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
 636                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 637                         &weight_op[6], &weight_avg[6],
 638                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 639                 }
 640             }
 641         }
 642     }
 643
 644     prefetch_motion(h, 1);
 645 }
 646
 647
 648 static void free_tables(H264Context *h){
 649     int i;
 650     H264Context *hx;
 651     av_freep(&h->intra4x4_pred_mode);
 652     av_freep(&h->chroma_pred_mode_table);
 653     av_freep(&h->cbp_table);
 654     av_freep(&h->mvd_table[0]);
 655     av_freep(&h->mvd_table[1]);
 656     av_freep(&h->direct_table);
 657     av_freep(&h->non_zero_count);
 658     av_freep(&h->slice_table_base);
 659     h->slice_table= NULL;
 660     av_freep(&h->list_counts);
 661
 662     av_freep(&h->mb2b_xy);
 663     av_freep(&h->mb2br_xy);
 664
 665     for(i = 0; i < MAX_THREADS; i++) {
 666         hx = h->thread_context[i];
 667         if(!hx) continue;
 668         av_freep(&hx->top_borders[1]);
 669         av_freep(&hx->top_borders[0]);
 670         av_freep(&hx->s.obmc_scratchpad);
 671         av_freep(&hx->rbsp_buffer[1]);
 672         av_freep(&hx->rbsp_buffer[0]);
 673         hx->rbsp_buffer_size[0] = 0;
 674         hx->rbsp_buffer_size[1] = 0;
 675         if (i) av_freep(&h->thread_context[i]);
 676     }
 677 }
 678
 679 static void init_dequant8_coeff_table(H264Context *h){
 680     int i,q,x;
 681     const int transpose = (h->h264dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
 682     h->dequant8_coeff[0] = h->dequant8_buffer[0];
 683     h->dequant8_coeff[1] = h->dequant8_buffer[1];
 684
 685     for(i=0; i<2; i++ ){
 686         if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
 687             h->dequant8_coeff[1] = h->dequant8_buffer[0];
 688             break;
 689         }
 690
 691         for(q=0; q<52; q++){
 692             int shift = div6[q];
 693             int idx = rem6[q];
 694             for(x=0; x<64; x++)
 695                 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
 696                     ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
 697                     h->pps.scaling_matrix8[i][x]) << shift;
 698         }
 699     }
 700 }
 701
 702 static void init_dequant4_coeff_table(H264Context *h){
 703     int i,j,q,x;
 704     const int transpose = (h->h264dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
 705     for(i=0; i<6; i++ ){
 706         h->dequant4_coeff[i] = h->dequant4_buffer[i];
 707         for(j=0; j<i; j++){
 708             if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
 709                 h->dequant4_coeff[i] = h->dequant4_buffer[j];
 710                 break;
 711             }
 712         }
 713         if(j<i)
 714             continue;
 715
 716         for(q=0; q<52; q++){
 717             int shift = div6[q] + 2;
 718             int idx = rem6[q];
 719             for(x=0; x<16; x++)
 720                 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
 721                     ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
 722                     h->pps.scaling_matrix4[i][x]) << shift;
 723         }
 724     }
 725 }
 726
 727 static void init_dequant_tables(H264Context *h){
 728     int i,x;
 729     init_dequant4_coeff_table(h);
 730     if(h->pps.transform_8x8_mode)
 731         init_dequant8_coeff_table(h);
 732     if(h->sps.transform_bypass){
 733         for(i=0; i<6; i++)
 734             for(x=0; x<16; x++)
 735                 h->dequant4_coeff[i][0][x] = 1<<6;
 736         if(h->pps.transform_8x8_mode)
 737             for(i=0; i<2; i++)
 738                 for(x=0; x<64; x++)
 739                     h->dequant8_coeff[i][0][x] = 1<<6;
 740     }
 741 }
 742
 743
 744 int ff_h264_alloc_tables(H264Context *h){
 745     MpegEncContext * const s = &h->s;
 746     const int big_mb_num= s->mb_stride * (s->mb_height+1);
 747     const int row_mb_num= 2*s->mb_stride*s->avctx->thread_count;
 748     int x,y;
 749
 750     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8  * sizeof(uint8_t), fail)
 751
 752     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 32 * sizeof(uint8_t), fail)
 753     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
 754     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
 755
 756     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
 757     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
 758     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
 759     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
 760     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
 761
 762     memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
 763     h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
 764
 765     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy  , big_mb_num * sizeof(uint32_t), fail);
 766     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
 767     for(y=0; y<s->mb_height; y++){
 768         for(x=0; x<s->mb_width; x++){
 769             const int mb_xy= x + y*s->mb_stride;
 770             const int b_xy = 4*x + 4*y*h->b_stride;
 771
 772             h->mb2b_xy [mb_xy]= b_xy;
 773             h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
 774         }
 775     }
 776
 777     s->obmc_scratchpad = NULL;
 778
 779     if(!h->dequant4_coeff[0])
 780         init_dequant_tables(h);
 781
 782     return 0;
 783 fail:
 784     free_tables(h);
 785     return -1;
 786 }
 787
 788 /**
 789  * Mimic alloc_tables(), but for every context thread.
 790  */
 791 static void clone_tables(H264Context *dst, H264Context *src, int i){
 792     MpegEncContext * const s = &src->s;
 793     dst->intra4x4_pred_mode       = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
 794     dst->non_zero_count           = src->non_zero_count;
 795     dst->slice_table              = src->slice_table;
 796     dst->cbp_table                = src->cbp_table;
 797     dst->mb2b_xy                  = src->mb2b_xy;
 798     dst->mb2br_xy                 = src->mb2br_xy;
 799     dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
 800     dst->mvd_table[0]             = src->mvd_table[0] + i*8*2*s->mb_stride;
 801     dst->mvd_table[1]             = src->mvd_table[1] + i*8*2*s->mb_stride;
 802     dst->direct_table             = src->direct_table;
 803     dst->list_counts              = src->list_counts;
 804
 805     dst->s.obmc_scratchpad = NULL;
 806     ff_h264_pred_init(&dst->hpc, src->s.codec_id);
 807 }
 808
 809 /**
 810  * Init context
 811  * Allocate buffers which are not shared amongst multiple threads.
 812  */
 813 static int context_init(H264Context *h){
 814     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
 815     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
 816
 817     h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
 818     h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
 819
 820     return 0;
 821 fail:
 822     return -1; // free_tables will clean up for us
 823 }
 824
 825 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
 826
 827 static av_cold void common_init(H264Context *h){
 828     MpegEncContext * const s = &h->s;
 829
 830     s->width = s->avctx->width;
 831     s->height = s->avctx->height;
 832     s->codec_id= s->avctx->codec->id;
 833
 834     ff_h264dsp_init(&h->h264dsp);
 835     ff_h264_pred_init(&h->hpc, s->codec_id);
 836
 837     h->dequant_coeff_pps= -1;
 838     s->unrestricted_mv=1;
 839     s->decode=1; //FIXME
 840
 841     dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
 842
 843     memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
 844     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
 845 }
 846
 847 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
 848     H264Context *h= avctx->priv_data;
 849     MpegEncContext * const s = &h->s;
 850
 851     MPV_decode_defaults(s);
 852
 853     s->avctx = avctx;
 854     common_init(h);
 855
 856     s->out_format = FMT_H264;
 857     s->workaround_bugs= avctx->workaround_bugs;
 858
 859     // set defaults
 860 //    s->decode_mb= ff_h263_decode_mb;
 861     s->quarter_sample = 1;
 862     if(!avctx->has_b_frames)
 863     s->low_delay= 1;
 864
 865     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
 866
 867     ff_h264_decode_init_vlc();
 868
 869     h->thread_context[0] = h;
 870     h->outputed_poc = INT_MIN;
 871     h->prev_poc_msb= 1<<16;
 872     h->x264_build = -1;
 873     ff_h264_reset_sei(h);
 874     if(avctx->codec_id == CODEC_ID_H264){
 875         if(avctx->ticks_per_frame == 1){
 876             s->avctx->time_base.den *=2;
 877         }
 878         avctx->ticks_per_frame = 2;
 879     }
 880
 881     if(avctx->extradata_size > 0 && avctx->extradata && *(char *)avctx->extradata == 1){
 882         int i, cnt, nalsize;
 883         unsigned char *p = avctx->extradata;
 884
 885         h->is_avc = 1;
 886
 887         if(avctx->extradata_size < 7) {
 888             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
 889             return -1;
 890         }
 891         /* sps and pps in the avcC always have length coded with 2 bytes,
 892            so put a fake nal_length_size = 2 while parsing them */
 893         h->nal_length_size = 2;
 894         // Decode sps from avcC
 895         cnt = *(p+5) & 0x1f; // Number of sps
 896         p += 6;
 897         for (i = 0; i < cnt; i++) {
 898             nalsize = AV_RB16(p) + 2;
 899             if(decode_nal_units(h, p, nalsize) < 0) {
 900                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
 901                 return -1;
 902             }
 903             p += nalsize;
 904         }
 905         // Decode pps from avcC
 906         cnt = *(p++); // Number of pps
 907         for (i = 0; i < cnt; i++) {
 908             nalsize = AV_RB16(p) + 2;
 909             if(decode_nal_units(h, p, nalsize)  != nalsize) {
 910                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
 911                 return -1;
 912             }
 913             p += nalsize;
 914         }
 915         // Now store right nal length size, that will be use to parse all other nals
 916         h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
 917     } else {
 918         h->is_avc = 0;
 919         if(decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) < 0)
 920             return -1;
 921     }
 922     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
 923         s->avctx->has_b_frames = h->sps.num_reorder_frames;
 924         s->low_delay = 0;
 925     }
 926
 927     return 0;
 928 }
 929
 930 int ff_h264_frame_start(H264Context *h){
 931     MpegEncContext * const s = &h->s;
 932     int i;
 933
 934     if(MPV_frame_start(s, s->avctx) < 0)
 935         return -1;
 936     ff_er_frame_start(s);
 937     /*
 938      * MPV_frame_start uses pict_type to derive key_frame.
 939      * This is incorrect for H.264; IDR markings must be used.
 940      * Zero here; IDR markings per slice in frame or fields are ORed in later.
 941      * See decode_nal_units().
 942      */
 943     s->current_picture_ptr->key_frame= 0;
 944     s->current_picture_ptr->mmco_reset= 0;
 945
 946     assert(s->linesize && s->uvlinesize);
 947
 948     for(i=0; i<16; i++){
 949         h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
 950         h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
 951     }
 952     for(i=0; i<4; i++){
 953         h->block_offset[16+i]=
 954         h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
 955         h->block_offset[24+16+i]=
 956         h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
 957     }
 958
 959     /* can't be in alloc_tables because linesize isn't known there.
 960      * FIXME: redo bipred weight to not require extra buffer? */
 961     for(i = 0; i < s->avctx->thread_count; i++)
 962         if(!h->thread_context[i]->s.obmc_scratchpad)
 963             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
 964
 965     /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
 966     memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
 967
 968 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
 969
 970     // We mark the current picture as non-reference after allocating it, so
 971     // that if we break out due to an error it can be released automatically
 972     // in the next MPV_frame_start().
 973     // SVQ3 as well as most other codecs have only last/next/current and thus
 974     // get released even with set reference, besides SVQ3 and others do not
 975     // mark frames as reference later "naturally".
 976     if(s->codec_id != CODEC_ID_SVQ3)
 977         s->current_picture_ptr->reference= 0;
 978
 979     s->current_picture_ptr->field_poc[0]=
 980     s->current_picture_ptr->field_poc[1]= INT_MAX;
 981     assert(s->current_picture_ptr->long_ref==0);
 982
 983     return 0;
 984 }
 985
 986 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){
 987     MpegEncContext * const s = &h->s;
 988     uint8_t *top_border;
 989     int top_idx = 1;
 990
 991     src_y  -=   linesize;
 992     src_cb -= uvlinesize;
 993     src_cr -= uvlinesize;
 994
 995     if(!simple && FRAME_MBAFF){
 996         if(s->mb_y&1){
 997             if(!MB_MBAFF){
 998                 top_border = h->top_borders[0][s->mb_x];
 999                 AV_COPY128(top_border, src_y + 15*linesize);
1000                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1001                     AV_COPY64(top_border+16, src_cb+7*uvlinesize);
1002                     AV_COPY64(top_border+24, src_cr+7*uvlinesize);
1003                 }
1004             }
1005         }else if(MB_MBAFF){
1006             top_idx = 0;
1007         }else
1008             return;
1009     }
1010
1011     top_border = h->top_borders[top_idx][s->mb_x];
1012     // There are two lines saved, the line above the the top macroblock of a pair,
1013     // and the line above the bottom macroblock
1014     AV_COPY128(top_border, src_y + 16*linesize);
1015
1016     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1017         AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1018         AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1019     }
1020 }
1021
1022 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){
1023     MpegEncContext * const s = &h->s;
1024     int deblock_left;
1025     int deblock_top;
1026     int top_idx = 1;
1027     uint8_t *top_border_m1;
1028     uint8_t *top_border;
1029
1030     if(!simple && FRAME_MBAFF){
1031         if(s->mb_y&1){
1032             if(!MB_MBAFF)
1033                 return;
1034         }else{
1035             top_idx = MB_MBAFF ? 0 : 1;
1036         }
1037     }
1038
1039     if(h->deblocking_filter == 2) {
1040         deblock_left = h->left_type[0];
1041         deblock_top  = h->top_type;
1042     } else {
1043         deblock_left = (s->mb_x > 0);
1044         deblock_top =  (s->mb_y > !!MB_FIELD);
1045     }
1046
1047     src_y  -=   linesize + 1;
1048     src_cb -= uvlinesize + 1;
1049     src_cr -= uvlinesize + 1;
1050
1051     top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1052     top_border    = h->top_borders[top_idx][s->mb_x];
1053
1054 #define XCHG(a,b,xchg)\
1055 if (xchg) AV_SWAP64(b,a);\
1056 else      AV_COPY64(b,a);
1057
1058     if(deblock_top){
1059         if(deblock_left){
1060             XCHG(top_border_m1+8, src_y -7, 1);
1061         }
1062         XCHG(top_border+0, src_y +1, xchg);
1063         XCHG(top_border+8, src_y +9, 1);
1064         if(s->mb_x+1 < s->mb_width){
1065             XCHG(h->top_borders[top_idx][s->mb_x+1], src_y +17, 1);
1066         }
1067     }
1068
1069     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1070         if(deblock_top){
1071             if(deblock_left){
1072                 XCHG(top_border_m1+16, src_cb -7, 1);
1073                 XCHG(top_border_m1+24, src_cr -7, 1);
1074             }
1075             XCHG(top_border+16, src_cb+1, 1);
1076             XCHG(top_border+24, src_cr+1, 1);
1077         }
1078     }
1079 }
1080
1081 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
1082     MpegEncContext * const s = &h->s;
1083     const int mb_x= s->mb_x;
1084     const int mb_y= s->mb_y;
1085     const int mb_xy= h->mb_xy;
1086     const int mb_type= s->current_picture.mb_type[mb_xy];
1087     uint8_t  *dest_y, *dest_cb, *dest_cr;
1088     int linesize, uvlinesize /*dct_offset*/;
1089     int i;
1090     int *block_offset = &h->block_offset[0];
1091     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1092     /* is_h264 should always be true if SVQ3 is disabled. */
1093     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1094     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1095     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1096
1097     dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
1098     dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
1099     dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
1100
1101     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
1102     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
1103
1104     h->list_counts[mb_xy]= h->list_count;
1105
1106     if (!simple && MB_FIELD) {
1107         linesize   = h->mb_linesize   = s->linesize * 2;
1108         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1109         block_offset = &h->block_offset[24];
1110         if(mb_y&1){ //FIXME move out of this function?
1111             dest_y -= s->linesize*15;
1112             dest_cb-= s->uvlinesize*7;
1113             dest_cr-= s->uvlinesize*7;
1114         }
1115         if(FRAME_MBAFF) {
1116             int list;
1117             for(list=0; list<h->list_count; list++){
1118                 if(!USES_LIST(mb_type, list))
1119                     continue;
1120                 if(IS_16X16(mb_type)){
1121                     int8_t *ref = &h->ref_cache[list][scan8[0]];
1122                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1123                 }else{
1124                     for(i=0; i<16; i+=4){
1125                         int ref = h->ref_cache[list][scan8[i]];
1126                         if(ref >= 0)
1127                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1128                     }
1129                 }
1130             }
1131         }
1132     } else {
1133         linesize   = h->mb_linesize   = s->linesize;
1134         uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1135 //        dct_offset = s->linesize * 16;
1136     }
1137
1138     if (!simple && IS_INTRA_PCM(mb_type)) {
1139         for (i=0; i<16; i++) {
1140             memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1141         }
1142         for (i=0; i<8; i++) {
1143             memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
1144             memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
1145         }
1146     } else {
1147         if(IS_INTRA(mb_type)){
1148             if(h->deblocking_filter)
1149                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
1150
1151             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1152                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1153                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1154             }
1155
1156             if(IS_INTRA4x4(mb_type)){
1157                 if(simple || !s->encoding){
1158                     if(IS_8x8DCT(mb_type)){
1159                         if(transform_bypass){
1160                             idct_dc_add =
1161                             idct_add    = s->dsp.add_pixels8;
1162                         }else{
1163                             idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1164                             idct_add    = h->h264dsp.h264_idct8_add;
1165                         }
1166                         for(i=0; i<16; i+=4){
1167                             uint8_t * const ptr= dest_y + block_offset[i];
1168                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1169                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1170                                 h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
1171                             }else{
1172                                 const int nnz = h->non_zero_count_cache[ scan8[i] ];
1173                                 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1174                                                             (h->topright_samples_available<<i)&0x4000, linesize);
1175                                 if(nnz){
1176                                     if(nnz == 1 && h->mb[i*16])
1177                                         idct_dc_add(ptr, h->mb + i*16, linesize);
1178                                     else
1179                                         idct_add   (ptr, h->mb + i*16, linesize);
1180                                 }
1181                             }
1182                         }
1183                     }else{
1184                         if(transform_bypass){
1185                             idct_dc_add =
1186                             idct_add    = s->dsp.add_pixels4;
1187                         }else{
1188                             idct_dc_add = h->h264dsp.h264_idct_dc_add;
1189                             idct_add    = h->h264dsp.h264_idct_add;
1190                         }
1191                         for(i=0; i<16; i++){
1192                             uint8_t * const ptr= dest_y + block_offset[i];
1193                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1194
1195                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1196                                 h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
1197                             }else{
1198                                 uint8_t *topright;
1199                                 int nnz, tr;
1200                                 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1201                                     const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1202                                     assert(mb_y || linesize <= block_offset[i]);
1203                                     if(!topright_avail){
1204                                         tr= ptr[3 - linesize]*0x01010101;
1205                                         topright= (uint8_t*) &tr;
1206                                     }else
1207                                         topright= ptr + 4 - linesize;
1208                                 }else
1209                                     topright= NULL;
1210
1211                                 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1212                                 nnz = h->non_zero_count_cache[ scan8[i] ];
1213                                 if(nnz){
1214                                     if(is_h264){
1215                                         if(nnz == 1 && h->mb[i*16])
1216                                             idct_dc_add(ptr, h->mb + i*16, linesize);
1217                                         else
1218                                             idct_add   (ptr, h->mb + i*16, linesize);
1219                                     }else
1220                                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1221                                 }
1222                             }
1223                         }
1224                     }
1225                 }
1226             }else{
1227                 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1228                 if(is_h264){
1229                     if(!transform_bypass)
1230                         h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
1231                 }else
1232                     ff_svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
1233             }
1234             if(h->deblocking_filter)
1235                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
1236         }else if(is_h264){
1237             hl_motion(h, dest_y, dest_cb, dest_cr,
1238                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1239                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1240                       h->h264dsp.weight_h264_pixels_tab, h->h264dsp.biweight_h264_pixels_tab);
1241         }
1242
1243
1244         if(!IS_INTRA4x4(mb_type)){
1245             if(is_h264){
1246                 if(IS_INTRA16x16(mb_type)){
1247                     if(transform_bypass){
1248                         if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1249                             h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1250                         }else{
1251                             for(i=0; i<16; i++){
1252                                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1253                                     s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
1254                             }
1255                         }
1256                     }else{
1257                          h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1258                     }
1259                 }else if(h->cbp&15){
1260                     if(transform_bypass){
1261                         const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1262                         idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1263                         for(i=0; i<16; i+=di){
1264                             if(h->non_zero_count_cache[ scan8[i] ]){
1265                                 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
1266                             }
1267                         }
1268                     }else{
1269                         if(IS_8x8DCT(mb_type)){
1270                             h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1271                         }else{
1272                             h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1273                         }
1274                     }
1275                 }
1276             }else{
1277                 for(i=0; i<16; i++){
1278                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1279                         uint8_t * const ptr= dest_y + block_offset[i];
1280                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1281                     }
1282                 }
1283             }
1284         }
1285
1286         if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1287             uint8_t *dest[2] = {dest_cb, dest_cr};
1288             if(transform_bypass){
1289                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1290                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
1291                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
1292                 }else{
1293                     idct_add = s->dsp.add_pixels4;
1294                     for(i=16; i<16+8; i++){
1295                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1296                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1297                     }
1298                 }
1299             }else{
1300                 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp[0], h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
1301                 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp[1], h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
1302                 if(is_h264){
1303                     idct_add = h->h264dsp.h264_idct_add;
1304                     idct_dc_add = h->h264dsp.h264_idct_dc_add;
1305                     for(i=16; i<16+8; i++){
1306                         if(h->non_zero_count_cache[ scan8[i] ])
1307                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1308                         else if(h->mb[i*16])
1309                             idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1310                     }
1311                 }else{
1312                     for(i=16; i<16+8; i++){
1313                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1314                             uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1315                             ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
1316                         }
1317                     }
1318                 }
1319             }
1320         }
1321     }
1322     if(h->cbp || IS_INTRA(mb_type))
1323         s->dsp.clear_blocks(h->mb);
1324 }
1325
1326 /**
1327  * Process a macroblock; this case avoids checks for expensive uncommon cases.
1328  */
1329 static void hl_decode_mb_simple(H264Context *h){
1330     hl_decode_mb_internal(h, 1);
1331 }
1332
1333 /**
1334  * Process a macroblock; this handles edge cases, such as interlacing.
1335  */
1336 static void av_noinline hl_decode_mb_complex(H264Context *h){
1337     hl_decode_mb_internal(h, 0);
1338 }
1339
1340 void ff_h264_hl_decode_mb(H264Context *h){
1341     MpegEncContext * const s = &h->s;
1342     const int mb_xy= h->mb_xy;
1343     const int mb_type= s->current_picture.mb_type[mb_xy];
1344     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1345
1346     if (is_complex)
1347         hl_decode_mb_complex(h);
1348     else hl_decode_mb_simple(h);
1349 }
1350
1351 static int pred_weight_table(H264Context *h){
1352     MpegEncContext * const s = &h->s;
1353     int list, i;
1354     int luma_def, chroma_def;
1355
1356     h->use_weight= 0;
1357     h->use_weight_chroma= 0;
1358     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1359     h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1360     luma_def = 1<<h->luma_log2_weight_denom;
1361     chroma_def = 1<<h->chroma_log2_weight_denom;
1362
1363     for(list=0; list<2; list++){
1364         h->luma_weight_flag[list]   = 0;
1365         h->chroma_weight_flag[list] = 0;
1366         for(i=0; i<h->ref_count[list]; i++){
1367             int luma_weight_flag, chroma_weight_flag;
1368
1369             luma_weight_flag= get_bits1(&s->gb);
1370             if(luma_weight_flag){
1371                 h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
1372                 h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
1373                 if(   h->luma_weight[i][list][0] != luma_def
1374                    || h->luma_weight[i][list][1] != 0) {
1375                     h->use_weight= 1;
1376                     h->luma_weight_flag[list]= 1;
1377                 }
1378             }else{
1379                 h->luma_weight[i][list][0]= luma_def;
1380                 h->luma_weight[i][list][1]= 0;
1381             }
1382
1383             if(CHROMA){
1384                 chroma_weight_flag= get_bits1(&s->gb);
1385                 if(chroma_weight_flag){
1386                     int j;
1387                     for(j=0; j<2; j++){
1388                         h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
1389                         h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
1390                         if(   h->chroma_weight[i][list][j][0] != chroma_def
1391                            || h->chroma_weight[i][list][j][1] != 0) {
1392                             h->use_weight_chroma= 1;
1393                             h->chroma_weight_flag[list]= 1;
1394                         }
1395                     }
1396                 }else{
1397                     int j;
1398                     for(j=0; j<2; j++){
1399                         h->chroma_weight[i][list][j][0]= chroma_def;
1400                         h->chroma_weight[i][list][j][1]= 0;
1401                     }
1402                 }
1403             }
1404         }
1405         if(h->slice_type_nos != FF_B_TYPE) break;
1406     }
1407     h->use_weight= h->use_weight || h->use_weight_chroma;
1408     return 0;
1409 }
1410
1411 /**
1412  * Initialize implicit_weight table.
1413  * @param field, 0/1 initialize the weight for interlaced MBAFF
1414  *                -1 initializes the rest
1415  */
1416 static void implicit_weight_table(H264Context *h, int field){
1417     MpegEncContext * const s = &h->s;
1418     int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
1419
1420     for (i = 0; i < 2; i++) {
1421         h->luma_weight_flag[i]   = 0;
1422         h->chroma_weight_flag[i] = 0;
1423     }
1424
1425     if(field < 0){
1426         cur_poc = s->current_picture_ptr->poc;
1427     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
1428        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1429         h->use_weight= 0;
1430         h->use_weight_chroma= 0;
1431         return;
1432     }
1433         ref_start= 0;
1434         ref_count0= h->ref_count[0];
1435         ref_count1= h->ref_count[1];
1436     }else{
1437         cur_poc = s->current_picture_ptr->field_poc[field];
1438         ref_start= 16;
1439         ref_count0= 16+2*h->ref_count[0];
1440         ref_count1= 16+2*h->ref_count[1];
1441     }
1442
1443     h->use_weight= 2;
1444     h->use_weight_chroma= 2;
1445     h->luma_log2_weight_denom= 5;
1446     h->chroma_log2_weight_denom= 5;
1447
1448     for(ref0=ref_start; ref0 < ref_count0; ref0++){
1449         int poc0 = h->ref_list[0][ref0].poc;
1450         for(ref1=ref_start; ref1 < ref_count1; ref1++){
1451             int poc1 = h->ref_list[1][ref1].poc;
1452             int td = av_clip(poc1 - poc0, -128, 127);
1453             int w= 32;
1454             if(td){
1455                 int tb = av_clip(cur_poc - poc0, -128, 127);
1456                 int tx = (16384 + (FFABS(td) >> 1)) / td;
1457                 int dist_scale_factor = (tb*tx + 32) >> 8;
1458                 if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
1459                     w = 64 - dist_scale_factor;
1460             }
1461             if(field<0){
1462                 h->implicit_weight[ref0][ref1][0]=
1463                 h->implicit_weight[ref0][ref1][1]= w;
1464             }else{
1465                 h->implicit_weight[ref0][ref1][field]=w;
1466             }
1467         }
1468     }
1469 }
1470
1471 /**
1472  * instantaneous decoder refresh.
1473  */
1474 static void idr(H264Context *h){
1475     ff_h264_remove_all_refs(h);
1476     h->prev_frame_num= 0;
1477     h->prev_frame_num_offset= 0;
1478     h->prev_poc_msb=
1479     h->prev_poc_lsb= 0;
1480 }
1481
1482 /* forget old pics after a seek */
1483 static void flush_dpb(AVCodecContext *avctx){
1484     H264Context *h= avctx->priv_data;
1485     int i;
1486     for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1487         if(h->delayed_pic[i])
1488             h->delayed_pic[i]->reference= 0;
1489         h->delayed_pic[i]= NULL;
1490     }
1491     h->outputed_poc= INT_MIN;
1492     h->prev_interlaced_frame = 1;
1493     idr(h);
1494     if(h->s.current_picture_ptr)
1495         h->s.current_picture_ptr->reference= 0;
1496     h->s.first_field= 0;
1497     ff_h264_reset_sei(h);
1498     ff_mpeg_flush(avctx);
1499 }
1500
1501 static int init_poc(H264Context *h){
1502     MpegEncContext * const s = &h->s;
1503     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1504     int field_poc[2];
1505     Picture *cur = s->current_picture_ptr;
1506
1507     h->frame_num_offset= h->prev_frame_num_offset;
1508     if(h->frame_num < h->prev_frame_num)
1509         h->frame_num_offset += max_frame_num;
1510
1511     if(h->sps.poc_type==0){
1512         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1513
1514         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1515             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1516         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1517             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1518         else
1519             h->poc_msb = h->prev_poc_msb;
1520 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1521         field_poc[0] =
1522         field_poc[1] = h->poc_msb + h->poc_lsb;
1523         if(s->picture_structure == PICT_FRAME)
1524             field_poc[1] += h->delta_poc_bottom;
1525     }else if(h->sps.poc_type==1){
1526         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1527         int i;
1528
1529         if(h->sps.poc_cycle_length != 0)
1530             abs_frame_num = h->frame_num_offset + h->frame_num;
1531         else
1532             abs_frame_num = 0;
1533
1534         if(h->nal_ref_idc==0 && abs_frame_num > 0)
1535             abs_frame_num--;
1536
1537         expected_delta_per_poc_cycle = 0;
1538         for(i=0; i < h->sps.poc_cycle_length; i++)
1539             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1540
1541         if(abs_frame_num > 0){
1542             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1543             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1544
1545             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1546             for(i = 0; i <= frame_num_in_poc_cycle; i++)
1547                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1548         } else
1549             expectedpoc = 0;
1550
1551         if(h->nal_ref_idc == 0)
1552             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1553
1554         field_poc[0] = expectedpoc + h->delta_poc[0];
1555         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1556
1557         if(s->picture_structure == PICT_FRAME)
1558             field_poc[1] += h->delta_poc[1];
1559     }else{
1560         int poc= 2*(h->frame_num_offset + h->frame_num);
1561
1562         if(!h->nal_ref_idc)
1563             poc--;
1564
1565         field_poc[0]= poc;
1566         field_poc[1]= poc;
1567     }
1568
1569     if(s->picture_structure != PICT_BOTTOM_FIELD)
1570         s->current_picture_ptr->field_poc[0]= field_poc[0];
1571     if(s->picture_structure != PICT_TOP_FIELD)
1572         s->current_picture_ptr->field_poc[1]= field_poc[1];
1573     cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1574
1575     return 0;
1576 }
1577
1578
1579 /**
1580  * initialize scan tables
1581  */
1582 static void init_scan_tables(H264Context *h){
1583     int i;
1584     if(h->h264dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
1585         memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
1586         memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
1587     }else{
1588         for(i=0; i<16; i++){
1589 #define T(x) (x>>2) | ((x<<2) & 0xF)
1590             h->zigzag_scan[i] = T(zigzag_scan[i]);
1591             h-> field_scan[i] = T( field_scan[i]);
1592 #undef T
1593         }
1594     }
1595     if(h->h264dsp.h264_idct8_add == ff_h264_idct8_add_c){
1596         memcpy(h->zigzag_scan8x8,       ff_zigzag_direct,     64*sizeof(uint8_t));
1597         memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
1598         memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
1599         memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
1600     }else{
1601         for(i=0; i<64; i++){
1602 #define T(x) (x>>3) | ((x&7)<<3)
1603             h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
1604             h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1605             h->field_scan8x8[i]        = T(field_scan8x8[i]);
1606             h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
1607 #undef T
1608         }
1609     }
1610     if(h->sps.transform_bypass){ //FIXME same ugly
1611         h->zigzag_scan_q0          = zigzag_scan;
1612         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
1613         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1614         h->field_scan_q0           = field_scan;
1615         h->field_scan8x8_q0        = field_scan8x8;
1616         h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
1617     }else{
1618         h->zigzag_scan_q0          = h->zigzag_scan;
1619         h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
1620         h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1621         h->field_scan_q0           = h->field_scan;
1622         h->field_scan8x8_q0        = h->field_scan8x8;
1623         h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
1624     }
1625 }
1626
1627 static void field_end(H264Context *h){
1628     MpegEncContext * const s = &h->s;
1629     AVCodecContext * const avctx= s->avctx;
1630     s->mb_y= 0;
1631
1632     s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1633     s->current_picture_ptr->pict_type= s->pict_type;
1634
1635     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1636         ff_vdpau_h264_set_reference_frames(s);
1637
1638     if(!s->dropable) {
1639         ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1640         h->prev_poc_msb= h->poc_msb;
1641         h->prev_poc_lsb= h->poc_lsb;
1642     }
1643     h->prev_frame_num_offset= h->frame_num_offset;
1644     h->prev_frame_num= h->frame_num;
1645
1646     if (avctx->hwaccel) {
1647         if (avctx->hwaccel->end_frame(avctx) < 0)
1648             av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1649     }
1650
1651     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1652         ff_vdpau_h264_picture_complete(s);
1653
1654     /*
1655      * FIXME: Error handling code does not seem to support interlaced
1656      * when slices span multiple rows
1657      * The ff_er_add_slice calls don't work right for bottom
1658      * fields; they cause massive erroneous error concealing
1659      * Error marking covers both fields (top and bottom).
1660      * This causes a mismatched s->error_count
1661      * and a bad error table. Further, the error count goes to
1662      * INT_MAX when called for bottom field, because mb_y is
1663      * past end by one (callers fault) and resync_mb_y != 0
1664      * causes problems for the first MB line, too.
1665      */
1666     if (!FIELD_PICTURE)
1667         ff_er_frame_end(s);
1668
1669     MPV_frame_end(s);
1670
1671     h->current_slice=0;
1672 }
1673
1674 /**
1675  * Replicates H264 "master" context to thread contexts.
1676  */
1677 static void clone_slice(H264Context *dst, H264Context *src)
1678 {
1679     memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
1680     dst->s.current_picture_ptr  = src->s.current_picture_ptr;
1681     dst->s.current_picture      = src->s.current_picture;
1682     dst->s.linesize             = src->s.linesize;
1683     dst->s.uvlinesize           = src->s.uvlinesize;
1684     dst->s.first_field          = src->s.first_field;
1685
1686     dst->prev_poc_msb           = src->prev_poc_msb;
1687     dst->prev_poc_lsb           = src->prev_poc_lsb;
1688     dst->prev_frame_num_offset  = src->prev_frame_num_offset;
1689     dst->prev_frame_num         = src->prev_frame_num;
1690     dst->short_ref_count        = src->short_ref_count;
1691
1692     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
1693     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
1694     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1695     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
1696
1697     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
1698     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
1699 }
1700
1701 /**
1702  * decodes a slice header.
1703  * This will also call MPV_common_init() and frame_start() as needed.
1704  *
1705  * @param h h264context
1706  * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1707  *
1708  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1709  */
1710 static int decode_slice_header(H264Context *h, H264Context *h0){
1711     MpegEncContext * const s = &h->s;
1712     MpegEncContext * const s0 = &h0->s;
1713     unsigned int first_mb_in_slice;
1714     unsigned int pps_id;
1715     int num_ref_idx_active_override_flag;
1716     unsigned int slice_type, tmp, i, j;
1717     int default_ref_list_done = 0;
1718     int last_pic_structure;
1719
1720     s->dropable= h->nal_ref_idc == 0;
1721
1722     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1723         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1724         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1725     }else{
1726         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1727         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1728     }
1729
1730     first_mb_in_slice= get_ue_golomb(&s->gb);
1731
1732     if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1733         if(h0->current_slice && FIELD_PICTURE){
1734             field_end(h);
1735         }
1736
1737         h0->current_slice = 0;
1738         if (!s0->first_field)
1739             s->current_picture_ptr= NULL;
1740     }
1741
1742     slice_type= get_ue_golomb_31(&s->gb);
1743     if(slice_type > 9){
1744         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);
1745         return -1;
1746     }
1747     if(slice_type > 4){
1748         slice_type -= 5;
1749         h->slice_type_fixed=1;
1750     }else
1751         h->slice_type_fixed=0;
1752
1753     slice_type= golomb_to_pict_type[ slice_type ];
1754     if (slice_type == FF_I_TYPE
1755         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
1756         default_ref_list_done = 1;
1757     }
1758     h->slice_type= slice_type;
1759     h->slice_type_nos= slice_type & 3;
1760
1761     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
1762
1763     pps_id= get_ue_golomb(&s->gb);
1764     if(pps_id>=MAX_PPS_COUNT){
1765         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
1766         return -1;
1767     }
1768     if(!h0->pps_buffers[pps_id]) {
1769         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
1770         return -1;
1771     }
1772     h->pps= *h0->pps_buffers[pps_id];
1773
1774     if(!h0->sps_buffers[h->pps.sps_id]) {
1775         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
1776         return -1;
1777     }
1778     h->sps = *h0->sps_buffers[h->pps.sps_id];
1779
1780     if(h == h0 && h->dequant_coeff_pps != pps_id){
1781         h->dequant_coeff_pps = pps_id;
1782         init_dequant_tables(h);
1783     }
1784
1785     s->mb_width= h->sps.mb_width;
1786     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1787
1788     h->b_stride=  s->mb_width*4;
1789
1790     s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
1791     if(h->sps.frame_mbs_only_flag)
1792         s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
1793     else
1794         s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
1795
1796     if (s->context_initialized
1797         && (   s->width != s->avctx->width || s->height != s->avctx->height
1798             || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
1799         if(h != h0)
1800             return -1;   // width / height changed during parallelized decoding
1801         free_tables(h);
1802         flush_dpb(s->avctx);
1803         MPV_common_end(s);
1804     }
1805     if (!s->context_initialized) {
1806         if(h != h0)
1807             return -1;  // we cant (re-)initialize context during parallel decoding
1808
1809         avcodec_set_dimensions(s->avctx, s->width, s->height);
1810         s->avctx->sample_aspect_ratio= h->sps.sar;
1811         if(!s->avctx->sample_aspect_ratio.den)
1812             s->avctx->sample_aspect_ratio.den = 1;
1813
1814         if(h->sps.video_signal_type_present_flag){
1815             s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1816             if(h->sps.colour_description_present_flag){
1817                 s->avctx->color_primaries = h->sps.color_primaries;
1818                 s->avctx->color_trc       = h->sps.color_trc;
1819                 s->avctx->colorspace      = h->sps.colorspace;
1820             }
1821         }
1822
1823         if(h->sps.timing_info_present_flag){
1824             int64_t den= h->sps.time_scale;
1825             if(h->x264_build < 44U)
1826                 den *= 2;
1827             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
1828                       h->sps.num_units_in_tick, den, 1<<30);
1829         }
1830         s->avctx->pix_fmt = s->avctx->get_format(s->avctx, s->avctx->codec->pix_fmts);
1831         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1832
1833         if (MPV_common_init(s) < 0)
1834             return -1;
1835         s->first_field = 0;
1836         h->prev_interlaced_frame = 1;
1837
1838         init_scan_tables(h);
1839         ff_h264_alloc_tables(h);
1840
1841         for(i = 1; i < s->avctx->thread_count; i++) {
1842             H264Context *c;
1843             c = h->thread_context[i] = av_malloc(sizeof(H264Context));
1844             memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
1845             memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
1846             c->h264dsp = h->h264dsp;
1847             c->sps = h->sps;
1848             c->pps = h->pps;
1849             init_scan_tables(c);
1850             clone_tables(c, h, i);
1851         }
1852
1853         for(i = 0; i < s->avctx->thread_count; i++)
1854             if(context_init(h->thread_context[i]) < 0)
1855                 return -1;
1856     }
1857
1858     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1859
1860     h->mb_mbaff = 0;
1861     h->mb_aff_frame = 0;
1862     last_pic_structure = s0->picture_structure;
1863     if(h->sps.frame_mbs_only_flag){
1864         s->picture_structure= PICT_FRAME;
1865     }else{
1866         if(get_bits1(&s->gb)) { //field_pic_flag
1867             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
1868         } else {
1869             s->picture_structure= PICT_FRAME;
1870             h->mb_aff_frame = h->sps.mb_aff;
1871         }
1872     }
1873     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
1874
1875     if(h0->current_slice == 0){
1876         while(h->frame_num !=  h->prev_frame_num &&
1877               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
1878             av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
1879             if (ff_h264_frame_start(h) < 0)
1880                 return -1;
1881             h->prev_frame_num++;
1882             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
1883             s->current_picture_ptr->frame_num= h->prev_frame_num;
1884             ff_h264_execute_ref_pic_marking(h, NULL, 0);
1885         }
1886
1887         /* See if we have a decoded first field looking for a pair... */
1888         if (s0->first_field) {
1889             assert(s0->current_picture_ptr);
1890             assert(s0->current_picture_ptr->data[0]);
1891             assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
1892
1893             /* figure out if we have a complementary field pair */
1894             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
1895                 /*
1896                  * Previous field is unmatched. Don't display it, but let it
1897                  * remain for reference if marked as such.
1898                  */
1899                 s0->current_picture_ptr = NULL;
1900                 s0->first_field = FIELD_PICTURE;
1901
1902             } else {
1903                 if (h->nal_ref_idc &&
1904                         s0->current_picture_ptr->reference &&
1905                         s0->current_picture_ptr->frame_num != h->frame_num) {
1906                     /*
1907                      * This and previous field were reference, but had
1908                      * different frame_nums. Consider this field first in
1909                      * pair. Throw away previous field except for reference
1910                      * purposes.
1911                      */
1912                     s0->first_field = 1;
1913                     s0->current_picture_ptr = NULL;
1914
1915                 } else {
1916                     /* Second field in complementary pair */
1917                     s0->first_field = 0;
1918                 }
1919             }
1920
1921         } else {
1922             /* Frame or first field in a potentially complementary pair */
1923             assert(!s0->current_picture_ptr);
1924             s0->first_field = FIELD_PICTURE;
1925         }
1926
1927         if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
1928             s0->first_field = 0;
1929             return -1;
1930         }
1931     }
1932     if(h != h0)
1933         clone_slice(h, h0);
1934
1935     s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
1936
1937     assert(s->mb_num == s->mb_width * s->mb_height);
1938     if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
1939        first_mb_in_slice                    >= s->mb_num){
1940         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1941         return -1;
1942     }
1943     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
1944     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
1945     if (s->picture_structure == PICT_BOTTOM_FIELD)
1946         s->resync_mb_y = s->mb_y = s->mb_y + 1;
1947     assert(s->mb_y < s->mb_height);
1948
1949     if(s->picture_structure==PICT_FRAME){
1950         h->curr_pic_num=   h->frame_num;
1951         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
1952     }else{
1953         h->curr_pic_num= 2*h->frame_num + 1;
1954         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
1955     }
1956
1957     if(h->nal_unit_type == NAL_IDR_SLICE){
1958         get_ue_golomb(&s->gb); /* idr_pic_id */
1959     }
1960
1961     if(h->sps.poc_type==0){
1962         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
1963
1964         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
1965             h->delta_poc_bottom= get_se_golomb(&s->gb);
1966         }
1967     }
1968
1969     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
1970         h->delta_poc[0]= get_se_golomb(&s->gb);
1971
1972         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
1973             h->delta_poc[1]= get_se_golomb(&s->gb);
1974     }
1975
1976     init_poc(h);
1977
1978     if(h->pps.redundant_pic_cnt_present){
1979         h->redundant_pic_count= get_ue_golomb(&s->gb);
1980     }
1981
1982     //set defaults, might be overridden a few lines later
1983     h->ref_count[0]= h->pps.ref_count[0];
1984     h->ref_count[1]= h->pps.ref_count[1];
1985
1986     if(h->slice_type_nos != FF_I_TYPE){
1987         if(h->slice_type_nos == FF_B_TYPE){
1988             h->direct_spatial_mv_pred= get_bits1(&s->gb);
1989         }
1990         num_ref_idx_active_override_flag= get_bits1(&s->gb);
1991
1992         if(num_ref_idx_active_override_flag){
1993             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
1994             if(h->slice_type_nos==FF_B_TYPE)
1995                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
1996
1997             if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
1998                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
1999                 h->ref_count[0]= h->ref_count[1]= 1;
2000                 return -1;
2001             }
2002         }
2003         if(h->slice_type_nos == FF_B_TYPE)
2004             h->list_count= 2;
2005         else
2006             h->list_count= 1;
2007     }else
2008         h->list_count= 0;
2009
2010     if(!default_ref_list_done){
2011         ff_h264_fill_default_ref_list(h);
2012     }
2013
2014     if(h->slice_type_nos!=FF_I_TYPE && ff_h264_decode_ref_pic_list_reordering(h) < 0)
2015         return -1;
2016
2017     if(h->slice_type_nos!=FF_I_TYPE){
2018         s->last_picture_ptr= &h->ref_list[0][0];
2019         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
2020     }
2021     if(h->slice_type_nos==FF_B_TYPE){
2022         s->next_picture_ptr= &h->ref_list[1][0];
2023         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
2024     }
2025
2026     if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
2027        ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
2028         pred_weight_table(h);
2029     else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE){
2030         implicit_weight_table(h, -1);
2031     }else {
2032         h->use_weight = 0;
2033         for (i = 0; i < 2; i++) {
2034             h->luma_weight_flag[i]   = 0;
2035             h->chroma_weight_flag[i] = 0;
2036         }
2037     }
2038
2039     if(h->nal_ref_idc)
2040         ff_h264_decode_ref_pic_marking(h0, &s->gb);
2041
2042     if(FRAME_MBAFF){
2043         ff_h264_fill_mbaff_ref_list(h);
2044
2045         if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE){
2046             implicit_weight_table(h, 0);
2047             implicit_weight_table(h, 1);
2048         }
2049     }
2050
2051     if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
2052         ff_h264_direct_dist_scale_factor(h);
2053     ff_h264_direct_ref_list_init(h);
2054
2055     if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
2056         tmp = get_ue_golomb_31(&s->gb);
2057         if(tmp > 2){
2058             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2059             return -1;
2060         }
2061         h->cabac_init_idc= tmp;
2062     }
2063
2064     h->last_qscale_diff = 0;
2065     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2066     if(tmp>51){
2067         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2068         return -1;
2069     }
2070     s->qscale= tmp;
2071     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2072     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2073     //FIXME qscale / qp ... stuff
2074     if(h->slice_type == FF_SP_TYPE){
2075         get_bits1(&s->gb); /* sp_for_switch_flag */
2076     }
2077     if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
2078         get_se_golomb(&s->gb); /* slice_qs_delta */
2079     }
2080
2081     h->deblocking_filter = 1;
2082     h->slice_alpha_c0_offset = 52;
2083     h->slice_beta_offset = 52;
2084     if( h->pps.deblocking_filter_parameters_present ) {
2085         tmp= get_ue_golomb_31(&s->gb);
2086         if(tmp > 2){
2087             av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2088             return -1;
2089         }
2090         h->deblocking_filter= tmp;
2091         if(h->deblocking_filter < 2)
2092             h->deblocking_filter^= 1; // 1<->0
2093
2094         if( h->deblocking_filter ) {
2095             h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2096             h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
2097             if(   h->slice_alpha_c0_offset > 104U
2098                || h->slice_beta_offset     > 104U){
2099                 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);
2100                 return -1;
2101             }
2102         }
2103     }
2104
2105     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
2106        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
2107        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
2108        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2109         h->deblocking_filter= 0;
2110
2111     if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2112         if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2113             /* Cheat slightly for speed:
2114                Do not bother to deblock across slices. */
2115             h->deblocking_filter = 2;
2116         } else {
2117             h0->max_contexts = 1;
2118             if(!h0->single_decode_warning) {
2119                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2120                 h0->single_decode_warning = 1;
2121             }
2122             if(h != h0)
2123                 return 1; // deblocking switched inside frame
2124         }
2125     }
2126     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]);
2127
2128 #if 0 //FMO
2129     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2130         slice_group_change_cycle= get_bits(&s->gb, ?);
2131 #endif
2132
2133     h0->last_slice_type = slice_type;
2134     h->slice_num = ++h0->current_slice;
2135     if(h->slice_num >= MAX_SLICES){
2136         av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2137     }
2138
2139     for(j=0; j<2; j++){
2140         int id_list[16];
2141         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2142         for(i=0; i<16; i++){
2143             id_list[i]= 60;
2144             if(h->ref_list[j][i].data[0]){
2145                 int k;
2146                 uint8_t *base= h->ref_list[j][i].base[0];
2147                 for(k=0; k<h->short_ref_count; k++)
2148                     if(h->short_ref[k]->base[0] == base){
2149                         id_list[i]= k;
2150                         break;
2151                     }
2152                 for(k=0; k<h->long_ref_count; k++)
2153                     if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2154                         id_list[i]= h->short_ref_count + k;
2155                         break;
2156                     }
2157             }
2158         }
2159
2160         ref2frm[0]=
2161         ref2frm[1]= -1;
2162         for(i=0; i<16; i++)
2163             ref2frm[i+2]= 4*id_list[i]
2164                           +(h->ref_list[j][i].reference&3);
2165         ref2frm[18+0]=
2166         ref2frm[18+1]= -1;
2167         for(i=16; i<48; i++)
2168             ref2frm[i+4]= 4*id_list[(i-16)>>1]
2169                           +(h->ref_list[j][i].reference&3);
2170     }
2171
2172     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
2173     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2174
2175     s->avctx->refs= h->sps.ref_frame_count;
2176
2177     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2178         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",
2179                h->slice_num,
2180                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2181                first_mb_in_slice,
2182                av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2183                pps_id, h->frame_num,
2184                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2185                h->ref_count[0], h->ref_count[1],
2186                s->qscale,
2187                h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
2188                h->use_weight,
2189                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2190                h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2191                );
2192     }
2193
2194     return 0;
2195 }
2196
2197 int ff_h264_get_slice_type(const H264Context *h)
2198 {
2199     switch (h->slice_type) {
2200     case FF_P_TYPE:  return 0;
2201     case FF_B_TYPE:  return 1;
2202     case FF_I_TYPE:  return 2;
2203     case FF_SP_TYPE: return 3;
2204     case FF_SI_TYPE: return 4;
2205     default:         return -1;
2206     }
2207 }
2208
2209 /**
2210  *
2211  * @return non zero if the loop filter can be skiped
2212  */
2213 static int fill_filter_caches(H264Context *h, int mb_type){
2214     MpegEncContext * const s = &h->s;
2215     const int mb_xy= h->mb_xy;
2216     int top_xy, left_xy[2];
2217     int top_type, left_type[2];
2218
2219     top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
2220
2221     //FIXME deblocking could skip the intra and nnz parts.
2222
2223     /* Wow, what a mess, why didn't they simplify the interlacing & intra
2224      * stuff, I can't imagine that these complex rules are worth it. */
2225
2226     left_xy[1] = left_xy[0] = mb_xy-1;
2227     if(FRAME_MBAFF){
2228         const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
2229         const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
2230         if(s->mb_y&1){
2231             if (left_mb_field_flag != curr_mb_field_flag) {
2232                 left_xy[0] -= s->mb_stride;
2233             }
2234         }else{
2235             if(curr_mb_field_flag){
2236                 top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
2237             }
2238             if (left_mb_field_flag != curr_mb_field_flag) {
2239                 left_xy[1] += s->mb_stride;
2240             }
2241         }
2242     }
2243
2244     h->top_mb_xy = top_xy;
2245     h->left_mb_xy[0] = left_xy[0];
2246     h->left_mb_xy[1] = left_xy[1];
2247     {
2248         //for sufficiently low qp, filtering wouldn't do anything
2249         //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
2250         int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
2251         int qp = s->current_picture.qscale_table[mb_xy];
2252         if(qp <= qp_thresh
2253            && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
2254            && (top_xy   < 0 || ((qp + s->current_picture.qscale_table[top_xy    ] + 1)>>1) <= qp_thresh)){
2255             if(!FRAME_MBAFF)
2256                 return 1;
2257             if(   (left_xy[0]< 0            || ((qp + s->current_picture.qscale_table[left_xy[1]             ] + 1)>>1) <= qp_thresh)
2258                && (top_xy    < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy    -s->mb_stride] + 1)>>1) <= qp_thresh))
2259                 return 1;
2260         }
2261     }
2262
2263     top_type     = s->current_picture.mb_type[top_xy]    ;
2264     left_type[0] = s->current_picture.mb_type[left_xy[0]];
2265     left_type[1] = s->current_picture.mb_type[left_xy[1]];
2266     if(h->deblocking_filter == 2){
2267         if(h->slice_table[top_xy     ] != h->slice_num) top_type= 0;
2268         if(h->slice_table[left_xy[0] ] != h->slice_num) left_type[0]= left_type[1]= 0;
2269     }else{
2270         if(h->slice_table[top_xy     ] == 0xFFFF) top_type= 0;
2271         if(h->slice_table[left_xy[0] ] == 0xFFFF) left_type[0]= left_type[1] =0;
2272     }
2273     h->top_type    = top_type    ;
2274     h->left_type[0]= left_type[0];
2275     h->left_type[1]= left_type[1];
2276
2277     if(IS_INTRA(mb_type))
2278         return 0;
2279
2280     AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
2281     AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
2282     AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]);
2283     AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]);
2284     AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
2285
2286     h->cbp= h->cbp_table[mb_xy];
2287
2288     {
2289         int list;
2290         for(list=0; list<h->list_count; list++){
2291             int8_t *ref;
2292             int y, b_stride;
2293             int16_t (*mv_dst)[2];
2294             int16_t (*mv_src)[2];
2295
2296             if(!USES_LIST(mb_type, list)){
2297                 fill_rectangle(  h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
2298                 AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2299                 AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2300                 AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2301                 AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2302                 continue;
2303             }
2304
2305             ref = &s->current_picture.ref_index[list][4*mb_xy];
2306             {
2307                 int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2308                 AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2309                 AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2310                 ref += 2;
2311                 AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2312                 AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2313             }
2314
2315             b_stride = h->b_stride;
2316             mv_dst   = &h->mv_cache[list][scan8[0]];
2317             mv_src   = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
2318             for(y=0; y<4; y++){
2319                 AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
2320             }
2321
2322         }
2323     }
2324
2325
2326 /*
2327 0 . T T. T T T T
2328 1 L . .L . . . .
2329 2 L . .L . . . .
2330 3 . T TL . . . .
2331 4 L . .L . . . .
2332 5 L . .. . . . .
2333 */
2334 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
2335     if(top_type){
2336         AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
2337     }
2338
2339     if(left_type[0]){
2340         h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
2341         h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
2342         h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
2343         h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
2344     }
2345
2346     // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
2347     if(!CABAC && h->pps.transform_8x8_mode){
2348         if(IS_8x8DCT(top_type)){
2349             h->non_zero_count_cache[4+8*0]=
2350             h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
2351             h->non_zero_count_cache[6+8*0]=
2352             h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
2353         }
2354         if(IS_8x8DCT(left_type[0])){
2355             h->non_zero_count_cache[3+8*1]=
2356             h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
2357         }
2358         if(IS_8x8DCT(left_type[1])){
2359             h->non_zero_count_cache[3+8*3]=
2360             h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
2361         }
2362
2363         if(IS_8x8DCT(mb_type)){
2364             h->non_zero_count_cache[scan8[0   ]]= h->non_zero_count_cache[scan8[1   ]]=
2365             h->non_zero_count_cache[scan8[2   ]]= h->non_zero_count_cache[scan8[3   ]]= h->cbp & 1;
2366
2367             h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
2368             h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
2369
2370             h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
2371             h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
2372
2373             h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
2374             h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
2375         }
2376     }
2377
2378     if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
2379         int list;
2380         for(list=0; list<h->list_count; list++){
2381             if(USES_LIST(top_type, list)){
2382                 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
2383                 const int b8_xy= 4*top_xy + 2;
2384                 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2385                 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
2386                 h->ref_cache[list][scan8[0] + 0 - 1*8]=
2387                 h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
2388                 h->ref_cache[list][scan8[0] + 2 - 1*8]=
2389                 h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
2390             }else{
2391                 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
2392                 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2393             }
2394
2395             if(!IS_INTERLACED(mb_type^left_type[0])){
2396                 if(USES_LIST(left_type[0], list)){
2397                     const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
2398                     const int b8_xy= 4*left_xy[0] + 1;
2399                     int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2400                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
2401                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
2402                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
2403                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
2404                     h->ref_cache[list][scan8[0] - 1 + 0 ]=
2405                     h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]];
2406                     h->ref_cache[list][scan8[0] - 1 +16 ]=
2407                     h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]];
2408                 }else{
2409                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
2410                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
2411                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
2412                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
2413                     h->ref_cache[list][scan8[0] - 1 + 0  ]=
2414                     h->ref_cache[list][scan8[0] - 1 + 8  ]=
2415                     h->ref_cache[list][scan8[0] - 1 + 16 ]=
2416                     h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
2417                 }
2418             }
2419         }
2420     }
2421
2422     return 0;
2423 }
2424
2425 static void loop_filter(H264Context *h){
2426     MpegEncContext * const s = &h->s;
2427     uint8_t  *dest_y, *dest_cb, *dest_cr;
2428     int linesize, uvlinesize, mb_x, mb_y;
2429     const int end_mb_y= s->mb_y + FRAME_MBAFF;
2430     const int old_slice_type= h->slice_type;
2431
2432     if(h->deblocking_filter) {
2433         for(mb_x= 0; mb_x<s->mb_width; mb_x++){
2434             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
2435                 int mb_xy, mb_type;
2436                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2437                 h->slice_num= h->slice_table[mb_xy];
2438                 mb_type= s->current_picture.mb_type[mb_xy];
2439                 h->list_count= h->list_counts[mb_xy];
2440
2441                 if(FRAME_MBAFF)
2442                     h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2443
2444                 s->mb_x= mb_x;
2445                 s->mb_y= mb_y;
2446                 dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2447                 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2448                 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2449                     //FIXME simplify above
2450
2451                 if (MB_FIELD) {
2452                     linesize   = h->mb_linesize   = s->linesize * 2;
2453                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2454                     if(mb_y&1){ //FIXME move out of this function?
2455                         dest_y -= s->linesize*15;
2456                         dest_cb-= s->uvlinesize*7;
2457                         dest_cr-= s->uvlinesize*7;
2458                     }
2459                 } else {
2460                     linesize   = h->mb_linesize   = s->linesize;
2461                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2462                 }
2463                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2464                 if(fill_filter_caches(h, mb_type))
2465                     continue;
2466                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2467                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2468
2469                 if (FRAME_MBAFF) {
2470                     ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2471                 } else {
2472                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2473                 }
2474             }
2475         }
2476     }
2477     h->slice_type= old_slice_type;
2478     s->mb_x= 0;
2479     s->mb_y= end_mb_y - FRAME_MBAFF;
2480     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2481     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2482 }
2483
2484 static void predict_field_decoding_flag(H264Context *h){
2485     MpegEncContext * const s = &h->s;
2486     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2487     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
2488                 ? s->current_picture.mb_type[mb_xy-1]
2489                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
2490                 ? s->current_picture.mb_type[mb_xy-s->mb_stride]
2491                 : 0;
2492     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2493 }
2494
2495 static int decode_slice(struct AVCodecContext *avctx, void *arg){
2496     H264Context *h = *(void**)arg;
2497     MpegEncContext * const s = &h->s;
2498     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2499
2500     s->mb_skip_run= -1;
2501
2502     h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
2503                     (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
2504
2505     if( h->pps.cabac ) {
2506         /* realign */
2507         align_get_bits( &s->gb );
2508
2509         /* init cabac */
2510         ff_init_cabac_states( &h->cabac);
2511         ff_init_cabac_decoder( &h->cabac,
2512                                s->gb.buffer + get_bits_count(&s->gb)/8,
2513                                (get_bits_left(&s->gb) + 7)/8);
2514
2515         ff_h264_init_cabac_states(h);
2516
2517         for(;;){
2518 //START_TIMER
2519             int ret = ff_h264_decode_mb_cabac(h);
2520             int eos;
2521 //STOP_TIMER("decode_mb_cabac")
2522
2523             if(ret>=0) ff_h264_hl_decode_mb(h);
2524
2525             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
2526                 s->mb_y++;
2527
2528                 ret = ff_h264_decode_mb_cabac(h);
2529
2530                 if(ret>=0) ff_h264_hl_decode_mb(h);
2531                 s->mb_y--;
2532             }
2533             eos = get_cabac_terminate( &h->cabac );
2534
2535             if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
2536                 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);
2537                 return 0;
2538             }
2539             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2540                 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);
2541                 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);
2542                 return -1;
2543             }
2544
2545             if( ++s->mb_x >= s->mb_width ) {
2546                 s->mb_x = 0;
2547                 loop_filter(h);
2548                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2549                 ++s->mb_y;
2550                 if(FIELD_OR_MBAFF_PICTURE) {
2551                     ++s->mb_y;
2552                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2553                         predict_field_decoding_flag(h);
2554                 }
2555             }
2556
2557             if( eos || s->mb_y >= s->mb_height ) {
2558                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2559                 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);
2560                 return 0;
2561             }
2562         }
2563
2564     } else {
2565         for(;;){
2566             int ret = ff_h264_decode_mb_cavlc(h);
2567
2568             if(ret>=0) ff_h264_hl_decode_mb(h);
2569
2570             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
2571                 s->mb_y++;
2572                 ret = ff_h264_decode_mb_cavlc(h);
2573
2574                 if(ret>=0) ff_h264_hl_decode_mb(h);
2575                 s->mb_y--;
2576             }
2577
2578             if(ret<0){
2579                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2580                 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);
2581
2582                 return -1;
2583             }
2584
2585             if(++s->mb_x >= s->mb_width){
2586                 s->mb_x=0;
2587                 loop_filter(h);
2588                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2589                 ++s->mb_y;
2590                 if(FIELD_OR_MBAFF_PICTURE) {
2591                     ++s->mb_y;
2592                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2593                         predict_field_decoding_flag(h);
2594                 }
2595                 if(s->mb_y >= s->mb_height){
2596                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2597
2598                     if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
2599                         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);
2600
2601                         return 0;
2602                     }else{
2603                         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);
2604
2605                         return -1;
2606                     }
2607                 }
2608             }
2609
2610             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
2611                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2612                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2613                     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);
2614
2615                     return 0;
2616                 }else{
2617                     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);
2618
2619                     return -1;
2620                 }
2621             }
2622         }
2623     }
2624
2625 #if 0
2626     for(;s->mb_y < s->mb_height; s->mb_y++){
2627         for(;s->mb_x < s->mb_width; s->mb_x++){
2628             int ret= decode_mb(h);
2629
2630             ff_h264_hl_decode_mb(h);
2631
2632             if(ret<0){
2633                 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2634                 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);
2635
2636                 return -1;
2637             }
2638
2639             if(++s->mb_x >= s->mb_width){
2640                 s->mb_x=0;
2641                 if(++s->mb_y >= s->mb_height){
2642                     if(get_bits_count(s->gb) == s->gb.size_in_bits){
2643                         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);
2644
2645                         return 0;
2646                     }else{
2647                         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);
2648
2649                         return -1;
2650                     }
2651                 }
2652             }
2653
2654             if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2655                 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2656                     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);
2657
2658                     return 0;
2659                 }else{
2660                     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);
2661
2662                     return -1;
2663                 }
2664             }
2665         }
2666         s->mb_x=0;
2667         ff_draw_horiz_band(s, 16*s->mb_y, 16);
2668     }
2669 #endif
2670     return -1; //not reached
2671 }
2672
2673 /**
2674  * Call decode_slice() for each context.
2675  *
2676  * @param h h264 master context
2677  * @param context_count number of contexts to execute
2678  */
2679 static void execute_decode_slices(H264Context *h, int context_count){
2680     MpegEncContext * const s = &h->s;
2681     AVCodecContext * const avctx= s->avctx;
2682     H264Context *hx;
2683     int i;
2684
2685     if (s->avctx->hwaccel)
2686         return;
2687     if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2688         return;
2689     if(context_count == 1) {
2690         decode_slice(avctx, &h);
2691     } else {
2692         for(i = 1; i < context_count; i++) {
2693             hx = h->thread_context[i];
2694             hx->s.error_recognition = avctx->error_recognition;
2695             hx->s.error_count = 0;
2696         }
2697
2698         avctx->execute(avctx, (void *)decode_slice,
2699                        h->thread_context, NULL, context_count, sizeof(void*));
2700
2701         /* pull back stuff from slices to master context */
2702         hx = h->thread_context[context_count - 1];
2703         s->mb_x = hx->s.mb_x;
2704         s->mb_y = hx->s.mb_y;
2705         s->dropable = hx->s.dropable;
2706         s->picture_structure = hx->s.picture_structure;
2707         for(i = 1; i < context_count; i++)
2708             h->s.error_count += h->thread_context[i]->s.error_count;
2709     }
2710 }
2711
2712
2713 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
2714     MpegEncContext * const s = &h->s;
2715     AVCodecContext * const avctx= s->avctx;
2716     int buf_index=0;
2717     H264Context *hx; ///< thread context
2718     int context_count = 0;
2719     int next_avc= h->is_avc ? 0 : buf_size;
2720
2721     h->max_contexts = avctx->thread_count;
2722 #if 0
2723     int i;
2724     for(i=0; i<50; i++){
2725         av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
2726     }
2727 #endif
2728     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
2729         h->current_slice = 0;
2730         if (!s->first_field)
2731             s->current_picture_ptr= NULL;
2732         ff_h264_reset_sei(h);
2733     }
2734
2735     for(;;){
2736         int consumed;
2737         int dst_length;
2738         int bit_length;
2739         const uint8_t *ptr;
2740         int i, nalsize = 0;
2741         int err;
2742
2743         if(buf_index >= next_avc) {
2744             if(buf_index >= buf_size) break;
2745             nalsize = 0;
2746             for(i = 0; i < h->nal_length_size; i++)
2747                 nalsize = (nalsize << 8) | buf[buf_index++];
2748             if(nalsize <= 1 || nalsize > buf_size - buf_index){
2749                 if(nalsize == 1){
2750                     buf_index++;
2751                     continue;
2752                 }else{
2753                     av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2754                     break;
2755                 }
2756             }
2757             next_avc= buf_index + nalsize;
2758         } else {
2759             // start code prefix search
2760             for(; buf_index + 3 < next_avc; buf_index++){
2761                 // This should always succeed in the first iteration.
2762                 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2763                     break;
2764             }
2765
2766             if(buf_index+3 >= buf_size) break;
2767
2768             buf_index+=3;
2769             if(buf_index >= next_avc) continue;
2770         }
2771
2772         hx = h->thread_context[context_count];
2773
2774         ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
2775         if (ptr==NULL || dst_length < 0){
2776             return -1;
2777         }
2778         i= buf_index + consumed;
2779         if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
2780            buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
2781             s->workaround_bugs |= FF_BUG_TRUNCATED;
2782
2783         if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
2784         while(ptr[dst_length - 1] == 0 && dst_length > 0)
2785             dst_length--;
2786         }
2787         bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
2788
2789         if(s->avctx->debug&FF_DEBUG_STARTCODE){
2790             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);
2791         }
2792
2793         if (h->is_avc && (nalsize != consumed) && nalsize){
2794             av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
2795         }
2796
2797         buf_index += consumed;
2798
2799         if(  (s->hurry_up == 1 && h->nal_ref_idc  == 0) //FIXME do not discard SEI id
2800            ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0))
2801             continue;
2802
2803       again:
2804         err = 0;
2805         switch(hx->nal_unit_type){
2806         case NAL_IDR_SLICE:
2807             if (h->nal_unit_type != NAL_IDR_SLICE) {
2808                 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2809                 return -1;
2810             }
2811             idr(h); //FIXME ensure we don't loose some frames if there is reordering
2812         case NAL_SLICE:
2813             init_get_bits(&hx->s.gb, ptr, bit_length);
2814             hx->intra_gb_ptr=
2815             hx->inter_gb_ptr= &hx->s.gb;
2816             hx->s.data_partitioning = 0;
2817
2818             if((err = decode_slice_header(hx, h)))
2819                break;
2820
2821             avctx->profile = hx->sps.profile_idc;
2822             avctx->level   = hx->sps.level_idc;
2823
2824             if (h->current_slice == 1) {
2825                 if (s->avctx->hwaccel && s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
2826                     return -1;
2827                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2828                     ff_vdpau_h264_picture_start(s);
2829             }
2830
2831             s->current_picture_ptr->key_frame |=
2832                     (hx->nal_unit_type == NAL_IDR_SLICE) ||
2833                     (h->sei_recovery_frame_cnt >= 0);
2834             if(hx->redundant_pic_count==0 && hx->s.hurry_up < 5
2835                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2836                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2837                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2838                && avctx->skip_frame < AVDISCARD_ALL){
2839                 if(avctx->hwaccel) {
2840                     if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
2841                         return -1;
2842                 }else
2843                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
2844                     static const uint8_t start_code[] = {0x00, 0x00, 0x01};
2845                     ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
2846                     ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
2847                 }else
2848                     context_count++;
2849             }
2850             break;
2851         case NAL_DPA:
2852             init_get_bits(&hx->s.gb, ptr, bit_length);
2853             hx->intra_gb_ptr=
2854             hx->inter_gb_ptr= NULL;
2855
2856             if ((err = decode_slice_header(hx, h)) < 0)
2857                 break;
2858
2859             avctx->profile = hx->sps.profile_idc;
2860             avctx->level   = hx->sps.level_idc;
2861
2862             hx->s.data_partitioning = 1;
2863
2864             break;
2865         case NAL_DPB:
2866             init_get_bits(&hx->intra_gb, ptr, bit_length);
2867             hx->intra_gb_ptr= &hx->intra_gb;
2868             break;
2869         case NAL_DPC:
2870             init_get_bits(&hx->inter_gb, ptr, bit_length);
2871             hx->inter_gb_ptr= &hx->inter_gb;
2872
2873             if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
2874                && s->context_initialized
2875                && s->hurry_up < 5
2876                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2877                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2878                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2879                && avctx->skip_frame < AVDISCARD_ALL)
2880                 context_count++;
2881             break;
2882         case NAL_SEI:
2883             init_get_bits(&s->gb, ptr, bit_length);
2884             ff_h264_decode_sei(h);
2885             break;
2886         case NAL_SPS:
2887             init_get_bits(&s->gb, ptr, bit_length);
2888             ff_h264_decode_seq_parameter_set(h);
2889
2890             if(s->flags& CODEC_FLAG_LOW_DELAY)
2891                 s->low_delay=1;
2892
2893             if(avctx->has_b_frames < 2)
2894                 avctx->has_b_frames= !s->low_delay;
2895             break;
2896         case NAL_PPS:
2897             init_get_bits(&s->gb, ptr, bit_length);
2898
2899             ff_h264_decode_picture_parameter_set(h, bit_length);
2900
2901             break;
2902         case NAL_AUD:
2903         case NAL_END_SEQUENCE:
2904         case NAL_END_STREAM:
2905         case NAL_FILLER_DATA:
2906         case NAL_SPS_EXT:
2907         case NAL_AUXILIARY_SLICE:
2908             break;
2909         default:
2910             av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
2911         }
2912
2913         if(context_count == h->max_contexts) {
2914             execute_decode_slices(h, context_count);
2915             context_count = 0;
2916         }
2917
2918         if (err < 0)
2919             av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
2920         else if(err == 1) {
2921             /* Slice could not be decoded in parallel mode, copy down
2922              * NAL unit stuff to context 0 and restart. Note that
2923              * rbsp_buffer is not transferred, but since we no longer
2924              * run in parallel mode this should not be an issue. */
2925             h->nal_unit_type = hx->nal_unit_type;
2926             h->nal_ref_idc   = hx->nal_ref_idc;
2927             hx = h;
2928             goto again;
2929         }
2930     }
2931     if(context_count)
2932         execute_decode_slices(h, context_count);
2933     return buf_index;
2934 }
2935
2936 /**
2937  * returns the number of bytes consumed for building the current frame
2938  */
2939 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
2940         if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
2941         if(pos+10>buf_size) pos=buf_size; // oops ;)
2942
2943         return pos;
2944 }
2945
2946 static int decode_frame(AVCodecContext *avctx,
2947                              void *data, int *data_size,
2948                              AVPacket *avpkt)
2949 {
2950     const uint8_t *buf = avpkt->data;
2951     int buf_size = avpkt->size;
2952     H264Context *h = avctx->priv_data;
2953     MpegEncContext *s = &h->s;
2954     AVFrame *pict = data;
2955     int buf_index;
2956
2957     s->flags= avctx->flags;
2958     s->flags2= avctx->flags2;
2959
2960    /* end of stream, output what is still in the buffers */
2961     if (buf_size == 0) {
2962         Picture *out;
2963         int i, out_idx;
2964
2965 //FIXME factorize this with the output code below
2966         out = h->delayed_pic[0];
2967         out_idx = 0;
2968         for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2969             if(h->delayed_pic[i]->poc < out->poc){
2970                 out = h->delayed_pic[i];
2971                 out_idx = i;
2972             }
2973
2974         for(i=out_idx; h->delayed_pic[i]; i++)
2975             h->delayed_pic[i] = h->delayed_pic[i+1];
2976
2977         if(out){
2978             *data_size = sizeof(AVFrame);
2979             *pict= *(AVFrame*)out;
2980         }
2981
2982         return 0;
2983     }
2984
2985     buf_index=decode_nal_units(h, buf, buf_size);
2986     if(buf_index < 0)
2987         return -1;
2988
2989     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
2990         if (avctx->skip_frame >= AVDISCARD_NONREF || s->hurry_up) return 0;
2991         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
2992         return -1;
2993     }
2994
2995     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
2996         Picture *out = s->current_picture_ptr;
2997         Picture *cur = s->current_picture_ptr;
2998         int i, pics, out_of_order, out_idx;
2999
3000         field_end(h);
3001
3002         if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
3003             /* Wait for second field. */
3004             *data_size = 0;
3005
3006         } else {
3007             cur->interlaced_frame = 0;
3008             cur->repeat_pict = 0;
3009
3010             /* Signal interlacing information externally. */
3011             /* Prioritize picture timing SEI information over used decoding process if it exists. */
3012
3013             if(h->sps.pic_struct_present_flag){
3014                 switch (h->sei_pic_struct)
3015                 {
3016                 case SEI_PIC_STRUCT_FRAME:
3017                     break;
3018                 case SEI_PIC_STRUCT_TOP_FIELD:
3019                 case SEI_PIC_STRUCT_BOTTOM_FIELD:
3020                     cur->interlaced_frame = 1;
3021                     break;
3022                 case SEI_PIC_STRUCT_TOP_BOTTOM:
3023                 case SEI_PIC_STRUCT_BOTTOM_TOP:
3024                     if (FIELD_OR_MBAFF_PICTURE)
3025                         cur->interlaced_frame = 1;
3026                     else
3027                         // try to flag soft telecine progressive
3028                         cur->interlaced_frame = h->prev_interlaced_frame;
3029                     break;
3030                 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
3031                 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
3032                     // Signal the possibility of telecined film externally (pic_struct 5,6)
3033                     // From these hints, let the applications decide if they apply deinterlacing.
3034                     cur->repeat_pict = 1;
3035                     break;
3036                 case SEI_PIC_STRUCT_FRAME_DOUBLING:
3037                     // Force progressive here, as doubling interlaced frame is a bad idea.
3038                     cur->repeat_pict = 2;
3039                     break;
3040                 case SEI_PIC_STRUCT_FRAME_TRIPLING:
3041                     cur->repeat_pict = 4;
3042                     break;
3043                 }
3044
3045                 if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
3046                     cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
3047             }else{
3048                 /* Derive interlacing flag from used decoding process. */
3049                 cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
3050             }
3051             h->prev_interlaced_frame = cur->interlaced_frame;
3052
3053             if (cur->field_poc[0] != cur->field_poc[1]){
3054                 /* Derive top_field_first from field pocs. */
3055                 cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
3056             }else{
3057                 if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
3058                     /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
3059                     if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
3060                       || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
3061                         cur->top_field_first = 1;
3062                     else
3063                         cur->top_field_first = 0;
3064                 }else{
3065                     /* Most likely progressive */
3066                     cur->top_field_first = 0;
3067                 }
3068             }
3069
3070         //FIXME do something with unavailable reference frames
3071
3072             /* Sort B-frames into display order */
3073
3074             if(h->sps.bitstream_restriction_flag
3075                && s->avctx->has_b_frames < h->sps.num_reorder_frames){
3076                 s->avctx->has_b_frames = h->sps.num_reorder_frames;
3077                 s->low_delay = 0;
3078             }
3079
3080             if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
3081                && !h->sps.bitstream_restriction_flag){
3082                 s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
3083                 s->low_delay= 0;
3084             }
3085
3086             pics = 0;
3087             while(h->delayed_pic[pics]) pics++;
3088
3089             assert(pics <= MAX_DELAYED_PIC_COUNT);
3090
3091             h->delayed_pic[pics++] = cur;
3092             if(cur->reference == 0)
3093                 cur->reference = DELAYED_PIC_REF;
3094
3095             out = h->delayed_pic[0];
3096             out_idx = 0;
3097             for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
3098                 if(h->delayed_pic[i]->poc < out->poc){
3099                     out = h->delayed_pic[i];
3100                     out_idx = i;
3101                 }
3102             if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
3103                 h->outputed_poc= INT_MIN;
3104             out_of_order = out->poc < h->outputed_poc;
3105
3106             if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
3107                 { }
3108             else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
3109                || (s->low_delay &&
3110                 ((h->outputed_poc != INT_MIN && out->poc > h->outputed_poc + 2)
3111                  || cur->pict_type == FF_B_TYPE)))
3112             {
3113                 s->low_delay = 0;
3114                 s->avctx->has_b_frames++;
3115             }
3116
3117             if(out_of_order || pics > s->avctx->has_b_frames){
3118                 out->reference &= ~DELAYED_PIC_REF;
3119                 for(i=out_idx; h->delayed_pic[i]; i++)
3120                     h->delayed_pic[i] = h->delayed_pic[i+1];
3121             }
3122             if(!out_of_order && pics > s->avctx->has_b_frames){
3123                 *data_size = sizeof(AVFrame);
3124
3125                 if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
3126                     h->outputed_poc = INT_MIN;
3127                 } else
3128                     h->outputed_poc = out->poc;
3129                 *pict= *(AVFrame*)out;
3130             }else{
3131                 av_log(avctx, AV_LOG_DEBUG, "no picture\n");
3132             }
3133         }
3134     }
3135
3136     assert(pict->data[0] || !*data_size);
3137     ff_print_debug_info(s, pict);
3138 //printf("out %d\n", (int)pict->data[0]);
3139
3140     return get_consumed_bytes(s, buf_index, buf_size);
3141 }
3142 #if 0
3143 static inline void fill_mb_avail(H264Context *h){
3144     MpegEncContext * const s = &h->s;
3145     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3146
3147     if(s->mb_y){
3148         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
3149         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
3150         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
3151     }else{
3152         h->mb_avail[0]=
3153         h->mb_avail[1]=
3154         h->mb_avail[2]= 0;
3155     }
3156     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
3157     h->mb_avail[4]= 1; //FIXME move out
3158     h->mb_avail[5]= 0; //FIXME move out
3159 }
3160 #endif
3161
3162 #ifdef TEST
3163 #undef printf
3164 #undef random
3165 #define COUNT 8000
3166 #define SIZE (COUNT*40)
3167 int main(void){
3168     int i;
3169     uint8_t temp[SIZE];
3170     PutBitContext pb;
3171     GetBitContext gb;
3172 //    int int_temp[10000];
3173     DSPContext dsp;
3174     AVCodecContext avctx;
3175
3176     dsputil_init(&dsp, &avctx);
3177
3178     init_put_bits(&pb, temp, SIZE);
3179     printf("testing unsigned exp golomb\n");
3180     for(i=0; i<COUNT; i++){
3181         START_TIMER
3182         set_ue_golomb(&pb, i);
3183         STOP_TIMER("set_ue_golomb");
3184     }
3185     flush_put_bits(&pb);
3186
3187     init_get_bits(&gb, temp, 8*SIZE);
3188     for(i=0; i<COUNT; i++){
3189         int j, s;
3190
3191         s= show_bits(&gb, 24);
3192
3193         START_TIMER
3194         j= get_ue_golomb(&gb);
3195         if(j != i){
3196             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
3197 //            return -1;
3198         }
3199         STOP_TIMER("get_ue_golomb");
3200     }
3201
3202
3203     init_put_bits(&pb, temp, SIZE);
3204     printf("testing signed exp golomb\n");
3205     for(i=0; i<COUNT; i++){
3206         START_TIMER
3207         set_se_golomb(&pb, i - COUNT/2);
3208         STOP_TIMER("set_se_golomb");
3209     }
3210     flush_put_bits(&pb);
3211
3212     init_get_bits(&gb, temp, 8*SIZE);
3213     for(i=0; i<COUNT; i++){
3214         int j, s;
3215
3216         s= show_bits(&gb, 24);
3217
3218         START_TIMER
3219         j= get_se_golomb(&gb);
3220         if(j != i - COUNT/2){
3221             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
3222 //            return -1;
3223         }
3224         STOP_TIMER("get_se_golomb");
3225     }
3226
3227 #if 0
3228     printf("testing 4x4 (I)DCT\n");
3229
3230     DCTELEM block[16];
3231     uint8_t src[16], ref[16];
3232     uint64_t error= 0, max_error=0;
3233
3234     for(i=0; i<COUNT; i++){
3235         int j;
3236 //        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
3237         for(j=0; j<16; j++){
3238             ref[j]= random()%255;
3239             src[j]= random()%255;
3240         }
3241
3242         h264_diff_dct_c(block, src, ref, 4);
3243
3244         //normalize
3245         for(j=0; j<16; j++){
3246 //            printf("%d ", block[j]);
3247             block[j]= block[j]*4;
3248             if(j&1) block[j]= (block[j]*4 + 2)/5;
3249             if(j&4) block[j]= (block[j]*4 + 2)/5;
3250         }
3251 //        printf("\n");
3252
3253         h->h264dsp.h264_idct_add(ref, block, 4);
3254 /*        for(j=0; j<16; j++){
3255             printf("%d ", ref[j]);
3256         }
3257         printf("\n");*/
3258
3259         for(j=0; j<16; j++){
3260             int diff= FFABS(src[j] - ref[j]);
3261
3262             error+= diff*diff;
3263             max_error= FFMAX(max_error, diff);
3264         }
3265     }
3266     printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
3267     printf("testing quantizer\n");
3268     for(qp=0; qp<52; qp++){
3269         for(i=0; i<16; i++)
3270             src1_block[i]= src2_block[i]= random()%255;
3271
3272     }
3273     printf("Testing NAL layer\n");
3274
3275     uint8_t bitstream[COUNT];
3276     uint8_t nal[COUNT*2];
3277     H264Context h;
3278     memset(&h, 0, sizeof(H264Context));
3279
3280     for(i=0; i<COUNT; i++){
3281         int zeros= i;
3282         int nal_length;
3283         int consumed;
3284         int out_length;
3285         uint8_t *out;
3286         int j;
3287
3288         for(j=0; j<COUNT; j++){
3289             bitstream[j]= (random() % 255) + 1;
3290         }
3291
3292         for(j=0; j<zeros; j++){
3293             int pos= random() % COUNT;
3294             while(bitstream[pos] == 0){
3295                 pos++;
3296                 pos %= COUNT;
3297             }
3298             bitstream[pos]=0;
3299         }
3300
3301         START_TIMER
3302
3303         nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
3304         if(nal_length<0){
3305             printf("encoding failed\n");
3306             return -1;
3307         }
3308
3309         out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
3310
3311         STOP_TIMER("NAL")
3312
3313         if(out_length != COUNT){
3314             printf("incorrect length %d %d\n", out_length, COUNT);
3315             return -1;
3316         }
3317
3318         if(consumed != nal_length){
3319             printf("incorrect consumed length %d %d\n", nal_length, consumed);
3320             return -1;
3321         }
3322
3323         if(memcmp(bitstream, out, COUNT)){
3324             printf("mismatch\n");
3325             return -1;
3326         }
3327     }
3328 #endif
3329
3330     printf("Testing RBSP\n");
3331
3332
3333     return 0;
3334 }
3335 #endif /* TEST */
3336
3337
3338 av_cold void ff_h264_free_context(H264Context *h)
3339 {
3340     int i;
3341
3342     free_tables(h); //FIXME cleanup init stuff perhaps
3343
3344     for(i = 0; i < MAX_SPS_COUNT; i++)
3345         av_freep(h->sps_buffers + i);
3346
3347     for(i = 0; i < MAX_PPS_COUNT; i++)
3348         av_freep(h->pps_buffers + i);
3349 }
3350
3351 av_cold int ff_h264_decode_end(AVCodecContext *avctx)
3352 {
3353     H264Context *h = avctx->priv_data;
3354     MpegEncContext *s = &h->s;
3355
3356     ff_h264_free_context(h);
3357
3358     MPV_common_end(s);
3359
3360 //    memset(h, 0, sizeof(H264Context));
3361
3362     return 0;
3363 }
3364
3365
3366 AVCodec h264_decoder = {
3367     "h264",
3368     AVMEDIA_TYPE_VIDEO,
3369     CODEC_ID_H264,
3370     sizeof(H264Context),
3371     ff_h264_decode_init,
3372     NULL,
3373     ff_h264_decode_end,
3374     decode_frame,
3375     /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3376     .flush= flush_dpb,
3377     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
3378     .pix_fmts= ff_hwaccel_pixfmt_list_420,
3379 };
3380
3381 #if CONFIG_H264_VDPAU_DECODER
3382 AVCodec h264_vdpau_decoder = {
3383     "h264_vdpau",
3384     AVMEDIA_TYPE_VIDEO,
3385     CODEC_ID_H264,
3386     sizeof(H264Context),
3387     ff_h264_decode_init,
3388     NULL,
3389     ff_h264_decode_end,
3390     decode_frame,
3391     CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3392     .flush= flush_dpb,
3393     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
3394     .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
3395 };
3396 #endif