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
  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 static const enum PixelFormat hwaccel_pixfmt_list_h264_jpeg_420[] = {
  55     PIX_FMT_DXVA2_VLD,
  56     PIX_FMT_VAAPI_VLD,
  57     PIX_FMT_YUVJ420P,
  58     PIX_FMT_NONE
  59 };
  60
  61 void ff_h264_write_back_intra_pred_mode(H264Context *h){
  62     int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
  63
  64     AV_COPY32(mode, h->intra4x4_pred_mode_cache + 4 + 8*4);
  65     mode[4]= h->intra4x4_pred_mode_cache[7+8*3];
  66     mode[5]= h->intra4x4_pred_mode_cache[7+8*2];
  67     mode[6]= h->intra4x4_pred_mode_cache[7+8*1];
  68 }
  69
  70 /**
  71  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
  72  */
  73 int ff_h264_check_intra4x4_pred_mode(H264Context *h){
  74     MpegEncContext * const s = &h->s;
  75     static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
  76     static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
  77     int i;
  78
  79     if(!(h->top_samples_available&0x8000)){
  80         for(i=0; i<4; i++){
  81             int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
  82             if(status<0){
  83                 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);
  84                 return -1;
  85             } else if(status){
  86                 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
  87             }
  88         }
  89     }
  90
  91     if((h->left_samples_available&0x8888)!=0x8888){
  92         static const int mask[4]={0x8000,0x2000,0x80,0x20};
  93         for(i=0; i<4; i++){
  94             if(!(h->left_samples_available&mask[i])){
  95                 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
  96                 if(status<0){
  97                     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);
  98                     return -1;
  99                 } else if(status){
 100                     h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
 101                 }
 102             }
 103         }
 104     }
 105
 106     return 0;
 107 } //FIXME cleanup like ff_h264_check_intra_pred_mode
 108
 109 /**
 110  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
 111  */
 112 int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
 113     MpegEncContext * const s = &h->s;
 114     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
 115     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
 116
 117     if(mode > 6U) {
 118         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);
 119         return -1;
 120     }
 121
 122     if(!(h->top_samples_available&0x8000)){
 123         mode= top[ mode ];
 124         if(mode<0){
 125             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);
 126             return -1;
 127         }
 128     }
 129
 130     if((h->left_samples_available&0x8080) != 0x8080){
 131         mode= left[ mode ];
 132         if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
 133             mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
 134         }
 135         if(mode<0){
 136             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);
 137             return -1;
 138         }
 139     }
 140
 141     return mode;
 142 }
 143
 144 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
 145     int i, si, di;
 146     uint8_t *dst;
 147     int bufidx;
 148
 149 //    src[0]&0x80;                //forbidden bit
 150     h->nal_ref_idc= src[0]>>5;
 151     h->nal_unit_type= src[0]&0x1F;
 152
 153     src++; length--;
 154 #if 0
 155     for(i=0; i<length; i++)
 156         printf("%2X ", src[i]);
 157 #endif
 158
 159 #if HAVE_FAST_UNALIGNED
 160 # if HAVE_FAST_64BIT
 161 #   define RS 7
 162     for(i=0; i+1<length; i+=9){
 163         if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
 164 # else
 165 #   define RS 3
 166     for(i=0; i+1<length; i+=5){
 167         if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
 168 # endif
 169             continue;
 170         if(i>0 && !src[i]) i--;
 171         while(src[i]) i++;
 172 #else
 173 #   define RS 0
 174     for(i=0; i+1<length; i+=2){
 175         if(src[i]) continue;
 176         if(i>0 && src[i-1]==0) i--;
 177 #endif
 178         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
 179             if(src[i+2]!=3){
 180                 /* startcode, so we must be past the end */
 181                 length=i;
 182             }
 183             break;
 184         }
 185         i-= RS;
 186     }
 187
 188     if(i>=length-1){ //no escaped 0
 189         *dst_length= length;
 190         *consumed= length+1; //+1 for the header
 191         return src;
 192     }
 193
 194     bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
 195     av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
 196     dst= h->rbsp_buffer[bufidx];
 197
 198     if (dst == NULL){
 199         return NULL;
 200     }
 201
 202 //printf("decoding esc\n");
 203     memcpy(dst, src, i);
 204     si=di=i;
 205     while(si+2<length){
 206         //remove escapes (very rare 1:2^22)
 207         if(src[si+2]>3){
 208             dst[di++]= src[si++];
 209             dst[di++]= src[si++];
 210         }else if(src[si]==0 && src[si+1]==0){
 211             if(src[si+2]==3){ //escape
 212                 dst[di++]= 0;
 213                 dst[di++]= 0;
 214                 si+=3;
 215                 continue;
 216             }else //next start code
 217                 goto nsc;
 218         }
 219
 220         dst[di++]= src[si++];
 221     }
 222     while(si<length)
 223         dst[di++]= src[si++];
 224 nsc:
 225
 226     memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
 227
 228     *dst_length= di;
 229     *consumed= si + 1;//+1 for the header
 230 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
 231     return dst;
 232 }
 233
 234 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
 235     int v= *src;
 236     int r;
 237
 238     tprintf(h->s.avctx, "rbsp trailing %X\n", v);
 239
 240     for(r=1; r<9; r++){
 241         if(v&1) return r;
 242         v>>=1;
 243     }
 244     return 0;
 245 }
 246
 247 /**
 248  * IDCT transforms the 16 dc values and dequantizes them.
 249  * @param qp quantization parameter
 250  */
 251 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
 252 #define stride 16
 253     int i;
 254     int temp[16]; //FIXME check if this is a good idea
 255     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
 256     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
 257
 258 //memset(block, 64, 2*256);
 259 //return;
 260     for(i=0; i<4; i++){
 261         const int offset= y_offset[i];
 262         const int z0= block[offset+stride*0] + block[offset+stride*4];
 263         const int z1= block[offset+stride*0] - block[offset+stride*4];
 264         const int z2= block[offset+stride*1] - block[offset+stride*5];
 265         const int z3= block[offset+stride*1] + block[offset+stride*5];
 266
 267         temp[4*i+0]= z0+z3;
 268         temp[4*i+1]= z1+z2;
 269         temp[4*i+2]= z1-z2;
 270         temp[4*i+3]= z0-z3;
 271     }
 272
 273     for(i=0; i<4; i++){
 274         const int offset= x_offset[i];
 275         const int z0= temp[4*0+i] + temp[4*2+i];
 276         const int z1= temp[4*0+i] - temp[4*2+i];
 277         const int z2= temp[4*1+i] - temp[4*3+i];
 278         const int z3= temp[4*1+i] + temp[4*3+i];
 279
 280         block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
 281         block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
 282         block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
 283         block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
 284     }
 285 }
 286
 287 #if 0
 288 /**
 289  * DCT transforms the 16 dc values.
 290  * @param qp quantization parameter ??? FIXME
 291  */
 292 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
 293 //    const int qmul= dequant_coeff[qp][0];
 294     int i;
 295     int temp[16]; //FIXME check if this is a good idea
 296     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
 297     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
 298
 299     for(i=0; i<4; i++){
 300         const int offset= y_offset[i];
 301         const int z0= block[offset+stride*0] + block[offset+stride*4];
 302         const int z1= block[offset+stride*0] - block[offset+stride*4];
 303         const int z2= block[offset+stride*1] - block[offset+stride*5];
 304         const int z3= block[offset+stride*1] + block[offset+stride*5];
 305
 306         temp[4*i+0]= z0+z3;
 307         temp[4*i+1]= z1+z2;
 308         temp[4*i+2]= z1-z2;
 309         temp[4*i+3]= z0-z3;
 310     }
 311
 312     for(i=0; i<4; i++){
 313         const int offset= x_offset[i];
 314         const int z0= temp[4*0+i] + temp[4*2+i];
 315         const int z1= temp[4*0+i] - temp[4*2+i];
 316         const int z2= temp[4*1+i] - temp[4*3+i];
 317         const int z3= temp[4*1+i] + temp[4*3+i];
 318
 319         block[stride*0 +offset]= (z0 + z3)>>1;
 320         block[stride*2 +offset]= (z1 + z2)>>1;
 321         block[stride*8 +offset]= (z1 - z2)>>1;
 322         block[stride*10+offset]= (z0 - z3)>>1;
 323     }
 324 }
 325 #endif
 326
 327 #undef xStride
 328 #undef stride
 329
 330 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
 331     const int stride= 16*2;
 332     const int xStride= 16;
 333     int a,b,c,d,e;
 334
 335     a= block[stride*0 + xStride*0];
 336     b= block[stride*0 + xStride*1];
 337     c= block[stride*1 + xStride*0];
 338     d= block[stride*1 + xStride*1];
 339
 340     e= a-b;
 341     a= a+b;
 342     b= c-d;
 343     c= c+d;
 344
 345     block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
 346     block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
 347     block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
 348     block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
 349 }
 350
 351 #if 0
 352 static void chroma_dc_dct_c(DCTELEM *block){
 353     const int stride= 16*2;
 354     const int xStride= 16;
 355     int a,b,c,d,e;
 356
 357     a= block[stride*0 + xStride*0];
 358     b= block[stride*0 + xStride*1];
 359     c= block[stride*1 + xStride*0];
 360     d= block[stride*1 + xStride*1];
 361
 362     e= a-b;
 363     a= a+b;
 364     b= c-d;
 365     c= c+d;
 366
 367     block[stride*0 + xStride*0]= (a+c);
 368     block[stride*0 + xStride*1]= (e+b);
 369     block[stride*1 + xStride*0]= (a-c);
 370     block[stride*1 + xStride*1]= (e-b);
 371 }
 372 #endif
 373
 374 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
 375                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 376                            int src_x_offset, int src_y_offset,
 377                            qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
 378     MpegEncContext * const s = &h->s;
 379     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
 380     int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
 381     const int luma_xy= (mx&3) + ((my&3)<<2);
 382     uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
 383     uint8_t * src_cb, * src_cr;
 384     int extra_width= h->emu_edge_width;
 385     int extra_height= h->emu_edge_height;
 386     int emu=0;
 387     const int full_mx= mx>>2;
 388     const int full_my= my>>2;
 389     const int pic_width  = 16*s->mb_width;
 390     const int pic_height = 16*s->mb_height >> MB_FIELD;
 391
 392     if(mx&7) extra_width -= 3;
 393     if(my&7) extra_height -= 3;
 394
 395     if(   full_mx < 0-extra_width
 396        || full_my < 0-extra_height
 397        || full_mx + 16/*FIXME*/ > pic_width + extra_width
 398        || full_my + 16/*FIXME*/ > pic_height + extra_height){
 399         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);
 400             src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
 401         emu=1;
 402     }
 403
 404     qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
 405     if(!square){
 406         qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
 407     }
 408
 409     if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
 410
 411     if(MB_FIELD){
 412         // chroma offset when predicting from a field of opposite parity
 413         my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
 414         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
 415     }
 416     src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
 417     src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
 418
 419     if(emu){
 420         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);
 421             src_cb= s->edge_emu_buffer;
 422     }
 423     chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
 424
 425     if(emu){
 426         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);
 427             src_cr= s->edge_emu_buffer;
 428     }
 429     chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
 430 }
 431
 432 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
 433                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 434                            int x_offset, int y_offset,
 435                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 436                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
 437                            int list0, int list1){
 438     MpegEncContext * const s = &h->s;
 439     qpel_mc_func *qpix_op=  qpix_put;
 440     h264_chroma_mc_func chroma_op= chroma_put;
 441
 442     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
 443     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
 444     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
 445     x_offset += 8*s->mb_x;
 446     y_offset += 8*(s->mb_y >> MB_FIELD);
 447
 448     if(list0){
 449         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
 450         mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
 451                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
 452                            qpix_op, chroma_op);
 453
 454         qpix_op=  qpix_avg;
 455         chroma_op= chroma_avg;
 456     }
 457
 458     if(list1){
 459         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
 460         mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
 461                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
 462                            qpix_op, chroma_op);
 463     }
 464 }
 465
 466 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
 467                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 468                            int x_offset, int y_offset,
 469                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 470                            h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
 471                            h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
 472                            int list0, int list1){
 473     MpegEncContext * const s = &h->s;
 474
 475     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
 476     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
 477     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
 478     x_offset += 8*s->mb_x;
 479     y_offset += 8*(s->mb_y >> MB_FIELD);
 480
 481     if(list0 && list1){
 482         /* don't optimize for luma-only case, since B-frames usually
 483          * use implicit weights => chroma too. */
 484         uint8_t *tmp_cb = s->obmc_scratchpad;
 485         uint8_t *tmp_cr = s->obmc_scratchpad + 8;
 486         uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
 487         int refn0 = h->ref_cache[0][ scan8[n] ];
 488         int refn1 = h->ref_cache[1][ scan8[n] ];
 489
 490         mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
 491                     dest_y, dest_cb, dest_cr,
 492                     x_offset, y_offset, qpix_put, chroma_put);
 493         mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
 494                     tmp_y, tmp_cb, tmp_cr,
 495                     x_offset, y_offset, qpix_put, chroma_put);
 496
 497         if(h->use_weight == 2){
 498             int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
 499             int weight1 = 64 - weight0;
 500             luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
 501             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
 502             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
 503         }else{
 504             luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
 505                             h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
 506                             h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
 507             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 508                             h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
 509                             h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
 510             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 511                             h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
 512                             h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
 513         }
 514     }else{
 515         int list = list1 ? 1 : 0;
 516         int refn = h->ref_cache[list][ scan8[n] ];
 517         Picture *ref= &h->ref_list[list][refn];
 518         mc_dir_part(h, ref, n, square, chroma_height, delta, list,
 519                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
 520                     qpix_put, chroma_put);
 521
 522         luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
 523                        h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
 524         if(h->use_weight_chroma){
 525             chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 526                              h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
 527             chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 528                              h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
 529         }
 530     }
 531 }
 532
 533 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
 534                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 535                            int x_offset, int y_offset,
 536                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 537                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
 538                            h264_weight_func *weight_op, h264_biweight_func *weight_avg,
 539                            int list0, int list1){
 540     if((h->use_weight==2 && list0 && list1
 541         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
 542        || h->use_weight==1)
 543         mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
 544                          x_offset, y_offset, qpix_put, chroma_put,
 545                          weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
 546     else
 547         mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
 548                     x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
 549 }
 550
 551 static inline void prefetch_motion(H264Context *h, int list){
 552     /* fetch pixels for estimated mv 4 macroblocks ahead
 553      * optimized for 64byte cache lines */
 554     MpegEncContext * const s = &h->s;
 555     const int refn = h->ref_cache[list][scan8[0]];
 556     if(refn >= 0){
 557         const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
 558         const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
 559         uint8_t **src= h->ref_list[list][refn].data;
 560         int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
 561         s->dsp.prefetch(src[0]+off, s->linesize, 4);
 562         off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
 563         s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
 564     }
 565 }
 566
 567 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 568                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
 569                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
 570                       h264_weight_func *weight_op, h264_biweight_func *weight_avg){
 571     MpegEncContext * const s = &h->s;
 572     const int mb_xy= h->mb_xy;
 573     const int mb_type= s->current_picture.mb_type[mb_xy];
 574
 575     assert(IS_INTER(mb_type));
 576
 577     prefetch_motion(h, 0);
 578
 579     if(IS_16X16(mb_type)){
 580         mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
 581                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
 582                 weight_op, weight_avg,
 583                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
 584     }else if(IS_16X8(mb_type)){
 585         mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
 586                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
 587                 &weight_op[1], &weight_avg[1],
 588                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
 589         mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
 590                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
 591                 &weight_op[1], &weight_avg[1],
 592                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
 593     }else if(IS_8X16(mb_type)){
 594         mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
 595                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 596                 &weight_op[2], &weight_avg[2],
 597                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
 598         mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
 599                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 600                 &weight_op[2], &weight_avg[2],
 601                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
 602     }else{
 603         int i;
 604
 605         assert(IS_8X8(mb_type));
 606
 607         for(i=0; i<4; i++){
 608             const int sub_mb_type= h->sub_mb_type[i];
 609             const int n= 4*i;
 610             int x_offset= (i&1)<<2;
 611             int y_offset= (i&2)<<1;
 612
 613             if(IS_SUB_8X8(sub_mb_type)){
 614                 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 615                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 616                     &weight_op[3], &weight_avg[3],
 617                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 618             }else if(IS_SUB_8X4(sub_mb_type)){
 619                 mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 620                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
 621                     &weight_op[4], &weight_avg[4],
 622                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 623                 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
 624                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
 625                     &weight_op[4], &weight_avg[4],
 626                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 627             }else if(IS_SUB_4X8(sub_mb_type)){
 628                 mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 629                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 630                     &weight_op[5], &weight_avg[5],
 631                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 632                 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
 633                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 634                     &weight_op[5], &weight_avg[5],
 635                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 636             }else{
 637                 int j;
 638                 assert(IS_SUB_4X4(sub_mb_type));
 639                 for(j=0; j<4; j++){
 640                     int sub_x_offset= x_offset + 2*(j&1);
 641                     int sub_y_offset= y_offset +   (j&2);
 642                     mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
 643                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 644                         &weight_op[6], &weight_avg[6],
 645                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 646                 }
 647             }
 648         }
 649     }
 650
 651     prefetch_motion(h, 1);
 652 }
 653
 654
 655 static void free_tables(H264Context *h){
 656     int i;
 657     H264Context *hx;
 658     av_freep(&h->intra4x4_pred_mode);
 659     av_freep(&h->chroma_pred_mode_table);
 660     av_freep(&h->cbp_table);
 661     av_freep(&h->mvd_table[0]);
 662     av_freep(&h->mvd_table[1]);
 663     av_freep(&h->direct_table);
 664     av_freep(&h->non_zero_count);
 665     av_freep(&h->slice_table_base);
 666     h->slice_table= NULL;
 667     av_freep(&h->list_counts);
 668
 669     av_freep(&h->mb2b_xy);
 670     av_freep(&h->mb2br_xy);
 671
 672     for(i = 0; i < MAX_THREADS; i++) {
 673         hx = h->thread_context[i];
 674         if(!hx) continue;
 675         av_freep(&hx->top_borders[1]);
 676         av_freep(&hx->top_borders[0]);
 677         av_freep(&hx->s.obmc_scratchpad);
 678         av_freep(&hx->rbsp_buffer[1]);
 679         av_freep(&hx->rbsp_buffer[0]);
 680         hx->rbsp_buffer_size[0] = 0;
 681         hx->rbsp_buffer_size[1] = 0;
 682         if (i) av_freep(&h->thread_context[i]);
 683     }
 684 }
 685
 686 static void init_dequant8_coeff_table(H264Context *h){
 687     int i,q,x;
 688     const int transpose = (h->h264dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
 689     h->dequant8_coeff[0] = h->dequant8_buffer[0];
 690     h->dequant8_coeff[1] = h->dequant8_buffer[1];
 691
 692     for(i=0; i<2; i++ ){
 693         if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
 694             h->dequant8_coeff[1] = h->dequant8_buffer[0];
 695             break;
 696         }
 697
 698         for(q=0; q<52; q++){
 699             int shift = div6[q];
 700             int idx = rem6[q];
 701             for(x=0; x<64; x++)
 702                 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
 703                     ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
 704                     h->pps.scaling_matrix8[i][x]) << shift;
 705         }
 706     }
 707 }
 708
 709 static void init_dequant4_coeff_table(H264Context *h){
 710     int i,j,q,x;
 711     const int transpose = (h->h264dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
 712     for(i=0; i<6; i++ ){
 713         h->dequant4_coeff[i] = h->dequant4_buffer[i];
 714         for(j=0; j<i; j++){
 715             if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
 716                 h->dequant4_coeff[i] = h->dequant4_buffer[j];
 717                 break;
 718             }
 719         }
 720         if(j<i)
 721             continue;
 722
 723         for(q=0; q<52; q++){
 724             int shift = div6[q] + 2;
 725             int idx = rem6[q];
 726             for(x=0; x<16; x++)
 727                 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
 728                     ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
 729                     h->pps.scaling_matrix4[i][x]) << shift;
 730         }
 731     }
 732 }
 733
 734 static void init_dequant_tables(H264Context *h){
 735     int i,x;
 736     init_dequant4_coeff_table(h);
 737     if(h->pps.transform_8x8_mode)
 738         init_dequant8_coeff_table(h);
 739     if(h->sps.transform_bypass){
 740         for(i=0; i<6; i++)
 741             for(x=0; x<16; x++)
 742                 h->dequant4_coeff[i][0][x] = 1<<6;
 743         if(h->pps.transform_8x8_mode)
 744             for(i=0; i<2; i++)
 745                 for(x=0; x<64; x++)
 746                     h->dequant8_coeff[i][0][x] = 1<<6;
 747     }
 748 }
 749
 750
 751 int ff_h264_alloc_tables(H264Context *h){
 752     MpegEncContext * const s = &h->s;
 753     const int big_mb_num= s->mb_stride * (s->mb_height+1);
 754     const int row_mb_num= 2*s->mb_stride*s->avctx->thread_count;
 755     int x,y;
 756
 757     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8  * sizeof(uint8_t), fail)
 758
 759     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 32 * sizeof(uint8_t), fail)
 760     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
 761     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
 762
 763     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
 764     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
 765     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
 766     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
 767     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
 768
 769     memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
 770     h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
 771
 772     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy  , big_mb_num * sizeof(uint32_t), fail);
 773     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
 774     for(y=0; y<s->mb_height; y++){
 775         for(x=0; x<s->mb_width; x++){
 776             const int mb_xy= x + y*s->mb_stride;
 777             const int b_xy = 4*x + 4*y*h->b_stride;
 778
 779             h->mb2b_xy [mb_xy]= b_xy;
 780             h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
 781         }
 782     }
 783
 784     s->obmc_scratchpad = NULL;
 785
 786     if(!h->dequant4_coeff[0])
 787         init_dequant_tables(h);
 788
 789     return 0;
 790 fail:
 791     free_tables(h);
 792     return -1;
 793 }
 794
 795 /**
 796  * Mimic alloc_tables(), but for every context thread.
 797  */
 798 static void clone_tables(H264Context *dst, H264Context *src, int i){
 799     MpegEncContext * const s = &src->s;
 800     dst->intra4x4_pred_mode       = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
 801     dst->non_zero_count           = src->non_zero_count;
 802     dst->slice_table              = src->slice_table;
 803     dst->cbp_table                = src->cbp_table;
 804     dst->mb2b_xy                  = src->mb2b_xy;
 805     dst->mb2br_xy                 = src->mb2br_xy;
 806     dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
 807     dst->mvd_table[0]             = src->mvd_table[0] + i*8*2*s->mb_stride;
 808     dst->mvd_table[1]             = src->mvd_table[1] + i*8*2*s->mb_stride;
 809     dst->direct_table             = src->direct_table;
 810     dst->list_counts              = src->list_counts;
 811
 812     dst->s.obmc_scratchpad = NULL;
 813     ff_h264_pred_init(&dst->hpc, src->s.codec_id);
 814 }
 815
 816 /**
 817  * Init context
 818  * Allocate buffers which are not shared amongst multiple threads.
 819  */
 820 static int context_init(H264Context *h){
 821     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
 822     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
 823
 824     h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
 825     h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
 826
 827     return 0;
 828 fail:
 829     return -1; // free_tables will clean up for us
 830 }
 831
 832 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
 833
 834 static av_cold void common_init(H264Context *h){
 835     MpegEncContext * const s = &h->s;
 836
 837     s->width = s->avctx->width;
 838     s->height = s->avctx->height;
 839     s->codec_id= s->avctx->codec->id;
 840
 841     ff_h264dsp_init(&h->h264dsp);
 842     ff_h264_pred_init(&h->hpc, s->codec_id);
 843
 844     h->dequant_coeff_pps= -1;
 845     s->unrestricted_mv=1;
 846     s->decode=1; //FIXME
 847
 848     dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
 849
 850     memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
 851     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
 852 }
 853
 854 int ff_h264_decode_extradata(H264Context *h)
 855 {
 856     AVCodecContext *avctx = h->s.avctx;
 857
 858     if(*(char *)avctx->extradata == 1){
 859         int i, cnt, nalsize;
 860         unsigned char *p = avctx->extradata;
 861
 862         h->is_avc = 1;
 863
 864         if(avctx->extradata_size < 7) {
 865             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
 866             return -1;
 867         }
 868         /* sps and pps in the avcC always have length coded with 2 bytes,
 869            so put a fake nal_length_size = 2 while parsing them */
 870         h->nal_length_size = 2;
 871         // Decode sps from avcC
 872         cnt = *(p+5) & 0x1f; // Number of sps
 873         p += 6;
 874         for (i = 0; i < cnt; i++) {
 875             nalsize = AV_RB16(p) + 2;
 876             if(decode_nal_units(h, p, nalsize) < 0) {
 877                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
 878                 return -1;
 879             }
 880             p += nalsize;
 881         }
 882         // Decode pps from avcC
 883         cnt = *(p++); // Number of pps
 884         for (i = 0; i < cnt; i++) {
 885             nalsize = AV_RB16(p) + 2;
 886             if(decode_nal_units(h, p, nalsize)  != nalsize) {
 887                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
 888                 return -1;
 889             }
 890             p += nalsize;
 891         }
 892         // Now store right nal length size, that will be use to parse all other nals
 893         h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
 894     } else {
 895         h->is_avc = 0;
 896         if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
 897             return -1;
 898     }
 899     return 0;
 900 }
 901
 902 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
 903     H264Context *h= avctx->priv_data;
 904     MpegEncContext * const s = &h->s;
 905
 906     MPV_decode_defaults(s);
 907
 908     s->avctx = avctx;
 909     common_init(h);
 910
 911     s->out_format = FMT_H264;
 912     s->workaround_bugs= avctx->workaround_bugs;
 913
 914     // set defaults
 915 //    s->decode_mb= ff_h263_decode_mb;
 916     s->quarter_sample = 1;
 917     if(!avctx->has_b_frames)
 918     s->low_delay= 1;
 919
 920     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
 921
 922     ff_h264_decode_init_vlc();
 923
 924     h->thread_context[0] = h;
 925     h->outputed_poc = INT_MIN;
 926     h->prev_poc_msb= 1<<16;
 927     h->x264_build = -1;
 928     ff_h264_reset_sei(h);
 929     if(avctx->codec_id == CODEC_ID_H264){
 930         if(avctx->ticks_per_frame == 1){
 931             s->avctx->time_base.den *=2;
 932         }
 933         avctx->ticks_per_frame = 2;
 934     }
 935
 936     if(avctx->extradata_size > 0 && avctx->extradata &&
 937         ff_h264_decode_extradata(h))
 938         return -1;
 939
 940     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
 941         s->avctx->has_b_frames = h->sps.num_reorder_frames;
 942         s->low_delay = 0;
 943     }
 944
 945     return 0;
 946 }
 947
 948 int ff_h264_frame_start(H264Context *h){
 949     MpegEncContext * const s = &h->s;
 950     int i;
 951
 952     if(MPV_frame_start(s, s->avctx) < 0)
 953         return -1;
 954     ff_er_frame_start(s);
 955     /*
 956      * MPV_frame_start uses pict_type to derive key_frame.
 957      * This is incorrect for H.264; IDR markings must be used.
 958      * Zero here; IDR markings per slice in frame or fields are ORed in later.
 959      * See decode_nal_units().
 960      */
 961     s->current_picture_ptr->key_frame= 0;
 962     s->current_picture_ptr->mmco_reset= 0;
 963
 964     assert(s->linesize && s->uvlinesize);
 965
 966     for(i=0; i<16; i++){
 967         h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
 968         h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
 969     }
 970     for(i=0; i<4; i++){
 971         h->block_offset[16+i]=
 972         h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
 973         h->block_offset[24+16+i]=
 974         h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
 975     }
 976
 977     /* can't be in alloc_tables because linesize isn't known there.
 978      * FIXME: redo bipred weight to not require extra buffer? */
 979     for(i = 0; i < s->avctx->thread_count; i++)
 980         if(!h->thread_context[i]->s.obmc_scratchpad)
 981             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
 982
 983     /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
 984     memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
 985
 986 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
 987
 988     // We mark the current picture as non-reference after allocating it, so
 989     // that if we break out due to an error it can be released automatically
 990     // in the next MPV_frame_start().
 991     // SVQ3 as well as most other codecs have only last/next/current and thus
 992     // get released even with set reference, besides SVQ3 and others do not
 993     // mark frames as reference later "naturally".
 994     if(s->codec_id != CODEC_ID_SVQ3)
 995         s->current_picture_ptr->reference= 0;
 996
 997     s->current_picture_ptr->field_poc[0]=
 998     s->current_picture_ptr->field_poc[1]= INT_MAX;
 999     assert(s->current_picture_ptr->long_ref==0);
1000
1001     return 0;
1002 }
1003
1004 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){
1005     MpegEncContext * const s = &h->s;
1006     uint8_t *top_border;
1007     int top_idx = 1;
1008
1009     src_y  -=   linesize;
1010     src_cb -= uvlinesize;
1011     src_cr -= uvlinesize;
1012
1013     if(!simple && FRAME_MBAFF){
1014         if(s->mb_y&1){
1015             if(!MB_MBAFF){
1016                 top_border = h->top_borders[0][s->mb_x];
1017                 AV_COPY128(top_border, src_y + 15*linesize);
1018                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1019                     AV_COPY64(top_border+16, src_cb+7*uvlinesize);
1020                     AV_COPY64(top_border+24, src_cr+7*uvlinesize);
1021                 }
1022             }
1023         }else if(MB_MBAFF){
1024             top_idx = 0;
1025         }else
1026             return;
1027     }
1028
1029     top_border = h->top_borders[top_idx][s->mb_x];
1030     // There are two lines saved, the line above the the top macroblock of a pair,
1031     // and the line above the bottom macroblock
1032     AV_COPY128(top_border, src_y + 16*linesize);
1033
1034     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1035         AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1036         AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1037     }
1038 }
1039
1040 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){
1041     MpegEncContext * const s = &h->s;
1042     int deblock_left;
1043     int deblock_top;
1044     int top_idx = 1;
1045     uint8_t *top_border_m1;
1046     uint8_t *top_border;
1047
1048     if(!simple && FRAME_MBAFF){
1049         if(s->mb_y&1){
1050             if(!MB_MBAFF)
1051                 return;
1052         }else{
1053             top_idx = MB_MBAFF ? 0 : 1;
1054         }
1055     }
1056
1057     if(h->deblocking_filter == 2) {
1058         deblock_left = h->left_type[0];
1059         deblock_top  = h->top_type;
1060     } else {
1061         deblock_left = (s->mb_x > 0);
1062         deblock_top =  (s->mb_y > !!MB_FIELD);
1063     }
1064
1065     src_y  -=   linesize + 1;
1066     src_cb -= uvlinesize + 1;
1067     src_cr -= uvlinesize + 1;
1068
1069     top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1070     top_border    = h->top_borders[top_idx][s->mb_x];
1071
1072 #define XCHG(a,b,xchg)\
1073 if (xchg) AV_SWAP64(b,a);\
1074 else      AV_COPY64(b,a);
1075
1076     if(deblock_top){
1077         if(deblock_left){
1078             XCHG(top_border_m1+8, src_y -7, 1);
1079         }
1080         XCHG(top_border+0, src_y +1, xchg);
1081         XCHG(top_border+8, src_y +9, 1);
1082         if(s->mb_x+1 < s->mb_width){
1083             XCHG(h->top_borders[top_idx][s->mb_x+1], src_y +17, 1);
1084         }
1085     }
1086
1087     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1088         if(deblock_top){
1089             if(deblock_left){
1090                 XCHG(top_border_m1+16, src_cb -7, 1);
1091                 XCHG(top_border_m1+24, src_cr -7, 1);
1092             }
1093             XCHG(top_border+16, src_cb+1, 1);
1094             XCHG(top_border+24, src_cr+1, 1);
1095         }
1096     }
1097 }
1098
1099 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
1100     MpegEncContext * const s = &h->s;
1101     const int mb_x= s->mb_x;
1102     const int mb_y= s->mb_y;
1103     const int mb_xy= h->mb_xy;
1104     const int mb_type= s->current_picture.mb_type[mb_xy];
1105     uint8_t  *dest_y, *dest_cb, *dest_cr;
1106     int linesize, uvlinesize /*dct_offset*/;
1107     int i;
1108     int *block_offset = &h->block_offset[0];
1109     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1110     /* is_h264 should always be true if SVQ3 is disabled. */
1111     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1112     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1113     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1114
1115     dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
1116     dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
1117     dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
1118
1119     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
1120     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
1121
1122     h->list_counts[mb_xy]= h->list_count;
1123
1124     if (!simple && MB_FIELD) {
1125         linesize   = h->mb_linesize   = s->linesize * 2;
1126         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1127         block_offset = &h->block_offset[24];
1128         if(mb_y&1){ //FIXME move out of this function?
1129             dest_y -= s->linesize*15;
1130             dest_cb-= s->uvlinesize*7;
1131             dest_cr-= s->uvlinesize*7;
1132         }
1133         if(FRAME_MBAFF) {
1134             int list;
1135             for(list=0; list<h->list_count; list++){
1136                 if(!USES_LIST(mb_type, list))
1137                     continue;
1138                 if(IS_16X16(mb_type)){
1139                     int8_t *ref = &h->ref_cache[list][scan8[0]];
1140                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1141                 }else{
1142                     for(i=0; i<16; i+=4){
1143                         int ref = h->ref_cache[list][scan8[i]];
1144                         if(ref >= 0)
1145                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1146                     }
1147                 }
1148             }
1149         }
1150     } else {
1151         linesize   = h->mb_linesize   = s->linesize;
1152         uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1153 //        dct_offset = s->linesize * 16;
1154     }
1155
1156     if (!simple && IS_INTRA_PCM(mb_type)) {
1157         for (i=0; i<16; i++) {
1158             memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1159         }
1160         for (i=0; i<8; i++) {
1161             memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
1162             memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
1163         }
1164     } else {
1165         if(IS_INTRA(mb_type)){
1166             if(h->deblocking_filter)
1167                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
1168
1169             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1170                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1171                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1172             }
1173
1174             if(IS_INTRA4x4(mb_type)){
1175                 if(simple || !s->encoding){
1176                     if(IS_8x8DCT(mb_type)){
1177                         if(transform_bypass){
1178                             idct_dc_add =
1179                             idct_add    = s->dsp.add_pixels8;
1180                         }else{
1181                             idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1182                             idct_add    = h->h264dsp.h264_idct8_add;
1183                         }
1184                         for(i=0; i<16; i+=4){
1185                             uint8_t * const ptr= dest_y + block_offset[i];
1186                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1187                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1188                                 h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
1189                             }else{
1190                                 const int nnz = h->non_zero_count_cache[ scan8[i] ];
1191                                 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1192                                                             (h->topright_samples_available<<i)&0x4000, linesize);
1193                                 if(nnz){
1194                                     if(nnz == 1 && h->mb[i*16])
1195                                         idct_dc_add(ptr, h->mb + i*16, linesize);
1196                                     else
1197                                         idct_add   (ptr, h->mb + i*16, linesize);
1198                                 }
1199                             }
1200                         }
1201                     }else{
1202                         if(transform_bypass){
1203                             idct_dc_add =
1204                             idct_add    = s->dsp.add_pixels4;
1205                         }else{
1206                             idct_dc_add = h->h264dsp.h264_idct_dc_add;
1207                             idct_add    = h->h264dsp.h264_idct_add;
1208                         }
1209                         for(i=0; i<16; i++){
1210                             uint8_t * const ptr= dest_y + block_offset[i];
1211                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1212
1213                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1214                                 h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
1215                             }else{
1216                                 uint8_t *topright;
1217                                 int nnz, tr;
1218                                 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1219                                     const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1220                                     assert(mb_y || linesize <= block_offset[i]);
1221                                     if(!topright_avail){
1222                                         tr= ptr[3 - linesize]*0x01010101;
1223                                         topright= (uint8_t*) &tr;
1224                                     }else
1225                                         topright= ptr + 4 - linesize;
1226                                 }else
1227                                     topright= NULL;
1228
1229                                 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1230                                 nnz = h->non_zero_count_cache[ scan8[i] ];
1231                                 if(nnz){
1232                                     if(is_h264){
1233                                         if(nnz == 1 && h->mb[i*16])
1234                                             idct_dc_add(ptr, h->mb + i*16, linesize);
1235                                         else
1236                                             idct_add   (ptr, h->mb + i*16, linesize);
1237                                     }else
1238                                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1239                                 }
1240                             }
1241                         }
1242                     }
1243                 }
1244             }else{
1245                 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1246                 if(is_h264){
1247                     if(!transform_bypass)
1248                         h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
1249                 }else
1250                     ff_svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
1251             }
1252             if(h->deblocking_filter)
1253                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
1254         }else if(is_h264){
1255             hl_motion(h, dest_y, dest_cb, dest_cr,
1256                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1257                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1258                       h->h264dsp.weight_h264_pixels_tab, h->h264dsp.biweight_h264_pixels_tab);
1259         }
1260
1261
1262         if(!IS_INTRA4x4(mb_type)){
1263             if(is_h264){
1264                 if(IS_INTRA16x16(mb_type)){
1265                     if(transform_bypass){
1266                         if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1267                             h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1268                         }else{
1269                             for(i=0; i<16; i++){
1270                                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1271                                     s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
1272                             }
1273                         }
1274                     }else{
1275                          h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1276                     }
1277                 }else if(h->cbp&15){
1278                     if(transform_bypass){
1279                         const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1280                         idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1281                         for(i=0; i<16; i+=di){
1282                             if(h->non_zero_count_cache[ scan8[i] ]){
1283                                 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
1284                             }
1285                         }
1286                     }else{
1287                         if(IS_8x8DCT(mb_type)){
1288                             h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1289                         }else{
1290                             h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1291                         }
1292                     }
1293                 }
1294             }else{
1295                 for(i=0; i<16; i++){
1296                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1297                         uint8_t * const ptr= dest_y + block_offset[i];
1298                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1299                     }
1300                 }
1301             }
1302         }
1303
1304         if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1305             uint8_t *dest[2] = {dest_cb, dest_cr};
1306             if(transform_bypass){
1307                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1308                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
1309                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
1310                 }else{
1311                     idct_add = s->dsp.add_pixels4;
1312                     for(i=16; i<16+8; i++){
1313                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1314                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1315                     }
1316                 }
1317             }else{
1318                 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]);
1319                 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]);
1320                 if(is_h264){
1321                     idct_add = h->h264dsp.h264_idct_add;
1322                     idct_dc_add = h->h264dsp.h264_idct_dc_add;
1323                     for(i=16; i<16+8; i++){
1324                         if(h->non_zero_count_cache[ scan8[i] ])
1325                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1326                         else if(h->mb[i*16])
1327                             idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1328                     }
1329                 }else{
1330                     for(i=16; i<16+8; i++){
1331                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1332                             uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1333                             ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
1334                         }
1335                     }
1336                 }
1337             }
1338         }
1339     }
1340     if(h->cbp || IS_INTRA(mb_type))
1341         s->dsp.clear_blocks(h->mb);
1342 }
1343
1344 /**
1345  * Process a macroblock; this case avoids checks for expensive uncommon cases.
1346  */
1347 static void hl_decode_mb_simple(H264Context *h){
1348     hl_decode_mb_internal(h, 1);
1349 }
1350
1351 /**
1352  * Process a macroblock; this handles edge cases, such as interlacing.
1353  */
1354 static void av_noinline hl_decode_mb_complex(H264Context *h){
1355     hl_decode_mb_internal(h, 0);
1356 }
1357
1358 void ff_h264_hl_decode_mb(H264Context *h){
1359     MpegEncContext * const s = &h->s;
1360     const int mb_xy= h->mb_xy;
1361     const int mb_type= s->current_picture.mb_type[mb_xy];
1362     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1363
1364     if (is_complex)
1365         hl_decode_mb_complex(h);
1366     else hl_decode_mb_simple(h);
1367 }
1368
1369 static int pred_weight_table(H264Context *h){
1370     MpegEncContext * const s = &h->s;
1371     int list, i;
1372     int luma_def, chroma_def;
1373
1374     h->use_weight= 0;
1375     h->use_weight_chroma= 0;
1376     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1377     h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1378     luma_def = 1<<h->luma_log2_weight_denom;
1379     chroma_def = 1<<h->chroma_log2_weight_denom;
1380
1381     for(list=0; list<2; list++){
1382         h->luma_weight_flag[list]   = 0;
1383         h->chroma_weight_flag[list] = 0;
1384         for(i=0; i<h->ref_count[list]; i++){
1385             int luma_weight_flag, chroma_weight_flag;
1386
1387             luma_weight_flag= get_bits1(&s->gb);
1388             if(luma_weight_flag){
1389                 h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
1390                 h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
1391                 if(   h->luma_weight[i][list][0] != luma_def
1392                    || h->luma_weight[i][list][1] != 0) {
1393                     h->use_weight= 1;
1394                     h->luma_weight_flag[list]= 1;
1395                 }
1396             }else{
1397                 h->luma_weight[i][list][0]= luma_def;
1398                 h->luma_weight[i][list][1]= 0;
1399             }
1400
1401             if(CHROMA){
1402                 chroma_weight_flag= get_bits1(&s->gb);
1403                 if(chroma_weight_flag){
1404                     int j;
1405                     for(j=0; j<2; j++){
1406                         h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
1407                         h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
1408                         if(   h->chroma_weight[i][list][j][0] != chroma_def
1409                            || h->chroma_weight[i][list][j][1] != 0) {
1410                             h->use_weight_chroma= 1;
1411                             h->chroma_weight_flag[list]= 1;
1412                         }
1413                     }
1414                 }else{
1415                     int j;
1416                     for(j=0; j<2; j++){
1417                         h->chroma_weight[i][list][j][0]= chroma_def;
1418                         h->chroma_weight[i][list][j][1]= 0;
1419                     }
1420                 }
1421             }
1422         }
1423         if(h->slice_type_nos != FF_B_TYPE) break;
1424     }
1425     h->use_weight= h->use_weight || h->use_weight_chroma;
1426     return 0;
1427 }
1428
1429 /**
1430  * Initialize implicit_weight table.
1431  * @param field  0/1 initialize the weight for interlaced MBAFF
1432  *                -1 initializes the rest
1433  */
1434 static void implicit_weight_table(H264Context *h, int field){
1435     MpegEncContext * const s = &h->s;
1436     int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
1437
1438     for (i = 0; i < 2; i++) {
1439         h->luma_weight_flag[i]   = 0;
1440         h->chroma_weight_flag[i] = 0;
1441     }
1442
1443     if(field < 0){
1444         cur_poc = s->current_picture_ptr->poc;
1445     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
1446        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1447         h->use_weight= 0;
1448         h->use_weight_chroma= 0;
1449         return;
1450     }
1451         ref_start= 0;
1452         ref_count0= h->ref_count[0];
1453         ref_count1= h->ref_count[1];
1454     }else{
1455         cur_poc = s->current_picture_ptr->field_poc[field];
1456         ref_start= 16;
1457         ref_count0= 16+2*h->ref_count[0];
1458         ref_count1= 16+2*h->ref_count[1];
1459     }
1460
1461     h->use_weight= 2;
1462     h->use_weight_chroma= 2;
1463     h->luma_log2_weight_denom= 5;
1464     h->chroma_log2_weight_denom= 5;
1465
1466     for(ref0=ref_start; ref0 < ref_count0; ref0++){
1467         int poc0 = h->ref_list[0][ref0].poc;
1468         for(ref1=ref_start; ref1 < ref_count1; ref1++){
1469             int poc1 = h->ref_list[1][ref1].poc;
1470             int td = av_clip(poc1 - poc0, -128, 127);
1471             int w= 32;
1472             if(td){
1473                 int tb = av_clip(cur_poc - poc0, -128, 127);
1474                 int tx = (16384 + (FFABS(td) >> 1)) / td;
1475                 int dist_scale_factor = (tb*tx + 32) >> 8;
1476                 if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
1477                     w = 64 - dist_scale_factor;
1478             }
1479             if(field<0){
1480                 h->implicit_weight[ref0][ref1][0]=
1481                 h->implicit_weight[ref0][ref1][1]= w;
1482             }else{
1483                 h->implicit_weight[ref0][ref1][field]=w;
1484             }
1485         }
1486     }
1487 }
1488
1489 /**
1490  * instantaneous decoder refresh.
1491  */
1492 static void idr(H264Context *h){
1493     ff_h264_remove_all_refs(h);
1494     h->prev_frame_num= 0;
1495     h->prev_frame_num_offset= 0;
1496     h->prev_poc_msb=
1497     h->prev_poc_lsb= 0;
1498 }
1499
1500 /* forget old pics after a seek */
1501 static void flush_dpb(AVCodecContext *avctx){
1502     H264Context *h= avctx->priv_data;
1503     int i;
1504     for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1505         if(h->delayed_pic[i])
1506             h->delayed_pic[i]->reference= 0;
1507         h->delayed_pic[i]= NULL;
1508     }
1509     h->outputed_poc= INT_MIN;
1510     h->prev_interlaced_frame = 1;
1511     idr(h);
1512     if(h->s.current_picture_ptr)
1513         h->s.current_picture_ptr->reference= 0;
1514     h->s.first_field= 0;
1515     ff_h264_reset_sei(h);
1516     ff_mpeg_flush(avctx);
1517 }
1518
1519 static int init_poc(H264Context *h){
1520     MpegEncContext * const s = &h->s;
1521     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1522     int field_poc[2];
1523     Picture *cur = s->current_picture_ptr;
1524
1525     h->frame_num_offset= h->prev_frame_num_offset;
1526     if(h->frame_num < h->prev_frame_num)
1527         h->frame_num_offset += max_frame_num;
1528
1529     if(h->sps.poc_type==0){
1530         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1531
1532         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1533             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1534         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1535             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1536         else
1537             h->poc_msb = h->prev_poc_msb;
1538 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1539         field_poc[0] =
1540         field_poc[1] = h->poc_msb + h->poc_lsb;
1541         if(s->picture_structure == PICT_FRAME)
1542             field_poc[1] += h->delta_poc_bottom;
1543     }else if(h->sps.poc_type==1){
1544         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1545         int i;
1546
1547         if(h->sps.poc_cycle_length != 0)
1548             abs_frame_num = h->frame_num_offset + h->frame_num;
1549         else
1550             abs_frame_num = 0;
1551
1552         if(h->nal_ref_idc==0 && abs_frame_num > 0)
1553             abs_frame_num--;
1554
1555         expected_delta_per_poc_cycle = 0;
1556         for(i=0; i < h->sps.poc_cycle_length; i++)
1557             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1558
1559         if(abs_frame_num > 0){
1560             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1561             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1562
1563             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1564             for(i = 0; i <= frame_num_in_poc_cycle; i++)
1565                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1566         } else
1567             expectedpoc = 0;
1568
1569         if(h->nal_ref_idc == 0)
1570             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1571
1572         field_poc[0] = expectedpoc + h->delta_poc[0];
1573         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1574
1575         if(s->picture_structure == PICT_FRAME)
1576             field_poc[1] += h->delta_poc[1];
1577     }else{
1578         int poc= 2*(h->frame_num_offset + h->frame_num);
1579
1580         if(!h->nal_ref_idc)
1581             poc--;
1582
1583         field_poc[0]= poc;
1584         field_poc[1]= poc;
1585     }
1586
1587     if(s->picture_structure != PICT_BOTTOM_FIELD)
1588         s->current_picture_ptr->field_poc[0]= field_poc[0];
1589     if(s->picture_structure != PICT_TOP_FIELD)
1590         s->current_picture_ptr->field_poc[1]= field_poc[1];
1591     cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1592
1593     return 0;
1594 }
1595
1596
1597 /**
1598  * initialize scan tables
1599  */
1600 static void init_scan_tables(H264Context *h){
1601     int i;
1602     if(h->h264dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
1603         memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
1604         memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
1605     }else{
1606         for(i=0; i<16; i++){
1607 #define T(x) (x>>2) | ((x<<2) & 0xF)
1608             h->zigzag_scan[i] = T(zigzag_scan[i]);
1609             h-> field_scan[i] = T( field_scan[i]);
1610 #undef T
1611         }
1612     }
1613     if(h->h264dsp.h264_idct8_add == ff_h264_idct8_add_c){
1614         memcpy(h->zigzag_scan8x8,       ff_zigzag_direct,     64*sizeof(uint8_t));
1615         memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
1616         memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
1617         memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
1618     }else{
1619         for(i=0; i<64; i++){
1620 #define T(x) (x>>3) | ((x&7)<<3)
1621             h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
1622             h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1623             h->field_scan8x8[i]        = T(field_scan8x8[i]);
1624             h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
1625 #undef T
1626         }
1627     }
1628     if(h->sps.transform_bypass){ //FIXME same ugly
1629         h->zigzag_scan_q0          = zigzag_scan;
1630         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
1631         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1632         h->field_scan_q0           = field_scan;
1633         h->field_scan8x8_q0        = field_scan8x8;
1634         h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
1635     }else{
1636         h->zigzag_scan_q0          = h->zigzag_scan;
1637         h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
1638         h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1639         h->field_scan_q0           = h->field_scan;
1640         h->field_scan8x8_q0        = h->field_scan8x8;
1641         h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
1642     }
1643 }
1644
1645 static void field_end(H264Context *h){
1646     MpegEncContext * const s = &h->s;
1647     AVCodecContext * const avctx= s->avctx;
1648     s->mb_y= 0;
1649
1650     s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1651     s->current_picture_ptr->pict_type= s->pict_type;
1652
1653     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1654         ff_vdpau_h264_set_reference_frames(s);
1655
1656     if(!s->dropable) {
1657         ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1658         h->prev_poc_msb= h->poc_msb;
1659         h->prev_poc_lsb= h->poc_lsb;
1660     }
1661     h->prev_frame_num_offset= h->frame_num_offset;
1662     h->prev_frame_num= h->frame_num;
1663
1664     if (avctx->hwaccel) {
1665         if (avctx->hwaccel->end_frame(avctx) < 0)
1666             av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1667     }
1668
1669     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1670         ff_vdpau_h264_picture_complete(s);
1671
1672     /*
1673      * FIXME: Error handling code does not seem to support interlaced
1674      * when slices span multiple rows
1675      * The ff_er_add_slice calls don't work right for bottom
1676      * fields; they cause massive erroneous error concealing
1677      * Error marking covers both fields (top and bottom).
1678      * This causes a mismatched s->error_count
1679      * and a bad error table. Further, the error count goes to
1680      * INT_MAX when called for bottom field, because mb_y is
1681      * past end by one (callers fault) and resync_mb_y != 0
1682      * causes problems for the first MB line, too.
1683      */
1684     if (!FIELD_PICTURE)
1685         ff_er_frame_end(s);
1686
1687     MPV_frame_end(s);
1688
1689     h->current_slice=0;
1690 }
1691
1692 /**
1693  * Replicate H264 "master" context to thread contexts.
1694  */
1695 static void clone_slice(H264Context *dst, H264Context *src)
1696 {
1697     memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
1698     dst->s.current_picture_ptr  = src->s.current_picture_ptr;
1699     dst->s.current_picture      = src->s.current_picture;
1700     dst->s.linesize             = src->s.linesize;
1701     dst->s.uvlinesize           = src->s.uvlinesize;
1702     dst->s.first_field          = src->s.first_field;
1703
1704     dst->prev_poc_msb           = src->prev_poc_msb;
1705     dst->prev_poc_lsb           = src->prev_poc_lsb;
1706     dst->prev_frame_num_offset  = src->prev_frame_num_offset;
1707     dst->prev_frame_num         = src->prev_frame_num;
1708     dst->short_ref_count        = src->short_ref_count;
1709
1710     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
1711     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
1712     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1713     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
1714
1715     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
1716     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
1717 }
1718
1719 /**
1720  * decodes a slice header.
1721  * This will also call MPV_common_init() and frame_start() as needed.
1722  *
1723  * @param h h264context
1724  * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1725  *
1726  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1727  */
1728 static int decode_slice_header(H264Context *h, H264Context *h0){
1729     MpegEncContext * const s = &h->s;
1730     MpegEncContext * const s0 = &h0->s;
1731     unsigned int first_mb_in_slice;
1732     unsigned int pps_id;
1733     int num_ref_idx_active_override_flag;
1734     unsigned int slice_type, tmp, i, j;
1735     int default_ref_list_done = 0;
1736     int last_pic_structure;
1737
1738     s->dropable= h->nal_ref_idc == 0;
1739
1740     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1741         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1742         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1743     }else{
1744         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1745         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1746     }
1747
1748     first_mb_in_slice= get_ue_golomb(&s->gb);
1749
1750     if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1751         if(h0->current_slice && FIELD_PICTURE){
1752             field_end(h);
1753         }
1754
1755         h0->current_slice = 0;
1756         if (!s0->first_field)
1757             s->current_picture_ptr= NULL;
1758     }
1759
1760     slice_type= get_ue_golomb_31(&s->gb);
1761     if(slice_type > 9){
1762         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);
1763         return -1;
1764     }
1765     if(slice_type > 4){
1766         slice_type -= 5;
1767         h->slice_type_fixed=1;
1768     }else
1769         h->slice_type_fixed=0;
1770
1771     slice_type= golomb_to_pict_type[ slice_type ];
1772     if (slice_type == FF_I_TYPE
1773         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
1774         default_ref_list_done = 1;
1775     }
1776     h->slice_type= slice_type;
1777     h->slice_type_nos= slice_type & 3;
1778
1779     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
1780
1781     pps_id= get_ue_golomb(&s->gb);
1782     if(pps_id>=MAX_PPS_COUNT){
1783         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
1784         return -1;
1785     }
1786     if(!h0->pps_buffers[pps_id]) {
1787         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
1788         return -1;
1789     }
1790     h->pps= *h0->pps_buffers[pps_id];
1791
1792     if(!h0->sps_buffers[h->pps.sps_id]) {
1793         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
1794         return -1;
1795     }
1796     h->sps = *h0->sps_buffers[h->pps.sps_id];
1797
1798     s->avctx->profile = h->sps.profile_idc;
1799     s->avctx->level   = h->sps.level_idc;
1800     s->avctx->refs    = h->sps.ref_frame_count;
1801
1802     if(h == h0 && h->dequant_coeff_pps != pps_id){
1803         h->dequant_coeff_pps = pps_id;
1804         init_dequant_tables(h);
1805     }
1806
1807     s->mb_width= h->sps.mb_width;
1808     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1809
1810     h->b_stride=  s->mb_width*4;
1811
1812     s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
1813     if(h->sps.frame_mbs_only_flag)
1814         s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
1815     else
1816         s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
1817
1818     if (s->context_initialized
1819         && (   s->width != s->avctx->width || s->height != s->avctx->height
1820             || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
1821         if(h != h0)
1822             return -1;   // width / height changed during parallelized decoding
1823         free_tables(h);
1824         flush_dpb(s->avctx);
1825         MPV_common_end(s);
1826     }
1827     if (!s->context_initialized) {
1828         if(h != h0)
1829             return -1;  // we cant (re-)initialize context during parallel decoding
1830
1831         avcodec_set_dimensions(s->avctx, s->width, s->height);
1832         s->avctx->sample_aspect_ratio= h->sps.sar;
1833         if(!s->avctx->sample_aspect_ratio.den)
1834             s->avctx->sample_aspect_ratio.den = 1;
1835
1836         if(h->sps.video_signal_type_present_flag){
1837             s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1838             if(h->sps.colour_description_present_flag){
1839                 s->avctx->color_primaries = h->sps.color_primaries;
1840                 s->avctx->color_trc       = h->sps.color_trc;
1841                 s->avctx->colorspace      = h->sps.colorspace;
1842             }
1843         }
1844
1845         if(h->sps.timing_info_present_flag){
1846             int64_t den= h->sps.time_scale;
1847             if(h->x264_build < 44U)
1848                 den *= 2;
1849             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
1850                       h->sps.num_units_in_tick, den, 1<<30);
1851         }
1852         s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
1853                                                  s->avctx->codec->pix_fmts ?
1854                                                  s->avctx->codec->pix_fmts :
1855                                                  s->avctx->color_range == AVCOL_RANGE_JPEG ?
1856                                                  hwaccel_pixfmt_list_h264_jpeg_420 :
1857                                                  ff_hwaccel_pixfmt_list_420);
1858         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1859
1860         if (MPV_common_init(s) < 0)
1861             return -1;
1862         s->first_field = 0;
1863         h->prev_interlaced_frame = 1;
1864
1865         init_scan_tables(h);
1866         ff_h264_alloc_tables(h);
1867
1868         for(i = 1; i < s->avctx->thread_count; i++) {
1869             H264Context *c;
1870             c = h->thread_context[i] = av_malloc(sizeof(H264Context));
1871             memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
1872             memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
1873             c->h264dsp = h->h264dsp;
1874             c->sps = h->sps;
1875             c->pps = h->pps;
1876             init_scan_tables(c);
1877             clone_tables(c, h, i);
1878         }
1879
1880         for(i = 0; i < s->avctx->thread_count; i++)
1881             if(context_init(h->thread_context[i]) < 0)
1882                 return -1;
1883     }
1884
1885     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1886
1887     h->mb_mbaff = 0;
1888     h->mb_aff_frame = 0;
1889     last_pic_structure = s0->picture_structure;
1890     if(h->sps.frame_mbs_only_flag){
1891         s->picture_structure= PICT_FRAME;
1892     }else{
1893         if(get_bits1(&s->gb)) { //field_pic_flag
1894             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
1895         } else {
1896             s->picture_structure= PICT_FRAME;
1897             h->mb_aff_frame = h->sps.mb_aff;
1898         }
1899     }
1900     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
1901
1902     if(h0->current_slice == 0){
1903         while(h->frame_num !=  h->prev_frame_num &&
1904               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
1905             av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
1906             if (ff_h264_frame_start(h) < 0)
1907                 return -1;
1908             h->prev_frame_num++;
1909             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
1910             s->current_picture_ptr->frame_num= h->prev_frame_num;
1911             ff_generate_sliding_window_mmcos(h);
1912             ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1913         }
1914
1915         /* See if we have a decoded first field looking for a pair... */
1916         if (s0->first_field) {
1917             assert(s0->current_picture_ptr);
1918             assert(s0->current_picture_ptr->data[0]);
1919             assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
1920
1921             /* figure out if we have a complementary field pair */
1922             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
1923                 /*
1924                  * Previous field is unmatched. Don't display it, but let it
1925                  * remain for reference if marked as such.
1926                  */
1927                 s0->current_picture_ptr = NULL;
1928                 s0->first_field = FIELD_PICTURE;
1929
1930             } else {
1931                 if (h->nal_ref_idc &&
1932                         s0->current_picture_ptr->reference &&
1933                         s0->current_picture_ptr->frame_num != h->frame_num) {
1934                     /*
1935                      * This and previous field were reference, but had
1936                      * different frame_nums. Consider this field first in
1937                      * pair. Throw away previous field except for reference
1938                      * purposes.
1939                      */
1940                     s0->first_field = 1;
1941                     s0->current_picture_ptr = NULL;
1942
1943                 } else {
1944                     /* Second field in complementary pair */
1945                     s0->first_field = 0;
1946                 }
1947             }
1948
1949         } else {
1950             /* Frame or first field in a potentially complementary pair */
1951             assert(!s0->current_picture_ptr);
1952             s0->first_field = FIELD_PICTURE;
1953         }
1954
1955         if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
1956             s0->first_field = 0;
1957             return -1;
1958         }
1959     }
1960     if(h != h0)
1961         clone_slice(h, h0);
1962
1963     s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
1964
1965     assert(s->mb_num == s->mb_width * s->mb_height);
1966     if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
1967        first_mb_in_slice                    >= s->mb_num){
1968         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1969         return -1;
1970     }
1971     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
1972     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
1973     if (s->picture_structure == PICT_BOTTOM_FIELD)
1974         s->resync_mb_y = s->mb_y = s->mb_y + 1;
1975     assert(s->mb_y < s->mb_height);
1976
1977     if(s->picture_structure==PICT_FRAME){
1978         h->curr_pic_num=   h->frame_num;
1979         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
1980     }else{
1981         h->curr_pic_num= 2*h->frame_num + 1;
1982         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
1983     }
1984
1985     if(h->nal_unit_type == NAL_IDR_SLICE){
1986         get_ue_golomb(&s->gb); /* idr_pic_id */
1987     }
1988
1989     if(h->sps.poc_type==0){
1990         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
1991
1992         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
1993             h->delta_poc_bottom= get_se_golomb(&s->gb);
1994         }
1995     }
1996
1997     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
1998         h->delta_poc[0]= get_se_golomb(&s->gb);
1999
2000         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2001             h->delta_poc[1]= get_se_golomb(&s->gb);
2002     }
2003
2004     init_poc(h);
2005
2006     if(h->pps.redundant_pic_cnt_present){
2007         h->redundant_pic_count= get_ue_golomb(&s->gb);
2008     }
2009
2010     //set defaults, might be overridden a few lines later
2011     h->ref_count[0]= h->pps.ref_count[0];
2012     h->ref_count[1]= h->pps.ref_count[1];
2013
2014     if(h->slice_type_nos != FF_I_TYPE){
2015         if(h->slice_type_nos == FF_B_TYPE){
2016             h->direct_spatial_mv_pred= get_bits1(&s->gb);
2017         }
2018         num_ref_idx_active_override_flag= get_bits1(&s->gb);
2019
2020         if(num_ref_idx_active_override_flag){
2021             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2022             if(h->slice_type_nos==FF_B_TYPE)
2023                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2024
2025             if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
2026                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
2027                 h->ref_count[0]= h->ref_count[1]= 1;
2028                 return -1;
2029             }
2030         }
2031         if(h->slice_type_nos == FF_B_TYPE)
2032             h->list_count= 2;
2033         else
2034             h->list_count= 1;
2035     }else
2036         h->list_count= 0;
2037
2038     if(!default_ref_list_done){
2039         ff_h264_fill_default_ref_list(h);
2040     }
2041
2042     if(h->slice_type_nos!=FF_I_TYPE && ff_h264_decode_ref_pic_list_reordering(h) < 0)
2043         return -1;
2044
2045     if(h->slice_type_nos!=FF_I_TYPE){
2046         s->last_picture_ptr= &h->ref_list[0][0];
2047         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
2048     }
2049     if(h->slice_type_nos==FF_B_TYPE){
2050         s->next_picture_ptr= &h->ref_list[1][0];
2051         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
2052     }
2053
2054     if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
2055        ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
2056         pred_weight_table(h);
2057     else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE){
2058         implicit_weight_table(h, -1);
2059     }else {
2060         h->use_weight = 0;
2061         for (i = 0; i < 2; i++) {
2062             h->luma_weight_flag[i]   = 0;
2063             h->chroma_weight_flag[i] = 0;
2064         }
2065     }
2066
2067     if(h->nal_ref_idc)
2068         ff_h264_decode_ref_pic_marking(h0, &s->gb);
2069
2070     if(FRAME_MBAFF){
2071         ff_h264_fill_mbaff_ref_list(h);
2072
2073         if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE){
2074             implicit_weight_table(h, 0);
2075             implicit_weight_table(h, 1);
2076         }
2077     }
2078
2079     if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
2080         ff_h264_direct_dist_scale_factor(h);
2081     ff_h264_direct_ref_list_init(h);
2082
2083     if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
2084         tmp = get_ue_golomb_31(&s->gb);
2085         if(tmp > 2){
2086             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2087             return -1;
2088         }
2089         h->cabac_init_idc= tmp;
2090     }
2091
2092     h->last_qscale_diff = 0;
2093     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2094     if(tmp>51){
2095         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2096         return -1;
2097     }
2098     s->qscale= tmp;
2099     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2100     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2101     //FIXME qscale / qp ... stuff
2102     if(h->slice_type == FF_SP_TYPE){
2103         get_bits1(&s->gb); /* sp_for_switch_flag */
2104     }
2105     if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
2106         get_se_golomb(&s->gb); /* slice_qs_delta */
2107     }
2108
2109     h->deblocking_filter = 1;
2110     h->slice_alpha_c0_offset = 52;
2111     h->slice_beta_offset = 52;
2112     if( h->pps.deblocking_filter_parameters_present ) {
2113         tmp= get_ue_golomb_31(&s->gb);
2114         if(tmp > 2){
2115             av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2116             return -1;
2117         }
2118         h->deblocking_filter= tmp;
2119         if(h->deblocking_filter < 2)
2120             h->deblocking_filter^= 1; // 1<->0
2121
2122         if( h->deblocking_filter ) {
2123             h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2124             h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
2125             if(   h->slice_alpha_c0_offset > 104U
2126                || h->slice_beta_offset     > 104U){
2127                 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);
2128                 return -1;
2129             }
2130         }
2131     }
2132
2133     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
2134        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
2135        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
2136        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2137         h->deblocking_filter= 0;
2138
2139     if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2140         if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2141             /* Cheat slightly for speed:
2142                Do not bother to deblock across slices. */
2143             h->deblocking_filter = 2;
2144         } else {
2145             h0->max_contexts = 1;
2146             if(!h0->single_decode_warning) {
2147                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2148                 h0->single_decode_warning = 1;
2149             }
2150             if(h != h0)
2151                 return 1; // deblocking switched inside frame
2152         }
2153     }
2154     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]);
2155
2156 #if 0 //FMO
2157     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2158         slice_group_change_cycle= get_bits(&s->gb, ?);
2159 #endif
2160
2161     h0->last_slice_type = slice_type;
2162     h->slice_num = ++h0->current_slice;
2163     if(h->slice_num >= MAX_SLICES){
2164         av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2165     }
2166
2167     for(j=0; j<2; j++){
2168         int id_list[16];
2169         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2170         for(i=0; i<16; i++){
2171             id_list[i]= 60;
2172             if(h->ref_list[j][i].data[0]){
2173                 int k;
2174                 uint8_t *base= h->ref_list[j][i].base[0];
2175                 for(k=0; k<h->short_ref_count; k++)
2176                     if(h->short_ref[k]->base[0] == base){
2177                         id_list[i]= k;
2178                         break;
2179                     }
2180                 for(k=0; k<h->long_ref_count; k++)
2181                     if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2182                         id_list[i]= h->short_ref_count + k;
2183                         break;
2184                     }
2185             }
2186         }
2187
2188         ref2frm[0]=
2189         ref2frm[1]= -1;
2190         for(i=0; i<16; i++)
2191             ref2frm[i+2]= 4*id_list[i]
2192                           +(h->ref_list[j][i].reference&3);
2193         ref2frm[18+0]=
2194         ref2frm[18+1]= -1;
2195         for(i=16; i<48; i++)
2196             ref2frm[i+4]= 4*id_list[(i-16)>>1]
2197                           +(h->ref_list[j][i].reference&3);
2198     }
2199
2200     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
2201     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2202
2203     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2204         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",
2205                h->slice_num,
2206                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2207                first_mb_in_slice,
2208                av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2209                pps_id, h->frame_num,
2210                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2211                h->ref_count[0], h->ref_count[1],
2212                s->qscale,
2213                h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
2214                h->use_weight,
2215                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2216                h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2217                );
2218     }
2219
2220     return 0;
2221 }
2222
2223 int ff_h264_get_slice_type(const H264Context *h)
2224 {
2225     switch (h->slice_type) {
2226     case FF_P_TYPE:  return 0;
2227     case FF_B_TYPE:  return 1;
2228     case FF_I_TYPE:  return 2;
2229     case FF_SP_TYPE: return 3;
2230     case FF_SI_TYPE: return 4;
2231     default:         return -1;
2232     }
2233 }
2234
2235 /**
2236  *
2237  * @return non zero if the loop filter can be skiped
2238  */
2239 static int fill_filter_caches(H264Context *h, int mb_type){
2240     MpegEncContext * const s = &h->s;
2241     const int mb_xy= h->mb_xy;
2242     int top_xy, left_xy[2];
2243     int top_type, left_type[2];
2244
2245     top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
2246
2247     //FIXME deblocking could skip the intra and nnz parts.
2248
2249     /* Wow, what a mess, why didn't they simplify the interlacing & intra
2250      * stuff, I can't imagine that these complex rules are worth it. */
2251
2252     left_xy[1] = left_xy[0] = mb_xy-1;
2253     if(FRAME_MBAFF){
2254         const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
2255         const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
2256         if(s->mb_y&1){
2257             if (left_mb_field_flag != curr_mb_field_flag) {
2258                 left_xy[0] -= s->mb_stride;
2259             }
2260         }else{
2261             if(curr_mb_field_flag){
2262                 top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
2263             }
2264             if (left_mb_field_flag != curr_mb_field_flag) {
2265                 left_xy[1] += s->mb_stride;
2266             }
2267         }
2268     }
2269
2270     h->top_mb_xy = top_xy;
2271     h->left_mb_xy[0] = left_xy[0];
2272     h->left_mb_xy[1] = left_xy[1];
2273     {
2274         //for sufficiently low qp, filtering wouldn't do anything
2275         //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
2276         int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
2277         int qp = s->current_picture.qscale_table[mb_xy];
2278         if(qp <= qp_thresh
2279            && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
2280            && (top_xy   < 0 || ((qp + s->current_picture.qscale_table[top_xy    ] + 1)>>1) <= qp_thresh)){
2281             if(!FRAME_MBAFF)
2282                 return 1;
2283             if(   (left_xy[0]< 0            || ((qp + s->current_picture.qscale_table[left_xy[1]             ] + 1)>>1) <= qp_thresh)
2284                && (top_xy    < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy    -s->mb_stride] + 1)>>1) <= qp_thresh))
2285                 return 1;
2286         }
2287     }
2288
2289     top_type     = s->current_picture.mb_type[top_xy]    ;
2290     left_type[0] = s->current_picture.mb_type[left_xy[0]];
2291     left_type[1] = s->current_picture.mb_type[left_xy[1]];
2292     if(h->deblocking_filter == 2){
2293         if(h->slice_table[top_xy     ] != h->slice_num) top_type= 0;
2294         if(h->slice_table[left_xy[0] ] != h->slice_num) left_type[0]= left_type[1]= 0;
2295     }else{
2296         if(h->slice_table[top_xy     ] == 0xFFFF) top_type= 0;
2297         if(h->slice_table[left_xy[0] ] == 0xFFFF) left_type[0]= left_type[1] =0;
2298     }
2299     h->top_type    = top_type    ;
2300     h->left_type[0]= left_type[0];
2301     h->left_type[1]= left_type[1];
2302
2303     if(IS_INTRA(mb_type))
2304         return 0;
2305
2306     AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
2307     AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
2308     AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]);
2309     AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]);
2310     AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
2311
2312     h->cbp= h->cbp_table[mb_xy];
2313
2314     {
2315         int list;
2316         for(list=0; list<h->list_count; list++){
2317             int8_t *ref;
2318             int y, b_stride;
2319             int16_t (*mv_dst)[2];
2320             int16_t (*mv_src)[2];
2321
2322             if(!USES_LIST(mb_type, list)){
2323                 fill_rectangle(  h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
2324                 AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2325                 AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2326                 AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2327                 AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2328                 continue;
2329             }
2330
2331             ref = &s->current_picture.ref_index[list][4*mb_xy];
2332             {
2333                 int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2334                 AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2335                 AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2336                 ref += 2;
2337                 AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2338                 AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2339             }
2340
2341             b_stride = h->b_stride;
2342             mv_dst   = &h->mv_cache[list][scan8[0]];
2343             mv_src   = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
2344             for(y=0; y<4; y++){
2345                 AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
2346             }
2347
2348         }
2349     }
2350
2351
2352 /*
2353 0 . T T. T T T T
2354 1 L . .L . . . .
2355 2 L . .L . . . .
2356 3 . T TL . . . .
2357 4 L . .L . . . .
2358 5 L . .. . . . .
2359 */
2360 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
2361     if(top_type){
2362         AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
2363     }
2364
2365     if(left_type[0]){
2366         h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
2367         h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
2368         h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
2369         h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
2370     }
2371
2372     // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
2373     if(!CABAC && h->pps.transform_8x8_mode){
2374         if(IS_8x8DCT(top_type)){
2375             h->non_zero_count_cache[4+8*0]=
2376             h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
2377             h->non_zero_count_cache[6+8*0]=
2378             h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
2379         }
2380         if(IS_8x8DCT(left_type[0])){
2381             h->non_zero_count_cache[3+8*1]=
2382             h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
2383         }
2384         if(IS_8x8DCT(left_type[1])){
2385             h->non_zero_count_cache[3+8*3]=
2386             h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
2387         }
2388
2389         if(IS_8x8DCT(mb_type)){
2390             h->non_zero_count_cache[scan8[0   ]]= h->non_zero_count_cache[scan8[1   ]]=
2391             h->non_zero_count_cache[scan8[2   ]]= h->non_zero_count_cache[scan8[3   ]]= h->cbp & 1;
2392
2393             h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
2394             h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
2395
2396             h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
2397             h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
2398
2399             h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
2400             h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
2401         }
2402     }
2403
2404     if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
2405         int list;
2406         for(list=0; list<h->list_count; list++){
2407             if(USES_LIST(top_type, list)){
2408                 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
2409                 const int b8_xy= 4*top_xy + 2;
2410                 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2411                 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
2412                 h->ref_cache[list][scan8[0] + 0 - 1*8]=
2413                 h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
2414                 h->ref_cache[list][scan8[0] + 2 - 1*8]=
2415                 h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
2416             }else{
2417                 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
2418                 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2419             }
2420
2421             if(!IS_INTERLACED(mb_type^left_type[0])){
2422                 if(USES_LIST(left_type[0], list)){
2423                     const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
2424                     const int b8_xy= 4*left_xy[0] + 1;
2425                     int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2426                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
2427                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
2428                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
2429                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
2430                     h->ref_cache[list][scan8[0] - 1 + 0 ]=
2431                     h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]];
2432                     h->ref_cache[list][scan8[0] - 1 +16 ]=
2433                     h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]];
2434                 }else{
2435                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
2436                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
2437                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
2438                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
2439                     h->ref_cache[list][scan8[0] - 1 + 0  ]=
2440                     h->ref_cache[list][scan8[0] - 1 + 8  ]=
2441                     h->ref_cache[list][scan8[0] - 1 + 16 ]=
2442                     h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
2443                 }
2444             }
2445         }
2446     }
2447
2448     return 0;
2449 }
2450
2451 static void loop_filter(H264Context *h){
2452     MpegEncContext * const s = &h->s;
2453     uint8_t  *dest_y, *dest_cb, *dest_cr;
2454     int linesize, uvlinesize, mb_x, mb_y;
2455     const int end_mb_y= s->mb_y + FRAME_MBAFF;
2456     const int old_slice_type= h->slice_type;
2457
2458     if(h->deblocking_filter) {
2459         for(mb_x= 0; mb_x<s->mb_width; mb_x++){
2460             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
2461                 int mb_xy, mb_type;
2462                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2463                 h->slice_num= h->slice_table[mb_xy];
2464                 mb_type= s->current_picture.mb_type[mb_xy];
2465                 h->list_count= h->list_counts[mb_xy];
2466
2467                 if(FRAME_MBAFF)
2468                     h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2469
2470                 s->mb_x= mb_x;
2471                 s->mb_y= mb_y;
2472                 dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2473                 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2474                 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2475                     //FIXME simplify above
2476
2477                 if (MB_FIELD) {
2478                     linesize   = h->mb_linesize   = s->linesize * 2;
2479                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2480                     if(mb_y&1){ //FIXME move out of this function?
2481                         dest_y -= s->linesize*15;
2482                         dest_cb-= s->uvlinesize*7;
2483                         dest_cr-= s->uvlinesize*7;
2484                     }
2485                 } else {
2486                     linesize   = h->mb_linesize   = s->linesize;
2487                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2488                 }
2489                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2490                 if(fill_filter_caches(h, mb_type))
2491                     continue;
2492                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2493                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2494
2495                 if (FRAME_MBAFF) {
2496                     ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2497                 } else {
2498                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2499                 }
2500             }
2501         }
2502     }
2503     h->slice_type= old_slice_type;
2504     s->mb_x= 0;
2505     s->mb_y= end_mb_y - FRAME_MBAFF;
2506     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2507     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2508 }
2509
2510 static void predict_field_decoding_flag(H264Context *h){
2511     MpegEncContext * const s = &h->s;
2512     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2513     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
2514                 ? s->current_picture.mb_type[mb_xy-1]
2515                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
2516                 ? s->current_picture.mb_type[mb_xy-s->mb_stride]
2517                 : 0;
2518     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2519 }
2520
2521 static int decode_slice(struct AVCodecContext *avctx, void *arg){
2522     H264Context *h = *(void**)arg;
2523     MpegEncContext * const s = &h->s;
2524     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2525
2526     s->mb_skip_run= -1;
2527
2528     h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
2529                     (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
2530
2531     if( h->pps.cabac ) {
2532         /* realign */
2533         align_get_bits( &s->gb );
2534
2535         /* init cabac */
2536         ff_init_cabac_states( &h->cabac);
2537         ff_init_cabac_decoder( &h->cabac,
2538                                s->gb.buffer + get_bits_count(&s->gb)/8,
2539                                (get_bits_left(&s->gb) + 7)/8);
2540
2541         ff_h264_init_cabac_states(h);
2542
2543         for(;;){
2544 //START_TIMER
2545             int ret = ff_h264_decode_mb_cabac(h);
2546             int eos;
2547 //STOP_TIMER("decode_mb_cabac")
2548
2549             if(ret>=0) ff_h264_hl_decode_mb(h);
2550
2551             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
2552                 s->mb_y++;
2553
2554                 ret = ff_h264_decode_mb_cabac(h);
2555
2556                 if(ret>=0) ff_h264_hl_decode_mb(h);
2557                 s->mb_y--;
2558             }
2559             eos = get_cabac_terminate( &h->cabac );
2560
2561             if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
2562                 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);
2563                 return 0;
2564             }
2565             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2566                 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);
2567                 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);
2568                 return -1;
2569             }
2570
2571             if( ++s->mb_x >= s->mb_width ) {
2572                 s->mb_x = 0;
2573                 loop_filter(h);
2574                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2575                 ++s->mb_y;
2576                 if(FIELD_OR_MBAFF_PICTURE) {
2577                     ++s->mb_y;
2578                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2579                         predict_field_decoding_flag(h);
2580                 }
2581             }
2582
2583             if( eos || s->mb_y >= s->mb_height ) {
2584                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2585                 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);
2586                 return 0;
2587             }
2588         }
2589
2590     } else {
2591         for(;;){
2592             int ret = ff_h264_decode_mb_cavlc(h);
2593
2594             if(ret>=0) ff_h264_hl_decode_mb(h);
2595
2596             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
2597                 s->mb_y++;
2598                 ret = ff_h264_decode_mb_cavlc(h);
2599
2600                 if(ret>=0) ff_h264_hl_decode_mb(h);
2601                 s->mb_y--;
2602             }
2603
2604             if(ret<0){
2605                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2606                 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);
2607
2608                 return -1;
2609             }
2610
2611             if(++s->mb_x >= s->mb_width){
2612                 s->mb_x=0;
2613                 loop_filter(h);
2614                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2615                 ++s->mb_y;
2616                 if(FIELD_OR_MBAFF_PICTURE) {
2617                     ++s->mb_y;
2618                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2619                         predict_field_decoding_flag(h);
2620                 }
2621                 if(s->mb_y >= s->mb_height){
2622                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2623
2624                     if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
2625                         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);
2626
2627                         return 0;
2628                     }else{
2629                         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);
2630
2631                         return -1;
2632                     }
2633                 }
2634             }
2635
2636             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
2637                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2638                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2639                     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);
2640
2641                     return 0;
2642                 }else{
2643                     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);
2644
2645                     return -1;
2646                 }
2647             }
2648         }
2649     }
2650
2651 #if 0
2652     for(;s->mb_y < s->mb_height; s->mb_y++){
2653         for(;s->mb_x < s->mb_width; s->mb_x++){
2654             int ret= decode_mb(h);
2655
2656             ff_h264_hl_decode_mb(h);
2657
2658             if(ret<0){
2659                 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
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             if(++s->mb_x >= s->mb_width){
2666                 s->mb_x=0;
2667                 if(++s->mb_y >= s->mb_height){
2668                     if(get_bits_count(s->gb) == s->gb.size_in_bits){
2669                         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);
2670
2671                         return 0;
2672                     }else{
2673                         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);
2674
2675                         return -1;
2676                     }
2677                 }
2678             }
2679
2680             if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2681                 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2682                     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);
2683
2684                     return 0;
2685                 }else{
2686                     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);
2687
2688                     return -1;
2689                 }
2690             }
2691         }
2692         s->mb_x=0;
2693         ff_draw_horiz_band(s, 16*s->mb_y, 16);
2694     }
2695 #endif
2696     return -1; //not reached
2697 }
2698
2699 /**
2700  * Call decode_slice() for each context.
2701  *
2702  * @param h h264 master context
2703  * @param context_count number of contexts to execute
2704  */
2705 static void execute_decode_slices(H264Context *h, int context_count){
2706     MpegEncContext * const s = &h->s;
2707     AVCodecContext * const avctx= s->avctx;
2708     H264Context *hx;
2709     int i;
2710
2711     if (s->avctx->hwaccel)
2712         return;
2713     if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2714         return;
2715     if(context_count == 1) {
2716         decode_slice(avctx, &h);
2717     } else {
2718         for(i = 1; i < context_count; i++) {
2719             hx = h->thread_context[i];
2720             hx->s.error_recognition = avctx->error_recognition;
2721             hx->s.error_count = 0;
2722         }
2723
2724         avctx->execute(avctx, (void *)decode_slice,
2725                        h->thread_context, NULL, context_count, sizeof(void*));
2726
2727         /* pull back stuff from slices to master context */
2728         hx = h->thread_context[context_count - 1];
2729         s->mb_x = hx->s.mb_x;
2730         s->mb_y = hx->s.mb_y;
2731         s->dropable = hx->s.dropable;
2732         s->picture_structure = hx->s.picture_structure;
2733         for(i = 1; i < context_count; i++)
2734             h->s.error_count += h->thread_context[i]->s.error_count;
2735     }
2736 }
2737
2738
2739 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
2740     MpegEncContext * const s = &h->s;
2741     AVCodecContext * const avctx= s->avctx;
2742     int buf_index=0;
2743     H264Context *hx; ///< thread context
2744     int context_count = 0;
2745     int next_avc= h->is_avc ? 0 : buf_size;
2746
2747     h->max_contexts = avctx->thread_count;
2748 #if 0
2749     int i;
2750     for(i=0; i<50; i++){
2751         av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
2752     }
2753 #endif
2754     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
2755         h->current_slice = 0;
2756         if (!s->first_field)
2757             s->current_picture_ptr= NULL;
2758         ff_h264_reset_sei(h);
2759     }
2760
2761     for(;;){
2762         int consumed;
2763         int dst_length;
2764         int bit_length;
2765         const uint8_t *ptr;
2766         int i, nalsize = 0;
2767         int err;
2768
2769         if(buf_index >= next_avc) {
2770             if(buf_index >= buf_size) break;
2771             nalsize = 0;
2772             for(i = 0; i < h->nal_length_size; i++)
2773                 nalsize = (nalsize << 8) | buf[buf_index++];
2774             if(nalsize <= 1 || nalsize > buf_size - buf_index){
2775                 if(nalsize == 1){
2776                     buf_index++;
2777                     continue;
2778                 }else{
2779                     av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2780                     break;
2781                 }
2782             }
2783             next_avc= buf_index + nalsize;
2784         } else {
2785             // start code prefix search
2786             for(; buf_index + 3 < next_avc; buf_index++){
2787                 // This should always succeed in the first iteration.
2788                 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2789                     break;
2790             }
2791
2792             if(buf_index+3 >= buf_size) break;
2793
2794             buf_index+=3;
2795             if(buf_index >= next_avc) continue;
2796         }
2797
2798         hx = h->thread_context[context_count];
2799
2800         ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
2801         if (ptr==NULL || dst_length < 0){
2802             return -1;
2803         }
2804         i= buf_index + consumed;
2805         if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
2806            buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
2807             s->workaround_bugs |= FF_BUG_TRUNCATED;
2808
2809         if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
2810         while(ptr[dst_length - 1] == 0 && dst_length > 0)
2811             dst_length--;
2812         }
2813         bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
2814
2815         if(s->avctx->debug&FF_DEBUG_STARTCODE){
2816             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);
2817         }
2818
2819         if (h->is_avc && (nalsize != consumed) && nalsize){
2820             av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
2821         }
2822
2823         buf_index += consumed;
2824
2825         if(  (s->hurry_up == 1 && h->nal_ref_idc  == 0) //FIXME do not discard SEI id
2826            ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0))
2827             continue;
2828
2829       again:
2830         err = 0;
2831         switch(hx->nal_unit_type){
2832         case NAL_IDR_SLICE:
2833             if (h->nal_unit_type != NAL_IDR_SLICE) {
2834                 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2835                 return -1;
2836             }
2837             idr(h); //FIXME ensure we don't loose some frames if there is reordering
2838         case NAL_SLICE:
2839             init_get_bits(&hx->s.gb, ptr, bit_length);
2840             hx->intra_gb_ptr=
2841             hx->inter_gb_ptr= &hx->s.gb;
2842             hx->s.data_partitioning = 0;
2843
2844             if((err = decode_slice_header(hx, h)))
2845                break;
2846
2847             if (h->current_slice == 1) {
2848                 if (s->avctx->hwaccel && s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
2849                     return -1;
2850                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2851                     ff_vdpau_h264_picture_start(s);
2852             }
2853
2854             s->current_picture_ptr->key_frame |=
2855                     (hx->nal_unit_type == NAL_IDR_SLICE) ||
2856                     (h->sei_recovery_frame_cnt >= 0);
2857             if(hx->redundant_pic_count==0 && hx->s.hurry_up < 5
2858                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2859                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2860                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2861                && avctx->skip_frame < AVDISCARD_ALL){
2862                 if(avctx->hwaccel) {
2863                     if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
2864                         return -1;
2865                 }else
2866                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
2867                     static const uint8_t start_code[] = {0x00, 0x00, 0x01};
2868                     ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
2869                     ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
2870                 }else
2871                     context_count++;
2872             }
2873             break;
2874         case NAL_DPA:
2875             init_get_bits(&hx->s.gb, ptr, bit_length);
2876             hx->intra_gb_ptr=
2877             hx->inter_gb_ptr= NULL;
2878
2879             if ((err = decode_slice_header(hx, h)) < 0)
2880                 break;
2881
2882             hx->s.data_partitioning = 1;
2883
2884             break;
2885         case NAL_DPB:
2886             init_get_bits(&hx->intra_gb, ptr, bit_length);
2887             hx->intra_gb_ptr= &hx->intra_gb;
2888             break;
2889         case NAL_DPC:
2890             init_get_bits(&hx->inter_gb, ptr, bit_length);
2891             hx->inter_gb_ptr= &hx->inter_gb;
2892
2893             if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
2894                && s->context_initialized
2895                && s->hurry_up < 5
2896                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2897                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2898                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2899                && avctx->skip_frame < AVDISCARD_ALL)
2900                 context_count++;
2901             break;
2902         case NAL_SEI:
2903             init_get_bits(&s->gb, ptr, bit_length);
2904             ff_h264_decode_sei(h);
2905             break;
2906         case NAL_SPS:
2907             init_get_bits(&s->gb, ptr, bit_length);
2908             ff_h264_decode_seq_parameter_set(h);
2909
2910             if(s->flags& CODEC_FLAG_LOW_DELAY)
2911                 s->low_delay=1;
2912
2913             if(avctx->has_b_frames < 2)
2914                 avctx->has_b_frames= !s->low_delay;
2915             break;
2916         case NAL_PPS:
2917             init_get_bits(&s->gb, ptr, bit_length);
2918
2919             ff_h264_decode_picture_parameter_set(h, bit_length);
2920
2921             break;
2922         case NAL_AUD:
2923         case NAL_END_SEQUENCE:
2924         case NAL_END_STREAM:
2925         case NAL_FILLER_DATA:
2926         case NAL_SPS_EXT:
2927         case NAL_AUXILIARY_SLICE:
2928             break;
2929         default:
2930             av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
2931         }
2932
2933         if(context_count == h->max_contexts) {
2934             execute_decode_slices(h, context_count);
2935             context_count = 0;
2936         }
2937
2938         if (err < 0)
2939             av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
2940         else if(err == 1) {
2941             /* Slice could not be decoded in parallel mode, copy down
2942              * NAL unit stuff to context 0 and restart. Note that
2943              * rbsp_buffer is not transferred, but since we no longer
2944              * run in parallel mode this should not be an issue. */
2945             h->nal_unit_type = hx->nal_unit_type;
2946             h->nal_ref_idc   = hx->nal_ref_idc;
2947             hx = h;
2948             goto again;
2949         }
2950     }
2951     if(context_count)
2952         execute_decode_slices(h, context_count);
2953     return buf_index;
2954 }
2955
2956 /**
2957  * returns the number of bytes consumed for building the current frame
2958  */
2959 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
2960         if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
2961         if(pos+10>buf_size) pos=buf_size; // oops ;)
2962
2963         return pos;
2964 }
2965
2966 static int decode_frame(AVCodecContext *avctx,
2967                              void *data, int *data_size,
2968                              AVPacket *avpkt)
2969 {
2970     const uint8_t *buf = avpkt->data;
2971     int buf_size = avpkt->size;
2972     H264Context *h = avctx->priv_data;
2973     MpegEncContext *s = &h->s;
2974     AVFrame *pict = data;
2975     int buf_index;
2976
2977     s->flags= avctx->flags;
2978     s->flags2= avctx->flags2;
2979
2980    /* end of stream, output what is still in the buffers */
2981     if (buf_size == 0) {
2982         Picture *out;
2983         int i, out_idx;
2984
2985 //FIXME factorize this with the output code below
2986         out = h->delayed_pic[0];
2987         out_idx = 0;
2988         for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2989             if(h->delayed_pic[i]->poc < out->poc){
2990                 out = h->delayed_pic[i];
2991                 out_idx = i;
2992             }
2993
2994         for(i=out_idx; h->delayed_pic[i]; i++)
2995             h->delayed_pic[i] = h->delayed_pic[i+1];
2996
2997         if(out){
2998             *data_size = sizeof(AVFrame);
2999             *pict= *(AVFrame*)out;
3000         }
3001
3002         return 0;
3003     }
3004
3005     buf_index=decode_nal_units(h, buf, buf_size);
3006     if(buf_index < 0)
3007         return -1;
3008
3009     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
3010         if (avctx->skip_frame >= AVDISCARD_NONREF || s->hurry_up) return 0;
3011         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
3012         return -1;
3013     }
3014
3015     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
3016         Picture *out = s->current_picture_ptr;
3017         Picture *cur = s->current_picture_ptr;
3018         int i, pics, out_of_order, out_idx;
3019
3020         field_end(h);
3021
3022         if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
3023             /* Wait for second field. */
3024             *data_size = 0;
3025
3026         } else {
3027             cur->interlaced_frame = 0;
3028             cur->repeat_pict = 0;
3029
3030             /* Signal interlacing information externally. */
3031             /* Prioritize picture timing SEI information over used decoding process if it exists. */
3032
3033             if(h->sps.pic_struct_present_flag){
3034                 switch (h->sei_pic_struct)
3035                 {
3036                 case SEI_PIC_STRUCT_FRAME:
3037                     break;
3038                 case SEI_PIC_STRUCT_TOP_FIELD:
3039                 case SEI_PIC_STRUCT_BOTTOM_FIELD:
3040                     cur->interlaced_frame = 1;
3041                     break;
3042                 case SEI_PIC_STRUCT_TOP_BOTTOM:
3043                 case SEI_PIC_STRUCT_BOTTOM_TOP:
3044                     if (FIELD_OR_MBAFF_PICTURE)
3045                         cur->interlaced_frame = 1;
3046                     else
3047                         // try to flag soft telecine progressive
3048                         cur->interlaced_frame = h->prev_interlaced_frame;
3049                     break;
3050                 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
3051                 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
3052                     // Signal the possibility of telecined film externally (pic_struct 5,6)
3053                     // From these hints, let the applications decide if they apply deinterlacing.
3054                     cur->repeat_pict = 1;
3055                     break;
3056                 case SEI_PIC_STRUCT_FRAME_DOUBLING:
3057                     // Force progressive here, as doubling interlaced frame is a bad idea.
3058                     cur->repeat_pict = 2;
3059                     break;
3060                 case SEI_PIC_STRUCT_FRAME_TRIPLING:
3061                     cur->repeat_pict = 4;
3062                     break;
3063                 }
3064
3065                 if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
3066                     cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
3067             }else{
3068                 /* Derive interlacing flag from used decoding process. */
3069                 cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
3070             }
3071             h->prev_interlaced_frame = cur->interlaced_frame;
3072
3073             if (cur->field_poc[0] != cur->field_poc[1]){
3074                 /* Derive top_field_first from field pocs. */
3075                 cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
3076             }else{
3077                 if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
3078                     /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
3079                     if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
3080                       || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
3081                         cur->top_field_first = 1;
3082                     else
3083                         cur->top_field_first = 0;
3084                 }else{
3085                     /* Most likely progressive */
3086                     cur->top_field_first = 0;
3087                 }
3088             }
3089
3090         //FIXME do something with unavailable reference frames
3091
3092             /* Sort B-frames into display order */
3093
3094             if(h->sps.bitstream_restriction_flag
3095                && s->avctx->has_b_frames < h->sps.num_reorder_frames){
3096                 s->avctx->has_b_frames = h->sps.num_reorder_frames;
3097                 s->low_delay = 0;
3098             }
3099
3100             if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
3101                && !h->sps.bitstream_restriction_flag){
3102                 s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
3103                 s->low_delay= 0;
3104             }
3105
3106             pics = 0;
3107             while(h->delayed_pic[pics]) pics++;
3108
3109             assert(pics <= MAX_DELAYED_PIC_COUNT);
3110
3111             h->delayed_pic[pics++] = cur;
3112             if(cur->reference == 0)
3113                 cur->reference = DELAYED_PIC_REF;
3114
3115             out = h->delayed_pic[0];
3116             out_idx = 0;
3117             for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
3118                 if(h->delayed_pic[i]->poc < out->poc){
3119                     out = h->delayed_pic[i];
3120                     out_idx = i;
3121                 }
3122             if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
3123                 h->outputed_poc= INT_MIN;
3124             out_of_order = out->poc < h->outputed_poc;
3125
3126             if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
3127                 { }
3128             else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
3129                || (s->low_delay &&
3130                 ((h->outputed_poc != INT_MIN && out->poc > h->outputed_poc + 2)
3131                  || cur->pict_type == FF_B_TYPE)))
3132             {
3133                 s->low_delay = 0;
3134                 s->avctx->has_b_frames++;
3135             }
3136
3137             if(out_of_order || pics > s->avctx->has_b_frames){
3138                 out->reference &= ~DELAYED_PIC_REF;
3139                 for(i=out_idx; h->delayed_pic[i]; i++)
3140                     h->delayed_pic[i] = h->delayed_pic[i+1];
3141             }
3142             if(!out_of_order && pics > s->avctx->has_b_frames){
3143                 *data_size = sizeof(AVFrame);
3144
3145                 if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
3146                     h->outputed_poc = INT_MIN;
3147                 } else
3148                     h->outputed_poc = out->poc;
3149                 *pict= *(AVFrame*)out;
3150             }else{
3151                 av_log(avctx, AV_LOG_DEBUG, "no picture\n");
3152             }
3153         }
3154     }
3155
3156     assert(pict->data[0] || !*data_size);
3157     ff_print_debug_info(s, pict);
3158 //printf("out %d\n", (int)pict->data[0]);
3159
3160     return get_consumed_bytes(s, buf_index, buf_size);
3161 }
3162 #if 0
3163 static inline void fill_mb_avail(H264Context *h){
3164     MpegEncContext * const s = &h->s;
3165     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3166
3167     if(s->mb_y){
3168         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
3169         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
3170         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
3171     }else{
3172         h->mb_avail[0]=
3173         h->mb_avail[1]=
3174         h->mb_avail[2]= 0;
3175     }
3176     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
3177     h->mb_avail[4]= 1; //FIXME move out
3178     h->mb_avail[5]= 0; //FIXME move out
3179 }
3180 #endif
3181
3182 #ifdef TEST
3183 #undef printf
3184 #undef random
3185 #define COUNT 8000
3186 #define SIZE (COUNT*40)
3187 int main(void){
3188     int i;
3189     uint8_t temp[SIZE];
3190     PutBitContext pb;
3191     GetBitContext gb;
3192 //    int int_temp[10000];
3193     DSPContext dsp;
3194     AVCodecContext avctx;
3195
3196     dsputil_init(&dsp, &avctx);
3197
3198     init_put_bits(&pb, temp, SIZE);
3199     printf("testing unsigned exp golomb\n");
3200     for(i=0; i<COUNT; i++){
3201         START_TIMER
3202         set_ue_golomb(&pb, i);
3203         STOP_TIMER("set_ue_golomb");
3204     }
3205     flush_put_bits(&pb);
3206
3207     init_get_bits(&gb, temp, 8*SIZE);
3208     for(i=0; i<COUNT; i++){
3209         int j, s;
3210
3211         s= show_bits(&gb, 24);
3212
3213         START_TIMER
3214         j= get_ue_golomb(&gb);
3215         if(j != i){
3216             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
3217 //            return -1;
3218         }
3219         STOP_TIMER("get_ue_golomb");
3220     }
3221
3222
3223     init_put_bits(&pb, temp, SIZE);
3224     printf("testing signed exp golomb\n");
3225     for(i=0; i<COUNT; i++){
3226         START_TIMER
3227         set_se_golomb(&pb, i - COUNT/2);
3228         STOP_TIMER("set_se_golomb");
3229     }
3230     flush_put_bits(&pb);
3231
3232     init_get_bits(&gb, temp, 8*SIZE);
3233     for(i=0; i<COUNT; i++){
3234         int j, s;
3235
3236         s= show_bits(&gb, 24);
3237
3238         START_TIMER
3239         j= get_se_golomb(&gb);
3240         if(j != i - COUNT/2){
3241             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
3242 //            return -1;
3243         }
3244         STOP_TIMER("get_se_golomb");
3245     }
3246
3247 #if 0
3248     printf("testing 4x4 (I)DCT\n");
3249
3250     DCTELEM block[16];
3251     uint8_t src[16], ref[16];
3252     uint64_t error= 0, max_error=0;
3253
3254     for(i=0; i<COUNT; i++){
3255         int j;
3256 //        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
3257         for(j=0; j<16; j++){
3258             ref[j]= random()%255;
3259             src[j]= random()%255;
3260         }
3261
3262         h264_diff_dct_c(block, src, ref, 4);
3263
3264         //normalize
3265         for(j=0; j<16; j++){
3266 //            printf("%d ", block[j]);
3267             block[j]= block[j]*4;
3268             if(j&1) block[j]= (block[j]*4 + 2)/5;
3269             if(j&4) block[j]= (block[j]*4 + 2)/5;
3270         }
3271 //        printf("\n");
3272
3273         h->h264dsp.h264_idct_add(ref, block, 4);
3274 /*        for(j=0; j<16; j++){
3275             printf("%d ", ref[j]);
3276         }
3277         printf("\n");*/
3278
3279         for(j=0; j<16; j++){
3280             int diff= FFABS(src[j] - ref[j]);
3281
3282             error+= diff*diff;
3283             max_error= FFMAX(max_error, diff);
3284         }
3285     }
3286     printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
3287     printf("testing quantizer\n");
3288     for(qp=0; qp<52; qp++){
3289         for(i=0; i<16; i++)
3290             src1_block[i]= src2_block[i]= random()%255;
3291
3292     }
3293     printf("Testing NAL layer\n");
3294
3295     uint8_t bitstream[COUNT];
3296     uint8_t nal[COUNT*2];
3297     H264Context h;
3298     memset(&h, 0, sizeof(H264Context));
3299
3300     for(i=0; i<COUNT; i++){
3301         int zeros= i;
3302         int nal_length;
3303         int consumed;
3304         int out_length;
3305         uint8_t *out;
3306         int j;
3307
3308         for(j=0; j<COUNT; j++){
3309             bitstream[j]= (random() % 255) + 1;
3310         }
3311
3312         for(j=0; j<zeros; j++){
3313             int pos= random() % COUNT;
3314             while(bitstream[pos] == 0){
3315                 pos++;
3316                 pos %= COUNT;
3317             }
3318             bitstream[pos]=0;
3319         }
3320
3321         START_TIMER
3322
3323         nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
3324         if(nal_length<0){
3325             printf("encoding failed\n");
3326             return -1;
3327         }
3328
3329         out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
3330
3331         STOP_TIMER("NAL")
3332
3333         if(out_length != COUNT){
3334             printf("incorrect length %d %d\n", out_length, COUNT);
3335             return -1;
3336         }
3337
3338         if(consumed != nal_length){
3339             printf("incorrect consumed length %d %d\n", nal_length, consumed);
3340             return -1;
3341         }
3342
3343         if(memcmp(bitstream, out, COUNT)){
3344             printf("mismatch\n");
3345             return -1;
3346         }
3347     }
3348 #endif
3349
3350     printf("Testing RBSP\n");
3351
3352
3353     return 0;
3354 }
3355 #endif /* TEST */
3356
3357
3358 av_cold void ff_h264_free_context(H264Context *h)
3359 {
3360     int i;
3361
3362     free_tables(h); //FIXME cleanup init stuff perhaps
3363
3364     for(i = 0; i < MAX_SPS_COUNT; i++)
3365         av_freep(h->sps_buffers + i);
3366
3367     for(i = 0; i < MAX_PPS_COUNT; i++)
3368         av_freep(h->pps_buffers + i);
3369 }
3370
3371 av_cold int ff_h264_decode_end(AVCodecContext *avctx)
3372 {
3373     H264Context *h = avctx->priv_data;
3374     MpegEncContext *s = &h->s;
3375
3376     ff_h264_free_context(h);
3377
3378     MPV_common_end(s);
3379
3380 //    memset(h, 0, sizeof(H264Context));
3381
3382     return 0;
3383 }
3384
3385
3386 AVCodec h264_decoder = {
3387     "h264",
3388     AVMEDIA_TYPE_VIDEO,
3389     CODEC_ID_H264,
3390     sizeof(H264Context),
3391     ff_h264_decode_init,
3392     NULL,
3393     ff_h264_decode_end,
3394     decode_frame,
3395     /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3396     .flush= flush_dpb,
3397     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
3398 };
3399
3400 #if CONFIG_H264_VDPAU_DECODER
3401 AVCodec h264_vdpau_decoder = {
3402     "h264_vdpau",
3403     AVMEDIA_TYPE_VIDEO,
3404     CODEC_ID_H264,
3405     sizeof(H264Context),
3406     ff_h264_decode_init,
3407     NULL,
3408     ff_h264_decode_end,
3409     decode_frame,
3410     CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3411     .flush= flush_dpb,
3412     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
3413     .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
3414 };
3415 #endif