libavcodec/h264.c

   1 /*
   2  * H.26L/H.264/AVC/JVT/14496-10/... decoder
   3  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
   4  *
   5  * This file is part of Libav.
   6  *
   7  * Libav is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2.1 of the License, or (at your option) any later version.
  11  *
  12  * Libav is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with Libav; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 /**
  23  * @file
  24  * H.264 / AVC / MPEG4 part10 codec.
  25  * @author Michael Niedermayer <michaelni@gmx.at>
  26  */
  27
  28 #include "libavutil/imgutils.h"
  29 #include "internal.h"
  30 #include "dsputil.h"
  31 #include "avcodec.h"
  32 #include "mpegvideo.h"
  33 #include "h264.h"
  34 #include "h264data.h"
  35 #include "h264_mvpred.h"
  36 #include "golomb.h"
  37 #include "mathops.h"
  38 #include "rectangle.h"
  39 #include "vdpau_internal.h"
  40 #include "libavutil/avassert.h"
  41
  42 #include "cabac.h"
  43
  44 //#undef NDEBUG
  45 #include <assert.h>
  46
  47 static const uint8_t rem6[52]={
  48 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
  49 };
  50
  51 static const uint8_t div6[52]={
  52 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8,
  53 };
  54
  55 static const enum PixelFormat hwaccel_pixfmt_list_h264_jpeg_420[] = {
  56     PIX_FMT_DXVA2_VLD,
  57     PIX_FMT_VAAPI_VLD,
  58     PIX_FMT_YUVJ420P,
  59     PIX_FMT_NONE
  60 };
  61
  62 void ff_h264_write_back_intra_pred_mode(H264Context *h){
  63     int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
  64
  65     AV_COPY32(mode, h->intra4x4_pred_mode_cache + 4 + 8*4);
  66     mode[4]= h->intra4x4_pred_mode_cache[7+8*3];
  67     mode[5]= h->intra4x4_pred_mode_cache[7+8*2];
  68     mode[6]= h->intra4x4_pred_mode_cache[7+8*1];
  69 }
  70
  71 /**
  72  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
  73  */
  74 int ff_h264_check_intra4x4_pred_mode(H264Context *h){
  75     MpegEncContext * const s = &h->s;
  76     static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
  77     static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
  78     int i;
  79
  80     if(!(h->top_samples_available&0x8000)){
  81         for(i=0; i<4; i++){
  82             int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
  83             if(status<0){
  84                 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
  85                 return -1;
  86             } else if(status){
  87                 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
  88             }
  89         }
  90     }
  91
  92     if((h->left_samples_available&0x8888)!=0x8888){
  93         static const int mask[4]={0x8000,0x2000,0x80,0x20};
  94         for(i=0; i<4; i++){
  95             if(!(h->left_samples_available&mask[i])){
  96                 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
  97                 if(status<0){
  98                     av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
  99                     return -1;
 100                 } else if(status){
 101                     h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
 102                 }
 103             }
 104         }
 105     }
 106
 107     return 0;
 108 } //FIXME cleanup like ff_h264_check_intra_pred_mode
 109
 110 /**
 111  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
 112  */
 113 int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
 114     MpegEncContext * const s = &h->s;
 115     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
 116     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
 117
 118     if(mode > 6U) {
 119         av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
 120         return -1;
 121     }
 122
 123     if(!(h->top_samples_available&0x8000)){
 124         mode= top[ mode ];
 125         if(mode<0){
 126             av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
 127             return -1;
 128         }
 129     }
 130
 131     if((h->left_samples_available&0x8080) != 0x8080){
 132         mode= left[ mode ];
 133         if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
 134             mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
 135         }
 136         if(mode<0){
 137             av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
 138             return -1;
 139         }
 140     }
 141
 142     return mode;
 143 }
 144
 145 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
 146     int i, si, di;
 147     uint8_t *dst;
 148     int bufidx;
 149
 150 //    src[0]&0x80;                //forbidden bit
 151     h->nal_ref_idc= src[0]>>5;
 152     h->nal_unit_type= src[0]&0x1F;
 153
 154     src++; length--;
 155 #if 0
 156     for(i=0; i<length; i++)
 157         printf("%2X ", src[i]);
 158 #endif
 159
 160 #if HAVE_FAST_UNALIGNED
 161 # if HAVE_FAST_64BIT
 162 #   define RS 7
 163     for(i=0; i+1<length; i+=9){
 164         if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
 165 # else
 166 #   define RS 3
 167     for(i=0; i+1<length; i+=5){
 168         if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
 169 # endif
 170             continue;
 171         if(i>0 && !src[i]) i--;
 172         while(src[i]) i++;
 173 #else
 174 #   define RS 0
 175     for(i=0; i+1<length; i+=2){
 176         if(src[i]) continue;
 177         if(i>0 && src[i-1]==0) i--;
 178 #endif
 179         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
 180             if(src[i+2]!=3){
 181                 /* startcode, so we must be past the end */
 182                 length=i;
 183             }
 184             break;
 185         }
 186         i-= RS;
 187     }
 188
 189     if(i>=length-1){ //no escaped 0
 190         *dst_length= length;
 191         *consumed= length+1; //+1 for the header
 192         return src;
 193     }
 194
 195     bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
 196     av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
 197     dst= h->rbsp_buffer[bufidx];
 198
 199     if (dst == NULL){
 200         return NULL;
 201     }
 202
 203 //printf("decoding esc\n");
 204     memcpy(dst, src, i);
 205     si=di=i;
 206     while(si+2<length){
 207         //remove escapes (very rare 1:2^22)
 208         if(src[si+2]>3){
 209             dst[di++]= src[si++];
 210             dst[di++]= src[si++];
 211         }else if(src[si]==0 && src[si+1]==0){
 212             if(src[si+2]==3){ //escape
 213                 dst[di++]= 0;
 214                 dst[di++]= 0;
 215                 si+=3;
 216                 continue;
 217             }else //next start code
 218                 goto nsc;
 219         }
 220
 221         dst[di++]= src[si++];
 222     }
 223     while(si<length)
 224         dst[di++]= src[si++];
 225 nsc:
 226
 227     memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
 228
 229     *dst_length= di;
 230     *consumed= si + 1;//+1 for the header
 231 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
 232     return dst;
 233 }
 234
 235 /**
 236  * Identify the exact end of the bitstream
 237  * @return the length of the trailing, or 0 if damaged
 238  */
 239 static int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
 240     int v= *src;
 241     int r;
 242
 243     tprintf(h->s.avctx, "rbsp trailing %X\n", v);
 244
 245     for(r=1; r<9; r++){
 246         if(v&1) return r;
 247         v>>=1;
 248     }
 249     return 0;
 250 }
 251
 252 #if 0
 253 /**
 254  * DCT transforms the 16 dc values.
 255  * @param qp quantization parameter ??? FIXME
 256  */
 257 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
 258 //    const int qmul= dequant_coeff[qp][0];
 259     int i;
 260     int temp[16]; //FIXME check if this is a good idea
 261     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
 262     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
 263
 264     for(i=0; i<4; i++){
 265         const int offset= y_offset[i];
 266         const int z0= block[offset+stride*0] + block[offset+stride*4];
 267         const int z1= block[offset+stride*0] - block[offset+stride*4];
 268         const int z2= block[offset+stride*1] - block[offset+stride*5];
 269         const int z3= block[offset+stride*1] + block[offset+stride*5];
 270
 271         temp[4*i+0]= z0+z3;
 272         temp[4*i+1]= z1+z2;
 273         temp[4*i+2]= z1-z2;
 274         temp[4*i+3]= z0-z3;
 275     }
 276
 277     for(i=0; i<4; i++){
 278         const int offset= x_offset[i];
 279         const int z0= temp[4*0+i] + temp[4*2+i];
 280         const int z1= temp[4*0+i] - temp[4*2+i];
 281         const int z2= temp[4*1+i] - temp[4*3+i];
 282         const int z3= temp[4*1+i] + temp[4*3+i];
 283
 284         block[stride*0 +offset]= (z0 + z3)>>1;
 285         block[stride*2 +offset]= (z1 + z2)>>1;
 286         block[stride*8 +offset]= (z1 - z2)>>1;
 287         block[stride*10+offset]= (z0 - z3)>>1;
 288     }
 289 }
 290 #endif
 291
 292 #undef xStride
 293 #undef stride
 294
 295 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qmul){
 296     const int stride= 16*2;
 297     const int xStride= 16;
 298     int a,b,c,d,e;
 299
 300     a= block[stride*0 + xStride*0];
 301     b= block[stride*0 + xStride*1];
 302     c= block[stride*1 + xStride*0];
 303     d= block[stride*1 + xStride*1];
 304
 305     e= a-b;
 306     a= a+b;
 307     b= c-d;
 308     c= c+d;
 309
 310     block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
 311     block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
 312     block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
 313     block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
 314 }
 315
 316 #if 0
 317 static void chroma_dc_dct_c(DCTELEM *block){
 318     const int stride= 16*2;
 319     const int xStride= 16;
 320     int a,b,c,d,e;
 321
 322     a= block[stride*0 + xStride*0];
 323     b= block[stride*0 + xStride*1];
 324     c= block[stride*1 + xStride*0];
 325     d= block[stride*1 + xStride*1];
 326
 327     e= a-b;
 328     a= a+b;
 329     b= c-d;
 330     c= c+d;
 331
 332     block[stride*0 + xStride*0]= (a+c);
 333     block[stride*0 + xStride*1]= (e+b);
 334     block[stride*1 + xStride*0]= (a-c);
 335     block[stride*1 + xStride*1]= (e-b);
 336 }
 337 #endif
 338
 339 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
 340                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 341                            int src_x_offset, int src_y_offset,
 342                            qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
 343     MpegEncContext * const s = &h->s;
 344     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
 345     int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
 346     const int luma_xy= (mx&3) + ((my&3)<<2);
 347     uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
 348     uint8_t * src_cb, * src_cr;
 349     int extra_width= h->emu_edge_width;
 350     int extra_height= h->emu_edge_height;
 351     int emu=0;
 352     const int full_mx= mx>>2;
 353     const int full_my= my>>2;
 354     const int pic_width  = 16*s->mb_width;
 355     const int pic_height = 16*s->mb_height >> MB_FIELD;
 356
 357     if(mx&7) extra_width -= 3;
 358     if(my&7) extra_height -= 3;
 359
 360     if(   full_mx < 0-extra_width
 361        || full_my < 0-extra_height
 362        || full_mx + 16/*FIXME*/ > pic_width + extra_width
 363        || full_my + 16/*FIXME*/ > pic_height + extra_height){
 364         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
 365             src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
 366         emu=1;
 367     }
 368
 369     qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
 370     if(!square){
 371         qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
 372     }
 373
 374     if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
 375
 376     if(MB_FIELD){
 377         // chroma offset when predicting from a field of opposite parity
 378         my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
 379         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
 380     }
 381     src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
 382     src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
 383
 384     if(emu){
 385         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
 386             src_cb= s->edge_emu_buffer;
 387     }
 388     chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
 389
 390     if(emu){
 391         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
 392             src_cr= s->edge_emu_buffer;
 393     }
 394     chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
 395 }
 396
 397 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
 398                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 399                            int x_offset, int y_offset,
 400                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 401                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
 402                            int list0, int list1){
 403     MpegEncContext * const s = &h->s;
 404     qpel_mc_func *qpix_op=  qpix_put;
 405     h264_chroma_mc_func chroma_op= chroma_put;
 406
 407     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
 408     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
 409     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
 410     x_offset += 8*s->mb_x;
 411     y_offset += 8*(s->mb_y >> MB_FIELD);
 412
 413     if(list0){
 414         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
 415         mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
 416                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
 417                            qpix_op, chroma_op);
 418
 419         qpix_op=  qpix_avg;
 420         chroma_op= chroma_avg;
 421     }
 422
 423     if(list1){
 424         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
 425         mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
 426                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
 427                            qpix_op, chroma_op);
 428     }
 429 }
 430
 431 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
 432                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 433                            int x_offset, int y_offset,
 434                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 435                            h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
 436                            h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
 437                            int list0, int list1){
 438     MpegEncContext * const s = &h->s;
 439
 440     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
 441     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
 442     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
 443     x_offset += 8*s->mb_x;
 444     y_offset += 8*(s->mb_y >> MB_FIELD);
 445
 446     if(list0 && list1){
 447         /* don't optimize for luma-only case, since B-frames usually
 448          * use implicit weights => chroma too. */
 449         uint8_t *tmp_cb = s->obmc_scratchpad;
 450         uint8_t *tmp_cr = s->obmc_scratchpad + 8;
 451         uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
 452         int refn0 = h->ref_cache[0][ scan8[n] ];
 453         int refn1 = h->ref_cache[1][ scan8[n] ];
 454
 455         mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
 456                     dest_y, dest_cb, dest_cr,
 457                     x_offset, y_offset, qpix_put, chroma_put);
 458         mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
 459                     tmp_y, tmp_cb, tmp_cr,
 460                     x_offset, y_offset, qpix_put, chroma_put);
 461
 462         if(h->use_weight == 2){
 463             int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
 464             int weight1 = 64 - weight0;
 465             luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
 466             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
 467             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
 468         }else{
 469             luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
 470                             h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
 471                             h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
 472             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 473                             h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
 474                             h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
 475             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 476                             h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
 477                             h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
 478         }
 479     }else{
 480         int list = list1 ? 1 : 0;
 481         int refn = h->ref_cache[list][ scan8[n] ];
 482         Picture *ref= &h->ref_list[list][refn];
 483         mc_dir_part(h, ref, n, square, chroma_height, delta, list,
 484                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
 485                     qpix_put, chroma_put);
 486
 487         luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
 488                        h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
 489         if(h->use_weight_chroma){
 490             chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 491                              h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
 492             chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 493                              h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
 494         }
 495     }
 496 }
 497
 498 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
 499                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 500                            int x_offset, int y_offset,
 501                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 502                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
 503                            h264_weight_func *weight_op, h264_biweight_func *weight_avg,
 504                            int list0, int list1){
 505     if((h->use_weight==2 && list0 && list1
 506         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
 507        || h->use_weight==1)
 508         mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
 509                          x_offset, y_offset, qpix_put, chroma_put,
 510                          weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
 511     else
 512         mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
 513                     x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
 514 }
 515
 516 static inline void prefetch_motion(H264Context *h, int list){
 517     /* fetch pixels for estimated mv 4 macroblocks ahead
 518      * optimized for 64byte cache lines */
 519     MpegEncContext * const s = &h->s;
 520     const int refn = h->ref_cache[list][scan8[0]];
 521     if(refn >= 0){
 522         const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
 523         const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
 524         uint8_t **src= h->ref_list[list][refn].data;
 525         int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
 526         s->dsp.prefetch(src[0]+off, s->linesize, 4);
 527         off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
 528         s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
 529     }
 530 }
 531
 532 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 533                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
 534                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
 535                       h264_weight_func *weight_op, h264_biweight_func *weight_avg){
 536     MpegEncContext * const s = &h->s;
 537     const int mb_xy= h->mb_xy;
 538     const int mb_type= s->current_picture.mb_type[mb_xy];
 539
 540     assert(IS_INTER(mb_type));
 541
 542     prefetch_motion(h, 0);
 543
 544     if(IS_16X16(mb_type)){
 545         mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
 546                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
 547                 weight_op, weight_avg,
 548                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
 549     }else if(IS_16X8(mb_type)){
 550         mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
 551                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
 552                 &weight_op[1], &weight_avg[1],
 553                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
 554         mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
 555                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
 556                 &weight_op[1], &weight_avg[1],
 557                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
 558     }else if(IS_8X16(mb_type)){
 559         mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
 560                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 561                 &weight_op[2], &weight_avg[2],
 562                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
 563         mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
 564                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 565                 &weight_op[2], &weight_avg[2],
 566                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
 567     }else{
 568         int i;
 569
 570         assert(IS_8X8(mb_type));
 571
 572         for(i=0; i<4; i++){
 573             const int sub_mb_type= h->sub_mb_type[i];
 574             const int n= 4*i;
 575             int x_offset= (i&1)<<2;
 576             int y_offset= (i&2)<<1;
 577
 578             if(IS_SUB_8X8(sub_mb_type)){
 579                 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 580                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 581                     &weight_op[3], &weight_avg[3],
 582                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 583             }else if(IS_SUB_8X4(sub_mb_type)){
 584                 mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 585                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
 586                     &weight_op[4], &weight_avg[4],
 587                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 588                 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
 589                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
 590                     &weight_op[4], &weight_avg[4],
 591                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 592             }else if(IS_SUB_4X8(sub_mb_type)){
 593                 mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 594                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 595                     &weight_op[5], &weight_avg[5],
 596                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 597                 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
 598                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 599                     &weight_op[5], &weight_avg[5],
 600                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 601             }else{
 602                 int j;
 603                 assert(IS_SUB_4X4(sub_mb_type));
 604                 for(j=0; j<4; j++){
 605                     int sub_x_offset= x_offset + 2*(j&1);
 606                     int sub_y_offset= y_offset +   (j&2);
 607                     mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
 608                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 609                         &weight_op[6], &weight_avg[6],
 610                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 611                 }
 612             }
 613         }
 614     }
 615
 616     prefetch_motion(h, 1);
 617 }
 618
 619
 620 static void free_tables(H264Context *h, int free_rbsp){
 621     int i;
 622     H264Context *hx;
 623     av_freep(&h->intra4x4_pred_mode);
 624     av_freep(&h->chroma_pred_mode_table);
 625     av_freep(&h->cbp_table);
 626     av_freep(&h->mvd_table[0]);
 627     av_freep(&h->mvd_table[1]);
 628     av_freep(&h->direct_table);
 629     av_freep(&h->non_zero_count);
 630     av_freep(&h->slice_table_base);
 631     h->slice_table= NULL;
 632     av_freep(&h->list_counts);
 633
 634     av_freep(&h->mb2b_xy);
 635     av_freep(&h->mb2br_xy);
 636
 637     for(i = 0; i < MAX_THREADS; i++) {
 638         hx = h->thread_context[i];
 639         if(!hx) continue;
 640         av_freep(&hx->top_borders[1]);
 641         av_freep(&hx->top_borders[0]);
 642         av_freep(&hx->s.obmc_scratchpad);
 643         if (free_rbsp){
 644             av_freep(&hx->rbsp_buffer[1]);
 645             av_freep(&hx->rbsp_buffer[0]);
 646             hx->rbsp_buffer_size[0] = 0;
 647             hx->rbsp_buffer_size[1] = 0;
 648         }
 649         if (i) av_freep(&h->thread_context[i]);
 650     }
 651 }
 652
 653 static void init_dequant8_coeff_table(H264Context *h){
 654     int i,q,x;
 655     h->dequant8_coeff[0] = h->dequant8_buffer[0];
 656     h->dequant8_coeff[1] = h->dequant8_buffer[1];
 657
 658     for(i=0; i<2; i++ ){
 659         if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
 660             h->dequant8_coeff[1] = h->dequant8_buffer[0];
 661             break;
 662         }
 663
 664         for(q=0; q<52; q++){
 665             int shift = div6[q];
 666             int idx = rem6[q];
 667             for(x=0; x<64; x++)
 668                 h->dequant8_coeff[i][q][(x>>3)|((x&7)<<3)] =
 669                     ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
 670                     h->pps.scaling_matrix8[i][x]) << shift;
 671         }
 672     }
 673 }
 674
 675 static void init_dequant4_coeff_table(H264Context *h){
 676     int i,j,q,x;
 677     for(i=0; i<6; i++ ){
 678         h->dequant4_coeff[i] = h->dequant4_buffer[i];
 679         for(j=0; j<i; j++){
 680             if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
 681                 h->dequant4_coeff[i] = h->dequant4_buffer[j];
 682                 break;
 683             }
 684         }
 685         if(j<i)
 686             continue;
 687
 688         for(q=0; q<52; q++){
 689             int shift = div6[q] + 2;
 690             int idx = rem6[q];
 691             for(x=0; x<16; x++)
 692                 h->dequant4_coeff[i][q][(x>>2)|((x<<2)&0xF)] =
 693                     ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
 694                     h->pps.scaling_matrix4[i][x]) << shift;
 695         }
 696     }
 697 }
 698
 699 static void init_dequant_tables(H264Context *h){
 700     int i,x;
 701     init_dequant4_coeff_table(h);
 702     if(h->pps.transform_8x8_mode)
 703         init_dequant8_coeff_table(h);
 704     if(h->sps.transform_bypass){
 705         for(i=0; i<6; i++)
 706             for(x=0; x<16; x++)
 707                 h->dequant4_coeff[i][0][x] = 1<<6;
 708         if(h->pps.transform_8x8_mode)
 709             for(i=0; i<2; i++)
 710                 for(x=0; x<64; x++)
 711                     h->dequant8_coeff[i][0][x] = 1<<6;
 712     }
 713 }
 714
 715
 716 int ff_h264_alloc_tables(H264Context *h){
 717     MpegEncContext * const s = &h->s;
 718     const int big_mb_num= s->mb_stride * (s->mb_height+1);
 719     const int row_mb_num= 2*s->mb_stride*s->avctx->thread_count;
 720     int x,y;
 721
 722     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8  * sizeof(uint8_t), fail)
 723
 724     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 32 * sizeof(uint8_t), fail)
 725     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
 726     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
 727
 728     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
 729     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
 730     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
 731     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
 732     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
 733
 734     memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
 735     h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
 736
 737     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy  , big_mb_num * sizeof(uint32_t), fail);
 738     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
 739     for(y=0; y<s->mb_height; y++){
 740         for(x=0; x<s->mb_width; x++){
 741             const int mb_xy= x + y*s->mb_stride;
 742             const int b_xy = 4*x + 4*y*h->b_stride;
 743
 744             h->mb2b_xy [mb_xy]= b_xy;
 745             h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
 746         }
 747     }
 748
 749     s->obmc_scratchpad = NULL;
 750
 751     if(!h->dequant4_coeff[0])
 752         init_dequant_tables(h);
 753
 754     return 0;
 755 fail:
 756     free_tables(h, 1);
 757     return -1;
 758 }
 759
 760 /**
 761  * Mimic alloc_tables(), but for every context thread.
 762  */
 763 static void clone_tables(H264Context *dst, H264Context *src, int i){
 764     MpegEncContext * const s = &src->s;
 765     dst->intra4x4_pred_mode       = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
 766     dst->non_zero_count           = src->non_zero_count;
 767     dst->slice_table              = src->slice_table;
 768     dst->cbp_table                = src->cbp_table;
 769     dst->mb2b_xy                  = src->mb2b_xy;
 770     dst->mb2br_xy                 = src->mb2br_xy;
 771     dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
 772     dst->mvd_table[0]             = src->mvd_table[0] + i*8*2*s->mb_stride;
 773     dst->mvd_table[1]             = src->mvd_table[1] + i*8*2*s->mb_stride;
 774     dst->direct_table             = src->direct_table;
 775     dst->list_counts              = src->list_counts;
 776
 777     dst->s.obmc_scratchpad = NULL;
 778     ff_h264_pred_init(&dst->hpc, src->s.codec_id);
 779 }
 780
 781 /**
 782  * Init context
 783  * Allocate buffers which are not shared amongst multiple threads.
 784  */
 785 static int context_init(H264Context *h){
 786     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
 787     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
 788
 789     h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
 790     h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
 791
 792     return 0;
 793 fail:
 794     return -1; // free_tables will clean up for us
 795 }
 796
 797 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
 798
 799 static av_cold void common_init(H264Context *h){
 800     MpegEncContext * const s = &h->s;
 801
 802     s->width = s->avctx->width;
 803     s->height = s->avctx->height;
 804     s->codec_id= s->avctx->codec->id;
 805
 806     ff_h264dsp_init(&h->h264dsp);
 807     ff_h264_pred_init(&h->hpc, s->codec_id);
 808
 809     h->dequant_coeff_pps= -1;
 810     s->unrestricted_mv=1;
 811     s->decode=1; //FIXME
 812
 813     dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
 814
 815     memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
 816     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
 817 }
 818
 819 int ff_h264_decode_extradata(H264Context *h)
 820 {
 821     AVCodecContext *avctx = h->s.avctx;
 822
 823     if(*(char *)avctx->extradata == 1){
 824         int i, cnt, nalsize;
 825         unsigned char *p = avctx->extradata;
 826
 827         h->is_avc = 1;
 828
 829         if(avctx->extradata_size < 7) {
 830             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
 831             return -1;
 832         }
 833         /* sps and pps in the avcC always have length coded with 2 bytes,
 834            so put a fake nal_length_size = 2 while parsing them */
 835         h->nal_length_size = 2;
 836         // Decode sps from avcC
 837         cnt = *(p+5) & 0x1f; // Number of sps
 838         p += 6;
 839         for (i = 0; i < cnt; i++) {
 840             nalsize = AV_RB16(p) + 2;
 841             if(decode_nal_units(h, p, nalsize) < 0) {
 842                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
 843                 return -1;
 844             }
 845             p += nalsize;
 846         }
 847         // Decode pps from avcC
 848         cnt = *(p++); // Number of pps
 849         for (i = 0; i < cnt; i++) {
 850             nalsize = AV_RB16(p) + 2;
 851             if(decode_nal_units(h, p, nalsize)  != nalsize) {
 852                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
 853                 return -1;
 854             }
 855             p += nalsize;
 856         }
 857         // Now store right nal length size, that will be use to parse all other nals
 858         h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
 859     } else {
 860         h->is_avc = 0;
 861         if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
 862             return -1;
 863     }
 864     return 0;
 865 }
 866
 867 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
 868     H264Context *h= avctx->priv_data;
 869     MpegEncContext * const s = &h->s;
 870
 871     MPV_decode_defaults(s);
 872
 873     s->avctx = avctx;
 874     common_init(h);
 875
 876     s->out_format = FMT_H264;
 877     s->workaround_bugs= avctx->workaround_bugs;
 878
 879     // set defaults
 880 //    s->decode_mb= ff_h263_decode_mb;
 881     s->quarter_sample = 1;
 882     if(!avctx->has_b_frames)
 883     s->low_delay= 1;
 884
 885     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
 886
 887     ff_h264_decode_init_vlc();
 888
 889     h->thread_context[0] = h;
 890     h->outputed_poc = INT_MIN;
 891     h->prev_poc_msb= 1<<16;
 892     h->x264_build = -1;
 893     ff_h264_reset_sei(h);
 894     if(avctx->codec_id == CODEC_ID_H264){
 895         if(avctx->ticks_per_frame == 1){
 896             s->avctx->time_base.den *=2;
 897         }
 898         avctx->ticks_per_frame = 2;
 899     }
 900
 901     if(avctx->extradata_size > 0 && avctx->extradata &&
 902         ff_h264_decode_extradata(h))
 903         return -1;
 904
 905     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
 906         s->avctx->has_b_frames = h->sps.num_reorder_frames;
 907         s->low_delay = 0;
 908     }
 909
 910     return 0;
 911 }
 912
 913 int ff_h264_frame_start(H264Context *h){
 914     MpegEncContext * const s = &h->s;
 915     int i;
 916
 917     if(MPV_frame_start(s, s->avctx) < 0)
 918         return -1;
 919     ff_er_frame_start(s);
 920     /*
 921      * MPV_frame_start uses pict_type to derive key_frame.
 922      * This is incorrect for H.264; IDR markings must be used.
 923      * Zero here; IDR markings per slice in frame or fields are ORed in later.
 924      * See decode_nal_units().
 925      */
 926     s->current_picture_ptr->key_frame= 0;
 927     s->current_picture_ptr->mmco_reset= 0;
 928
 929     assert(s->linesize && s->uvlinesize);
 930
 931     for(i=0; i<16; i++){
 932         h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
 933         h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
 934     }
 935     for(i=0; i<4; i++){
 936         h->block_offset[16+i]=
 937         h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
 938         h->block_offset[24+16+i]=
 939         h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
 940     }
 941
 942     /* can't be in alloc_tables because linesize isn't known there.
 943      * FIXME: redo bipred weight to not require extra buffer? */
 944     for(i = 0; i < s->avctx->thread_count; i++)
 945         if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
 946             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
 947
 948     /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
 949     memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
 950
 951 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
 952
 953     // We mark the current picture as non-reference after allocating it, so
 954     // that if we break out due to an error it can be released automatically
 955     // in the next MPV_frame_start().
 956     // SVQ3 as well as most other codecs have only last/next/current and thus
 957     // get released even with set reference, besides SVQ3 and others do not
 958     // mark frames as reference later "naturally".
 959     if(s->codec_id != CODEC_ID_SVQ3)
 960         s->current_picture_ptr->reference= 0;
 961
 962     s->current_picture_ptr->field_poc[0]=
 963     s->current_picture_ptr->field_poc[1]= INT_MAX;
 964     assert(s->current_picture_ptr->long_ref==0);
 965
 966     return 0;
 967 }
 968
 969 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){
 970     MpegEncContext * const s = &h->s;
 971     uint8_t *top_border;
 972     int top_idx = 1;
 973
 974     src_y  -=   linesize;
 975     src_cb -= uvlinesize;
 976     src_cr -= uvlinesize;
 977
 978     if(!simple && FRAME_MBAFF){
 979         if(s->mb_y&1){
 980             if(!MB_MBAFF){
 981                 top_border = h->top_borders[0][s->mb_x];
 982                 AV_COPY128(top_border, src_y + 15*linesize);
 983                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
 984                     AV_COPY64(top_border+16, src_cb+7*uvlinesize);
 985                     AV_COPY64(top_border+24, src_cr+7*uvlinesize);
 986                 }
 987             }
 988         }else if(MB_MBAFF){
 989             top_idx = 0;
 990         }else
 991             return;
 992     }
 993
 994     top_border = h->top_borders[top_idx][s->mb_x];
 995     // There are two lines saved, the line above the the top macroblock of a pair,
 996     // and the line above the bottom macroblock
 997     AV_COPY128(top_border, src_y + 16*linesize);
 998
 999     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1000         AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1001         AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1002     }
1003 }
1004
1005 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){
1006     MpegEncContext * const s = &h->s;
1007     int deblock_left;
1008     int deblock_top;
1009     int top_idx = 1;
1010     uint8_t *top_border_m1;
1011     uint8_t *top_border;
1012
1013     if(!simple && FRAME_MBAFF){
1014         if(s->mb_y&1){
1015             if(!MB_MBAFF)
1016                 return;
1017         }else{
1018             top_idx = MB_MBAFF ? 0 : 1;
1019         }
1020     }
1021
1022     if(h->deblocking_filter == 2) {
1023         deblock_left = h->left_type[0];
1024         deblock_top  = h->top_type;
1025     } else {
1026         deblock_left = (s->mb_x > 0);
1027         deblock_top =  (s->mb_y > !!MB_FIELD);
1028     }
1029
1030     src_y  -=   linesize + 1;
1031     src_cb -= uvlinesize + 1;
1032     src_cr -= uvlinesize + 1;
1033
1034     top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1035     top_border    = h->top_borders[top_idx][s->mb_x];
1036
1037 #define XCHG(a,b,xchg)\
1038 if (xchg) AV_SWAP64(b,a);\
1039 else      AV_COPY64(b,a);
1040
1041     if(deblock_top){
1042         if(deblock_left){
1043             XCHG(top_border_m1+8, src_y -7, 1);
1044         }
1045         XCHG(top_border+0, src_y +1, xchg);
1046         XCHG(top_border+8, src_y +9, 1);
1047         if(s->mb_x+1 < s->mb_width){
1048             XCHG(h->top_borders[top_idx][s->mb_x+1], src_y +17, 1);
1049         }
1050     }
1051
1052     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1053         if(deblock_top){
1054             if(deblock_left){
1055                 XCHG(top_border_m1+16, src_cb -7, 1);
1056                 XCHG(top_border_m1+24, src_cr -7, 1);
1057             }
1058             XCHG(top_border+16, src_cb+1, 1);
1059             XCHG(top_border+24, src_cr+1, 1);
1060         }
1061     }
1062 }
1063
1064 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
1065     MpegEncContext * const s = &h->s;
1066     const int mb_x= s->mb_x;
1067     const int mb_y= s->mb_y;
1068     const int mb_xy= h->mb_xy;
1069     const int mb_type= s->current_picture.mb_type[mb_xy];
1070     uint8_t  *dest_y, *dest_cb, *dest_cr;
1071     int linesize, uvlinesize /*dct_offset*/;
1072     int i;
1073     int *block_offset = &h->block_offset[0];
1074     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1075     /* is_h264 should always be true if SVQ3 is disabled. */
1076     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1077     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1078     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1079
1080     dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
1081     dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
1082     dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
1083
1084     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
1085     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
1086
1087     h->list_counts[mb_xy]= h->list_count;
1088
1089     if (!simple && MB_FIELD) {
1090         linesize   = h->mb_linesize   = s->linesize * 2;
1091         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1092         block_offset = &h->block_offset[24];
1093         if(mb_y&1){ //FIXME move out of this function?
1094             dest_y -= s->linesize*15;
1095             dest_cb-= s->uvlinesize*7;
1096             dest_cr-= s->uvlinesize*7;
1097         }
1098         if(FRAME_MBAFF) {
1099             int list;
1100             for(list=0; list<h->list_count; list++){
1101                 if(!USES_LIST(mb_type, list))
1102                     continue;
1103                 if(IS_16X16(mb_type)){
1104                     int8_t *ref = &h->ref_cache[list][scan8[0]];
1105                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1106                 }else{
1107                     for(i=0; i<16; i+=4){
1108                         int ref = h->ref_cache[list][scan8[i]];
1109                         if(ref >= 0)
1110                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1111                     }
1112                 }
1113             }
1114         }
1115     } else {
1116         linesize   = h->mb_linesize   = s->linesize;
1117         uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1118 //        dct_offset = s->linesize * 16;
1119     }
1120
1121     if (!simple && IS_INTRA_PCM(mb_type)) {
1122         for (i=0; i<16; i++) {
1123             memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1124         }
1125         for (i=0; i<8; i++) {
1126             memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
1127             memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
1128         }
1129     } else {
1130         if(IS_INTRA(mb_type)){
1131             if(h->deblocking_filter)
1132                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
1133
1134             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1135                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1136                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1137             }
1138
1139             if(IS_INTRA4x4(mb_type)){
1140                 if(simple || !s->encoding){
1141                     if(IS_8x8DCT(mb_type)){
1142                         if(transform_bypass){
1143                             idct_dc_add =
1144                             idct_add    = s->dsp.add_pixels8;
1145                         }else{
1146                             idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1147                             idct_add    = h->h264dsp.h264_idct8_add;
1148                         }
1149                         for(i=0; i<16; i+=4){
1150                             uint8_t * const ptr= dest_y + block_offset[i];
1151                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1152                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1153                                 h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
1154                             }else{
1155                                 const int nnz = h->non_zero_count_cache[ scan8[i] ];
1156                                 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1157                                                             (h->topright_samples_available<<i)&0x4000, linesize);
1158                                 if(nnz){
1159                                     if(nnz == 1 && h->mb[i*16])
1160                                         idct_dc_add(ptr, h->mb + i*16, linesize);
1161                                     else
1162                                         idct_add   (ptr, h->mb + i*16, linesize);
1163                                 }
1164                             }
1165                         }
1166                     }else{
1167                         if(transform_bypass){
1168                             idct_dc_add =
1169                             idct_add    = s->dsp.add_pixels4;
1170                         }else{
1171                             idct_dc_add = h->h264dsp.h264_idct_dc_add;
1172                             idct_add    = h->h264dsp.h264_idct_add;
1173                         }
1174                         for(i=0; i<16; i++){
1175                             uint8_t * const ptr= dest_y + block_offset[i];
1176                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1177
1178                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1179                                 h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
1180                             }else{
1181                                 uint8_t *topright;
1182                                 int nnz, tr;
1183                                 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1184                                     const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1185                                     assert(mb_y || linesize <= block_offset[i]);
1186                                     if(!topright_avail){
1187                                         tr= ptr[3 - linesize]*0x01010101;
1188                                         topright= (uint8_t*) &tr;
1189                                     }else
1190                                         topright= ptr + 4 - linesize;
1191                                 }else
1192                                     topright= NULL;
1193
1194                                 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1195                                 nnz = h->non_zero_count_cache[ scan8[i] ];
1196                                 if(nnz){
1197                                     if(is_h264){
1198                                         if(nnz == 1 && h->mb[i*16])
1199                                             idct_dc_add(ptr, h->mb + i*16, linesize);
1200                                         else
1201                                             idct_add   (ptr, h->mb + i*16, linesize);
1202                                     }else
1203                                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1204                                 }
1205                             }
1206                         }
1207                     }
1208                 }
1209             }else{
1210                 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1211                 if(is_h264){
1212                     if(h->non_zero_count_cache[ scan8[LUMA_DC_BLOCK_INDEX] ]){
1213                         if(!transform_bypass)
1214                             h->h264dsp.h264_luma_dc_dequant_idct(h->mb, h->mb_luma_dc, h->dequant4_coeff[0][s->qscale][0]);
1215                         else{
1216                             static const uint8_t dc_mapping[16] = { 0*16, 1*16, 4*16, 5*16, 2*16, 3*16, 6*16, 7*16,
1217                                                                     8*16, 9*16,12*16,13*16,10*16,11*16,14*16,15*16};
1218                             for(i = 0; i < 16; i++)
1219                                 h->mb[dc_mapping[i]] = h->mb_luma_dc[i];
1220                         }
1221                     }
1222                 }else
1223                     ff_svq3_luma_dc_dequant_idct_c(h->mb, h->mb_luma_dc, s->qscale);
1224             }
1225             if(h->deblocking_filter)
1226                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
1227         }else if(is_h264){
1228             hl_motion(h, dest_y, dest_cb, dest_cr,
1229                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1230                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1231                       h->h264dsp.weight_h264_pixels_tab, h->h264dsp.biweight_h264_pixels_tab);
1232         }
1233
1234
1235         if(!IS_INTRA4x4(mb_type)){
1236             if(is_h264){
1237                 if(IS_INTRA16x16(mb_type)){
1238                     if(transform_bypass){
1239                         if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1240                             h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1241                         }else{
1242                             for(i=0; i<16; i++){
1243                                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1244                                     s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
1245                             }
1246                         }
1247                     }else{
1248                          h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1249                     }
1250                 }else if(h->cbp&15){
1251                     if(transform_bypass){
1252                         const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1253                         idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1254                         for(i=0; i<16; i+=di){
1255                             if(h->non_zero_count_cache[ scan8[i] ]){
1256                                 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
1257                             }
1258                         }
1259                     }else{
1260                         if(IS_8x8DCT(mb_type)){
1261                             h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1262                         }else{
1263                             h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1264                         }
1265                     }
1266                 }
1267             }else{
1268                 for(i=0; i<16; i++){
1269                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1270                         uint8_t * const ptr= dest_y + block_offset[i];
1271                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1272                     }
1273                 }
1274             }
1275         }
1276
1277         if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1278             uint8_t *dest[2] = {dest_cb, dest_cr};
1279             if(transform_bypass){
1280                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1281                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
1282                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
1283                 }else{
1284                     idct_add = s->dsp.add_pixels4;
1285                     for(i=16; i<16+8; i++){
1286                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1287                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1288                     }
1289                 }
1290             }else{
1291                 if(is_h264){
1292                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+0] ])
1293                         chroma_dc_dequant_idct_c(h->mb + 16*16     , h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
1294                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+1] ])
1295                         chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
1296                     h->h264dsp.h264_idct_add8(dest, block_offset,
1297                                               h->mb, uvlinesize,
1298                                               h->non_zero_count_cache);
1299                 }else{
1300                     chroma_dc_dequant_idct_c(h->mb + 16*16     , h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
1301                     chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
1302                     for(i=16; i<16+8; i++){
1303                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1304                             uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1305                             ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
1306                         }
1307                     }
1308                 }
1309             }
1310         }
1311     }
1312     if(h->cbp || IS_INTRA(mb_type))
1313         s->dsp.clear_blocks(h->mb);
1314 }
1315
1316 /**
1317  * Process a macroblock; this case avoids checks for expensive uncommon cases.
1318  */
1319 static void hl_decode_mb_simple(H264Context *h){
1320     hl_decode_mb_internal(h, 1);
1321 }
1322
1323 /**
1324  * Process a macroblock; this handles edge cases, such as interlacing.
1325  */
1326 static void av_noinline hl_decode_mb_complex(H264Context *h){
1327     hl_decode_mb_internal(h, 0);
1328 }
1329
1330 void ff_h264_hl_decode_mb(H264Context *h){
1331     MpegEncContext * const s = &h->s;
1332     const int mb_xy= h->mb_xy;
1333     const int mb_type= s->current_picture.mb_type[mb_xy];
1334     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1335
1336     if (is_complex)
1337         hl_decode_mb_complex(h);
1338     else hl_decode_mb_simple(h);
1339 }
1340
1341 static int pred_weight_table(H264Context *h){
1342     MpegEncContext * const s = &h->s;
1343     int list, i;
1344     int luma_def, chroma_def;
1345
1346     h->use_weight= 0;
1347     h->use_weight_chroma= 0;
1348     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1349     if(CHROMA)
1350         h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1351     luma_def = 1<<h->luma_log2_weight_denom;
1352     chroma_def = 1<<h->chroma_log2_weight_denom;
1353
1354     for(list=0; list<2; list++){
1355         h->luma_weight_flag[list]   = 0;
1356         h->chroma_weight_flag[list] = 0;
1357         for(i=0; i<h->ref_count[list]; i++){
1358             int luma_weight_flag, chroma_weight_flag;
1359
1360             luma_weight_flag= get_bits1(&s->gb);
1361             if(luma_weight_flag){
1362                 h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
1363                 h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
1364                 if(   h->luma_weight[i][list][0] != luma_def
1365                    || h->luma_weight[i][list][1] != 0) {
1366                     h->use_weight= 1;
1367                     h->luma_weight_flag[list]= 1;
1368                 }
1369             }else{
1370                 h->luma_weight[i][list][0]= luma_def;
1371                 h->luma_weight[i][list][1]= 0;
1372             }
1373
1374             if(CHROMA){
1375                 chroma_weight_flag= get_bits1(&s->gb);
1376                 if(chroma_weight_flag){
1377                     int j;
1378                     for(j=0; j<2; j++){
1379                         h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
1380                         h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
1381                         if(   h->chroma_weight[i][list][j][0] != chroma_def
1382                            || h->chroma_weight[i][list][j][1] != 0) {
1383                             h->use_weight_chroma= 1;
1384                             h->chroma_weight_flag[list]= 1;
1385                         }
1386                     }
1387                 }else{
1388                     int j;
1389                     for(j=0; j<2; j++){
1390                         h->chroma_weight[i][list][j][0]= chroma_def;
1391                         h->chroma_weight[i][list][j][1]= 0;
1392                     }
1393                 }
1394             }
1395         }
1396         if(h->slice_type_nos != FF_B_TYPE) break;
1397     }
1398     h->use_weight= h->use_weight || h->use_weight_chroma;
1399     return 0;
1400 }
1401
1402 /**
1403  * Initialize implicit_weight table.
1404  * @param field  0/1 initialize the weight for interlaced MBAFF
1405  *                -1 initializes the rest
1406  */
1407 static void implicit_weight_table(H264Context *h, int field){
1408     MpegEncContext * const s = &h->s;
1409     int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
1410
1411     for (i = 0; i < 2; i++) {
1412         h->luma_weight_flag[i]   = 0;
1413         h->chroma_weight_flag[i] = 0;
1414     }
1415
1416     if(field < 0){
1417         cur_poc = s->current_picture_ptr->poc;
1418     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
1419        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1420         h->use_weight= 0;
1421         h->use_weight_chroma= 0;
1422         return;
1423     }
1424         ref_start= 0;
1425         ref_count0= h->ref_count[0];
1426         ref_count1= h->ref_count[1];
1427     }else{
1428         cur_poc = s->current_picture_ptr->field_poc[field];
1429         ref_start= 16;
1430         ref_count0= 16+2*h->ref_count[0];
1431         ref_count1= 16+2*h->ref_count[1];
1432     }
1433
1434     h->use_weight= 2;
1435     h->use_weight_chroma= 2;
1436     h->luma_log2_weight_denom= 5;
1437     h->chroma_log2_weight_denom= 5;
1438
1439     for(ref0=ref_start; ref0 < ref_count0; ref0++){
1440         int poc0 = h->ref_list[0][ref0].poc;
1441         for(ref1=ref_start; ref1 < ref_count1; ref1++){
1442             int poc1 = h->ref_list[1][ref1].poc;
1443             int td = av_clip(poc1 - poc0, -128, 127);
1444             int w= 32;
1445             if(td){
1446                 int tb = av_clip(cur_poc - poc0, -128, 127);
1447                 int tx = (16384 + (FFABS(td) >> 1)) / td;
1448                 int dist_scale_factor = (tb*tx + 32) >> 8;
1449                 if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
1450                     w = 64 - dist_scale_factor;
1451             }
1452             if(field<0){
1453                 h->implicit_weight[ref0][ref1][0]=
1454                 h->implicit_weight[ref0][ref1][1]= w;
1455             }else{
1456                 h->implicit_weight[ref0][ref1][field]=w;
1457             }
1458         }
1459     }
1460 }
1461
1462 /**
1463  * instantaneous decoder refresh.
1464  */
1465 static void idr(H264Context *h){
1466     ff_h264_remove_all_refs(h);
1467     h->prev_frame_num= 0;
1468     h->prev_frame_num_offset= 0;
1469     h->prev_poc_msb=
1470     h->prev_poc_lsb= 0;
1471 }
1472
1473 /* forget old pics after a seek */
1474 static void flush_dpb(AVCodecContext *avctx){
1475     H264Context *h= avctx->priv_data;
1476     int i;
1477     for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1478         if(h->delayed_pic[i])
1479             h->delayed_pic[i]->reference= 0;
1480         h->delayed_pic[i]= NULL;
1481     }
1482     h->outputed_poc= INT_MIN;
1483     h->prev_interlaced_frame = 1;
1484     idr(h);
1485     if(h->s.current_picture_ptr)
1486         h->s.current_picture_ptr->reference= 0;
1487     h->s.first_field= 0;
1488     ff_h264_reset_sei(h);
1489     ff_mpeg_flush(avctx);
1490 }
1491
1492 static int init_poc(H264Context *h){
1493     MpegEncContext * const s = &h->s;
1494     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1495     int field_poc[2];
1496     Picture *cur = s->current_picture_ptr;
1497
1498     h->frame_num_offset= h->prev_frame_num_offset;
1499     if(h->frame_num < h->prev_frame_num)
1500         h->frame_num_offset += max_frame_num;
1501
1502     if(h->sps.poc_type==0){
1503         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1504
1505         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1506             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1507         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1508             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1509         else
1510             h->poc_msb = h->prev_poc_msb;
1511 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1512         field_poc[0] =
1513         field_poc[1] = h->poc_msb + h->poc_lsb;
1514         if(s->picture_structure == PICT_FRAME)
1515             field_poc[1] += h->delta_poc_bottom;
1516     }else if(h->sps.poc_type==1){
1517         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1518         int i;
1519
1520         if(h->sps.poc_cycle_length != 0)
1521             abs_frame_num = h->frame_num_offset + h->frame_num;
1522         else
1523             abs_frame_num = 0;
1524
1525         if(h->nal_ref_idc==0 && abs_frame_num > 0)
1526             abs_frame_num--;
1527
1528         expected_delta_per_poc_cycle = 0;
1529         for(i=0; i < h->sps.poc_cycle_length; i++)
1530             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1531
1532         if(abs_frame_num > 0){
1533             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1534             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1535
1536             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1537             for(i = 0; i <= frame_num_in_poc_cycle; i++)
1538                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1539         } else
1540             expectedpoc = 0;
1541
1542         if(h->nal_ref_idc == 0)
1543             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1544
1545         field_poc[0] = expectedpoc + h->delta_poc[0];
1546         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1547
1548         if(s->picture_structure == PICT_FRAME)
1549             field_poc[1] += h->delta_poc[1];
1550     }else{
1551         int poc= 2*(h->frame_num_offset + h->frame_num);
1552
1553         if(!h->nal_ref_idc)
1554             poc--;
1555
1556         field_poc[0]= poc;
1557         field_poc[1]= poc;
1558     }
1559
1560     if(s->picture_structure != PICT_BOTTOM_FIELD)
1561         s->current_picture_ptr->field_poc[0]= field_poc[0];
1562     if(s->picture_structure != PICT_TOP_FIELD)
1563         s->current_picture_ptr->field_poc[1]= field_poc[1];
1564     cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1565
1566     return 0;
1567 }
1568
1569
1570 /**
1571  * initialize scan tables
1572  */
1573 static void init_scan_tables(H264Context *h){
1574     int i;
1575     for(i=0; i<16; i++){
1576 #define T(x) (x>>2) | ((x<<2) & 0xF)
1577         h->zigzag_scan[i] = T(zigzag_scan[i]);
1578         h-> field_scan[i] = T( field_scan[i]);
1579 #undef T
1580     }
1581     for(i=0; i<64; i++){
1582 #define T(x) (x>>3) | ((x&7)<<3)
1583         h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
1584         h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1585         h->field_scan8x8[i]        = T(field_scan8x8[i]);
1586         h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
1587 #undef T
1588     }
1589     if(h->sps.transform_bypass){ //FIXME same ugly
1590         h->zigzag_scan_q0          = zigzag_scan;
1591         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
1592         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1593         h->field_scan_q0           = field_scan;
1594         h->field_scan8x8_q0        = field_scan8x8;
1595         h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
1596     }else{
1597         h->zigzag_scan_q0          = h->zigzag_scan;
1598         h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
1599         h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1600         h->field_scan_q0           = h->field_scan;
1601         h->field_scan8x8_q0        = h->field_scan8x8;
1602         h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
1603     }
1604 }
1605
1606 static void field_end(H264Context *h){
1607     MpegEncContext * const s = &h->s;
1608     AVCodecContext * const avctx= s->avctx;
1609     s->mb_y= 0;
1610
1611     s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1612     s->current_picture_ptr->pict_type= s->pict_type;
1613
1614     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1615         ff_vdpau_h264_set_reference_frames(s);
1616
1617     if(!s->dropable) {
1618         ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1619         h->prev_poc_msb= h->poc_msb;
1620         h->prev_poc_lsb= h->poc_lsb;
1621     }
1622     h->prev_frame_num_offset= h->frame_num_offset;
1623     h->prev_frame_num= h->frame_num;
1624
1625     if (avctx->hwaccel) {
1626         if (avctx->hwaccel->end_frame(avctx) < 0)
1627             av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1628     }
1629
1630     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1631         ff_vdpau_h264_picture_complete(s);
1632
1633     /*
1634      * FIXME: Error handling code does not seem to support interlaced
1635      * when slices span multiple rows
1636      * The ff_er_add_slice calls don't work right for bottom
1637      * fields; they cause massive erroneous error concealing
1638      * Error marking covers both fields (top and bottom).
1639      * This causes a mismatched s->error_count
1640      * and a bad error table. Further, the error count goes to
1641      * INT_MAX when called for bottom field, because mb_y is
1642      * past end by one (callers fault) and resync_mb_y != 0
1643      * causes problems for the first MB line, too.
1644      */
1645     if (!FIELD_PICTURE)
1646         ff_er_frame_end(s);
1647
1648     MPV_frame_end(s);
1649
1650     h->current_slice=0;
1651 }
1652
1653 /**
1654  * Replicate H264 "master" context to thread contexts.
1655  */
1656 static void clone_slice(H264Context *dst, H264Context *src)
1657 {
1658     memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
1659     dst->s.current_picture_ptr  = src->s.current_picture_ptr;
1660     dst->s.current_picture      = src->s.current_picture;
1661     dst->s.linesize             = src->s.linesize;
1662     dst->s.uvlinesize           = src->s.uvlinesize;
1663     dst->s.first_field          = src->s.first_field;
1664
1665     dst->prev_poc_msb           = src->prev_poc_msb;
1666     dst->prev_poc_lsb           = src->prev_poc_lsb;
1667     dst->prev_frame_num_offset  = src->prev_frame_num_offset;
1668     dst->prev_frame_num         = src->prev_frame_num;
1669     dst->short_ref_count        = src->short_ref_count;
1670
1671     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
1672     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
1673     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1674     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
1675
1676     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
1677     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
1678 }
1679
1680 /**
1681  * computes profile from profile_idc and constraint_set?_flags
1682  *
1683  * @param sps SPS
1684  *
1685  * @return profile as defined by FF_PROFILE_H264_*
1686  */
1687 int ff_h264_get_profile(SPS *sps)
1688 {
1689     int profile = sps->profile_idc;
1690
1691     switch(sps->profile_idc) {
1692     case FF_PROFILE_H264_BASELINE:
1693         // constraint_set1_flag set to 1
1694         profile |= (sps->constraint_set_flags & 1<<1) ? FF_PROFILE_H264_CONSTRAINED : 0;
1695         break;
1696     case FF_PROFILE_H264_HIGH_10:
1697     case FF_PROFILE_H264_HIGH_422:
1698     case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
1699         // constraint_set3_flag set to 1
1700         profile |= (sps->constraint_set_flags & 1<<3) ? FF_PROFILE_H264_INTRA : 0;
1701         break;
1702     }
1703
1704     return profile;
1705 }
1706
1707 /**
1708  * decodes a slice header.
1709  * This will also call MPV_common_init() and frame_start() as needed.
1710  *
1711  * @param h h264context
1712  * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1713  *
1714  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1715  */
1716 static int decode_slice_header(H264Context *h, H264Context *h0){
1717     MpegEncContext * const s = &h->s;
1718     MpegEncContext * const s0 = &h0->s;
1719     unsigned int first_mb_in_slice;
1720     unsigned int pps_id;
1721     int num_ref_idx_active_override_flag;
1722     unsigned int slice_type, tmp, i, j;
1723     int default_ref_list_done = 0;
1724     int last_pic_structure;
1725
1726     s->dropable= h->nal_ref_idc == 0;
1727
1728     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1729         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1730         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1731     }else{
1732         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1733         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1734     }
1735
1736     first_mb_in_slice= get_ue_golomb(&s->gb);
1737
1738     if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1739         if(h0->current_slice && FIELD_PICTURE){
1740             field_end(h);
1741         }
1742
1743         h0->current_slice = 0;
1744         if (!s0->first_field)
1745             s->current_picture_ptr= NULL;
1746     }
1747
1748     slice_type= get_ue_golomb_31(&s->gb);
1749     if(slice_type > 9){
1750         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);
1751         return -1;
1752     }
1753     if(slice_type > 4){
1754         slice_type -= 5;
1755         h->slice_type_fixed=1;
1756     }else
1757         h->slice_type_fixed=0;
1758
1759     slice_type= golomb_to_pict_type[ slice_type ];
1760     if (slice_type == FF_I_TYPE
1761         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
1762         default_ref_list_done = 1;
1763     }
1764     h->slice_type= slice_type;
1765     h->slice_type_nos= slice_type & 3;
1766
1767     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
1768
1769     pps_id= get_ue_golomb(&s->gb);
1770     if(pps_id>=MAX_PPS_COUNT){
1771         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
1772         return -1;
1773     }
1774     if(!h0->pps_buffers[pps_id]) {
1775         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
1776         return -1;
1777     }
1778     h->pps= *h0->pps_buffers[pps_id];
1779
1780     if(!h0->sps_buffers[h->pps.sps_id]) {
1781         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
1782         return -1;
1783     }
1784     h->sps = *h0->sps_buffers[h->pps.sps_id];
1785
1786     s->avctx->profile = ff_h264_get_profile(&h->sps);
1787     s->avctx->level   = h->sps.level_idc;
1788     s->avctx->refs    = h->sps.ref_frame_count;
1789
1790     if(h == h0 && h->dequant_coeff_pps != pps_id){
1791         h->dequant_coeff_pps = pps_id;
1792         init_dequant_tables(h);
1793     }
1794
1795     s->mb_width= h->sps.mb_width;
1796     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1797
1798     h->b_stride=  s->mb_width*4;
1799
1800     s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
1801     if(h->sps.frame_mbs_only_flag)
1802         s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
1803     else
1804         s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 7);
1805
1806     if (s->context_initialized
1807         && (   s->width != s->avctx->width || s->height != s->avctx->height
1808             || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
1809         if(h != h0)
1810             return -1;   // width / height changed during parallelized decoding
1811         free_tables(h, 0);
1812         flush_dpb(s->avctx);
1813         MPV_common_end(s);
1814     }
1815     if (!s->context_initialized) {
1816         if(h != h0)
1817             return -1;  // we cant (re-)initialize context during parallel decoding
1818
1819         avcodec_set_dimensions(s->avctx, s->width, s->height);
1820         s->avctx->sample_aspect_ratio= h->sps.sar;
1821         av_assert0(s->avctx->sample_aspect_ratio.den);
1822
1823         if(h->sps.video_signal_type_present_flag){
1824             s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1825             if(h->sps.colour_description_present_flag){
1826                 s->avctx->color_primaries = h->sps.color_primaries;
1827                 s->avctx->color_trc       = h->sps.color_trc;
1828                 s->avctx->colorspace      = h->sps.colorspace;
1829             }
1830         }
1831
1832         if(h->sps.timing_info_present_flag){
1833             int64_t den= h->sps.time_scale;
1834             if(h->x264_build < 44U)
1835                 den *= 2;
1836             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
1837                       h->sps.num_units_in_tick, den, 1<<30);
1838         }
1839         s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
1840                                                  s->avctx->codec->pix_fmts ?
1841                                                  s->avctx->codec->pix_fmts :
1842                                                  s->avctx->color_range == AVCOL_RANGE_JPEG ?
1843                                                  hwaccel_pixfmt_list_h264_jpeg_420 :
1844                                                  ff_hwaccel_pixfmt_list_420);
1845         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1846
1847         if (MPV_common_init(s) < 0)
1848             return -1;
1849         s->first_field = 0;
1850         h->prev_interlaced_frame = 1;
1851
1852         init_scan_tables(h);
1853         ff_h264_alloc_tables(h);
1854
1855         for(i = 1; i < s->avctx->thread_count; i++) {
1856             H264Context *c;
1857             c = h->thread_context[i] = av_malloc(sizeof(H264Context));
1858             memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
1859             memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
1860             c->h264dsp = h->h264dsp;
1861             c->sps = h->sps;
1862             c->pps = h->pps;
1863             init_scan_tables(c);
1864             clone_tables(c, h, i);
1865         }
1866
1867         for(i = 0; i < s->avctx->thread_count; i++)
1868             if(context_init(h->thread_context[i]) < 0)
1869                 return -1;
1870     }
1871
1872     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1873
1874     h->mb_mbaff = 0;
1875     h->mb_aff_frame = 0;
1876     last_pic_structure = s0->picture_structure;
1877     if(h->sps.frame_mbs_only_flag){
1878         s->picture_structure= PICT_FRAME;
1879     }else{
1880         if(get_bits1(&s->gb)) { //field_pic_flag
1881             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
1882         } else {
1883             s->picture_structure= PICT_FRAME;
1884             h->mb_aff_frame = h->sps.mb_aff;
1885         }
1886     }
1887     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
1888
1889     if(h0->current_slice == 0){
1890         while(h->frame_num !=  h->prev_frame_num &&
1891               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
1892             Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1893             av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
1894             if (ff_h264_frame_start(h) < 0)
1895                 return -1;
1896             h->prev_frame_num++;
1897             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
1898             s->current_picture_ptr->frame_num= h->prev_frame_num;
1899             ff_generate_sliding_window_mmcos(h);
1900             ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1901             /* Error concealment: if a ref is missing, copy the previous ref in its place.
1902              * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
1903              * about there being no actual duplicates.
1904              * FIXME: this doesn't copy padding for out-of-frame motion vectors.  Given we're
1905              * concealing a lost frame, this probably isn't noticable by comparison, but it should
1906              * be fixed. */
1907             if (h->short_ref_count) {
1908                 if (prev) {
1909                     av_image_copy(h->short_ref[0]->data, h->short_ref[0]->linesize,
1910                                   (const uint8_t**)prev->data, prev->linesize,
1911                                   s->avctx->pix_fmt, s->mb_width*16, s->mb_height*16);
1912                     h->short_ref[0]->poc = prev->poc+2;
1913                 }
1914                 h->short_ref[0]->frame_num = h->prev_frame_num;
1915             }
1916         }
1917
1918         /* See if we have a decoded first field looking for a pair... */
1919         if (s0->first_field) {
1920             assert(s0->current_picture_ptr);
1921             assert(s0->current_picture_ptr->data[0]);
1922             assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
1923
1924             /* figure out if we have a complementary field pair */
1925             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
1926                 /*
1927                  * Previous field is unmatched. Don't display it, but let it
1928                  * remain for reference if marked as such.
1929                  */
1930                 s0->current_picture_ptr = NULL;
1931                 s0->first_field = FIELD_PICTURE;
1932
1933             } else {
1934                 if (h->nal_ref_idc &&
1935                         s0->current_picture_ptr->reference &&
1936                         s0->current_picture_ptr->frame_num != h->frame_num) {
1937                     /*
1938                      * This and previous field were reference, but had
1939                      * different frame_nums. Consider this field first in
1940                      * pair. Throw away previous field except for reference
1941                      * purposes.
1942                      */
1943                     s0->first_field = 1;
1944                     s0->current_picture_ptr = NULL;
1945
1946                 } else {
1947                     /* Second field in complementary pair */
1948                     s0->first_field = 0;
1949                 }
1950             }
1951
1952         } else {
1953             /* Frame or first field in a potentially complementary pair */
1954             assert(!s0->current_picture_ptr);
1955             s0->first_field = FIELD_PICTURE;
1956         }
1957
1958         if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
1959             s0->first_field = 0;
1960             return -1;
1961         }
1962     }
1963     if(h != h0)
1964         clone_slice(h, h0);
1965
1966     s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
1967
1968     assert(s->mb_num == s->mb_width * s->mb_height);
1969     if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
1970        first_mb_in_slice                    >= s->mb_num){
1971         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1972         return -1;
1973     }
1974     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
1975     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
1976     if (s->picture_structure == PICT_BOTTOM_FIELD)
1977         s->resync_mb_y = s->mb_y = s->mb_y + 1;
1978     assert(s->mb_y < s->mb_height);
1979
1980     if(s->picture_structure==PICT_FRAME){
1981         h->curr_pic_num=   h->frame_num;
1982         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
1983     }else{
1984         h->curr_pic_num= 2*h->frame_num + 1;
1985         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
1986     }
1987
1988     if(h->nal_unit_type == NAL_IDR_SLICE){
1989         get_ue_golomb(&s->gb); /* idr_pic_id */
1990     }
1991
1992     if(h->sps.poc_type==0){
1993         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
1994
1995         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
1996             h->delta_poc_bottom= get_se_golomb(&s->gb);
1997         }
1998     }
1999
2000     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
2001         h->delta_poc[0]= get_se_golomb(&s->gb);
2002
2003         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2004             h->delta_poc[1]= get_se_golomb(&s->gb);
2005     }
2006
2007     init_poc(h);
2008
2009     if(h->pps.redundant_pic_cnt_present){
2010         h->redundant_pic_count= get_ue_golomb(&s->gb);
2011     }
2012
2013     //set defaults, might be overridden a few lines later
2014     h->ref_count[0]= h->pps.ref_count[0];
2015     h->ref_count[1]= h->pps.ref_count[1];
2016
2017     if(h->slice_type_nos != FF_I_TYPE){
2018         if(h->slice_type_nos == FF_B_TYPE){
2019             h->direct_spatial_mv_pred= get_bits1(&s->gb);
2020         }
2021         num_ref_idx_active_override_flag= get_bits1(&s->gb);
2022
2023         if(num_ref_idx_active_override_flag){
2024             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2025             if(h->slice_type_nos==FF_B_TYPE)
2026                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2027
2028             if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
2029                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
2030                 h->ref_count[0]= h->ref_count[1]= 1;
2031                 return -1;
2032             }
2033         }
2034         if(h->slice_type_nos == FF_B_TYPE)
2035             h->list_count= 2;
2036         else
2037             h->list_count= 1;
2038     }else
2039         h->list_count= 0;
2040
2041     if(!default_ref_list_done){
2042         ff_h264_fill_default_ref_list(h);
2043     }
2044
2045     if(h->slice_type_nos!=FF_I_TYPE && ff_h264_decode_ref_pic_list_reordering(h) < 0)
2046         return -1;
2047
2048     if(h->slice_type_nos!=FF_I_TYPE){
2049         s->last_picture_ptr= &h->ref_list[0][0];
2050         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
2051     }
2052     if(h->slice_type_nos==FF_B_TYPE){
2053         s->next_picture_ptr= &h->ref_list[1][0];
2054         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
2055     }
2056
2057     if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
2058        ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
2059         pred_weight_table(h);
2060     else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE){
2061         implicit_weight_table(h, -1);
2062     }else {
2063         h->use_weight = 0;
2064         for (i = 0; i < 2; i++) {
2065             h->luma_weight_flag[i]   = 0;
2066             h->chroma_weight_flag[i] = 0;
2067         }
2068     }
2069
2070     if(h->nal_ref_idc)
2071         ff_h264_decode_ref_pic_marking(h0, &s->gb);
2072
2073     if(FRAME_MBAFF){
2074         ff_h264_fill_mbaff_ref_list(h);
2075
2076         if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE){
2077             implicit_weight_table(h, 0);
2078             implicit_weight_table(h, 1);
2079         }
2080     }
2081
2082     if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
2083         ff_h264_direct_dist_scale_factor(h);
2084     ff_h264_direct_ref_list_init(h);
2085
2086     if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
2087         tmp = get_ue_golomb_31(&s->gb);
2088         if(tmp > 2){
2089             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2090             return -1;
2091         }
2092         h->cabac_init_idc= tmp;
2093     }
2094
2095     h->last_qscale_diff = 0;
2096     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2097     if(tmp>51){
2098         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2099         return -1;
2100     }
2101     s->qscale= tmp;
2102     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2103     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2104     //FIXME qscale / qp ... stuff
2105     if(h->slice_type == FF_SP_TYPE){
2106         get_bits1(&s->gb); /* sp_for_switch_flag */
2107     }
2108     if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
2109         get_se_golomb(&s->gb); /* slice_qs_delta */
2110     }
2111
2112     h->deblocking_filter = 1;
2113     h->slice_alpha_c0_offset = 52;
2114     h->slice_beta_offset = 52;
2115     if( h->pps.deblocking_filter_parameters_present ) {
2116         tmp= get_ue_golomb_31(&s->gb);
2117         if(tmp > 2){
2118             av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2119             return -1;
2120         }
2121         h->deblocking_filter= tmp;
2122         if(h->deblocking_filter < 2)
2123             h->deblocking_filter^= 1; // 1<->0
2124
2125         if( h->deblocking_filter ) {
2126             h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2127             h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
2128             if(   h->slice_alpha_c0_offset > 104U
2129                || h->slice_beta_offset     > 104U){
2130                 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);
2131                 return -1;
2132             }
2133         }
2134     }
2135
2136     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
2137        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
2138        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
2139        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2140         h->deblocking_filter= 0;
2141
2142     if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2143         if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2144             /* Cheat slightly for speed:
2145                Do not bother to deblock across slices. */
2146             h->deblocking_filter = 2;
2147         } else {
2148             h0->max_contexts = 1;
2149             if(!h0->single_decode_warning) {
2150                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2151                 h0->single_decode_warning = 1;
2152             }
2153             if(h != h0)
2154                 return 1; // deblocking switched inside frame
2155         }
2156     }
2157     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]);
2158
2159 #if 0 //FMO
2160     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2161         slice_group_change_cycle= get_bits(&s->gb, ?);
2162 #endif
2163
2164     h0->last_slice_type = slice_type;
2165     h->slice_num = ++h0->current_slice;
2166     if(h->slice_num >= MAX_SLICES){
2167         av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2168     }
2169
2170     for(j=0; j<2; j++){
2171         int id_list[16];
2172         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2173         for(i=0; i<16; i++){
2174             id_list[i]= 60;
2175             if(h->ref_list[j][i].data[0]){
2176                 int k;
2177                 uint8_t *base= h->ref_list[j][i].base[0];
2178                 for(k=0; k<h->short_ref_count; k++)
2179                     if(h->short_ref[k]->base[0] == base){
2180                         id_list[i]= k;
2181                         break;
2182                     }
2183                 for(k=0; k<h->long_ref_count; k++)
2184                     if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2185                         id_list[i]= h->short_ref_count + k;
2186                         break;
2187                     }
2188             }
2189         }
2190
2191         ref2frm[0]=
2192         ref2frm[1]= -1;
2193         for(i=0; i<16; i++)
2194             ref2frm[i+2]= 4*id_list[i]
2195                           +(h->ref_list[j][i].reference&3);
2196         ref2frm[18+0]=
2197         ref2frm[18+1]= -1;
2198         for(i=16; i<48; i++)
2199             ref2frm[i+4]= 4*id_list[(i-16)>>1]
2200                           +(h->ref_list[j][i].reference&3);
2201     }
2202
2203     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
2204     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2205
2206     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2207         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",
2208                h->slice_num,
2209                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2210                first_mb_in_slice,
2211                av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2212                pps_id, h->frame_num,
2213                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2214                h->ref_count[0], h->ref_count[1],
2215                s->qscale,
2216                h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
2217                h->use_weight,
2218                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2219                h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2220                );
2221     }
2222
2223     return 0;
2224 }
2225
2226 int ff_h264_get_slice_type(const H264Context *h)
2227 {
2228     switch (h->slice_type) {
2229     case FF_P_TYPE:  return 0;
2230     case FF_B_TYPE:  return 1;
2231     case FF_I_TYPE:  return 2;
2232     case FF_SP_TYPE: return 3;
2233     case FF_SI_TYPE: return 4;
2234     default:         return -1;
2235     }
2236 }
2237
2238 /**
2239  *
2240  * @return non zero if the loop filter can be skiped
2241  */
2242 static int fill_filter_caches(H264Context *h, int mb_type){
2243     MpegEncContext * const s = &h->s;
2244     const int mb_xy= h->mb_xy;
2245     int top_xy, left_xy[2];
2246     int top_type, left_type[2];
2247
2248     top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
2249
2250     //FIXME deblocking could skip the intra and nnz parts.
2251
2252     /* Wow, what a mess, why didn't they simplify the interlacing & intra
2253      * stuff, I can't imagine that these complex rules are worth it. */
2254
2255     left_xy[1] = left_xy[0] = mb_xy-1;
2256     if(FRAME_MBAFF){
2257         const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
2258         const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
2259         if(s->mb_y&1){
2260             if (left_mb_field_flag != curr_mb_field_flag) {
2261                 left_xy[0] -= s->mb_stride;
2262             }
2263         }else{
2264             if(curr_mb_field_flag){
2265                 top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
2266             }
2267             if (left_mb_field_flag != curr_mb_field_flag) {
2268                 left_xy[1] += s->mb_stride;
2269             }
2270         }
2271     }
2272
2273     h->top_mb_xy = top_xy;
2274     h->left_mb_xy[0] = left_xy[0];
2275     h->left_mb_xy[1] = left_xy[1];
2276     {
2277         //for sufficiently low qp, filtering wouldn't do anything
2278         //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
2279         int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
2280         int qp = s->current_picture.qscale_table[mb_xy];
2281         if(qp <= qp_thresh
2282            && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
2283            && (top_xy   < 0 || ((qp + s->current_picture.qscale_table[top_xy    ] + 1)>>1) <= qp_thresh)){
2284             if(!FRAME_MBAFF)
2285                 return 1;
2286             if(   (left_xy[0]< 0            || ((qp + s->current_picture.qscale_table[left_xy[1]             ] + 1)>>1) <= qp_thresh)
2287                && (top_xy    < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy    -s->mb_stride] + 1)>>1) <= qp_thresh))
2288                 return 1;
2289         }
2290     }
2291
2292     top_type     = s->current_picture.mb_type[top_xy]    ;
2293     left_type[0] = s->current_picture.mb_type[left_xy[0]];
2294     left_type[1] = s->current_picture.mb_type[left_xy[1]];
2295     if(h->deblocking_filter == 2){
2296         if(h->slice_table[top_xy     ] != h->slice_num) top_type= 0;
2297         if(h->slice_table[left_xy[0] ] != h->slice_num) left_type[0]= left_type[1]= 0;
2298     }else{
2299         if(h->slice_table[top_xy     ] == 0xFFFF) top_type= 0;
2300         if(h->slice_table[left_xy[0] ] == 0xFFFF) left_type[0]= left_type[1] =0;
2301     }
2302     h->top_type    = top_type    ;
2303     h->left_type[0]= left_type[0];
2304     h->left_type[1]= left_type[1];
2305
2306     if(IS_INTRA(mb_type))
2307         return 0;
2308
2309     AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
2310     AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
2311     AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]);
2312     AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]);
2313     AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
2314
2315     h->cbp= h->cbp_table[mb_xy];
2316
2317     {
2318         int list;
2319         for(list=0; list<h->list_count; list++){
2320             int8_t *ref;
2321             int y, b_stride;
2322             int16_t (*mv_dst)[2];
2323             int16_t (*mv_src)[2];
2324
2325             if(!USES_LIST(mb_type, list)){
2326                 fill_rectangle(  h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
2327                 AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2328                 AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2329                 AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2330                 AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2331                 continue;
2332             }
2333
2334             ref = &s->current_picture.ref_index[list][4*mb_xy];
2335             {
2336                 int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2337                 AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2338                 AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2339                 ref += 2;
2340                 AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2341                 AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2342             }
2343
2344             b_stride = h->b_stride;
2345             mv_dst   = &h->mv_cache[list][scan8[0]];
2346             mv_src   = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
2347             for(y=0; y<4; y++){
2348                 AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
2349             }
2350
2351         }
2352     }
2353
2354
2355 /*
2356 0 . T T. T T T T
2357 1 L . .L . . . .
2358 2 L . .L . . . .
2359 3 . T TL . . . .
2360 4 L . .L . . . .
2361 5 L . .. . . . .
2362 */
2363 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
2364     if(top_type){
2365         AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
2366     }
2367
2368     if(left_type[0]){
2369         h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
2370         h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
2371         h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
2372         h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
2373     }
2374
2375     // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
2376     if(!CABAC && h->pps.transform_8x8_mode){
2377         if(IS_8x8DCT(top_type)){
2378             h->non_zero_count_cache[4+8*0]=
2379             h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
2380             h->non_zero_count_cache[6+8*0]=
2381             h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
2382         }
2383         if(IS_8x8DCT(left_type[0])){
2384             h->non_zero_count_cache[3+8*1]=
2385             h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
2386         }
2387         if(IS_8x8DCT(left_type[1])){
2388             h->non_zero_count_cache[3+8*3]=
2389             h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
2390         }
2391
2392         if(IS_8x8DCT(mb_type)){
2393             h->non_zero_count_cache[scan8[0   ]]= h->non_zero_count_cache[scan8[1   ]]=
2394             h->non_zero_count_cache[scan8[2   ]]= h->non_zero_count_cache[scan8[3   ]]= h->cbp & 1;
2395
2396             h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
2397             h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
2398
2399             h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
2400             h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
2401
2402             h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
2403             h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
2404         }
2405     }
2406
2407     if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
2408         int list;
2409         for(list=0; list<h->list_count; list++){
2410             if(USES_LIST(top_type, list)){
2411                 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
2412                 const int b8_xy= 4*top_xy + 2;
2413                 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2414                 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
2415                 h->ref_cache[list][scan8[0] + 0 - 1*8]=
2416                 h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
2417                 h->ref_cache[list][scan8[0] + 2 - 1*8]=
2418                 h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
2419             }else{
2420                 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
2421                 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2422             }
2423
2424             if(!IS_INTERLACED(mb_type^left_type[0])){
2425                 if(USES_LIST(left_type[0], list)){
2426                     const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
2427                     const int b8_xy= 4*left_xy[0] + 1;
2428                     int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2429                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
2430                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
2431                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
2432                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
2433                     h->ref_cache[list][scan8[0] - 1 + 0 ]=
2434                     h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]];
2435                     h->ref_cache[list][scan8[0] - 1 +16 ]=
2436                     h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]];
2437                 }else{
2438                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
2439                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
2440                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
2441                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
2442                     h->ref_cache[list][scan8[0] - 1 + 0  ]=
2443                     h->ref_cache[list][scan8[0] - 1 + 8  ]=
2444                     h->ref_cache[list][scan8[0] - 1 + 16 ]=
2445                     h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
2446                 }
2447             }
2448         }
2449     }
2450
2451     return 0;
2452 }
2453
2454 static void loop_filter(H264Context *h){
2455     MpegEncContext * const s = &h->s;
2456     uint8_t  *dest_y, *dest_cb, *dest_cr;
2457     int linesize, uvlinesize, mb_x, mb_y;
2458     const int end_mb_y= s->mb_y + FRAME_MBAFF;
2459     const int old_slice_type= h->slice_type;
2460
2461     if(h->deblocking_filter) {
2462         for(mb_x= 0; mb_x<s->mb_width; mb_x++){
2463             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
2464                 int mb_xy, mb_type;
2465                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2466                 h->slice_num= h->slice_table[mb_xy];
2467                 mb_type= s->current_picture.mb_type[mb_xy];
2468                 h->list_count= h->list_counts[mb_xy];
2469
2470                 if(FRAME_MBAFF)
2471                     h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2472
2473                 s->mb_x= mb_x;
2474                 s->mb_y= mb_y;
2475                 dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2476                 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2477                 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2478                     //FIXME simplify above
2479
2480                 if (MB_FIELD) {
2481                     linesize   = h->mb_linesize   = s->linesize * 2;
2482                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2483                     if(mb_y&1){ //FIXME move out of this function?
2484                         dest_y -= s->linesize*15;
2485                         dest_cb-= s->uvlinesize*7;
2486                         dest_cr-= s->uvlinesize*7;
2487                     }
2488                 } else {
2489                     linesize   = h->mb_linesize   = s->linesize;
2490                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2491                 }
2492                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2493                 if(fill_filter_caches(h, mb_type))
2494                     continue;
2495                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2496                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2497
2498                 if (FRAME_MBAFF) {
2499                     ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2500                 } else {
2501                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2502                 }
2503             }
2504         }
2505     }
2506     h->slice_type= old_slice_type;
2507     s->mb_x= 0;
2508     s->mb_y= end_mb_y - FRAME_MBAFF;
2509     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2510     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2511 }
2512
2513 static void predict_field_decoding_flag(H264Context *h){
2514     MpegEncContext * const s = &h->s;
2515     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2516     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
2517                 ? s->current_picture.mb_type[mb_xy-1]
2518                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
2519                 ? s->current_picture.mb_type[mb_xy-s->mb_stride]
2520                 : 0;
2521     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2522 }
2523
2524 static int decode_slice(struct AVCodecContext *avctx, void *arg){
2525     H264Context *h = *(void**)arg;
2526     MpegEncContext * const s = &h->s;
2527     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2528
2529     s->mb_skip_run= -1;
2530
2531     h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
2532                     (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
2533
2534     if( h->pps.cabac ) {
2535         /* realign */
2536         align_get_bits( &s->gb );
2537
2538         /* init cabac */
2539         ff_init_cabac_states( &h->cabac);
2540         ff_init_cabac_decoder( &h->cabac,
2541                                s->gb.buffer + get_bits_count(&s->gb)/8,
2542                                (get_bits_left(&s->gb) + 7)/8);
2543
2544         ff_h264_init_cabac_states(h);
2545
2546         for(;;){
2547 //START_TIMER
2548             int ret = ff_h264_decode_mb_cabac(h);
2549             int eos;
2550 //STOP_TIMER("decode_mb_cabac")
2551
2552             if(ret>=0) ff_h264_hl_decode_mb(h);
2553
2554             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
2555                 s->mb_y++;
2556
2557                 ret = ff_h264_decode_mb_cabac(h);
2558
2559                 if(ret>=0) ff_h264_hl_decode_mb(h);
2560                 s->mb_y--;
2561             }
2562             eos = get_cabac_terminate( &h->cabac );
2563
2564             if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
2565                 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);
2566                 return 0;
2567             }
2568             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2569                 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);
2570                 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);
2571                 return -1;
2572             }
2573
2574             if( ++s->mb_x >= s->mb_width ) {
2575                 s->mb_x = 0;
2576                 loop_filter(h);
2577                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2578                 ++s->mb_y;
2579                 if(FIELD_OR_MBAFF_PICTURE) {
2580                     ++s->mb_y;
2581                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2582                         predict_field_decoding_flag(h);
2583                 }
2584             }
2585
2586             if( eos || s->mb_y >= s->mb_height ) {
2587                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2588                 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);
2589                 return 0;
2590             }
2591         }
2592
2593     } else {
2594         for(;;){
2595             int ret = ff_h264_decode_mb_cavlc(h);
2596
2597             if(ret>=0) ff_h264_hl_decode_mb(h);
2598
2599             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
2600                 s->mb_y++;
2601                 ret = ff_h264_decode_mb_cavlc(h);
2602
2603                 if(ret>=0) ff_h264_hl_decode_mb(h);
2604                 s->mb_y--;
2605             }
2606
2607             if(ret<0){
2608                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2609                 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);
2610
2611                 return -1;
2612             }
2613
2614             if(++s->mb_x >= s->mb_width){
2615                 s->mb_x=0;
2616                 loop_filter(h);
2617                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
2618                 ++s->mb_y;
2619                 if(FIELD_OR_MBAFF_PICTURE) {
2620                     ++s->mb_y;
2621                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2622                         predict_field_decoding_flag(h);
2623                 }
2624                 if(s->mb_y >= s->mb_height){
2625                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2626
2627                     if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
2628                         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);
2629
2630                         return 0;
2631                     }else{
2632                         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);
2633
2634                         return -1;
2635                     }
2636                 }
2637             }
2638
2639             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
2640                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2641                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2642                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2643
2644                     return 0;
2645                 }else{
2646                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
2647
2648                     return -1;
2649                 }
2650             }
2651         }
2652     }
2653
2654 #if 0
2655     for(;s->mb_y < s->mb_height; s->mb_y++){
2656         for(;s->mb_x < s->mb_width; s->mb_x++){
2657             int ret= decode_mb(h);
2658
2659             ff_h264_hl_decode_mb(h);
2660
2661             if(ret<0){
2662                 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2663                 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);
2664
2665                 return -1;
2666             }
2667
2668             if(++s->mb_x >= s->mb_width){
2669                 s->mb_x=0;
2670                 if(++s->mb_y >= s->mb_height){
2671                     if(get_bits_count(s->gb) == s->gb.size_in_bits){
2672                         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);
2673
2674                         return 0;
2675                     }else{
2676                         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);
2677
2678                         return -1;
2679                     }
2680                 }
2681             }
2682
2683             if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2684                 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2685                     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);
2686
2687                     return 0;
2688                 }else{
2689                     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);
2690
2691                     return -1;
2692                 }
2693             }
2694         }
2695         s->mb_x=0;
2696         ff_draw_horiz_band(s, 16*s->mb_y, 16);
2697     }
2698 #endif
2699     return -1; //not reached
2700 }
2701
2702 /**
2703  * Call decode_slice() for each context.
2704  *
2705  * @param h h264 master context
2706  * @param context_count number of contexts to execute
2707  */
2708 static void execute_decode_slices(H264Context *h, int context_count){
2709     MpegEncContext * const s = &h->s;
2710     AVCodecContext * const avctx= s->avctx;
2711     H264Context *hx;
2712     int i;
2713
2714     if (s->avctx->hwaccel)
2715         return;
2716     if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2717         return;
2718     if(context_count == 1) {
2719         decode_slice(avctx, &h);
2720     } else {
2721         for(i = 1; i < context_count; i++) {
2722             hx = h->thread_context[i];
2723             hx->s.error_recognition = avctx->error_recognition;
2724             hx->s.error_count = 0;
2725         }
2726
2727         avctx->execute(avctx, (void *)decode_slice,
2728                        h->thread_context, NULL, context_count, sizeof(void*));
2729
2730         /* pull back stuff from slices to master context */
2731         hx = h->thread_context[context_count - 1];
2732         s->mb_x = hx->s.mb_x;
2733         s->mb_y = hx->s.mb_y;
2734         s->dropable = hx->s.dropable;
2735         s->picture_structure = hx->s.picture_structure;
2736         for(i = 1; i < context_count; i++)
2737             h->s.error_count += h->thread_context[i]->s.error_count;
2738     }
2739 }
2740
2741
2742 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
2743     MpegEncContext * const s = &h->s;
2744     AVCodecContext * const avctx= s->avctx;
2745     int buf_index=0;
2746     H264Context *hx; ///< thread context
2747     int context_count = 0;
2748     int next_avc= h->is_avc ? 0 : buf_size;
2749
2750     h->max_contexts = avctx->thread_count;
2751 #if 0
2752     int i;
2753     for(i=0; i<50; i++){
2754         av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
2755     }
2756 #endif
2757     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
2758         h->current_slice = 0;
2759         if (!s->first_field)
2760             s->current_picture_ptr= NULL;
2761         ff_h264_reset_sei(h);
2762     }
2763
2764     for(;;){
2765         int consumed;
2766         int dst_length;
2767         int bit_length;
2768         const uint8_t *ptr;
2769         int i, nalsize = 0;
2770         int err;
2771
2772         if(buf_index >= next_avc) {
2773             if(buf_index >= buf_size) break;
2774             nalsize = 0;
2775             for(i = 0; i < h->nal_length_size; i++)
2776                 nalsize = (nalsize << 8) | buf[buf_index++];
2777             if(nalsize <= 0 || nalsize > buf_size - buf_index){
2778                 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2779                 break;
2780             }
2781             next_avc= buf_index + nalsize;
2782         } else {
2783             // start code prefix search
2784             for(; buf_index + 3 < next_avc; buf_index++){
2785                 // This should always succeed in the first iteration.
2786                 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2787                     break;
2788             }
2789
2790             if(buf_index+3 >= buf_size) break;
2791
2792             buf_index+=3;
2793             if(buf_index >= next_avc) continue;
2794         }
2795
2796         hx = h->thread_context[context_count];
2797
2798         ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
2799         if (ptr==NULL || dst_length < 0){
2800             return -1;
2801         }
2802         i= buf_index + consumed;
2803         if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
2804            buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
2805             s->workaround_bugs |= FF_BUG_TRUNCATED;
2806
2807         if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
2808         while(ptr[dst_length - 1] == 0 && dst_length > 0)
2809             dst_length--;
2810         }
2811         bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
2812
2813         if(s->avctx->debug&FF_DEBUG_STARTCODE){
2814             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);
2815         }
2816
2817         if (h->is_avc && (nalsize != consumed) && nalsize){
2818             av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
2819         }
2820
2821         buf_index += consumed;
2822
2823         //FIXME do not discard SEI id
2824         if(
2825 #if FF_API_HURRY_UP
2826            (s->hurry_up == 1 && h->nal_ref_idc  == 0) ||
2827 #endif
2828            (avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0))
2829             continue;
2830
2831       again:
2832         err = 0;
2833         switch(hx->nal_unit_type){
2834         case NAL_IDR_SLICE:
2835             if (h->nal_unit_type != NAL_IDR_SLICE) {
2836                 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2837                 return -1;
2838             }
2839             idr(h); //FIXME ensure we don't loose some frames if there is reordering
2840         case NAL_SLICE:
2841             init_get_bits(&hx->s.gb, ptr, bit_length);
2842             hx->intra_gb_ptr=
2843             hx->inter_gb_ptr= &hx->s.gb;
2844             hx->s.data_partitioning = 0;
2845
2846             if((err = decode_slice_header(hx, h)))
2847                break;
2848
2849             if (h->current_slice == 1) {
2850                 if (s->avctx->hwaccel && s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
2851                     return -1;
2852                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2853                     ff_vdpau_h264_picture_start(s);
2854             }
2855
2856             s->current_picture_ptr->key_frame |=
2857                     (hx->nal_unit_type == NAL_IDR_SLICE) ||
2858                     (h->sei_recovery_frame_cnt >= 0);
2859             if(hx->redundant_pic_count==0
2860 #if FF_API_HURRY_UP
2861                && hx->s.hurry_up < 5
2862 #endif
2863                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2864                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2865                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2866                && avctx->skip_frame < AVDISCARD_ALL){
2867                 if(avctx->hwaccel) {
2868                     if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
2869                         return -1;
2870                 }else
2871                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
2872                     static const uint8_t start_code[] = {0x00, 0x00, 0x01};
2873                     ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
2874                     ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
2875                 }else
2876                     context_count++;
2877             }
2878             break;
2879         case NAL_DPA:
2880             init_get_bits(&hx->s.gb, ptr, bit_length);
2881             hx->intra_gb_ptr=
2882             hx->inter_gb_ptr= NULL;
2883
2884             if ((err = decode_slice_header(hx, h)) < 0)
2885                 break;
2886
2887             hx->s.data_partitioning = 1;
2888
2889             break;
2890         case NAL_DPB:
2891             init_get_bits(&hx->intra_gb, ptr, bit_length);
2892             hx->intra_gb_ptr= &hx->intra_gb;
2893             break;
2894         case NAL_DPC:
2895             init_get_bits(&hx->inter_gb, ptr, bit_length);
2896             hx->inter_gb_ptr= &hx->inter_gb;
2897
2898             if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
2899                && s->context_initialized
2900 #if FF_API_HURRY_UP
2901                && s->hurry_up < 5
2902 #endif
2903                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2904                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2905                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2906                && avctx->skip_frame < AVDISCARD_ALL)
2907                 context_count++;
2908             break;
2909         case NAL_SEI:
2910             init_get_bits(&s->gb, ptr, bit_length);
2911             ff_h264_decode_sei(h);
2912             break;
2913         case NAL_SPS:
2914             init_get_bits(&s->gb, ptr, bit_length);
2915             ff_h264_decode_seq_parameter_set(h);
2916
2917             if(s->flags& CODEC_FLAG_LOW_DELAY)
2918                 s->low_delay=1;
2919
2920             if(avctx->has_b_frames < 2)
2921                 avctx->has_b_frames= !s->low_delay;
2922             break;
2923         case NAL_PPS:
2924             init_get_bits(&s->gb, ptr, bit_length);
2925
2926             ff_h264_decode_picture_parameter_set(h, bit_length);
2927
2928             break;
2929         case NAL_AUD:
2930         case NAL_END_SEQUENCE:
2931         case NAL_END_STREAM:
2932         case NAL_FILLER_DATA:
2933         case NAL_SPS_EXT:
2934         case NAL_AUXILIARY_SLICE:
2935             break;
2936         default:
2937             av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
2938         }
2939
2940         if(context_count == h->max_contexts) {
2941             execute_decode_slices(h, context_count);
2942             context_count = 0;
2943         }
2944
2945         if (err < 0)
2946             av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
2947         else if(err == 1) {
2948             /* Slice could not be decoded in parallel mode, copy down
2949              * NAL unit stuff to context 0 and restart. Note that
2950              * rbsp_buffer is not transferred, but since we no longer
2951              * run in parallel mode this should not be an issue. */
2952             h->nal_unit_type = hx->nal_unit_type;
2953             h->nal_ref_idc   = hx->nal_ref_idc;
2954             hx = h;
2955             goto again;
2956         }
2957     }
2958     if(context_count)
2959         execute_decode_slices(h, context_count);
2960     return buf_index;
2961 }
2962
2963 /**
2964  * returns the number of bytes consumed for building the current frame
2965  */
2966 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
2967         if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
2968         if(pos+10>buf_size) pos=buf_size; // oops ;)
2969
2970         return pos;
2971 }
2972
2973 static int decode_frame(AVCodecContext *avctx,
2974                              void *data, int *data_size,
2975                              AVPacket *avpkt)
2976 {
2977     const uint8_t *buf = avpkt->data;
2978     int buf_size = avpkt->size;
2979     H264Context *h = avctx->priv_data;
2980     MpegEncContext *s = &h->s;
2981     AVFrame *pict = data;
2982     int buf_index;
2983
2984     s->flags= avctx->flags;
2985     s->flags2= avctx->flags2;
2986
2987    /* end of stream, output what is still in the buffers */
2988  out:
2989     if (buf_size == 0) {
2990         Picture *out;
2991         int i, out_idx;
2992
2993 //FIXME factorize this with the output code below
2994         out = h->delayed_pic[0];
2995         out_idx = 0;
2996         for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2997             if(h->delayed_pic[i]->poc < out->poc){
2998                 out = h->delayed_pic[i];
2999                 out_idx = i;
3000             }
3001
3002         for(i=out_idx; h->delayed_pic[i]; i++)
3003             h->delayed_pic[i] = h->delayed_pic[i+1];
3004
3005         if(out){
3006             *data_size = sizeof(AVFrame);
3007             *pict= *(AVFrame*)out;
3008         }
3009
3010         return 0;
3011     }
3012
3013     buf_index=decode_nal_units(h, buf, buf_size);
3014     if(buf_index < 0)
3015         return -1;
3016
3017     if (!s->current_picture_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
3018         buf_size = 0;
3019         goto out;
3020     }
3021
3022     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
3023         if (avctx->skip_frame >= AVDISCARD_NONREF
3024 #if FF_API_HURRY_UP
3025                 || s->hurry_up
3026 #endif
3027            )
3028             return 0;
3029         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
3030         return -1;
3031     }
3032
3033     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
3034         Picture *out = s->current_picture_ptr;
3035         Picture *cur = s->current_picture_ptr;
3036         int i, pics, out_of_order, out_idx;
3037
3038         field_end(h);
3039
3040         if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
3041             /* Wait for second field. */
3042             *data_size = 0;
3043
3044         } else {
3045             cur->interlaced_frame = 0;
3046             cur->repeat_pict = 0;
3047
3048             /* Signal interlacing information externally. */
3049             /* Prioritize picture timing SEI information over used decoding process if it exists. */
3050
3051             if(h->sps.pic_struct_present_flag){
3052                 switch (h->sei_pic_struct)
3053                 {
3054                 case SEI_PIC_STRUCT_FRAME:
3055                     break;
3056                 case SEI_PIC_STRUCT_TOP_FIELD:
3057                 case SEI_PIC_STRUCT_BOTTOM_FIELD:
3058                     cur->interlaced_frame = 1;
3059                     break;
3060                 case SEI_PIC_STRUCT_TOP_BOTTOM:
3061                 case SEI_PIC_STRUCT_BOTTOM_TOP:
3062                     if (FIELD_OR_MBAFF_PICTURE)
3063                         cur->interlaced_frame = 1;
3064                     else
3065                         // try to flag soft telecine progressive
3066                         cur->interlaced_frame = h->prev_interlaced_frame;
3067                     break;
3068                 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
3069                 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
3070                     // Signal the possibility of telecined film externally (pic_struct 5,6)
3071                     // From these hints, let the applications decide if they apply deinterlacing.
3072                     cur->repeat_pict = 1;
3073                     break;
3074                 case SEI_PIC_STRUCT_FRAME_DOUBLING:
3075                     // Force progressive here, as doubling interlaced frame is a bad idea.
3076                     cur->repeat_pict = 2;
3077                     break;
3078                 case SEI_PIC_STRUCT_FRAME_TRIPLING:
3079                     cur->repeat_pict = 4;
3080                     break;
3081                 }
3082
3083                 if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
3084                     cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
3085             }else{
3086                 /* Derive interlacing flag from used decoding process. */
3087                 cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
3088             }
3089             h->prev_interlaced_frame = cur->interlaced_frame;
3090
3091             if (cur->field_poc[0] != cur->field_poc[1]){
3092                 /* Derive top_field_first from field pocs. */
3093                 cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
3094             }else{
3095                 if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
3096                     /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
3097                     if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
3098                       || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
3099                         cur->top_field_first = 1;
3100                     else
3101                         cur->top_field_first = 0;
3102                 }else{
3103                     /* Most likely progressive */
3104                     cur->top_field_first = 0;
3105                 }
3106             }
3107
3108         //FIXME do something with unavailable reference frames
3109
3110             /* Sort B-frames into display order */
3111
3112             if(h->sps.bitstream_restriction_flag
3113                && s->avctx->has_b_frames < h->sps.num_reorder_frames){
3114                 s->avctx->has_b_frames = h->sps.num_reorder_frames;
3115                 s->low_delay = 0;
3116             }
3117
3118             if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
3119                && !h->sps.bitstream_restriction_flag){
3120                 s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
3121                 s->low_delay= 0;
3122             }
3123
3124             pics = 0;
3125             while(h->delayed_pic[pics]) pics++;
3126
3127             assert(pics <= MAX_DELAYED_PIC_COUNT);
3128
3129             h->delayed_pic[pics++] = cur;
3130             if(cur->reference == 0)
3131                 cur->reference = DELAYED_PIC_REF;
3132
3133             out = h->delayed_pic[0];
3134             out_idx = 0;
3135             for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
3136                 if(h->delayed_pic[i]->poc < out->poc){
3137                     out = h->delayed_pic[i];
3138                     out_idx = i;
3139                 }
3140             if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
3141                 h->outputed_poc= INT_MIN;
3142             out_of_order = out->poc < h->outputed_poc;
3143
3144             if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
3145                 { }
3146             else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
3147                || (s->low_delay &&
3148                 ((h->outputed_poc != INT_MIN && out->poc > h->outputed_poc + 2)
3149                  || cur->pict_type == FF_B_TYPE)))
3150             {
3151                 s->low_delay = 0;
3152                 s->avctx->has_b_frames++;
3153             }
3154
3155             if(out_of_order || pics > s->avctx->has_b_frames){
3156                 out->reference &= ~DELAYED_PIC_REF;
3157                 for(i=out_idx; h->delayed_pic[i]; i++)
3158                     h->delayed_pic[i] = h->delayed_pic[i+1];
3159             }
3160             if(!out_of_order && pics > s->avctx->has_b_frames){
3161                 *data_size = sizeof(AVFrame);
3162
3163                 if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
3164                     h->outputed_poc = INT_MIN;
3165                 } else
3166                     h->outputed_poc = out->poc;
3167                 *pict= *(AVFrame*)out;
3168             }else{
3169                 av_log(avctx, AV_LOG_DEBUG, "no picture\n");
3170             }
3171         }
3172     }
3173
3174     assert(pict->data[0] || !*data_size);
3175     ff_print_debug_info(s, pict);
3176 //printf("out %d\n", (int)pict->data[0]);
3177
3178     return get_consumed_bytes(s, buf_index, buf_size);
3179 }
3180 #if 0
3181 static inline void fill_mb_avail(H264Context *h){
3182     MpegEncContext * const s = &h->s;
3183     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3184
3185     if(s->mb_y){
3186         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
3187         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
3188         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
3189     }else{
3190         h->mb_avail[0]=
3191         h->mb_avail[1]=
3192         h->mb_avail[2]= 0;
3193     }
3194     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
3195     h->mb_avail[4]= 1; //FIXME move out
3196     h->mb_avail[5]= 0; //FIXME move out
3197 }
3198 #endif
3199
3200 #ifdef TEST
3201 #undef printf
3202 #undef random
3203 #define COUNT 8000
3204 #define SIZE (COUNT*40)
3205 int main(void){
3206     int i;
3207     uint8_t temp[SIZE];
3208     PutBitContext pb;
3209     GetBitContext gb;
3210 //    int int_temp[10000];
3211     DSPContext dsp;
3212     AVCodecContext avctx;
3213
3214     dsputil_init(&dsp, &avctx);
3215
3216     init_put_bits(&pb, temp, SIZE);
3217     printf("testing unsigned exp golomb\n");
3218     for(i=0; i<COUNT; i++){
3219         START_TIMER
3220         set_ue_golomb(&pb, i);
3221         STOP_TIMER("set_ue_golomb");
3222     }
3223     flush_put_bits(&pb);
3224
3225     init_get_bits(&gb, temp, 8*SIZE);
3226     for(i=0; i<COUNT; i++){
3227         int j, s;
3228
3229         s= show_bits(&gb, 24);
3230
3231         START_TIMER
3232         j= get_ue_golomb(&gb);
3233         if(j != i){
3234             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
3235 //            return -1;
3236         }
3237         STOP_TIMER("get_ue_golomb");
3238     }
3239
3240
3241     init_put_bits(&pb, temp, SIZE);
3242     printf("testing signed exp golomb\n");
3243     for(i=0; i<COUNT; i++){
3244         START_TIMER
3245         set_se_golomb(&pb, i - COUNT/2);
3246         STOP_TIMER("set_se_golomb");
3247     }
3248     flush_put_bits(&pb);
3249
3250     init_get_bits(&gb, temp, 8*SIZE);
3251     for(i=0; i<COUNT; i++){
3252         int j, s;
3253
3254         s= show_bits(&gb, 24);
3255
3256         START_TIMER
3257         j= get_se_golomb(&gb);
3258         if(j != i - COUNT/2){
3259             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
3260 //            return -1;
3261         }
3262         STOP_TIMER("get_se_golomb");
3263     }
3264
3265 #if 0
3266     printf("testing 4x4 (I)DCT\n");
3267
3268     DCTELEM block[16];
3269     uint8_t src[16], ref[16];
3270     uint64_t error= 0, max_error=0;
3271
3272     for(i=0; i<COUNT; i++){
3273         int j;
3274 //        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
3275         for(j=0; j<16; j++){
3276             ref[j]= random()%255;
3277             src[j]= random()%255;
3278         }
3279
3280         h264_diff_dct_c(block, src, ref, 4);
3281
3282         //normalize
3283         for(j=0; j<16; j++){
3284 //            printf("%d ", block[j]);
3285             block[j]= block[j]*4;
3286             if(j&1) block[j]= (block[j]*4 + 2)/5;
3287             if(j&4) block[j]= (block[j]*4 + 2)/5;
3288         }
3289 //        printf("\n");
3290
3291         h->h264dsp.h264_idct_add(ref, block, 4);
3292 /*        for(j=0; j<16; j++){
3293             printf("%d ", ref[j]);
3294         }
3295         printf("\n");*/
3296
3297         for(j=0; j<16; j++){
3298             int diff= FFABS(src[j] - ref[j]);
3299
3300             error+= diff*diff;
3301             max_error= FFMAX(max_error, diff);
3302         }
3303     }
3304     printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
3305     printf("testing quantizer\n");
3306     for(qp=0; qp<52; qp++){
3307         for(i=0; i<16; i++)
3308             src1_block[i]= src2_block[i]= random()%255;
3309
3310     }
3311     printf("Testing NAL layer\n");
3312
3313     uint8_t bitstream[COUNT];
3314     uint8_t nal[COUNT*2];
3315     H264Context h;
3316     memset(&h, 0, sizeof(H264Context));
3317
3318     for(i=0; i<COUNT; i++){
3319         int zeros= i;
3320         int nal_length;
3321         int consumed;
3322         int out_length;
3323         uint8_t *out;
3324         int j;
3325
3326         for(j=0; j<COUNT; j++){
3327             bitstream[j]= (random() % 255) + 1;
3328         }
3329
3330         for(j=0; j<zeros; j++){
3331             int pos= random() % COUNT;
3332             while(bitstream[pos] == 0){
3333                 pos++;
3334                 pos %= COUNT;
3335             }
3336             bitstream[pos]=0;
3337         }
3338
3339         START_TIMER
3340
3341         nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
3342         if(nal_length<0){
3343             printf("encoding failed\n");
3344             return -1;
3345         }
3346
3347         out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
3348
3349         STOP_TIMER("NAL")
3350
3351         if(out_length != COUNT){
3352             printf("incorrect length %d %d\n", out_length, COUNT);
3353             return -1;
3354         }
3355
3356         if(consumed != nal_length){
3357             printf("incorrect consumed length %d %d\n", nal_length, consumed);
3358             return -1;
3359         }
3360
3361         if(memcmp(bitstream, out, COUNT)){
3362             printf("mismatch\n");
3363             return -1;
3364         }
3365     }
3366 #endif
3367
3368     printf("Testing RBSP\n");
3369
3370
3371     return 0;
3372 }
3373 #endif /* TEST */
3374
3375
3376 av_cold void ff_h264_free_context(H264Context *h)
3377 {
3378     int i;
3379
3380     free_tables(h, 1); //FIXME cleanup init stuff perhaps
3381
3382     for(i = 0; i < MAX_SPS_COUNT; i++)
3383         av_freep(h->sps_buffers + i);
3384
3385     for(i = 0; i < MAX_PPS_COUNT; i++)
3386         av_freep(h->pps_buffers + i);
3387 }
3388
3389 av_cold int ff_h264_decode_end(AVCodecContext *avctx)
3390 {
3391     H264Context *h = avctx->priv_data;
3392     MpegEncContext *s = &h->s;
3393
3394     ff_h264_free_context(h);
3395
3396     MPV_common_end(s);
3397
3398 //    memset(h, 0, sizeof(H264Context));
3399
3400     return 0;
3401 }
3402
3403 static const AVProfile profiles[] = {
3404     { FF_PROFILE_H264_BASELINE,             "Baseline"              },
3405     { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline"  },
3406     { FF_PROFILE_H264_MAIN,                 "Main"                  },
3407     { FF_PROFILE_H264_EXTENDED,             "Extended"              },
3408     { FF_PROFILE_H264_HIGH,                 "High"                  },
3409     { FF_PROFILE_H264_HIGH_10,              "High 10"               },
3410     { FF_PROFILE_H264_HIGH_10_INTRA,        "High 10 Intra"         },
3411     { FF_PROFILE_H264_HIGH_422,             "High 4:2:2"            },
3412     { FF_PROFILE_H264_HIGH_422_INTRA,       "High 4:2:2 Intra"      },
3413     { FF_PROFILE_H264_HIGH_444,             "High 4:4:4"            },
3414     { FF_PROFILE_H264_HIGH_444_PREDICTIVE,  "High 4:4:4 Predictive" },
3415     { FF_PROFILE_H264_HIGH_444_INTRA,       "High 4:4:4 Intra"      },
3416     { FF_PROFILE_H264_CAVLC_444,            "CAVLC 4:4:4"           },
3417     { FF_PROFILE_UNKNOWN },
3418 };
3419
3420 AVCodec ff_h264_decoder = {
3421     "h264",
3422     AVMEDIA_TYPE_VIDEO,
3423     CODEC_ID_H264,
3424     sizeof(H264Context),
3425     ff_h264_decode_init,
3426     NULL,
3427     ff_h264_decode_end,
3428     decode_frame,
3429     /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3430     .flush= flush_dpb,
3431     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
3432     .profiles = NULL_IF_CONFIG_SMALL(profiles),
3433 };
3434
3435 #if CONFIG_H264_VDPAU_DECODER
3436 AVCodec ff_h264_vdpau_decoder = {
3437     "h264_vdpau",
3438     AVMEDIA_TYPE_VIDEO,
3439     CODEC_ID_H264,
3440     sizeof(H264Context),
3441     ff_h264_decode_init,
3442     NULL,
3443     ff_h264_decode_end,
3444     decode_frame,
3445     CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3446     .flush= flush_dpb,
3447     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
3448     .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
3449     .profiles = NULL_IF_CONFIG_SMALL(profiles),
3450 };
3451 #endif