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

   1 /*
   2  * H.26L/H.264/AVC/JVT/14496-10/... decoder
   3  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
   4  *
   5  * This file is part of FFmpeg.
   6  *
   7  * FFmpeg is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2.1 of the License, or (at your option) any later version.
  11  *
  12  * FFmpeg is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with FFmpeg; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 /**
  23  * @file
  24  * H.264 / AVC / MPEG4 part10 codec.
  25  * @author Michael Niedermayer <michaelni@gmx.at>
  26  */
  27
  28 #include "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 "thread.h"
  40 #include "vdpau_internal.h"
  41 #include "libavutil/avassert.h"
  42
  43 #include "cabac.h"
  44
  45 //#undef NDEBUG
  46 #include <assert.h>
  47
  48 static const uint8_t rem6[QP_MAX_MAX+1]={
  49 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
  50 };
  51
  52 static const uint8_t div6[QP_MAX_MAX+1]={
  53 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9,10,10,10,10,
  54 };
  55
  56 static const enum PixelFormat hwaccel_pixfmt_list_h264_jpeg_420[] = {
  57     PIX_FMT_DXVA2_VLD,
  58     PIX_FMT_VAAPI_VLD,
  59     PIX_FMT_YUVJ420P,
  60     PIX_FMT_NONE
  61 };
  62
  63 void ff_h264_write_back_intra_pred_mode(H264Context *h){
  64     int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
  65
  66     AV_COPY32(mode, h->intra4x4_pred_mode_cache + 4 + 8*4);
  67     mode[4]= h->intra4x4_pred_mode_cache[7+8*3];
  68     mode[5]= h->intra4x4_pred_mode_cache[7+8*2];
  69     mode[6]= h->intra4x4_pred_mode_cache[7+8*1];
  70 }
  71
  72 /**
  73  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
  74  */
  75 int ff_h264_check_intra4x4_pred_mode(H264Context *h){
  76     MpegEncContext * const s = &h->s;
  77     static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
  78     static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
  79     int i;
  80
  81     if(!(h->top_samples_available&0x8000)){
  82         for(i=0; i<4; i++){
  83             int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
  84             if(status<0){
  85                 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
  86                 return -1;
  87             } else if(status){
  88                 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
  89             }
  90         }
  91     }
  92
  93     if((h->left_samples_available&0x8888)!=0x8888){
  94         static const int mask[4]={0x8000,0x2000,0x80,0x20};
  95         for(i=0; i<4; i++){
  96             if(!(h->left_samples_available&mask[i])){
  97                 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
  98                 if(status<0){
  99                     av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
 100                     return -1;
 101                 } else if(status){
 102                     h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
 103                 }
 104             }
 105         }
 106     }
 107
 108     return 0;
 109 } //FIXME cleanup like ff_h264_check_intra_pred_mode
 110
 111 /**
 112  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
 113  */
 114 int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
 115     MpegEncContext * const s = &h->s;
 116     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
 117     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
 118
 119     if(mode > 6U) {
 120         av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
 121         return -1;
 122     }
 123
 124     if(!(h->top_samples_available&0x8000)){
 125         mode= top[ mode ];
 126         if(mode<0){
 127             av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
 128             return -1;
 129         }
 130     }
 131
 132     if((h->left_samples_available&0x8080) != 0x8080){
 133         mode= left[ mode ];
 134         if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
 135             mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
 136         }
 137         if(mode<0){
 138             av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
 139             return -1;
 140         }
 141     }
 142
 143     return mode;
 144 }
 145
 146 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
 147     int i, si, di;
 148     uint8_t *dst;
 149     int bufidx;
 150
 151 //    src[0]&0x80;                //forbidden bit
 152     h->nal_ref_idc= src[0]>>5;
 153     h->nal_unit_type= src[0]&0x1F;
 154
 155     src++; length--;
 156
 157 #if HAVE_FAST_UNALIGNED
 158 # if HAVE_FAST_64BIT
 159 #   define RS 7
 160     for(i=0; i+1<length; i+=9){
 161         if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
 162 # else
 163 #   define RS 3
 164     for(i=0; i+1<length; i+=5){
 165         if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
 166 # endif
 167             continue;
 168         if(i>0 && !src[i]) i--;
 169         while(src[i]) i++;
 170 #else
 171 #   define RS 0
 172     for(i=0; i+1<length; i+=2){
 173         if(src[i]) continue;
 174         if(i>0 && src[i-1]==0) i--;
 175 #endif
 176         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
 177             if(src[i+2]!=3){
 178                 /* startcode, so we must be past the end */
 179                 length=i;
 180             }
 181             break;
 182         }
 183         i-= RS;
 184     }
 185
 186     if(i>=length-1){ //no escaped 0
 187         *dst_length= length;
 188         *consumed= length+1; //+1 for the header
 189         return src;
 190     }
 191
 192     bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
 193     av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
 194     dst= h->rbsp_buffer[bufidx];
 195
 196     if (dst == NULL){
 197         return NULL;
 198     }
 199
 200 //printf("decoding esc\n");
 201     memcpy(dst, src, i);
 202     si=di=i;
 203     while(si+2<length){
 204         //remove escapes (very rare 1:2^22)
 205         if(src[si+2]>3){
 206             dst[di++]= src[si++];
 207             dst[di++]= src[si++];
 208         }else if(src[si]==0 && src[si+1]==0){
 209             if(src[si+2]==3){ //escape
 210                 dst[di++]= 0;
 211                 dst[di++]= 0;
 212                 si+=3;
 213                 continue;
 214             }else //next start code
 215                 goto nsc;
 216         }
 217
 218         dst[di++]= src[si++];
 219     }
 220     while(si<length)
 221         dst[di++]= src[si++];
 222 nsc:
 223
 224     memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
 225
 226     *dst_length= di;
 227     *consumed= si + 1;//+1 for the header
 228 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
 229     return dst;
 230 }
 231
 232 /**
 233  * Identify the exact end of the bitstream
 234  * @return the length of the trailing, or 0 if damaged
 235  */
 236 static int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
 237     int v= *src;
 238     int r;
 239
 240     tprintf(h->s.avctx, "rbsp trailing %X\n", v);
 241
 242     for(r=1; r<9; r++){
 243         if(v&1) return r;
 244         v>>=1;
 245     }
 246     return 0;
 247 }
 248
 249 #if 0
 250 /**
 251  * DCT transforms the 16 dc values.
 252  * @param qp quantization parameter ??? FIXME
 253  */
 254 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
 255 //    const int qmul= dequant_coeff[qp][0];
 256     int i;
 257     int temp[16]; //FIXME check if this is a good idea
 258     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
 259     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
 260
 261     for(i=0; i<4; i++){
 262         const int offset= y_offset[i];
 263         const int z0= block[offset+stride*0] + block[offset+stride*4];
 264         const int z1= block[offset+stride*0] - block[offset+stride*4];
 265         const int z2= block[offset+stride*1] - block[offset+stride*5];
 266         const int z3= block[offset+stride*1] + block[offset+stride*5];
 267
 268         temp[4*i+0]= z0+z3;
 269         temp[4*i+1]= z1+z2;
 270         temp[4*i+2]= z1-z2;
 271         temp[4*i+3]= z0-z3;
 272     }
 273
 274     for(i=0; i<4; i++){
 275         const int offset= x_offset[i];
 276         const int z0= temp[4*0+i] + temp[4*2+i];
 277         const int z1= temp[4*0+i] - temp[4*2+i];
 278         const int z2= temp[4*1+i] - temp[4*3+i];
 279         const int z3= temp[4*1+i] + temp[4*3+i];
 280
 281         block[stride*0 +offset]= (z0 + z3)>>1;
 282         block[stride*2 +offset]= (z1 + z2)>>1;
 283         block[stride*8 +offset]= (z1 - z2)>>1;
 284         block[stride*10+offset]= (z0 - z3)>>1;
 285     }
 286 }
 287 #endif
 288
 289 #undef xStride
 290 #undef stride
 291
 292 #if 0
 293 static void chroma_dc_dct_c(DCTELEM *block){
 294     const int stride= 16*2;
 295     const int xStride= 16;
 296     int a,b,c,d,e;
 297
 298     a= block[stride*0 + xStride*0];
 299     b= block[stride*0 + xStride*1];
 300     c= block[stride*1 + xStride*0];
 301     d= block[stride*1 + xStride*1];
 302
 303     e= a-b;
 304     a= a+b;
 305     b= c-d;
 306     c= c+d;
 307
 308     block[stride*0 + xStride*0]= (a+c);
 309     block[stride*0 + xStride*1]= (e+b);
 310     block[stride*1 + xStride*0]= (a-c);
 311     block[stride*1 + xStride*1]= (e-b);
 312 }
 313 #endif
 314
 315
 316 static void free_tables(H264Context *h, int free_rbsp){
 317     int i;
 318     H264Context *hx;
 319     av_freep(&h->intra4x4_pred_mode);
 320     av_freep(&h->chroma_pred_mode_table);
 321     av_freep(&h->cbp_table);
 322     av_freep(&h->mvd_table[0]);
 323     av_freep(&h->mvd_table[1]);
 324     av_freep(&h->direct_table);
 325     av_freep(&h->non_zero_count);
 326     av_freep(&h->slice_table_base);
 327     h->slice_table= NULL;
 328     av_freep(&h->list_counts);
 329
 330     av_freep(&h->mb2b_xy);
 331     av_freep(&h->mb2br_xy);
 332
 333     for(i = 0; i < MAX_THREADS; i++) {
 334         hx = h->thread_context[i];
 335         if(!hx) continue;
 336         av_freep(&hx->top_borders[1]);
 337         av_freep(&hx->top_borders[0]);
 338         av_freep(&hx->s.obmc_scratchpad);
 339         if (free_rbsp){
 340             av_freep(&hx->rbsp_buffer[1]);
 341             av_freep(&hx->rbsp_buffer[0]);
 342             hx->rbsp_buffer_size[0] = 0;
 343             hx->rbsp_buffer_size[1] = 0;
 344         }
 345         if (i) av_freep(&h->thread_context[i]);
 346     }
 347 }
 348
 349 static void init_dequant8_coeff_table(H264Context *h){
 350     int i,q,x;
 351     const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
 352     h->dequant8_coeff[0] = h->dequant8_buffer[0];
 353     h->dequant8_coeff[1] = h->dequant8_buffer[1];
 354
 355     for(i=0; i<2; i++ ){
 356         if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
 357             h->dequant8_coeff[1] = h->dequant8_buffer[0];
 358             break;
 359         }
 360
 361         for(q=0; q<max_qp+1; q++){
 362             int shift = div6[q];
 363             int idx = rem6[q];
 364             for(x=0; x<64; x++)
 365                 h->dequant8_coeff[i][q][(x>>3)|((x&7)<<3)] =
 366                     ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
 367                     h->pps.scaling_matrix8[i][x]) << shift;
 368         }
 369     }
 370 }
 371
 372 static void init_dequant4_coeff_table(H264Context *h){
 373     int i,j,q,x;
 374     const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
 375     for(i=0; i<6; i++ ){
 376         h->dequant4_coeff[i] = h->dequant4_buffer[i];
 377         for(j=0; j<i; j++){
 378             if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
 379                 h->dequant4_coeff[i] = h->dequant4_buffer[j];
 380                 break;
 381             }
 382         }
 383         if(j<i)
 384             continue;
 385
 386         for(q=0; q<max_qp+1; q++){
 387             int shift = div6[q] + 2;
 388             int idx = rem6[q];
 389             for(x=0; x<16; x++)
 390                 h->dequant4_coeff[i][q][(x>>2)|((x<<2)&0xF)] =
 391                     ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
 392                     h->pps.scaling_matrix4[i][x]) << shift;
 393         }
 394     }
 395 }
 396
 397 static void init_dequant_tables(H264Context *h){
 398     int i,x;
 399     init_dequant4_coeff_table(h);
 400     if(h->pps.transform_8x8_mode)
 401         init_dequant8_coeff_table(h);
 402     if(h->sps.transform_bypass){
 403         for(i=0; i<6; i++)
 404             for(x=0; x<16; x++)
 405                 h->dequant4_coeff[i][0][x] = 1<<6;
 406         if(h->pps.transform_8x8_mode)
 407             for(i=0; i<2; i++)
 408                 for(x=0; x<64; x++)
 409                     h->dequant8_coeff[i][0][x] = 1<<6;
 410     }
 411 }
 412
 413
 414 int ff_h264_alloc_tables(H264Context *h){
 415     MpegEncContext * const s = &h->s;
 416     const int big_mb_num= s->mb_stride * (s->mb_height+1);
 417     const int row_mb_num= 2*s->mb_stride*s->avctx->thread_count;
 418     int x,y;
 419
 420     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8  * sizeof(uint8_t), fail)
 421
 422     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 32 * sizeof(uint8_t), fail)
 423     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
 424     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
 425
 426     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
 427     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
 428     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
 429     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
 430     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
 431
 432     memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
 433     h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
 434
 435     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy  , big_mb_num * sizeof(uint32_t), fail);
 436     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
 437     for(y=0; y<s->mb_height; y++){
 438         for(x=0; x<s->mb_width; x++){
 439             const int mb_xy= x + y*s->mb_stride;
 440             const int b_xy = 4*x + 4*y*h->b_stride;
 441
 442             h->mb2b_xy [mb_xy]= b_xy;
 443             h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
 444         }
 445     }
 446
 447     s->obmc_scratchpad = NULL;
 448
 449     if(!h->dequant4_coeff[0])
 450         init_dequant_tables(h);
 451
 452     return 0;
 453 fail:
 454     free_tables(h, 1);
 455     return -1;
 456 }
 457
 458 /**
 459  * Mimic alloc_tables(), but for every context thread.
 460  */
 461 static void clone_tables(H264Context *dst, H264Context *src, int i){
 462     MpegEncContext * const s = &src->s;
 463     dst->intra4x4_pred_mode       = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
 464     dst->non_zero_count           = src->non_zero_count;
 465     dst->slice_table              = src->slice_table;
 466     dst->cbp_table                = src->cbp_table;
 467     dst->mb2b_xy                  = src->mb2b_xy;
 468     dst->mb2br_xy                 = src->mb2br_xy;
 469     dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
 470     dst->mvd_table[0]             = src->mvd_table[0] + i*8*2*s->mb_stride;
 471     dst->mvd_table[1]             = src->mvd_table[1] + i*8*2*s->mb_stride;
 472     dst->direct_table             = src->direct_table;
 473     dst->list_counts              = src->list_counts;
 474
 475     dst->s.obmc_scratchpad = NULL;
 476     ff_h264_pred_init(&dst->hpc, src->s.codec_id, src->sps.bit_depth_luma);
 477 }
 478
 479 /**
 480  * Init context
 481  * Allocate buffers which are not shared amongst multiple threads.
 482  */
 483 static int context_init(H264Context *h){
 484     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t)*2, fail)
 485     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t)*2, fail)
 486
 487     h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
 488     h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
 489
 490     return 0;
 491 fail:
 492     return -1; // free_tables will clean up for us
 493 }
 494
 495 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
 496
 497 static av_cold void common_init(H264Context *h){
 498     MpegEncContext * const s = &h->s;
 499
 500     s->width = s->avctx->width;
 501     s->height = s->avctx->height;
 502     s->codec_id= s->avctx->codec->id;
 503
 504     ff_h264dsp_init(&h->h264dsp, 8);
 505     ff_h264_pred_init(&h->hpc, s->codec_id, 8);
 506
 507     h->dequant_coeff_pps= -1;
 508     s->unrestricted_mv=1;
 509     s->decode=1; //FIXME
 510
 511     dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
 512
 513     memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
 514     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
 515 }
 516
 517 int ff_h264_decode_extradata(H264Context *h)
 518 {
 519     AVCodecContext *avctx = h->s.avctx;
 520
 521     if(*(char *)avctx->extradata == 1){
 522         int i, cnt, nalsize;
 523         unsigned char *p = avctx->extradata;
 524
 525         h->is_avc = 1;
 526
 527         if(avctx->extradata_size < 7) {
 528             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
 529             return -1;
 530         }
 531         /* sps and pps in the avcC always have length coded with 2 bytes,
 532            so put a fake nal_length_size = 2 while parsing them */
 533         h->nal_length_size = 2;
 534         // Decode sps from avcC
 535         cnt = *(p+5) & 0x1f; // Number of sps
 536         p += 6;
 537         for (i = 0; i < cnt; i++) {
 538             nalsize = AV_RB16(p) + 2;
 539             if(decode_nal_units(h, p, nalsize) < 0) {
 540                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
 541                 return -1;
 542             }
 543             p += nalsize;
 544         }
 545         // Decode pps from avcC
 546         cnt = *(p++); // Number of pps
 547         for (i = 0; i < cnt; i++) {
 548             nalsize = AV_RB16(p) + 2;
 549             if(decode_nal_units(h, p, nalsize) < 0) {
 550                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
 551                 return -1;
 552             }
 553             p += nalsize;
 554         }
 555         // Now store right nal length size, that will be use to parse all other nals
 556         h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
 557     } else {
 558         h->is_avc = 0;
 559         if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
 560             return -1;
 561     }
 562     return 0;
 563 }
 564
 565 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
 566     H264Context *h= avctx->priv_data;
 567     MpegEncContext * const s = &h->s;
 568
 569     MPV_decode_defaults(s);
 570
 571     s->avctx = avctx;
 572     common_init(h);
 573
 574     s->out_format = FMT_H264;
 575     s->workaround_bugs= avctx->workaround_bugs;
 576
 577     // set defaults
 578 //    s->decode_mb= ff_h263_decode_mb;
 579     s->quarter_sample = 1;
 580     if(!avctx->has_b_frames)
 581     s->low_delay= 1;
 582
 583     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
 584
 585     ff_h264_decode_init_vlc();
 586
 587     h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
 588     h->pixel_shift = 0;
 589
 590     h->thread_context[0] = h;
 591     h->outputed_poc = h->next_outputed_poc = INT_MIN;
 592     h->prev_poc_msb= 1<<16;
 593     h->x264_build = -1;
 594     ff_h264_reset_sei(h);
 595     if(avctx->codec_id == CODEC_ID_H264){
 596         if(avctx->ticks_per_frame == 1){
 597             s->avctx->time_base.den *=2;
 598         }
 599         avctx->ticks_per_frame = 2;
 600     }
 601
 602     if(avctx->extradata_size > 0 && avctx->extradata &&
 603         ff_h264_decode_extradata(h))
 604         return -1;
 605
 606     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
 607         s->avctx->has_b_frames = h->sps.num_reorder_frames;
 608         s->low_delay = 0;
 609     }
 610
 611     return 0;
 612 }
 613
 614 static void copy_picture_range(Picture **to, Picture **from, int count, MpegEncContext *new_base, MpegEncContext *old_base)
 615 {
 616     int i;
 617
 618     for (i=0; i<count; i++){
 619         to[i] = REBASE_PICTURE(from[i], new_base, old_base);
 620     }
 621 }
 622
 623 static void copy_parameter_set(void **to, void **from, int count, int size)
 624 {
 625     int i;
 626
 627     for (i=0; i<count; i++){
 628         if (to[i] && !from[i]) av_freep(&to[i]);
 629         else if (from[i] && !to[i]) to[i] = av_malloc(size);
 630
 631         if (from[i]) memcpy(to[i], from[i], size);
 632     }
 633 }
 634
 635 static int decode_init_thread_copy(AVCodecContext *avctx){
 636     H264Context *h= avctx->priv_data;
 637
 638     if (!avctx->is_copy) return 0;
 639     memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
 640     memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
 641
 642     return 0;
 643 }
 644
 645 #define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
 646 static int decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src){
 647     H264Context *h= dst->priv_data, *h1= src->priv_data;
 648     MpegEncContext * const s = &h->s, * const s1 = &h1->s;
 649     int inited = s->context_initialized, err;
 650     int i;
 651
 652     if(dst == src || !s1->context_initialized) return 0;
 653
 654     err = ff_mpeg_update_thread_context(dst, src);
 655     if(err) return err;
 656
 657     //FIXME handle width/height changing
 658     if(!inited){
 659         for(i = 0; i < MAX_SPS_COUNT; i++)
 660             av_freep(h->sps_buffers + i);
 661
 662         for(i = 0; i < MAX_PPS_COUNT; i++)
 663             av_freep(h->pps_buffers + i);
 664
 665         memcpy(&h->s + 1, &h1->s + 1, sizeof(H264Context) - sizeof(MpegEncContext)); //copy all fields after MpegEnc
 666         memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
 667         memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
 668         ff_h264_alloc_tables(h);
 669         context_init(h);
 670
 671         for(i=0; i<2; i++){
 672             h->rbsp_buffer[i] = NULL;
 673             h->rbsp_buffer_size[i] = 0;
 674         }
 675
 676         h->thread_context[0] = h;
 677
 678         // frame_start may not be called for the next thread (if it's decoding a bottom field)
 679         // so this has to be allocated here
 680         h->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
 681
 682         s->dsp.clear_blocks(h->mb);
 683     }
 684
 685     //extradata/NAL handling
 686     h->is_avc          = h1->is_avc;
 687
 688     //SPS/PPS
 689     copy_parameter_set((void**)h->sps_buffers, (void**)h1->sps_buffers, MAX_SPS_COUNT, sizeof(SPS));
 690     h->sps             = h1->sps;
 691     copy_parameter_set((void**)h->pps_buffers, (void**)h1->pps_buffers, MAX_PPS_COUNT, sizeof(PPS));
 692     h->pps             = h1->pps;
 693
 694     //Dequantization matrices
 695     //FIXME these are big - can they be only copied when PPS changes?
 696     copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
 697
 698     for(i=0; i<6; i++)
 699         h->dequant4_coeff[i] = h->dequant4_buffer[0] + (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
 700
 701     for(i=0; i<2; i++)
 702         h->dequant8_coeff[i] = h->dequant8_buffer[0] + (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
 703
 704     h->dequant_coeff_pps = h1->dequant_coeff_pps;
 705
 706     //POC timing
 707     copy_fields(h, h1, poc_lsb, redundant_pic_count);
 708
 709     //reference lists
 710     copy_fields(h, h1, ref_count, intra_gb);
 711     copy_fields(h, h1, short_ref, cabac_init_idc);
 712
 713     copy_picture_range(h->short_ref,   h1->short_ref,   32, s, s1);
 714     copy_picture_range(h->long_ref,    h1->long_ref,    32, s, s1);
 715     copy_picture_range(h->delayed_pic, h1->delayed_pic, MAX_DELAYED_PIC_COUNT+2, s, s1);
 716
 717     h->last_slice_type = h1->last_slice_type;
 718
 719     if(!s->current_picture_ptr) return 0;
 720
 721     if(!s->dropable) {
 722         ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
 723         h->prev_poc_msb     = h->poc_msb;
 724         h->prev_poc_lsb     = h->poc_lsb;
 725     }
 726     h->prev_frame_num_offset= h->frame_num_offset;
 727     h->prev_frame_num       = h->frame_num;
 728     h->outputed_poc         = h->next_outputed_poc;
 729
 730     return 0;
 731 }
 732
 733 int ff_h264_frame_start(H264Context *h){
 734     MpegEncContext * const s = &h->s;
 735     int i;
 736
 737     if(MPV_frame_start(s, s->avctx) < 0)
 738         return -1;
 739     ff_er_frame_start(s);
 740     /*
 741      * MPV_frame_start uses pict_type to derive key_frame.
 742      * This is incorrect for H.264; IDR markings must be used.
 743      * Zero here; IDR markings per slice in frame or fields are ORed in later.
 744      * See decode_nal_units().
 745      */
 746     s->current_picture_ptr->key_frame= 0;
 747     s->current_picture_ptr->mmco_reset= 0;
 748
 749     assert(s->linesize && s->uvlinesize);
 750
 751     for(i=0; i<16; i++){
 752         h->block_offset[i]= (4*((scan8[i] - scan8[0])&7)<<h->pixel_shift) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
 753         h->block_offset[24+i]= (4*((scan8[i] - scan8[0])&7)<<h->pixel_shift) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
 754     }
 755     for(i=0; i<4; i++){
 756         h->block_offset[16+i]=
 757         h->block_offset[20+i]= (4*((scan8[i] - scan8[0])&7)<<h->pixel_shift) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
 758         h->block_offset[24+16+i]=
 759         h->block_offset[24+20+i]= (4*((scan8[i] - scan8[0])&7)<<h->pixel_shift) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
 760     }
 761
 762     /* can't be in alloc_tables because linesize isn't known there.
 763      * FIXME: redo bipred weight to not require extra buffer? */
 764     for(i = 0; i < s->avctx->thread_count; i++)
 765         if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
 766             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
 767
 768     /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
 769     memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
 770
 771 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
 772
 773     // We mark the current picture as non-reference after allocating it, so
 774     // that if we break out due to an error it can be released automatically
 775     // in the next MPV_frame_start().
 776     // SVQ3 as well as most other codecs have only last/next/current and thus
 777     // get released even with set reference, besides SVQ3 and others do not
 778     // mark frames as reference later "naturally".
 779     if(s->codec_id != CODEC_ID_SVQ3)
 780         s->current_picture_ptr->reference= 0;
 781
 782     s->current_picture_ptr->field_poc[0]=
 783     s->current_picture_ptr->field_poc[1]= INT_MAX;
 784
 785     h->next_output_pic = NULL;
 786
 787     assert(s->current_picture_ptr->long_ref==0);
 788
 789     return 0;
 790 }
 791
 792 /**
 793   * Run setup operations that must be run after slice header decoding.
 794   * This includes finding the next displayed frame.
 795   *
 796   * @param h h264 master context
 797   */
 798 static void decode_postinit(H264Context *h){
 799     MpegEncContext * const s = &h->s;
 800     Picture *out = s->current_picture_ptr;
 801     Picture *cur = s->current_picture_ptr;
 802     int i, pics, out_of_order, out_idx;
 803
 804     s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
 805     s->current_picture_ptr->pict_type= s->pict_type;
 806
 807     if (h->next_output_pic) return;
 808
 809     if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
 810         //FIXME this allows the next thread to start once we encounter the first field of a PAFF packet
 811         //This works if the next packet contains the second field. It does not work if both fields are
 812         //in the same packet.
 813         //ff_thread_finish_setup(s->avctx);
 814         return;
 815     }
 816
 817     cur->interlaced_frame = 0;
 818     cur->repeat_pict = 0;
 819
 820     /* Signal interlacing information externally. */
 821     /* Prioritize picture timing SEI information over used decoding process if it exists. */
 822
 823     if(h->sps.pic_struct_present_flag){
 824         switch (h->sei_pic_struct)
 825         {
 826         case SEI_PIC_STRUCT_FRAME:
 827             break;
 828         case SEI_PIC_STRUCT_TOP_FIELD:
 829         case SEI_PIC_STRUCT_BOTTOM_FIELD:
 830             cur->interlaced_frame = 1;
 831             break;
 832         case SEI_PIC_STRUCT_TOP_BOTTOM:
 833         case SEI_PIC_STRUCT_BOTTOM_TOP:
 834             if (FIELD_OR_MBAFF_PICTURE)
 835                 cur->interlaced_frame = 1;
 836             else
 837                 // try to flag soft telecine progressive
 838                 cur->interlaced_frame = h->prev_interlaced_frame;
 839             break;
 840         case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
 841         case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
 842             // Signal the possibility of telecined film externally (pic_struct 5,6)
 843             // From these hints, let the applications decide if they apply deinterlacing.
 844             cur->repeat_pict = 1;
 845             break;
 846         case SEI_PIC_STRUCT_FRAME_DOUBLING:
 847             // Force progressive here, as doubling interlaced frame is a bad idea.
 848             cur->repeat_pict = 2;
 849             break;
 850         case SEI_PIC_STRUCT_FRAME_TRIPLING:
 851             cur->repeat_pict = 4;
 852             break;
 853         }
 854
 855         if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
 856             cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
 857     }else{
 858         /* Derive interlacing flag from used decoding process. */
 859         cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
 860     }
 861     h->prev_interlaced_frame = cur->interlaced_frame;
 862
 863     if (cur->field_poc[0] != cur->field_poc[1]){
 864         /* Derive top_field_first from field pocs. */
 865         cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
 866     }else{
 867         if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
 868             /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
 869             if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
 870               || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
 871                 cur->top_field_first = 1;
 872             else
 873                 cur->top_field_first = 0;
 874         }else{
 875             /* Most likely progressive */
 876             cur->top_field_first = 0;
 877         }
 878     }
 879
 880     //FIXME do something with unavailable reference frames
 881
 882     /* Sort B-frames into display order */
 883
 884     if(h->sps.bitstream_restriction_flag
 885        && s->avctx->has_b_frames < h->sps.num_reorder_frames){
 886         s->avctx->has_b_frames = h->sps.num_reorder_frames;
 887         s->low_delay = 0;
 888     }
 889
 890     if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
 891        && !h->sps.bitstream_restriction_flag){
 892         s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
 893         s->low_delay= 0;
 894     }
 895
 896     pics = 0;
 897     while(h->delayed_pic[pics]) pics++;
 898
 899     assert(pics <= MAX_DELAYED_PIC_COUNT);
 900
 901     h->delayed_pic[pics++] = cur;
 902     if(cur->reference == 0)
 903         cur->reference = DELAYED_PIC_REF;
 904
 905     out = h->delayed_pic[0];
 906     out_idx = 0;
 907     for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
 908         if(h->delayed_pic[i]->poc < out->poc){
 909             out = h->delayed_pic[i];
 910             out_idx = i;
 911         }
 912     if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
 913         h->next_outputed_poc= INT_MIN;
 914     out_of_order = out->poc < h->next_outputed_poc;
 915
 916     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
 917         { }
 918     else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
 919        || (s->low_delay &&
 920         ((h->next_outputed_poc != INT_MIN && out->poc > h->next_outputed_poc + 2)
 921          || cur->pict_type == AV_PICTURE_TYPE_B)))
 922     {
 923         s->low_delay = 0;
 924         s->avctx->has_b_frames++;
 925     }
 926
 927     if(out_of_order || pics > s->avctx->has_b_frames){
 928         out->reference &= ~DELAYED_PIC_REF;
 929         out->owner2 = s; // for frame threading, the owner must be the second field's thread
 930                          // or else the first thread can release the picture and reuse it unsafely
 931         for(i=out_idx; h->delayed_pic[i]; i++)
 932             h->delayed_pic[i] = h->delayed_pic[i+1];
 933     }
 934     if(!out_of_order && pics > s->avctx->has_b_frames){
 935         h->next_output_pic = out;
 936         if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
 937             h->next_outputed_poc = INT_MIN;
 938         } else
 939             h->next_outputed_poc = out->poc;
 940     }else{
 941         av_log(s->avctx, AV_LOG_DEBUG, "no picture\n");
 942     }
 943
 944     ff_thread_finish_setup(s->avctx);
 945 }
 946
 947 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){
 948     MpegEncContext * const s = &h->s;
 949     uint8_t *top_border;
 950     int top_idx = 1;
 951
 952     src_y  -=   linesize;
 953     src_cb -= uvlinesize;
 954     src_cr -= uvlinesize;
 955
 956     if(!simple && FRAME_MBAFF){
 957         if(s->mb_y&1){
 958             if(!MB_MBAFF){
 959                 top_border = h->top_borders[0][s->mb_x];
 960                 AV_COPY128(top_border, src_y + 15*linesize);
 961                 if (h->pixel_shift)
 962                     AV_COPY128(top_border+16, src_y+15*linesize+16);
 963                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
 964                     if (h->pixel_shift) {
 965                         AV_COPY128(top_border+32, src_cb+7*uvlinesize);
 966                         AV_COPY128(top_border+48, src_cr+7*uvlinesize);
 967                     } else {
 968                     AV_COPY64(top_border+16, src_cb+7*uvlinesize);
 969                     AV_COPY64(top_border+24, src_cr+7*uvlinesize);
 970                     }
 971                 }
 972             }
 973         }else if(MB_MBAFF){
 974             top_idx = 0;
 975         }else
 976             return;
 977     }
 978
 979     top_border = h->top_borders[top_idx][s->mb_x];
 980     // There are two lines saved, the line above the the top macroblock of a pair,
 981     // and the line above the bottom macroblock
 982     AV_COPY128(top_border, src_y + 16*linesize);
 983     if (h->pixel_shift)
 984         AV_COPY128(top_border+16, src_y+16*linesize+16);
 985
 986     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
 987         if (h->pixel_shift) {
 988             AV_COPY128(top_border+32, src_cb+8*uvlinesize);
 989             AV_COPY128(top_border+48, src_cr+8*uvlinesize);
 990         } else {
 991         AV_COPY64(top_border+16, src_cb+8*uvlinesize);
 992         AV_COPY64(top_border+24, src_cr+8*uvlinesize);
 993         }
 994     }
 995 }
 996
 997 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, int pixel_shift){
 998     MpegEncContext * const s = &h->s;
 999     int deblock_left;
1000     int deblock_top;
1001     int top_idx = 1;
1002     uint8_t *top_border_m1;
1003     uint8_t *top_border;
1004
1005     if(!simple && FRAME_MBAFF){
1006         if(s->mb_y&1){
1007             if(!MB_MBAFF)
1008                 return;
1009         }else{
1010             top_idx = MB_MBAFF ? 0 : 1;
1011         }
1012     }
1013
1014     if(h->deblocking_filter == 2) {
1015         deblock_left = h->left_type[0];
1016         deblock_top  = h->top_type;
1017     } else {
1018         deblock_left = (s->mb_x > 0);
1019         deblock_top =  (s->mb_y > !!MB_FIELD);
1020     }
1021
1022     src_y  -=   linesize + 1 + pixel_shift;
1023     src_cb -= uvlinesize + 1 + pixel_shift;
1024     src_cr -= uvlinesize + 1 + pixel_shift;
1025
1026     top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1027     top_border    = h->top_borders[top_idx][s->mb_x];
1028
1029 #define XCHG(a,b,xchg)\
1030     if (pixel_shift) {\
1031         if (xchg) {\
1032             AV_SWAP64(b+0,a+0);\
1033             AV_SWAP64(b+8,a+8);\
1034         } else {\
1035             AV_COPY128(b,a); \
1036         }\
1037     } else \
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<<pixel_shift), src_y -(7<<h->pixel_shift), 1);
1044         }
1045         XCHG(top_border+(0<<pixel_shift), src_y +(1<<pixel_shift), xchg);
1046         XCHG(top_border+(8<<pixel_shift), src_y +(9<<pixel_shift), 1);
1047         if(s->mb_x+1 < s->mb_width){
1048             XCHG(h->top_borders[top_idx][s->mb_x+1], src_y +(17<<pixel_shift), 1);
1049         }
1050     }
1051     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1052         if(deblock_top){
1053             if(deblock_left){
1054                 XCHG(top_border_m1+(16<<pixel_shift), src_cb -(7<<pixel_shift), 1);
1055                 XCHG(top_border_m1+(24<<pixel_shift), src_cr -(7<<pixel_shift), 1);
1056             }
1057             XCHG(top_border+(16<<pixel_shift), src_cb+1+pixel_shift, 1);
1058             XCHG(top_border+(24<<pixel_shift), src_cr+1+pixel_shift, 1);
1059         }
1060     }
1061 }
1062
1063 static av_always_inline int dctcoef_get(H264Context *h, DCTELEM *mb, int index, int pixel_shift) {
1064     if (!pixel_shift)
1065         return mb[index];
1066     else
1067         return ((int32_t*)mb)[index];
1068 }
1069
1070 static av_always_inline void dctcoef_set(H264Context *h, DCTELEM *mb, int index, int value, int pixel_shift) {
1071     if (!pixel_shift)
1072         mb[index] = value;
1073     else
1074         ((int32_t*)mb)[index] = value;
1075 }
1076
1077 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, int pixel_shift){
1078     MpegEncContext * const s = &h->s;
1079     const int mb_x= s->mb_x;
1080     const int mb_y= s->mb_y;
1081     const int mb_xy= h->mb_xy;
1082     const int mb_type= s->current_picture.mb_type[mb_xy];
1083     uint8_t  *dest_y, *dest_cb, *dest_cr;
1084     int linesize, uvlinesize /*dct_offset*/;
1085     int i;
1086     int *block_offset = &h->block_offset[0];
1087     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1088     /* is_h264 should always be true if SVQ3 is disabled. */
1089     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1090     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1091     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1092
1093     dest_y  = s->current_picture.data[0] + ((mb_x<<pixel_shift) + mb_y * s->linesize  ) * 16;
1094     dest_cb = s->current_picture.data[1] + ((mb_x<<pixel_shift) + mb_y * s->uvlinesize) * 8;
1095     dest_cr = s->current_picture.data[2] + ((mb_x<<pixel_shift) + mb_y * s->uvlinesize) * 8;
1096
1097     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + (64<<pixel_shift), s->linesize, 4);
1098     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + (64<<pixel_shift), dest_cr - dest_cb, 2);
1099
1100     h->list_counts[mb_xy]= h->list_count;
1101
1102     if (!simple && MB_FIELD) {
1103         linesize   = h->mb_linesize   = s->linesize * 2;
1104         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1105         block_offset = &h->block_offset[24];
1106         if(mb_y&1){ //FIXME move out of this function?
1107             dest_y -= s->linesize*15;
1108             dest_cb-= s->uvlinesize*7;
1109             dest_cr-= s->uvlinesize*7;
1110         }
1111         if(FRAME_MBAFF) {
1112             int list;
1113             for(list=0; list<h->list_count; list++){
1114                 if(!USES_LIST(mb_type, list))
1115                     continue;
1116                 if(IS_16X16(mb_type)){
1117                     int8_t *ref = &h->ref_cache[list][scan8[0]];
1118                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1119                 }else{
1120                     for(i=0; i<16; i+=4){
1121                         int ref = h->ref_cache[list][scan8[i]];
1122                         if(ref >= 0)
1123                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1124                     }
1125                 }
1126             }
1127         }
1128     } else {
1129         linesize   = h->mb_linesize   = s->linesize;
1130         uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1131 //        dct_offset = s->linesize * 16;
1132     }
1133
1134     if (!simple && IS_INTRA_PCM(mb_type)) {
1135         if (pixel_shift) {
1136             const int bit_depth = h->sps.bit_depth_luma;
1137             int j;
1138             GetBitContext gb;
1139             init_get_bits(&gb, (uint8_t*)h->mb, 384*bit_depth);
1140
1141             for (i = 0; i < 16; i++) {
1142                 uint16_t *tmp_y  = (uint16_t*)(dest_y  + i*linesize);
1143                 for (j = 0; j < 16; j++)
1144                     tmp_y[j] = get_bits(&gb, bit_depth);
1145             }
1146             for (i = 0; i < 8; i++) {
1147                 uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1148                 for (j = 0; j < 8; j++)
1149                     tmp_cb[j] = get_bits(&gb, bit_depth);
1150             }
1151             for (i = 0; i < 8; i++) {
1152                 uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1153                 for (j = 0; j < 8; j++)
1154                     tmp_cr[j] = get_bits(&gb, bit_depth);
1155             }
1156         } else {
1157         for (i=0; i<16; i++) {
1158             memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1159         }
1160         for (i=0; i<8; i++) {
1161             memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
1162             memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
1163         }
1164         }
1165     } else {
1166         if(IS_INTRA(mb_type)){
1167             if(h->deblocking_filter)
1168                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple, pixel_shift);
1169
1170             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1171                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1172                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1173             }
1174
1175             if(IS_INTRA4x4(mb_type)){
1176                 if(simple || !s->encoding){
1177                     if(IS_8x8DCT(mb_type)){
1178                         if(transform_bypass){
1179                             idct_dc_add =
1180                             idct_add    = s->dsp.add_pixels8;
1181                         }else{
1182                             idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1183                             idct_add    = h->h264dsp.h264_idct8_add;
1184                         }
1185                         for(i=0; i<16; i+=4){
1186                             uint8_t * const ptr= dest_y + block_offset[i];
1187                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1188                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1189                                 h->hpc.pred8x8l_add[dir](ptr, h->mb + (i*16<<pixel_shift), linesize);
1190                             }else{
1191                                 const int nnz = h->non_zero_count_cache[ scan8[i] ];
1192                                 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1193                                                             (h->topright_samples_available<<i)&0x4000, linesize);
1194                                 if(nnz){
1195                                     if(nnz == 1 && dctcoef_get(h, h->mb, i*16, pixel_shift))
1196                                         idct_dc_add(ptr, h->mb + (i*16<<pixel_shift), linesize);
1197                                     else
1198                                         idct_add   (ptr, h->mb + (i*16<<pixel_shift), linesize);
1199                                 }
1200                             }
1201                         }
1202                     }else{
1203                         if(transform_bypass){
1204                             idct_dc_add =
1205                             idct_add    = s->dsp.add_pixels4;
1206                         }else{
1207                             idct_dc_add = h->h264dsp.h264_idct_dc_add;
1208                             idct_add    = h->h264dsp.h264_idct_add;
1209                         }
1210                         for(i=0; i<16; i++){
1211                             uint8_t * const ptr= dest_y + block_offset[i];
1212                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1213
1214                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1215                                 h->hpc.pred4x4_add[dir](ptr, h->mb + (i*16<<pixel_shift), linesize);
1216                             }else{
1217                                 uint8_t *topright;
1218                                 int nnz, tr;
1219                                 uint64_t tr_high;
1220                                 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1221                                     const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1222                                     assert(mb_y || linesize <= block_offset[i]);
1223                                     if(!topright_avail){
1224                                         if (pixel_shift) {
1225                                             tr_high= ((uint16_t*)ptr)[3 - linesize/2]*0x0001000100010001ULL;
1226                                             topright= (uint8_t*) &tr_high;
1227                                         } else {
1228                                         tr= ptr[3 - linesize]*0x01010101;
1229                                         topright= (uint8_t*) &tr;
1230                                         }
1231                                     }else
1232                                         topright= ptr + (4<<pixel_shift) - linesize;
1233                                 }else
1234                                     topright= NULL;
1235
1236                                 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1237                                 nnz = h->non_zero_count_cache[ scan8[i] ];
1238                                 if(nnz){
1239                                     if(is_h264){
1240                                         if(nnz == 1 && dctcoef_get(h, h->mb, i*16, pixel_shift))
1241                                             idct_dc_add(ptr, h->mb + (i*16<<pixel_shift), linesize);
1242                                         else
1243                                             idct_add   (ptr, h->mb + (i*16<<pixel_shift), linesize);
1244                                     }
1245 #if CONFIG_SVQ3_DECODER
1246                                     else
1247                                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1248 #endif
1249                                 }
1250                             }
1251                         }
1252                     }
1253                 }
1254             }else{
1255                 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1256                 if(is_h264){
1257                     if(h->non_zero_count_cache[ scan8[LUMA_DC_BLOCK_INDEX] ]){
1258                         if(!transform_bypass)
1259                             h->h264dsp.h264_luma_dc_dequant_idct(h->mb, h->mb_luma_dc, h->dequant4_coeff[0][s->qscale][0]);
1260                         else{
1261                             static const uint8_t dc_mapping[16] = { 0*16, 1*16, 4*16, 5*16, 2*16, 3*16, 6*16, 7*16,
1262                                                                     8*16, 9*16,12*16,13*16,10*16,11*16,14*16,15*16};
1263                             for(i = 0; i < 16; i++)
1264                                 dctcoef_set(h, h->mb, dc_mapping[i], dctcoef_get(h, h->mb_luma_dc, i,pixel_shift),pixel_shift);
1265                         }
1266                     }
1267                 }
1268 #if CONFIG_SVQ3_DECODER
1269                 else
1270                     ff_svq3_luma_dc_dequant_idct_c(h->mb, h->mb_luma_dc, s->qscale);
1271 #endif
1272             }
1273             if(h->deblocking_filter)
1274                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple, pixel_shift);
1275         }else if(is_h264){
1276             ff_hl_motion(h, dest_y, dest_cb, dest_cr,
1277                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1278                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1279                       h->h264dsp.weight_h264_pixels_tab, h->h264dsp.biweight_h264_pixels_tab);
1280         }
1281
1282
1283         if(!IS_INTRA4x4(mb_type)){
1284             if(is_h264){
1285                 if(IS_INTRA16x16(mb_type)){
1286                     if(transform_bypass){
1287                         if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1288                             h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1289                         }else{
1290                             for(i=0; i<16; i++){
1291                                 if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h, h->mb, i*16,pixel_shift))
1292                                     s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + (i*16<<pixel_shift), linesize);
1293                             }
1294                         }
1295                     }else{
1296                          h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1297                     }
1298                 }else if(h->cbp&15){
1299                     if(transform_bypass){
1300                         const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1301                         idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1302                         for(i=0; i<16; i+=di){
1303                             if(h->non_zero_count_cache[ scan8[i] ]){
1304                                 idct_add(dest_y + block_offset[i], h->mb + (i*16<<pixel_shift), linesize);
1305                             }
1306                         }
1307                     }else{
1308                         if(IS_8x8DCT(mb_type)){
1309                             h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1310                         }else{
1311                             h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1312                         }
1313                     }
1314                 }
1315             }
1316 #if CONFIG_SVQ3_DECODER
1317             else{
1318                 for(i=0; i<16; i++){
1319                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1320                         uint8_t * const ptr= dest_y + block_offset[i];
1321                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1322                     }
1323                 }
1324             }
1325 #endif
1326         }
1327
1328         if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1329             uint8_t *dest[2] = {dest_cb, dest_cr};
1330             if(transform_bypass){
1331                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1332                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + (16*16<<pixel_shift), uvlinesize);
1333                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + (20*16<<pixel_shift), uvlinesize);
1334                 }else{
1335                     idct_add = s->dsp.add_pixels4;
1336                     for(i=16; i<16+8; i++){
1337                         if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h, h->mb, i*16,pixel_shift))
1338                             idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + (i*16<<pixel_shift), uvlinesize);
1339                     }
1340                 }
1341             }else{
1342                 if(is_h264){
1343                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+0] ])
1344                         h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16<<pixel_shift)       , h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
1345                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+1] ])
1346                         h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + ((16*16+4*16)<<pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
1347                     h->h264dsp.h264_idct_add8(dest, block_offset,
1348                                               h->mb, uvlinesize,
1349                                               h->non_zero_count_cache);
1350                 }
1351 #if CONFIG_SVQ3_DECODER
1352                 else{
1353                     h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16     , h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
1354                     h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16+4*16, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
1355                     for(i=16; i<16+8; i++){
1356                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1357                             uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1358                             ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[0][s->qscale + 12] - 12, 2);
1359                         }
1360                     }
1361                 }
1362 #endif
1363             }
1364         }
1365     }
1366     if(h->cbp || IS_INTRA(mb_type))
1367         s->dsp.clear_blocks(h->mb);
1368 }
1369
1370 /**
1371  * Process a macroblock; this case avoids checks for expensive uncommon cases.
1372  */
1373 static void hl_decode_mb_simple8(H264Context *h){
1374     hl_decode_mb_internal(h, 1, 0);
1375 }
1376
1377 /**
1378  * Process a macroblock; this handles edge cases, such as interlacing.
1379  */
1380 static void av_noinline hl_decode_mb_complex(H264Context *h){
1381     hl_decode_mb_internal(h, 0, h->pixel_shift);
1382 }
1383
1384 void ff_h264_hl_decode_mb(H264Context *h){
1385     MpegEncContext * const s = &h->s;
1386     const int mb_xy= h->mb_xy;
1387     const int mb_type= s->current_picture.mb_type[mb_xy];
1388     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1389
1390     if (is_complex || h->pixel_shift)
1391         hl_decode_mb_complex(h);
1392     else{
1393         hl_decode_mb_simple8(h);
1394     }
1395 }
1396
1397 static int pred_weight_table(H264Context *h){
1398     MpegEncContext * const s = &h->s;
1399     int list, i;
1400     int luma_def, chroma_def;
1401
1402     h->use_weight= 0;
1403     h->use_weight_chroma= 0;
1404     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1405     if(CHROMA)
1406         h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1407     luma_def = 1<<h->luma_log2_weight_denom;
1408     chroma_def = 1<<h->chroma_log2_weight_denom;
1409
1410     for(list=0; list<2; list++){
1411         h->luma_weight_flag[list]   = 0;
1412         h->chroma_weight_flag[list] = 0;
1413         for(i=0; i<h->ref_count[list]; i++){
1414             int luma_weight_flag, chroma_weight_flag;
1415
1416             luma_weight_flag= get_bits1(&s->gb);
1417             if(luma_weight_flag){
1418                 h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
1419                 h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
1420                 if(   h->luma_weight[i][list][0] != luma_def
1421                    || h->luma_weight[i][list][1] != 0) {
1422                     h->use_weight= 1;
1423                     h->luma_weight_flag[list]= 1;
1424                 }
1425             }else{
1426                 h->luma_weight[i][list][0]= luma_def;
1427                 h->luma_weight[i][list][1]= 0;
1428             }
1429
1430             if(CHROMA){
1431                 chroma_weight_flag= get_bits1(&s->gb);
1432                 if(chroma_weight_flag){
1433                     int j;
1434                     for(j=0; j<2; j++){
1435                         h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
1436                         h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
1437                         if(   h->chroma_weight[i][list][j][0] != chroma_def
1438                            || h->chroma_weight[i][list][j][1] != 0) {
1439                             h->use_weight_chroma= 1;
1440                             h->chroma_weight_flag[list]= 1;
1441                         }
1442                     }
1443                 }else{
1444                     int j;
1445                     for(j=0; j<2; j++){
1446                         h->chroma_weight[i][list][j][0]= chroma_def;
1447                         h->chroma_weight[i][list][j][1]= 0;
1448                     }
1449                 }
1450             }
1451         }
1452         if(h->slice_type_nos != AV_PICTURE_TYPE_B) break;
1453     }
1454     h->use_weight= h->use_weight || h->use_weight_chroma;
1455     return 0;
1456 }
1457
1458 /**
1459  * Initialize implicit_weight table.
1460  * @param field  0/1 initialize the weight for interlaced MBAFF
1461  *                -1 initializes the rest
1462  */
1463 static void implicit_weight_table(H264Context *h, int field){
1464     MpegEncContext * const s = &h->s;
1465     int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
1466
1467     for (i = 0; i < 2; i++) {
1468         h->luma_weight_flag[i]   = 0;
1469         h->chroma_weight_flag[i] = 0;
1470     }
1471
1472     if(field < 0){
1473         cur_poc = s->current_picture_ptr->poc;
1474     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
1475        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1476         h->use_weight= 0;
1477         h->use_weight_chroma= 0;
1478         return;
1479     }
1480         ref_start= 0;
1481         ref_count0= h->ref_count[0];
1482         ref_count1= h->ref_count[1];
1483     }else{
1484         cur_poc = s->current_picture_ptr->field_poc[field];
1485         ref_start= 16;
1486         ref_count0= 16+2*h->ref_count[0];
1487         ref_count1= 16+2*h->ref_count[1];
1488     }
1489
1490     h->use_weight= 2;
1491     h->use_weight_chroma= 2;
1492     h->luma_log2_weight_denom= 5;
1493     h->chroma_log2_weight_denom= 5;
1494
1495     for(ref0=ref_start; ref0 < ref_count0; ref0++){
1496         int poc0 = h->ref_list[0][ref0].poc;
1497         for(ref1=ref_start; ref1 < ref_count1; ref1++){
1498             int poc1 = h->ref_list[1][ref1].poc;
1499             int td = av_clip(poc1 - poc0, -128, 127);
1500             int w= 32;
1501             if(td){
1502                 int tb = av_clip(cur_poc - poc0, -128, 127);
1503                 int tx = (16384 + (FFABS(td) >> 1)) / td;
1504                 int dist_scale_factor = (tb*tx + 32) >> 8;
1505                 if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
1506                     w = 64 - dist_scale_factor;
1507             }
1508             if(field<0){
1509                 h->implicit_weight[ref0][ref1][0]=
1510                 h->implicit_weight[ref0][ref1][1]= w;
1511             }else{
1512                 h->implicit_weight[ref0][ref1][field]=w;
1513             }
1514         }
1515     }
1516 }
1517
1518 /**
1519  * instantaneous decoder refresh.
1520  */
1521 static void idr(H264Context *h){
1522     ff_h264_remove_all_refs(h);
1523     h->prev_frame_num= 0;
1524     h->prev_frame_num_offset= 0;
1525     h->prev_poc_msb=
1526     h->prev_poc_lsb= 0;
1527 }
1528
1529 /* forget old pics after a seek */
1530 static void flush_dpb(AVCodecContext *avctx){
1531     H264Context *h= avctx->priv_data;
1532     int i;
1533     for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1534         if(h->delayed_pic[i])
1535             h->delayed_pic[i]->reference= 0;
1536         h->delayed_pic[i]= NULL;
1537     }
1538     h->outputed_poc=h->next_outputed_poc= INT_MIN;
1539     h->prev_interlaced_frame = 1;
1540     idr(h);
1541     if(h->s.current_picture_ptr)
1542         h->s.current_picture_ptr->reference= 0;
1543     h->s.first_field= 0;
1544     ff_h264_reset_sei(h);
1545     ff_mpeg_flush(avctx);
1546 }
1547
1548 static int init_poc(H264Context *h){
1549     MpegEncContext * const s = &h->s;
1550     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1551     int field_poc[2];
1552     Picture *cur = s->current_picture_ptr;
1553
1554     h->frame_num_offset= h->prev_frame_num_offset;
1555     if(h->frame_num < h->prev_frame_num)
1556         h->frame_num_offset += max_frame_num;
1557
1558     if(h->sps.poc_type==0){
1559         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1560
1561         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1562             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1563         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1564             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1565         else
1566             h->poc_msb = h->prev_poc_msb;
1567 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1568         field_poc[0] =
1569         field_poc[1] = h->poc_msb + h->poc_lsb;
1570         if(s->picture_structure == PICT_FRAME)
1571             field_poc[1] += h->delta_poc_bottom;
1572     }else if(h->sps.poc_type==1){
1573         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1574         int i;
1575
1576         if(h->sps.poc_cycle_length != 0)
1577             abs_frame_num = h->frame_num_offset + h->frame_num;
1578         else
1579             abs_frame_num = 0;
1580
1581         if(h->nal_ref_idc==0 && abs_frame_num > 0)
1582             abs_frame_num--;
1583
1584         expected_delta_per_poc_cycle = 0;
1585         for(i=0; i < h->sps.poc_cycle_length; i++)
1586             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1587
1588         if(abs_frame_num > 0){
1589             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1590             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1591
1592             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1593             for(i = 0; i <= frame_num_in_poc_cycle; i++)
1594                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1595         } else
1596             expectedpoc = 0;
1597
1598         if(h->nal_ref_idc == 0)
1599             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1600
1601         field_poc[0] = expectedpoc + h->delta_poc[0];
1602         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1603
1604         if(s->picture_structure == PICT_FRAME)
1605             field_poc[1] += h->delta_poc[1];
1606     }else{
1607         int poc= 2*(h->frame_num_offset + h->frame_num);
1608
1609         if(!h->nal_ref_idc)
1610             poc--;
1611
1612         field_poc[0]= poc;
1613         field_poc[1]= poc;
1614     }
1615
1616     if(s->picture_structure != PICT_BOTTOM_FIELD)
1617         s->current_picture_ptr->field_poc[0]= field_poc[0];
1618     if(s->picture_structure != PICT_TOP_FIELD)
1619         s->current_picture_ptr->field_poc[1]= field_poc[1];
1620     cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1621
1622     return 0;
1623 }
1624
1625
1626 /**
1627  * initialize scan tables
1628  */
1629 static void init_scan_tables(H264Context *h){
1630     int i;
1631     for(i=0; i<16; i++){
1632 #define T(x) (x>>2) | ((x<<2) & 0xF)
1633         h->zigzag_scan[i] = T(zigzag_scan[i]);
1634         h-> field_scan[i] = T( field_scan[i]);
1635 #undef T
1636     }
1637     for(i=0; i<64; i++){
1638 #define T(x) (x>>3) | ((x&7)<<3)
1639         h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
1640         h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1641         h->field_scan8x8[i]        = T(field_scan8x8[i]);
1642         h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
1643 #undef T
1644     }
1645     if(h->sps.transform_bypass){ //FIXME same ugly
1646         h->zigzag_scan_q0          = zigzag_scan;
1647         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
1648         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1649         h->field_scan_q0           = field_scan;
1650         h->field_scan8x8_q0        = field_scan8x8;
1651         h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
1652     }else{
1653         h->zigzag_scan_q0          = h->zigzag_scan;
1654         h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
1655         h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1656         h->field_scan_q0           = h->field_scan;
1657         h->field_scan8x8_q0        = h->field_scan8x8;
1658         h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
1659     }
1660 }
1661
1662 static void field_end(H264Context *h, int in_setup){
1663     MpegEncContext * const s = &h->s;
1664     AVCodecContext * const avctx= s->avctx;
1665     s->mb_y= 0;
1666
1667     if (!in_setup && !s->dropable)
1668         ff_thread_report_progress((AVFrame*)s->current_picture_ptr, (16*s->mb_height >> FIELD_PICTURE) - 1,
1669                                  s->picture_structure==PICT_BOTTOM_FIELD);
1670
1671     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1672         ff_vdpau_h264_set_reference_frames(s);
1673
1674     if(in_setup || !(avctx->active_thread_type&FF_THREAD_FRAME)){
1675         if(!s->dropable) {
1676             ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1677             h->prev_poc_msb= h->poc_msb;
1678             h->prev_poc_lsb= h->poc_lsb;
1679         }
1680         h->prev_frame_num_offset= h->frame_num_offset;
1681         h->prev_frame_num= h->frame_num;
1682         h->outputed_poc = h->next_outputed_poc;
1683     }
1684
1685     if (avctx->hwaccel) {
1686         if (avctx->hwaccel->end_frame(avctx) < 0)
1687             av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1688     }
1689
1690     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1691         ff_vdpau_h264_picture_complete(s);
1692
1693     /*
1694      * FIXME: Error handling code does not seem to support interlaced
1695      * when slices span multiple rows
1696      * The ff_er_add_slice calls don't work right for bottom
1697      * fields; they cause massive erroneous error concealing
1698      * Error marking covers both fields (top and bottom).
1699      * This causes a mismatched s->error_count
1700      * and a bad error table. Further, the error count goes to
1701      * INT_MAX when called for bottom field, because mb_y is
1702      * past end by one (callers fault) and resync_mb_y != 0
1703      * causes problems for the first MB line, too.
1704      */
1705     if (!FIELD_PICTURE)
1706         ff_er_frame_end(s);
1707
1708     MPV_frame_end(s);
1709
1710     h->current_slice=0;
1711 }
1712
1713 /**
1714  * Replicate H264 "master" context to thread contexts.
1715  */
1716 static void clone_slice(H264Context *dst, H264Context *src)
1717 {
1718     memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
1719     dst->s.current_picture_ptr  = src->s.current_picture_ptr;
1720     dst->s.current_picture      = src->s.current_picture;
1721     dst->s.linesize             = src->s.linesize;
1722     dst->s.uvlinesize           = src->s.uvlinesize;
1723     dst->s.first_field          = src->s.first_field;
1724
1725     dst->prev_poc_msb           = src->prev_poc_msb;
1726     dst->prev_poc_lsb           = src->prev_poc_lsb;
1727     dst->prev_frame_num_offset  = src->prev_frame_num_offset;
1728     dst->prev_frame_num         = src->prev_frame_num;
1729     dst->short_ref_count        = src->short_ref_count;
1730
1731     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
1732     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
1733     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1734     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
1735
1736     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
1737     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
1738 }
1739
1740 /**
1741  * computes profile from profile_idc and constraint_set?_flags
1742  *
1743  * @param sps SPS
1744  *
1745  * @return profile as defined by FF_PROFILE_H264_*
1746  */
1747 int ff_h264_get_profile(SPS *sps)
1748 {
1749     int profile = sps->profile_idc;
1750
1751     switch(sps->profile_idc) {
1752     case FF_PROFILE_H264_BASELINE:
1753         // constraint_set1_flag set to 1
1754         profile |= (sps->constraint_set_flags & 1<<1) ? FF_PROFILE_H264_CONSTRAINED : 0;
1755         break;
1756     case FF_PROFILE_H264_HIGH_10:
1757     case FF_PROFILE_H264_HIGH_422:
1758     case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
1759         // constraint_set3_flag set to 1
1760         profile |= (sps->constraint_set_flags & 1<<3) ? FF_PROFILE_H264_INTRA : 0;
1761         break;
1762     }
1763
1764     return profile;
1765 }
1766
1767 /**
1768  * decodes a slice header.
1769  * This will also call MPV_common_init() and frame_start() as needed.
1770  *
1771  * @param h h264context
1772  * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1773  *
1774  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1775  */
1776 static int decode_slice_header(H264Context *h, H264Context *h0){
1777     MpegEncContext * const s = &h->s;
1778     MpegEncContext * const s0 = &h0->s;
1779     unsigned int first_mb_in_slice;
1780     unsigned int pps_id;
1781     int num_ref_idx_active_override_flag;
1782     unsigned int slice_type, tmp, i, j;
1783     int default_ref_list_done = 0;
1784     int last_pic_structure;
1785
1786     s->dropable= h->nal_ref_idc == 0;
1787
1788     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1789         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1790         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1791     }else{
1792         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1793         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1794     }
1795
1796     first_mb_in_slice= get_ue_golomb(&s->gb);
1797
1798     if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1799         if(h0->current_slice && FIELD_PICTURE){
1800             field_end(h, 1);
1801         }
1802
1803         h0->current_slice = 0;
1804         if (!s0->first_field)
1805             s->current_picture_ptr= NULL;
1806     }
1807
1808     slice_type= get_ue_golomb_31(&s->gb);
1809     if(slice_type > 9){
1810         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);
1811         return -1;
1812     }
1813     if(slice_type > 4){
1814         slice_type -= 5;
1815         h->slice_type_fixed=1;
1816     }else
1817         h->slice_type_fixed=0;
1818
1819     slice_type= golomb_to_pict_type[ slice_type ];
1820     if (slice_type == AV_PICTURE_TYPE_I
1821         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
1822         default_ref_list_done = 1;
1823     }
1824     h->slice_type= slice_type;
1825     h->slice_type_nos= slice_type & 3;
1826
1827     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
1828
1829     pps_id= get_ue_golomb(&s->gb);
1830     if(pps_id>=MAX_PPS_COUNT){
1831         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
1832         return -1;
1833     }
1834     if(!h0->pps_buffers[pps_id]) {
1835         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
1836         return -1;
1837     }
1838     h->pps= *h0->pps_buffers[pps_id];
1839
1840     if(!h0->sps_buffers[h->pps.sps_id]) {
1841         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
1842         return -1;
1843     }
1844     h->sps = *h0->sps_buffers[h->pps.sps_id];
1845
1846     s->avctx->profile = ff_h264_get_profile(&h->sps);
1847     s->avctx->level   = h->sps.level_idc;
1848     s->avctx->refs    = h->sps.ref_frame_count;
1849
1850     if(h == h0 && h->dequant_coeff_pps != pps_id){
1851         h->dequant_coeff_pps = pps_id;
1852         init_dequant_tables(h);
1853     }
1854
1855     s->mb_width= h->sps.mb_width;
1856     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1857
1858     h->b_stride=  s->mb_width*4;
1859
1860     s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
1861     if(h->sps.frame_mbs_only_flag)
1862         s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
1863     else
1864         s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 7);
1865
1866     if (s->context_initialized
1867         && (   s->width != s->avctx->width || s->height != s->avctx->height
1868             || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
1869         if(h != h0) {
1870             av_log_missing_feature(s->avctx, "Width/height changing with threads is", 0);
1871             return -1;   // width / height changed during parallelized decoding
1872         }
1873         free_tables(h, 0);
1874         flush_dpb(s->avctx);
1875         MPV_common_end(s);
1876     }
1877     if (!s->context_initialized) {
1878         if(h != h0){
1879             av_log(h->s.avctx, AV_LOG_ERROR, "we cant (re-)initialize context during parallel decoding\n");
1880             return -1;
1881         }
1882
1883         avcodec_set_dimensions(s->avctx, s->width, s->height);
1884         s->avctx->sample_aspect_ratio= h->sps.sar;
1885         av_assert0(s->avctx->sample_aspect_ratio.den);
1886
1887         if(h->sps.video_signal_type_present_flag){
1888             s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1889             if(h->sps.colour_description_present_flag){
1890                 s->avctx->color_primaries = h->sps.color_primaries;
1891                 s->avctx->color_trc       = h->sps.color_trc;
1892                 s->avctx->colorspace      = h->sps.colorspace;
1893             }
1894         }
1895
1896         if(h->sps.timing_info_present_flag){
1897             int64_t den= h->sps.time_scale;
1898             if(h->x264_build < 44U)
1899                 den *= 2;
1900             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
1901                       h->sps.num_units_in_tick, den, 1<<30);
1902         }
1903
1904         switch (h->sps.bit_depth_luma) {
1905             case 9 :
1906                 s->avctx->pix_fmt = PIX_FMT_YUV420P9;
1907                 break;
1908             case 10 :
1909                 s->avctx->pix_fmt = PIX_FMT_YUV420P10;
1910                 break;
1911             default:
1912         s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
1913                                                  s->avctx->codec->pix_fmts ?
1914                                                  s->avctx->codec->pix_fmts :
1915                                                  s->avctx->color_range == AVCOL_RANGE_JPEG ?
1916                                                  hwaccel_pixfmt_list_h264_jpeg_420 :
1917                                                  ff_hwaccel_pixfmt_list_420);
1918         }
1919
1920         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1921
1922         if (MPV_common_init(s) < 0){
1923             av_log(h->s.avctx, AV_LOG_ERROR, "MPV_common_init() failed\n");
1924             return -1;
1925         }
1926         s->first_field = 0;
1927         h->prev_interlaced_frame = 1;
1928
1929         init_scan_tables(h);
1930         ff_h264_alloc_tables(h);
1931
1932         if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_SLICE)) {
1933             if (context_init(h) < 0){
1934                 av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed\n");
1935                 return -1;
1936             }
1937         } else {
1938             for(i = 1; i < s->avctx->thread_count; i++) {
1939                 H264Context *c;
1940                 c = h->thread_context[i] = av_malloc(sizeof(H264Context));
1941                 memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
1942                 memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
1943                 c->h264dsp = h->h264dsp;
1944                 c->sps = h->sps;
1945                 c->pps = h->pps;
1946                 init_scan_tables(c);
1947                 clone_tables(c, h, i);
1948             }
1949
1950             for(i = 0; i < s->avctx->thread_count; i++)
1951                 if(context_init(h->thread_context[i]) < 0){
1952                     av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed\n");
1953                     return -1;
1954                 }
1955         }
1956     }
1957
1958     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1959
1960     h->mb_mbaff = 0;
1961     h->mb_aff_frame = 0;
1962     last_pic_structure = s0->picture_structure;
1963     if(h->sps.frame_mbs_only_flag){
1964         s->picture_structure= PICT_FRAME;
1965     }else{
1966         if(get_bits1(&s->gb)) { //field_pic_flag
1967             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
1968         } else {
1969             s->picture_structure= PICT_FRAME;
1970             h->mb_aff_frame = h->sps.mb_aff;
1971         }
1972     }
1973     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
1974
1975     if(h0->current_slice == 0){
1976         if(h->frame_num != h->prev_frame_num &&
1977           (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num) < (h->frame_num - h->sps.ref_frame_count))
1978             h->prev_frame_num = h->frame_num - h->sps.ref_frame_count - 1;
1979
1980         while(h->frame_num !=  h->prev_frame_num &&
1981               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
1982             Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1983             av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
1984             if (ff_h264_frame_start(h) < 0)
1985                 return -1;
1986             h->prev_frame_num++;
1987             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
1988             s->current_picture_ptr->frame_num= h->prev_frame_num;
1989             ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 0);
1990             ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 1);
1991             ff_generate_sliding_window_mmcos(h);
1992             ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1993             /* Error concealment: if a ref is missing, copy the previous ref in its place.
1994              * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
1995              * about there being no actual duplicates.
1996              * FIXME: this doesn't copy padding for out-of-frame motion vectors.  Given we're
1997              * concealing a lost frame, this probably isn't noticable by comparison, but it should
1998              * be fixed. */
1999             if (h->short_ref_count) {
2000                 if (prev) {
2001                     av_image_copy(h->short_ref[0]->data, h->short_ref[0]->linesize,
2002                                   (const uint8_t**)prev->data, prev->linesize,
2003                                   s->avctx->pix_fmt, s->mb_width*16, s->mb_height*16);
2004                     h->short_ref[0]->poc = prev->poc+2;
2005                 }
2006                 h->short_ref[0]->frame_num = h->prev_frame_num;
2007             }
2008         }
2009
2010         /* See if we have a decoded first field looking for a pair... */
2011         if (s0->first_field) {
2012             assert(s0->current_picture_ptr);
2013             assert(s0->current_picture_ptr->data[0]);
2014             assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
2015
2016             /* figure out if we have a complementary field pair */
2017             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2018                 /*
2019                  * Previous field is unmatched. Don't display it, but let it
2020                  * remain for reference if marked as such.
2021                  */
2022                 s0->current_picture_ptr = NULL;
2023                 s0->first_field = FIELD_PICTURE;
2024
2025             } else {
2026                 if (h->nal_ref_idc &&
2027                         s0->current_picture_ptr->reference &&
2028                         s0->current_picture_ptr->frame_num != h->frame_num) {
2029                     /*
2030                      * This and previous field were reference, but had
2031                      * different frame_nums. Consider this field first in
2032                      * pair. Throw away previous field except for reference
2033                      * purposes.
2034                      */
2035                     s0->first_field = 1;
2036                     s0->current_picture_ptr = NULL;
2037
2038                 } else {
2039                     /* Second field in complementary pair */
2040                     s0->first_field = 0;
2041                 }
2042             }
2043
2044         } else {
2045             /* Frame or first field in a potentially complementary pair */
2046             assert(!s0->current_picture_ptr);
2047             s0->first_field = FIELD_PICTURE;
2048         }
2049
2050         if(!FIELD_PICTURE || s0->first_field) {
2051             if (ff_h264_frame_start(h) < 0) {
2052                 s0->first_field = 0;
2053                 return -1;
2054             }
2055         } else {
2056             ff_release_unused_pictures(s, 0);
2057         }
2058     }
2059     if(h != h0)
2060         clone_slice(h, h0);
2061
2062     s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
2063
2064     assert(s->mb_num == s->mb_width * s->mb_height);
2065     if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
2066        first_mb_in_slice                    >= s->mb_num){
2067         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
2068         return -1;
2069     }
2070     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
2071     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
2072     if (s->picture_structure == PICT_BOTTOM_FIELD)
2073         s->resync_mb_y = s->mb_y = s->mb_y + 1;
2074     assert(s->mb_y < s->mb_height);
2075
2076     if(s->picture_structure==PICT_FRAME){
2077         h->curr_pic_num=   h->frame_num;
2078         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
2079     }else{
2080         h->curr_pic_num= 2*h->frame_num + 1;
2081         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
2082     }
2083
2084     if(h->nal_unit_type == NAL_IDR_SLICE){
2085         get_ue_golomb(&s->gb); /* idr_pic_id */
2086     }
2087
2088     if(h->sps.poc_type==0){
2089         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2090
2091         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
2092             h->delta_poc_bottom= get_se_golomb(&s->gb);
2093         }
2094     }
2095
2096     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
2097         h->delta_poc[0]= get_se_golomb(&s->gb);
2098
2099         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2100             h->delta_poc[1]= get_se_golomb(&s->gb);
2101     }
2102
2103     init_poc(h);
2104
2105     if(h->pps.redundant_pic_cnt_present){
2106         h->redundant_pic_count= get_ue_golomb(&s->gb);
2107     }
2108
2109     //set defaults, might be overridden a few lines later
2110     h->ref_count[0]= h->pps.ref_count[0];
2111     h->ref_count[1]= h->pps.ref_count[1];
2112
2113     if(h->slice_type_nos != AV_PICTURE_TYPE_I){
2114         if(h->slice_type_nos == AV_PICTURE_TYPE_B){
2115             h->direct_spatial_mv_pred= get_bits1(&s->gb);
2116         }
2117         num_ref_idx_active_override_flag= get_bits1(&s->gb);
2118
2119         if(num_ref_idx_active_override_flag){
2120             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2121             if(h->slice_type_nos==AV_PICTURE_TYPE_B)
2122                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2123
2124             if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
2125                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
2126                 h->ref_count[0]= h->ref_count[1]= 1;
2127                 return -1;
2128             }
2129         }
2130         if(h->slice_type_nos == AV_PICTURE_TYPE_B)
2131             h->list_count= 2;
2132         else
2133             h->list_count= 1;
2134     }else
2135         h->list_count= 0;
2136
2137     if(!default_ref_list_done){
2138         ff_h264_fill_default_ref_list(h);
2139     }
2140
2141     if(h->slice_type_nos!=AV_PICTURE_TYPE_I && ff_h264_decode_ref_pic_list_reordering(h) < 0)
2142         return -1;
2143
2144     if(h->slice_type_nos!=AV_PICTURE_TYPE_I){
2145         s->last_picture_ptr= &h->ref_list[0][0];
2146         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
2147     }
2148     if(h->slice_type_nos==AV_PICTURE_TYPE_B){
2149         s->next_picture_ptr= &h->ref_list[1][0];
2150         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
2151     }
2152
2153     if(   (h->pps.weighted_pred          && h->slice_type_nos == AV_PICTURE_TYPE_P )
2154        ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== AV_PICTURE_TYPE_B ) )
2155         pred_weight_table(h);
2156     else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
2157         implicit_weight_table(h, -1);
2158     }else {
2159         h->use_weight = 0;
2160         for (i = 0; i < 2; i++) {
2161             h->luma_weight_flag[i]   = 0;
2162             h->chroma_weight_flag[i] = 0;
2163         }
2164     }
2165
2166     if(h->nal_ref_idc)
2167         ff_h264_decode_ref_pic_marking(h0, &s->gb);
2168
2169     if(FRAME_MBAFF){
2170         ff_h264_fill_mbaff_ref_list(h);
2171
2172         if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
2173             implicit_weight_table(h, 0);
2174             implicit_weight_table(h, 1);
2175         }
2176     }
2177
2178     if(h->slice_type_nos==AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
2179         ff_h264_direct_dist_scale_factor(h);
2180     ff_h264_direct_ref_list_init(h);
2181
2182     if( h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac ){
2183         tmp = get_ue_golomb_31(&s->gb);
2184         if(tmp > 2){
2185             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2186             return -1;
2187         }
2188         h->cabac_init_idc= tmp;
2189     }
2190
2191     h->last_qscale_diff = 0;
2192     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2193     if(tmp>51+6*(h->sps.bit_depth_luma-8)){
2194         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2195         return -1;
2196     }
2197     s->qscale= tmp;
2198     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2199     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2200     //FIXME qscale / qp ... stuff
2201     if(h->slice_type == AV_PICTURE_TYPE_SP){
2202         get_bits1(&s->gb); /* sp_for_switch_flag */
2203     }
2204     if(h->slice_type==AV_PICTURE_TYPE_SP || h->slice_type == AV_PICTURE_TYPE_SI){
2205         get_se_golomb(&s->gb); /* slice_qs_delta */
2206     }
2207
2208     h->deblocking_filter = 1;
2209     h->slice_alpha_c0_offset = 52;
2210     h->slice_beta_offset = 52;
2211     if( h->pps.deblocking_filter_parameters_present ) {
2212         tmp= get_ue_golomb_31(&s->gb);
2213         if(tmp > 2){
2214             av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2215             return -1;
2216         }
2217         h->deblocking_filter= tmp;
2218         if(h->deblocking_filter < 2)
2219             h->deblocking_filter^= 1; // 1<->0
2220
2221         if( h->deblocking_filter ) {
2222             h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2223             h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
2224             if(   h->slice_alpha_c0_offset > 104U
2225                || h->slice_beta_offset     > 104U){
2226                 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);
2227                 return -1;
2228             }
2229         }
2230     }
2231
2232     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
2233        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != AV_PICTURE_TYPE_I)
2234        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == AV_PICTURE_TYPE_B)
2235        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2236         h->deblocking_filter= 0;
2237
2238     if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2239         if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2240             /* Cheat slightly for speed:
2241                Do not bother to deblock across slices. */
2242             h->deblocking_filter = 2;
2243         } else {
2244             h0->max_contexts = 1;
2245             if(!h0->single_decode_warning) {
2246                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2247                 h0->single_decode_warning = 1;
2248             }
2249             if(h != h0){
2250                 av_log(h->s.avctx, AV_LOG_ERROR, "deblocking switched inside frame\n");
2251                 return 1;
2252             }
2253         }
2254     }
2255     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]);
2256
2257 #if 0 //FMO
2258     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2259         slice_group_change_cycle= get_bits(&s->gb, ?);
2260 #endif
2261
2262     h0->last_slice_type = slice_type;
2263     h->slice_num = ++h0->current_slice;
2264     if(h->slice_num >= MAX_SLICES){
2265         av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2266     }
2267
2268     for(j=0; j<2; j++){
2269         int id_list[16];
2270         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2271         for(i=0; i<16; i++){
2272             id_list[i]= 60;
2273             if(h->ref_list[j][i].data[0]){
2274                 int k;
2275                 uint8_t *base= h->ref_list[j][i].base[0];
2276                 for(k=0; k<h->short_ref_count; k++)
2277                     if(h->short_ref[k]->base[0] == base){
2278                         id_list[i]= k;
2279                         break;
2280                     }
2281                 for(k=0; k<h->long_ref_count; k++)
2282                     if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2283                         id_list[i]= h->short_ref_count + k;
2284                         break;
2285                     }
2286             }
2287         }
2288
2289         ref2frm[0]=
2290         ref2frm[1]= -1;
2291         for(i=0; i<16; i++)
2292             ref2frm[i+2]= 4*id_list[i]
2293                           +(h->ref_list[j][i].reference&3);
2294         ref2frm[18+0]=
2295         ref2frm[18+1]= -1;
2296         for(i=16; i<48; i++)
2297             ref2frm[i+4]= 4*id_list[(i-16)>>1]
2298                           +(h->ref_list[j][i].reference&3);
2299     }
2300
2301     //FIXME: fix draw_edges+PAFF+frame threads
2302     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE || (!h->sps.frame_mbs_only_flag && s->avctx->active_thread_type&FF_THREAD_FRAME)) ? 0 : 16;
2303     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2304
2305     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2306         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",
2307                h->slice_num,
2308                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2309                first_mb_in_slice,
2310                av_get_picture_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2311                pps_id, h->frame_num,
2312                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2313                h->ref_count[0], h->ref_count[1],
2314                s->qscale,
2315                h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
2316                h->use_weight,
2317                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2318                h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2319                );
2320     }
2321
2322     return 0;
2323 }
2324
2325 int ff_h264_get_slice_type(const H264Context *h)
2326 {
2327     switch (h->slice_type) {
2328     case AV_PICTURE_TYPE_P:  return 0;
2329     case AV_PICTURE_TYPE_B:  return 1;
2330     case AV_PICTURE_TYPE_I:  return 2;
2331     case AV_PICTURE_TYPE_SP: return 3;
2332     case AV_PICTURE_TYPE_SI: return 4;
2333     default:         return -1;
2334     }
2335 }
2336
2337 /**
2338  *
2339  * @return non zero if the loop filter can be skiped
2340  */
2341 static int fill_filter_caches(H264Context *h, int mb_type){
2342     MpegEncContext * const s = &h->s;
2343     const int mb_xy= h->mb_xy;
2344     int top_xy, left_xy[2];
2345     int top_type, left_type[2];
2346
2347     top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
2348
2349     //FIXME deblocking could skip the intra and nnz parts.
2350
2351     /* Wow, what a mess, why didn't they simplify the interlacing & intra
2352      * stuff, I can't imagine that these complex rules are worth it. */
2353
2354     left_xy[1] = left_xy[0] = mb_xy-1;
2355     if(FRAME_MBAFF){
2356         const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
2357         const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
2358         if(s->mb_y&1){
2359             if (left_mb_field_flag != curr_mb_field_flag) {
2360                 left_xy[0] -= s->mb_stride;
2361             }
2362         }else{
2363             if(curr_mb_field_flag){
2364                 top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
2365             }
2366             if (left_mb_field_flag != curr_mb_field_flag) {
2367                 left_xy[1] += s->mb_stride;
2368             }
2369         }
2370     }
2371
2372     h->top_mb_xy = top_xy;
2373     h->left_mb_xy[0] = left_xy[0];
2374     h->left_mb_xy[1] = left_xy[1];
2375     {
2376         //for sufficiently low qp, filtering wouldn't do anything
2377         //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
2378         int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
2379         int qp = s->current_picture.qscale_table[mb_xy];
2380         if(qp <= qp_thresh
2381            && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
2382            && (top_xy   < 0 || ((qp + s->current_picture.qscale_table[top_xy    ] + 1)>>1) <= qp_thresh)){
2383             if(!FRAME_MBAFF)
2384                 return 1;
2385             if(   (left_xy[0]< 0            || ((qp + s->current_picture.qscale_table[left_xy[1]             ] + 1)>>1) <= qp_thresh)
2386                && (top_xy    < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy    -s->mb_stride] + 1)>>1) <= qp_thresh))
2387                 return 1;
2388         }
2389     }
2390
2391     top_type     = s->current_picture.mb_type[top_xy]    ;
2392     left_type[0] = s->current_picture.mb_type[left_xy[0]];
2393     left_type[1] = s->current_picture.mb_type[left_xy[1]];
2394     if(h->deblocking_filter == 2){
2395         if(h->slice_table[top_xy     ] != h->slice_num) top_type= 0;
2396         if(h->slice_table[left_xy[0] ] != h->slice_num) left_type[0]= left_type[1]= 0;
2397     }else{
2398         if(h->slice_table[top_xy     ] == 0xFFFF) top_type= 0;
2399         if(h->slice_table[left_xy[0] ] == 0xFFFF) left_type[0]= left_type[1] =0;
2400     }
2401     h->top_type    = top_type    ;
2402     h->left_type[0]= left_type[0];
2403     h->left_type[1]= left_type[1];
2404
2405     if(IS_INTRA(mb_type))
2406         return 0;
2407
2408     AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
2409     AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
2410     AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]);
2411     AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]);
2412     AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
2413
2414     h->cbp= h->cbp_table[mb_xy];
2415
2416     {
2417         int list;
2418         for(list=0; list<h->list_count; list++){
2419             int8_t *ref;
2420             int y, b_stride;
2421             int16_t (*mv_dst)[2];
2422             int16_t (*mv_src)[2];
2423
2424             if(!USES_LIST(mb_type, list)){
2425                 fill_rectangle(  h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
2426                 AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2427                 AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2428                 AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2429                 AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2430                 continue;
2431             }
2432
2433             ref = &s->current_picture.ref_index[list][4*mb_xy];
2434             {
2435                 int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2436                 AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2437                 AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2438                 ref += 2;
2439                 AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2440                 AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2441             }
2442
2443             b_stride = h->b_stride;
2444             mv_dst   = &h->mv_cache[list][scan8[0]];
2445             mv_src   = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
2446             for(y=0; y<4; y++){
2447                 AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
2448             }
2449
2450         }
2451     }
2452
2453
2454 /*
2455 0 . T T. T T T T
2456 1 L . .L . . . .
2457 2 L . .L . . . .
2458 3 . T TL . . . .
2459 4 L . .L . . . .
2460 5 L . .. . . . .
2461 */
2462 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
2463     if(top_type){
2464         AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
2465     }
2466
2467     if(left_type[0]){
2468         h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
2469         h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
2470         h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
2471         h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
2472     }
2473
2474     // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
2475     if(!CABAC && h->pps.transform_8x8_mode){
2476         if(IS_8x8DCT(top_type)){
2477             h->non_zero_count_cache[4+8*0]=
2478             h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
2479             h->non_zero_count_cache[6+8*0]=
2480             h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
2481         }
2482         if(IS_8x8DCT(left_type[0])){
2483             h->non_zero_count_cache[3+8*1]=
2484             h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
2485         }
2486         if(IS_8x8DCT(left_type[1])){
2487             h->non_zero_count_cache[3+8*3]=
2488             h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
2489         }
2490
2491         if(IS_8x8DCT(mb_type)){
2492             h->non_zero_count_cache[scan8[0   ]]= h->non_zero_count_cache[scan8[1   ]]=
2493             h->non_zero_count_cache[scan8[2   ]]= h->non_zero_count_cache[scan8[3   ]]= h->cbp & 1;
2494
2495             h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
2496             h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
2497
2498             h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
2499             h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
2500
2501             h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
2502             h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
2503         }
2504     }
2505
2506     if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
2507         int list;
2508         for(list=0; list<h->list_count; list++){
2509             if(USES_LIST(top_type, list)){
2510                 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
2511                 const int b8_xy= 4*top_xy + 2;
2512                 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2513                 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
2514                 h->ref_cache[list][scan8[0] + 0 - 1*8]=
2515                 h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
2516                 h->ref_cache[list][scan8[0] + 2 - 1*8]=
2517                 h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
2518             }else{
2519                 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
2520                 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2521             }
2522
2523             if(!IS_INTERLACED(mb_type^left_type[0])){
2524                 if(USES_LIST(left_type[0], list)){
2525                     const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
2526                     const int b8_xy= 4*left_xy[0] + 1;
2527                     int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2528                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
2529                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
2530                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
2531                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
2532                     h->ref_cache[list][scan8[0] - 1 + 0 ]=
2533                     h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]];
2534                     h->ref_cache[list][scan8[0] - 1 +16 ]=
2535                     h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]];
2536                 }else{
2537                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
2538                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
2539                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
2540                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
2541                     h->ref_cache[list][scan8[0] - 1 + 0  ]=
2542                     h->ref_cache[list][scan8[0] - 1 + 8  ]=
2543                     h->ref_cache[list][scan8[0] - 1 + 16 ]=
2544                     h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
2545                 }
2546             }
2547         }
2548     }
2549
2550     return 0;
2551 }
2552
2553 static void loop_filter(H264Context *h){
2554     MpegEncContext * const s = &h->s;
2555     uint8_t  *dest_y, *dest_cb, *dest_cr;
2556     int linesize, uvlinesize, mb_x, mb_y;
2557     const int end_mb_y= s->mb_y + FRAME_MBAFF;
2558     const int old_slice_type= h->slice_type;
2559     const int end_mb_x  = s->mb_x;
2560
2561     if(h->deblocking_filter) {
2562         int start_x= s->resync_mb_y == s->mb_y ? s->resync_mb_x : 0;
2563         for(mb_x= start_x; mb_x<end_mb_x; mb_x++){
2564             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
2565                 int mb_xy, mb_type;
2566                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2567                 h->slice_num= h->slice_table[mb_xy];
2568                 mb_type= s->current_picture.mb_type[mb_xy];
2569                 h->list_count= h->list_counts[mb_xy];
2570
2571                 if(FRAME_MBAFF)
2572                     h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2573
2574                 s->mb_x= mb_x;
2575                 s->mb_y= mb_y;
2576                 dest_y  = s->current_picture.data[0] + ((mb_x<<h->pixel_shift) + mb_y * s->linesize  ) * 16;
2577                 dest_cb = s->current_picture.data[1] + ((mb_x<<h->pixel_shift) + mb_y * s->uvlinesize) * 8;
2578                 dest_cr = s->current_picture.data[2] + ((mb_x<<h->pixel_shift) + mb_y * s->uvlinesize) * 8;
2579                     //FIXME simplify above
2580
2581                 if (MB_FIELD) {
2582                     linesize   = h->mb_linesize   = s->linesize * 2;
2583                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2584                     if(mb_y&1){ //FIXME move out of this function?
2585                         dest_y -= s->linesize*15;
2586                         dest_cb-= s->uvlinesize*7;
2587                         dest_cr-= s->uvlinesize*7;
2588                     }
2589                 } else {
2590                     linesize   = h->mb_linesize   = s->linesize;
2591                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2592                 }
2593                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2594                 if(fill_filter_caches(h, mb_type))
2595                     continue;
2596                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2597                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2598
2599                 if (FRAME_MBAFF) {
2600                     ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2601                 } else {
2602                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2603                 }
2604             }
2605         }
2606     }
2607     h->slice_type= old_slice_type;
2608     s->mb_x= end_mb_x;
2609     s->mb_y= end_mb_y - FRAME_MBAFF;
2610     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2611     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2612 }
2613
2614 static void predict_field_decoding_flag(H264Context *h){
2615     MpegEncContext * const s = &h->s;
2616     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2617     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
2618                 ? s->current_picture.mb_type[mb_xy-1]
2619                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
2620                 ? s->current_picture.mb_type[mb_xy-s->mb_stride]
2621                 : 0;
2622     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2623 }
2624
2625 /**
2626  * Draw edges and report progress for the last MB row.
2627  */
2628 static void decode_finish_row(H264Context *h){
2629     MpegEncContext * const s = &h->s;
2630     int top = 16*(s->mb_y >> FIELD_PICTURE);
2631     int height = 16 << FRAME_MBAFF;
2632     int deblock_border = (16 + 4) << FRAME_MBAFF;
2633     int pic_height = 16*s->mb_height >> FIELD_PICTURE;
2634
2635     if (h->deblocking_filter) {
2636         if((top + height) >= pic_height)
2637             height += deblock_border;
2638
2639         top -= deblock_border;
2640     }
2641
2642     if (top >= pic_height || (top + height) < h->emu_edge_height)
2643         return;
2644
2645     height = FFMIN(height, pic_height - top);
2646     if (top < h->emu_edge_height) {
2647         height = top+height;
2648         top = 0;
2649     }
2650
2651     ff_draw_horiz_band(s, top, height);
2652
2653     if (s->dropable) return;
2654
2655     ff_thread_report_progress((AVFrame*)s->current_picture_ptr, top + height - 1,
2656                              s->picture_structure==PICT_BOTTOM_FIELD);
2657 }
2658
2659 static int decode_slice(struct AVCodecContext *avctx, void *arg){
2660     H264Context *h = *(void**)arg;
2661     MpegEncContext * const s = &h->s;
2662     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2663
2664     s->mb_skip_run= -1;
2665
2666     h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
2667                     (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
2668
2669     if( h->pps.cabac ) {
2670         /* realign */
2671         align_get_bits( &s->gb );
2672
2673         /* init cabac */
2674         ff_init_cabac_states( &h->cabac);
2675         ff_init_cabac_decoder( &h->cabac,
2676                                s->gb.buffer + get_bits_count(&s->gb)/8,
2677                                (get_bits_left(&s->gb) + 7)/8);
2678
2679         ff_h264_init_cabac_states(h);
2680
2681         for(;;){
2682 //START_TIMER
2683             int ret = ff_h264_decode_mb_cabac(h);
2684             int eos;
2685 //STOP_TIMER("decode_mb_cabac")
2686
2687             if(ret>=0) ff_h264_hl_decode_mb(h);
2688
2689             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
2690                 s->mb_y++;
2691
2692                 ret = ff_h264_decode_mb_cabac(h);
2693
2694                 if(ret>=0) ff_h264_hl_decode_mb(h);
2695                 s->mb_y--;
2696             }
2697             eos = get_cabac_terminate( &h->cabac );
2698
2699             if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
2700                 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);
2701                 return 0;
2702             }
2703             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2704                 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);
2705                 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);
2706                 return -1;
2707             }
2708
2709             if( ++s->mb_x >= s->mb_width ) {
2710                 loop_filter(h);
2711                 s->mb_x = 0;
2712                 decode_finish_row(h);
2713                 ++s->mb_y;
2714                 if(FIELD_OR_MBAFF_PICTURE) {
2715                     ++s->mb_y;
2716                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2717                         predict_field_decoding_flag(h);
2718                 }
2719             }
2720
2721             if( eos || s->mb_y >= s->mb_height ) {
2722                 if(s->mb_x)
2723                     loop_filter(h);
2724                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2725                 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);
2726                 return 0;
2727             }
2728         }
2729
2730     } else {
2731         for(;;){
2732             int ret = ff_h264_decode_mb_cavlc(h);
2733
2734             if(ret>=0) ff_h264_hl_decode_mb(h);
2735
2736             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
2737                 s->mb_y++;
2738                 ret = ff_h264_decode_mb_cavlc(h);
2739
2740                 if(ret>=0) ff_h264_hl_decode_mb(h);
2741                 s->mb_y--;
2742             }
2743
2744             if(ret<0){
2745                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2746                 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);
2747
2748                 return -1;
2749             }
2750
2751             if(++s->mb_x >= s->mb_width){
2752                 loop_filter(h);
2753                 s->mb_x=0;
2754                 decode_finish_row(h);
2755                 ++s->mb_y;
2756                 if(FIELD_OR_MBAFF_PICTURE) {
2757                     ++s->mb_y;
2758                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
2759                         predict_field_decoding_flag(h);
2760                 }
2761                 if(s->mb_y >= s->mb_height){
2762                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2763
2764                     if(   get_bits_count(&s->gb) == s->gb.size_in_bits
2765                        || get_bits_count(&s->gb) <  s->gb.size_in_bits && s->avctx->error_recognition < FF_ER_AGGRESSIVE) {
2766                         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);
2767
2768                         return 0;
2769                     }else{
2770                         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);
2771
2772                         return -1;
2773                     }
2774                 }
2775             }
2776
2777             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
2778                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2779                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2780                     if(s->mb_x)
2781                         loop_filter(h);
2782                     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);
2783
2784                     return 0;
2785                 }else{
2786                     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);
2787
2788                     return -1;
2789                 }
2790             }
2791         }
2792     }
2793
2794 #if 0
2795     for(;s->mb_y < s->mb_height; s->mb_y++){
2796         for(;s->mb_x < s->mb_width; s->mb_x++){
2797             int ret= decode_mb(h);
2798
2799             ff_h264_hl_decode_mb(h);
2800
2801             if(ret<0){
2802                 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2803                 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);
2804
2805                 return -1;
2806             }
2807
2808             if(++s->mb_x >= s->mb_width){
2809                 s->mb_x=0;
2810                 if(++s->mb_y >= s->mb_height){
2811                     if(get_bits_count(s->gb) == s->gb.size_in_bits){
2812                         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);
2813
2814                         return 0;
2815                     }else{
2816                         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);
2817
2818                         return -1;
2819                     }
2820                 }
2821             }
2822
2823             if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2824                 if(get_bits_count(s->gb) == s->gb.size_in_bits){
2825                     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);
2826
2827                     return 0;
2828                 }else{
2829                     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);
2830
2831                     return -1;
2832                 }
2833             }
2834         }
2835         s->mb_x=0;
2836         ff_draw_horiz_band(s, 16*s->mb_y, 16);
2837     }
2838 #endif
2839     return -1; //not reached
2840 }
2841
2842 /**
2843  * Call decode_slice() for each context.
2844  *
2845  * @param h h264 master context
2846  * @param context_count number of contexts to execute
2847  */
2848 static void execute_decode_slices(H264Context *h, int context_count){
2849     MpegEncContext * const s = &h->s;
2850     AVCodecContext * const avctx= s->avctx;
2851     H264Context *hx;
2852     int i;
2853
2854     if (s->avctx->hwaccel)
2855         return;
2856     if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2857         return;
2858     if(context_count == 1) {
2859         decode_slice(avctx, &h);
2860     } else {
2861         for(i = 1; i < context_count; i++) {
2862             hx = h->thread_context[i];
2863             hx->s.error_recognition = avctx->error_recognition;
2864             hx->s.error_count = 0;
2865             hx->x264_build= h->x264_build;
2866         }
2867
2868         avctx->execute(avctx, (void *)decode_slice,
2869                        h->thread_context, NULL, context_count, sizeof(void*));
2870
2871         /* pull back stuff from slices to master context */
2872         hx = h->thread_context[context_count - 1];
2873         s->mb_x = hx->s.mb_x;
2874         s->mb_y = hx->s.mb_y;
2875         s->dropable = hx->s.dropable;
2876         s->picture_structure = hx->s.picture_structure;
2877         for(i = 1; i < context_count; i++)
2878             h->s.error_count += h->thread_context[i]->s.error_count;
2879     }
2880 }
2881
2882
2883 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
2884     MpegEncContext * const s = &h->s;
2885     AVCodecContext * const avctx= s->avctx;
2886     int buf_index=0;
2887     H264Context *hx; ///< thread context
2888     int context_count = 0;
2889     int next_avc= h->is_avc ? 0 : buf_size;
2890
2891     h->max_contexts = (HAVE_THREADS && (s->avctx->active_thread_type&FF_THREAD_SLICE)) ? avctx->thread_count : 1;
2892 #if 0
2893     int i;
2894     for(i=0; i<50; i++){
2895         av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
2896     }
2897 #endif
2898     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
2899         h->current_slice = 0;
2900         if (!s->first_field)
2901             s->current_picture_ptr= NULL;
2902         ff_h264_reset_sei(h);
2903     }
2904
2905     for(;;){
2906         int consumed;
2907         int dst_length;
2908         int bit_length;
2909         const uint8_t *ptr;
2910         int i, nalsize = 0;
2911         int err;
2912
2913         if(buf_index >= next_avc) {
2914             if(buf_index >= buf_size) break;
2915             nalsize = 0;
2916             for(i = 0; i < h->nal_length_size; i++)
2917                 nalsize = (nalsize << 8) | buf[buf_index++];
2918             if(nalsize <= 0 || nalsize > buf_size - buf_index){
2919                 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2920                 break;
2921             }
2922             next_avc= buf_index + nalsize;
2923         } else {
2924             // start code prefix search
2925             for(; buf_index + 3 < next_avc; buf_index++){
2926                 // This should always succeed in the first iteration.
2927                 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2928                     break;
2929             }
2930
2931             if(buf_index+3 >= buf_size) break;
2932
2933             buf_index+=3;
2934             if(buf_index >= next_avc) continue;
2935         }
2936
2937         hx = h->thread_context[context_count];
2938
2939         ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
2940         if (ptr==NULL || dst_length < 0){
2941             return -1;
2942         }
2943         i= buf_index + consumed;
2944         if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
2945            buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
2946             s->workaround_bugs |= FF_BUG_TRUNCATED;
2947
2948         if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
2949         while(ptr[dst_length - 1] == 0 && dst_length > 0)
2950             dst_length--;
2951         }
2952         bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
2953
2954         if(s->avctx->debug&FF_DEBUG_STARTCODE){
2955             av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d/%d at %d/%d length %d\n", hx->nal_unit_type, hx->nal_ref_idc, buf_index, buf_size, dst_length);
2956         }
2957
2958         if (h->is_avc && (nalsize != consumed) && nalsize){
2959             av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
2960         }
2961
2962         buf_index += consumed;
2963
2964         //FIXME do not discard SEI id
2965         if(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0)
2966             continue;
2967
2968       again:
2969         err = 0;
2970         switch(hx->nal_unit_type){
2971         case NAL_IDR_SLICE:
2972             if (h->nal_unit_type != NAL_IDR_SLICE) {
2973                 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2974                 return -1;
2975             }
2976             idr(h); //FIXME ensure we don't loose some frames if there is reordering
2977         case NAL_SLICE:
2978             init_get_bits(&hx->s.gb, ptr, bit_length);
2979             hx->intra_gb_ptr=
2980             hx->inter_gb_ptr= &hx->s.gb;
2981             hx->s.data_partitioning = 0;
2982
2983             if((err = decode_slice_header(hx, h)))
2984                break;
2985
2986             s->current_picture_ptr->key_frame |=
2987                     (hx->nal_unit_type == NAL_IDR_SLICE) ||
2988                     (h->sei_recovery_frame_cnt >= 0);
2989
2990             if (h->current_slice == 1) {
2991                 if(!(s->flags2 & CODEC_FLAG2_CHUNKS)) {
2992                     decode_postinit(h);
2993                 }
2994
2995                 if (s->avctx->hwaccel && s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
2996                     return -1;
2997                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2998                     ff_vdpau_h264_picture_start(s);
2999             }
3000
3001             if(hx->redundant_pic_count==0
3002                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
3003                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
3004                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
3005                && avctx->skip_frame < AVDISCARD_ALL){
3006                 if(avctx->hwaccel) {
3007                     if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
3008                         return -1;
3009                 }else
3010                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
3011                     static const uint8_t start_code[] = {0x00, 0x00, 0x01};
3012                     ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
3013                     ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
3014                 }else
3015                     context_count++;
3016             }
3017             break;
3018         case NAL_DPA:
3019             init_get_bits(&hx->s.gb, ptr, bit_length);
3020             hx->intra_gb_ptr=
3021             hx->inter_gb_ptr= NULL;
3022
3023             if ((err = decode_slice_header(hx, h)) < 0)
3024                 break;
3025
3026             hx->s.data_partitioning = 1;
3027
3028             break;
3029         case NAL_DPB:
3030             init_get_bits(&hx->intra_gb, ptr, bit_length);
3031             hx->intra_gb_ptr= &hx->intra_gb;
3032             break;
3033         case NAL_DPC:
3034             init_get_bits(&hx->inter_gb, ptr, bit_length);
3035             hx->inter_gb_ptr= &hx->inter_gb;
3036
3037             if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
3038                && s->context_initialized
3039                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
3040                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
3041                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
3042                && avctx->skip_frame < AVDISCARD_ALL)
3043                 context_count++;
3044             break;
3045         case NAL_SEI:
3046             init_get_bits(&s->gb, ptr, bit_length);
3047             ff_h264_decode_sei(h);
3048             break;
3049         case NAL_SPS:
3050             init_get_bits(&s->gb, ptr, bit_length);
3051             ff_h264_decode_seq_parameter_set(h);
3052
3053             if(s->flags& CODEC_FLAG_LOW_DELAY ||
3054               (h->sps.bitstream_restriction_flag && !h->sps.num_reorder_frames))
3055                 s->low_delay=1;
3056
3057             if(avctx->has_b_frames < 2)
3058                 avctx->has_b_frames= !s->low_delay;
3059
3060             if (avctx->bits_per_raw_sample != h->sps.bit_depth_luma) {
3061                 if (h->sps.bit_depth_luma >= 8 && h->sps.bit_depth_luma <= 10) {
3062                     avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
3063                     h->pixel_shift = h->sps.bit_depth_luma/9;
3064
3065                     ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma);
3066                     ff_h264_pred_init(&h->hpc, s->codec_id, h->sps.bit_depth_luma);
3067                     dsputil_init(&s->dsp, s->avctx);
3068                 } else {
3069                     av_log(avctx, AV_LOG_DEBUG, "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
3070                     return -1;
3071                 }
3072             }
3073             break;
3074         case NAL_PPS:
3075             init_get_bits(&s->gb, ptr, bit_length);
3076
3077             ff_h264_decode_picture_parameter_set(h, bit_length);
3078
3079             break;
3080         case NAL_AUD:
3081         case NAL_END_SEQUENCE:
3082         case NAL_END_STREAM:
3083         case NAL_FILLER_DATA:
3084         case NAL_SPS_EXT:
3085         case NAL_AUXILIARY_SLICE:
3086             break;
3087         default:
3088             av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
3089         }
3090
3091         if(context_count == h->max_contexts) {
3092             execute_decode_slices(h, context_count);
3093             context_count = 0;
3094         }
3095
3096         if (err < 0)
3097             av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
3098         else if(err == 1) {
3099             /* Slice could not be decoded in parallel mode, copy down
3100              * NAL unit stuff to context 0 and restart. Note that
3101              * rbsp_buffer is not transferred, but since we no longer
3102              * run in parallel mode this should not be an issue. */
3103             h->nal_unit_type = hx->nal_unit_type;
3104             h->nal_ref_idc   = hx->nal_ref_idc;
3105             hx = h;
3106             goto again;
3107         }
3108     }
3109     if(context_count)
3110         execute_decode_slices(h, context_count);
3111     return buf_index;
3112 }
3113
3114 /**
3115  * returns the number of bytes consumed for building the current frame
3116  */
3117 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
3118         if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
3119         if(pos+10>buf_size) pos=buf_size; // oops ;)
3120
3121         return pos;
3122 }
3123
3124 static int decode_frame(AVCodecContext *avctx,
3125                              void *data, int *data_size,
3126                              AVPacket *avpkt)
3127 {
3128     const uint8_t *buf = avpkt->data;
3129     int buf_size = avpkt->size;
3130     H264Context *h = avctx->priv_data;
3131     MpegEncContext *s = &h->s;
3132     AVFrame *pict = data;
3133     int buf_index;
3134
3135     s->flags= avctx->flags;
3136     s->flags2= avctx->flags2;
3137
3138    /* end of stream, output what is still in the buffers */
3139  out:
3140     if (buf_size == 0) {
3141         Picture *out;
3142         int i, out_idx;
3143
3144         s->current_picture_ptr = NULL;
3145
3146 //FIXME factorize this with the output code below
3147         out = h->delayed_pic[0];
3148         out_idx = 0;
3149         for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
3150             if(h->delayed_pic[i]->poc < out->poc){
3151                 out = h->delayed_pic[i];
3152                 out_idx = i;
3153             }
3154
3155         for(i=out_idx; h->delayed_pic[i]; i++)
3156             h->delayed_pic[i] = h->delayed_pic[i+1];
3157
3158         if(out){
3159             *data_size = sizeof(AVFrame);
3160             *pict= *(AVFrame*)out;
3161         }
3162
3163         return 0;
3164     }
3165
3166     buf_index=decode_nal_units(h, buf, buf_size);
3167     if(buf_index < 0)
3168         return -1;
3169
3170     if (!s->current_picture_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
3171         buf_size = 0;
3172         goto out;
3173     }
3174
3175     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
3176         if (avctx->skip_frame >= AVDISCARD_NONREF)
3177             return 0;
3178         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
3179         return -1;
3180     }
3181
3182     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
3183
3184         if(s->flags2 & CODEC_FLAG2_CHUNKS) decode_postinit(h);
3185
3186         field_end(h, 0);
3187
3188         if (!h->next_output_pic) {
3189             /* Wait for second field. */
3190             *data_size = 0;
3191
3192         } else {
3193             *data_size = sizeof(AVFrame);
3194             *pict = *(AVFrame*)h->next_output_pic;
3195         }
3196     }
3197
3198     assert(pict->data[0] || !*data_size);
3199     ff_print_debug_info(s, pict);
3200 //printf("out %d\n", (int)pict->data[0]);
3201
3202     return get_consumed_bytes(s, buf_index, buf_size);
3203 }
3204 #if 0
3205 static inline void fill_mb_avail(H264Context *h){
3206     MpegEncContext * const s = &h->s;
3207     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3208
3209     if(s->mb_y){
3210         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
3211         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
3212         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
3213     }else{
3214         h->mb_avail[0]=
3215         h->mb_avail[1]=
3216         h->mb_avail[2]= 0;
3217     }
3218     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
3219     h->mb_avail[4]= 1; //FIXME move out
3220     h->mb_avail[5]= 0; //FIXME move out
3221 }
3222 #endif
3223
3224 #ifdef TEST
3225 #undef printf
3226 #undef random
3227 #define COUNT 8000
3228 #define SIZE (COUNT*40)
3229 int main(void){
3230     int i;
3231     uint8_t temp[SIZE];
3232     PutBitContext pb;
3233     GetBitContext gb;
3234 //    int int_temp[10000];
3235     DSPContext dsp;
3236     AVCodecContext avctx;
3237
3238     dsputil_init(&dsp, &avctx);
3239
3240     init_put_bits(&pb, temp, SIZE);
3241     printf("testing unsigned exp golomb\n");
3242     for(i=0; i<COUNT; i++){
3243         START_TIMER
3244         set_ue_golomb(&pb, i);
3245         STOP_TIMER("set_ue_golomb");
3246     }
3247     flush_put_bits(&pb);
3248
3249     init_get_bits(&gb, temp, 8*SIZE);
3250     for(i=0; i<COUNT; i++){
3251         int j, s;
3252
3253         s= show_bits(&gb, 24);
3254
3255         START_TIMER
3256         j= get_ue_golomb(&gb);
3257         if(j != i){
3258             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
3259 //            return -1;
3260         }
3261         STOP_TIMER("get_ue_golomb");
3262     }
3263
3264
3265     init_put_bits(&pb, temp, SIZE);
3266     printf("testing signed exp golomb\n");
3267     for(i=0; i<COUNT; i++){
3268         START_TIMER
3269         set_se_golomb(&pb, i - COUNT/2);
3270         STOP_TIMER("set_se_golomb");
3271     }
3272     flush_put_bits(&pb);
3273
3274     init_get_bits(&gb, temp, 8*SIZE);
3275     for(i=0; i<COUNT; i++){
3276         int j, s;
3277
3278         s= show_bits(&gb, 24);
3279
3280         START_TIMER
3281         j= get_se_golomb(&gb);
3282         if(j != i - COUNT/2){
3283             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
3284 //            return -1;
3285         }
3286         STOP_TIMER("get_se_golomb");
3287     }
3288
3289 #if 0
3290     printf("testing 4x4 (I)DCT\n");
3291
3292     DCTELEM block[16];
3293     uint8_t src[16], ref[16];
3294     uint64_t error= 0, max_error=0;
3295
3296     for(i=0; i<COUNT; i++){
3297         int j;
3298 //        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
3299         for(j=0; j<16; j++){
3300             ref[j]= random()%255;
3301             src[j]= random()%255;
3302         }
3303
3304         h264_diff_dct_c(block, src, ref, 4);
3305
3306         //normalize
3307         for(j=0; j<16; j++){
3308 //            printf("%d ", block[j]);
3309             block[j]= block[j]*4;
3310             if(j&1) block[j]= (block[j]*4 + 2)/5;
3311             if(j&4) block[j]= (block[j]*4 + 2)/5;
3312         }
3313 //        printf("\n");
3314
3315         h->h264dsp.h264_idct_add(ref, block, 4);
3316 /*        for(j=0; j<16; j++){
3317             printf("%d ", ref[j]);
3318         }
3319         printf("\n");*/
3320
3321         for(j=0; j<16; j++){
3322             int diff= FFABS(src[j] - ref[j]);
3323
3324             error+= diff*diff;
3325             max_error= FFMAX(max_error, diff);
3326         }
3327     }
3328     printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
3329     printf("testing quantizer\n");
3330     for(qp=0; qp<52; qp++){
3331         for(i=0; i<16; i++)
3332             src1_block[i]= src2_block[i]= random()%255;
3333
3334     }
3335     printf("Testing NAL layer\n");
3336
3337     uint8_t bitstream[COUNT];
3338     uint8_t nal[COUNT*2];
3339     H264Context h;
3340     memset(&h, 0, sizeof(H264Context));
3341
3342     for(i=0; i<COUNT; i++){
3343         int zeros= i;
3344         int nal_length;
3345         int consumed;
3346         int out_length;
3347         uint8_t *out;
3348         int j;
3349
3350         for(j=0; j<COUNT; j++){
3351             bitstream[j]= (random() % 255) + 1;
3352         }
3353
3354         for(j=0; j<zeros; j++){
3355             int pos= random() % COUNT;
3356             while(bitstream[pos] == 0){
3357                 pos++;
3358                 pos %= COUNT;
3359             }
3360             bitstream[pos]=0;
3361         }
3362
3363         START_TIMER
3364
3365         nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
3366         if(nal_length<0){
3367             printf("encoding failed\n");
3368             return -1;
3369         }
3370
3371         out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
3372
3373         STOP_TIMER("NAL")
3374
3375         if(out_length != COUNT){
3376             printf("incorrect length %d %d\n", out_length, COUNT);
3377             return -1;
3378         }
3379
3380         if(consumed != nal_length){
3381             printf("incorrect consumed length %d %d\n", nal_length, consumed);
3382             return -1;
3383         }
3384
3385         if(memcmp(bitstream, out, COUNT)){
3386             printf("mismatch\n");
3387             return -1;
3388         }
3389     }
3390 #endif
3391
3392     printf("Testing RBSP\n");
3393
3394
3395     return 0;
3396 }
3397 #endif /* TEST */
3398
3399
3400 av_cold void ff_h264_free_context(H264Context *h)
3401 {
3402     int i;
3403
3404     free_tables(h, 1); //FIXME cleanup init stuff perhaps
3405
3406     for(i = 0; i < MAX_SPS_COUNT; i++)
3407         av_freep(h->sps_buffers + i);
3408
3409     for(i = 0; i < MAX_PPS_COUNT; i++)
3410         av_freep(h->pps_buffers + i);
3411 }
3412
3413 av_cold int ff_h264_decode_end(AVCodecContext *avctx)
3414 {
3415     H264Context *h = avctx->priv_data;
3416     MpegEncContext *s = &h->s;
3417
3418     ff_h264_free_context(h);
3419
3420     MPV_common_end(s);
3421
3422 //    memset(h, 0, sizeof(H264Context));
3423
3424     return 0;
3425 }
3426
3427 static const AVProfile profiles[] = {
3428     { FF_PROFILE_H264_BASELINE,             "Baseline"              },
3429     { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline"  },
3430     { FF_PROFILE_H264_MAIN,                 "Main"                  },
3431     { FF_PROFILE_H264_EXTENDED,             "Extended"              },
3432     { FF_PROFILE_H264_HIGH,                 "High"                  },
3433     { FF_PROFILE_H264_HIGH_10,              "High 10"               },
3434     { FF_PROFILE_H264_HIGH_10_INTRA,        "High 10 Intra"         },
3435     { FF_PROFILE_H264_HIGH_422,             "High 4:2:2"            },
3436     { FF_PROFILE_H264_HIGH_422_INTRA,       "High 4:2:2 Intra"      },
3437     { FF_PROFILE_H264_HIGH_444,             "High 4:4:4"            },
3438     { FF_PROFILE_H264_HIGH_444_PREDICTIVE,  "High 4:4:4 Predictive" },
3439     { FF_PROFILE_H264_HIGH_444_INTRA,       "High 4:4:4 Intra"      },
3440     { FF_PROFILE_H264_CAVLC_444,            "CAVLC 4:4:4"           },
3441     { FF_PROFILE_UNKNOWN },
3442 };
3443
3444 AVCodec ff_h264_decoder = {
3445     "h264",
3446     AVMEDIA_TYPE_VIDEO,
3447     CODEC_ID_H264,
3448     sizeof(H264Context),
3449     ff_h264_decode_init,
3450     NULL,
3451     ff_h264_decode_end,
3452     decode_frame,
3453     /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY |
3454         CODEC_CAP_FRAME_THREADS |
3455         CODEC_CAP_SLICE_THREADS,
3456     .flush= flush_dpb,
3457     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
3458     .init_thread_copy      = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
3459     .update_thread_context = ONLY_IF_THREADS_ENABLED(decode_update_thread_context),
3460     .profiles = NULL_IF_CONFIG_SMALL(profiles),
3461 };
3462
3463 #if CONFIG_H264_VDPAU_DECODER
3464 AVCodec ff_h264_vdpau_decoder = {
3465     "h264_vdpau",
3466     AVMEDIA_TYPE_VIDEO,
3467     CODEC_ID_H264,
3468     sizeof(H264Context),
3469     ff_h264_decode_init,
3470     NULL,
3471     ff_h264_decode_end,
3472     decode_frame,
3473     CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3474     .flush= flush_dpb,
3475     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
3476     .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
3477     .profiles = NULL_IF_CONFIG_SMALL(profiles),
3478 };
3479 #endif