tizen/distrib/libav/libavcodec/h264.c

   1 /*
   2  * H.26L/H.264/AVC/JVT/14496-10/... decoder
   3  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
   4  *
   5  * This file is part of Libav.
   6  *
   7  * Libav is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2.1 of the License, or (at your option) any later version.
  11  *
  12  * Libav is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with Libav; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 /**
  23  * @file
  24  * H.264 / AVC / MPEG4 part10 codec.
  25  * @author Michael Niedermayer <michaelni@gmx.at>
  26  */
  27
  28 #include "libavutil/imgutils.h"
  29 #include "internal.h"
  30 #include "dsputil.h"
  31 #include "avcodec.h"
  32 #include "mpegvideo.h"
  33 #include "h264.h"
  34 #include "h264data.h"
  35 #include "h264_mvpred.h"
  36 #include "golomb.h"
  37 #include "mathops.h"
  38 #include "rectangle.h"
  39 #include "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_NUM+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_NUM+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 static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n, int height,
 250                                  int y_offset, int list){
 251     int raw_my= h->mv_cache[list][ scan8[n] ][1];
 252     int filter_height= (raw_my&3) ? 2 : 0;
 253     int full_my= (raw_my>>2) + y_offset;
 254     int top = full_my - filter_height, bottom = full_my + height + filter_height;
 255
 256     return FFMAX(abs(top), bottom);
 257 }
 258
 259 static inline void get_lowest_part_y(H264Context *h, int refs[2][48], int n, int height,
 260                                int y_offset, int list0, int list1, int *nrefs){
 261     MpegEncContext * const s = &h->s;
 262     int my;
 263
 264     y_offset += 16*(s->mb_y >> MB_FIELD);
 265
 266     if(list0){
 267         int ref_n = h->ref_cache[0][ scan8[n] ];
 268         Picture *ref= &h->ref_list[0][ref_n];
 269
 270         // Error resilience puts the current picture in the ref list.
 271         // Don't try to wait on these as it will cause a deadlock.
 272         // Fields can wait on each other, though.
 273         if(ref->thread_opaque != s->current_picture.thread_opaque ||
 274            (ref->reference&3) != s->picture_structure) {
 275             my = get_lowest_part_list_y(h, ref, n, height, y_offset, 0);
 276             if (refs[0][ref_n] < 0) nrefs[0] += 1;
 277             refs[0][ref_n] = FFMAX(refs[0][ref_n], my);
 278         }
 279     }
 280
 281     if(list1){
 282         int ref_n = h->ref_cache[1][ scan8[n] ];
 283         Picture *ref= &h->ref_list[1][ref_n];
 284
 285         if(ref->thread_opaque != s->current_picture.thread_opaque ||
 286            (ref->reference&3) != s->picture_structure) {
 287             my = get_lowest_part_list_y(h, ref, n, height, y_offset, 1);
 288             if (refs[1][ref_n] < 0) nrefs[1] += 1;
 289             refs[1][ref_n] = FFMAX(refs[1][ref_n], my);
 290         }
 291     }
 292 }
 293
 294 /**
 295  * Wait until all reference frames are available for MC operations.
 296  *
 297  * @param h the H264 context
 298  */
 299 static void await_references(H264Context *h){
 300     MpegEncContext * const s = &h->s;
 301     const int mb_xy= h->mb_xy;
 302     const int mb_type= s->current_picture.mb_type[mb_xy];
 303     int refs[2][48];
 304     int nrefs[2] = {0};
 305     int ref, list;
 306
 307     memset(refs, -1, sizeof(refs));
 308
 309     if(IS_16X16(mb_type)){
 310         get_lowest_part_y(h, refs, 0, 16, 0,
 311                   IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
 312     }else if(IS_16X8(mb_type)){
 313         get_lowest_part_y(h, refs, 0, 8, 0,
 314                   IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
 315         get_lowest_part_y(h, refs, 8, 8, 8,
 316                   IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
 317     }else if(IS_8X16(mb_type)){
 318         get_lowest_part_y(h, refs, 0, 16, 0,
 319                   IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
 320         get_lowest_part_y(h, refs, 4, 16, 0,
 321                   IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
 322     }else{
 323         int i;
 324
 325         assert(IS_8X8(mb_type));
 326
 327         for(i=0; i<4; i++){
 328             const int sub_mb_type= h->sub_mb_type[i];
 329             const int n= 4*i;
 330             int y_offset= (i&2)<<2;
 331
 332             if(IS_SUB_8X8(sub_mb_type)){
 333                 get_lowest_part_y(h, refs, n  , 8, y_offset,
 334                           IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
 335             }else if(IS_SUB_8X4(sub_mb_type)){
 336                 get_lowest_part_y(h, refs, n  , 4, y_offset,
 337                           IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
 338                 get_lowest_part_y(h, refs, n+2, 4, y_offset+4,
 339                           IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
 340             }else if(IS_SUB_4X8(sub_mb_type)){
 341                 get_lowest_part_y(h, refs, n  , 8, y_offset,
 342                           IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
 343                 get_lowest_part_y(h, refs, n+1, 8, y_offset,
 344                           IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
 345             }else{
 346                 int j;
 347                 assert(IS_SUB_4X4(sub_mb_type));
 348                 for(j=0; j<4; j++){
 349                     int sub_y_offset= y_offset + 2*(j&2);
 350                     get_lowest_part_y(h, refs, n+j, 4, sub_y_offset,
 351                               IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
 352                 }
 353             }
 354         }
 355     }
 356
 357     for(list=h->list_count-1; list>=0; list--){
 358         for(ref=0; ref<48 && nrefs[list]; ref++){
 359             int row = refs[list][ref];
 360             if(row >= 0){
 361                 Picture *ref_pic = &h->ref_list[list][ref];
 362                 int ref_field = ref_pic->reference - 1;
 363                 int ref_field_picture = ref_pic->field_picture;
 364                 int pic_height = 16*s->mb_height >> ref_field_picture;
 365
 366                 row <<= MB_MBAFF;
 367                 nrefs[list]--;
 368
 369                 if(!FIELD_PICTURE && ref_field_picture){ // frame referencing two fields
 370                     ff_thread_await_progress((AVFrame*)ref_pic, FFMIN((row >> 1) - !(row&1), pic_height-1), 1);
 371                     ff_thread_await_progress((AVFrame*)ref_pic, FFMIN((row >> 1)           , pic_height-1), 0);
 372                 }else if(FIELD_PICTURE && !ref_field_picture){ // field referencing one field of a frame
 373                     ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row*2 + ref_field    , pic_height-1), 0);
 374                 }else if(FIELD_PICTURE){
 375                     ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row, pic_height-1), ref_field);
 376                 }else{
 377                     ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row, pic_height-1), 0);
 378                 }
 379             }
 380         }
 381     }
 382 }
 383
 384 #if 0
 385 /**
 386  * DCT transforms the 16 dc values.
 387  * @param qp quantization parameter ??? FIXME
 388  */
 389 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
 390 //    const int qmul= dequant_coeff[qp][0];
 391     int i;
 392     int temp[16]; //FIXME check if this is a good idea
 393     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
 394     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
 395
 396     for(i=0; i<4; i++){
 397         const int offset= y_offset[i];
 398         const int z0= block[offset+stride*0] + block[offset+stride*4];
 399         const int z1= block[offset+stride*0] - block[offset+stride*4];
 400         const int z2= block[offset+stride*1] - block[offset+stride*5];
 401         const int z3= block[offset+stride*1] + block[offset+stride*5];
 402
 403         temp[4*i+0]= z0+z3;
 404         temp[4*i+1]= z1+z2;
 405         temp[4*i+2]= z1-z2;
 406         temp[4*i+3]= z0-z3;
 407     }
 408
 409     for(i=0; i<4; i++){
 410         const int offset= x_offset[i];
 411         const int z0= temp[4*0+i] + temp[4*2+i];
 412         const int z1= temp[4*0+i] - temp[4*2+i];
 413         const int z2= temp[4*1+i] - temp[4*3+i];
 414         const int z3= temp[4*1+i] + temp[4*3+i];
 415
 416         block[stride*0 +offset]= (z0 + z3)>>1;
 417         block[stride*2 +offset]= (z1 + z2)>>1;
 418         block[stride*8 +offset]= (z1 - z2)>>1;
 419         block[stride*10+offset]= (z0 - z3)>>1;
 420     }
 421 }
 422 #endif
 423
 424 #undef xStride
 425 #undef stride
 426
 427 #if 0
 428 static void chroma_dc_dct_c(DCTELEM *block){
 429     const int stride= 16*2;
 430     const int xStride= 16;
 431     int a,b,c,d,e;
 432
 433     a= block[stride*0 + xStride*0];
 434     b= block[stride*0 + xStride*1];
 435     c= block[stride*1 + xStride*0];
 436     d= block[stride*1 + xStride*1];
 437
 438     e= a-b;
 439     a= a+b;
 440     b= c-d;
 441     c= c+d;
 442
 443     block[stride*0 + xStride*0]= (a+c);
 444     block[stride*0 + xStride*1]= (e+b);
 445     block[stride*1 + xStride*0]= (a-c);
 446     block[stride*1 + xStride*1]= (e-b);
 447 }
 448 #endif
 449
 450 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
 451                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 452                            int src_x_offset, int src_y_offset,
 453                            qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op,
 454                            int pixel_shift, int chroma444){
 455     MpegEncContext * const s = &h->s;
 456     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
 457     int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
 458     const int luma_xy= (mx&3) + ((my&3)<<2);
 459     int offset = ((mx>>2) << pixel_shift) + (my>>2)*h->mb_linesize;
 460     uint8_t * src_y = pic->data[0] + offset;
 461     uint8_t * src_cb, * src_cr;
 462     int extra_width= h->emu_edge_width;
 463     int extra_height= h->emu_edge_height;
 464     int emu=0;
 465     const int full_mx= mx>>2;
 466     const int full_my= my>>2;
 467     const int pic_width  = 16*s->mb_width;
 468     const int pic_height = 16*s->mb_height >> MB_FIELD;
 469
 470     if(mx&7) extra_width -= 3;
 471     if(my&7) extra_height -= 3;
 472
 473     if(   full_mx < 0-extra_width
 474        || full_my < 0-extra_height
 475        || full_mx + 16/*FIXME*/ > pic_width + extra_width
 476        || full_my + 16/*FIXME*/ > pic_height + extra_height){
 477         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_y - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
 478             src_y= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
 479         emu=1;
 480     }
 481
 482     qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
 483     if(!square){
 484         qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
 485     }
 486
 487     if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
 488
 489     if(chroma444){
 490         src_cb = pic->data[1] + offset;
 491         if(emu){
 492             s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
 493                                     16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
 494             src_cb= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
 495         }
 496         qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); //FIXME try variable height perhaps?
 497         if(!square){
 498             qpix_op[luma_xy](dest_cb + delta, src_cb + delta, h->mb_linesize);
 499         }
 500
 501         src_cr = pic->data[2] + offset;
 502         if(emu){
 503             s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
 504                                     16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
 505             src_cr= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
 506         }
 507         qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); //FIXME try variable height perhaps?
 508         if(!square){
 509             qpix_op[luma_xy](dest_cr + delta, src_cr + delta, h->mb_linesize);
 510         }
 511         return;
 512     }
 513
 514     if(MB_FIELD){
 515         // chroma offset when predicting from a field of opposite parity
 516         my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
 517         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
 518     }
 519     src_cb= pic->data[1] + ((mx>>3) << pixel_shift) + (my>>3)*h->mb_uvlinesize;
 520     src_cr= pic->data[2] + ((mx>>3) << pixel_shift) + (my>>3)*h->mb_uvlinesize;
 521
 522     if(emu){
 523         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
 524             src_cb= s->edge_emu_buffer;
 525     }
 526     chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
 527
 528     if(emu){
 529         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
 530             src_cr= s->edge_emu_buffer;
 531     }
 532     chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
 533 }
 534
 535 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
 536                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 537                            int x_offset, int y_offset,
 538                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 539                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
 540                            int list0, int list1, int pixel_shift, int chroma444){
 541     MpegEncContext * const s = &h->s;
 542     qpel_mc_func *qpix_op=  qpix_put;
 543     h264_chroma_mc_func chroma_op= chroma_put;
 544
 545     dest_y  += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
 546     if(chroma444){
 547         dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
 548         dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
 549     }else{
 550         dest_cb += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
 551         dest_cr += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
 552     }
 553     x_offset += 8*s->mb_x;
 554     y_offset += 8*(s->mb_y >> MB_FIELD);
 555
 556     if(list0){
 557         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
 558         mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
 559                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
 560                            qpix_op, chroma_op, pixel_shift, chroma444);
 561
 562         qpix_op=  qpix_avg;
 563         chroma_op= chroma_avg;
 564     }
 565
 566     if(list1){
 567         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
 568         mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
 569                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
 570                            qpix_op, chroma_op, pixel_shift, chroma444);
 571     }
 572 }
 573
 574 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
 575                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 576                            int x_offset, int y_offset,
 577                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 578                            h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
 579                            h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
 580                            int list0, int list1, int pixel_shift, int chroma444){
 581     MpegEncContext * const s = &h->s;
 582
 583     dest_y += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
 584     if(chroma444){
 585         chroma_weight_avg = luma_weight_avg;
 586         chroma_weight_op = luma_weight_op;
 587         dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
 588         dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
 589     }else{
 590         dest_cb += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
 591         dest_cr += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
 592     }
 593     x_offset += 8*s->mb_x;
 594     y_offset += 8*(s->mb_y >> MB_FIELD);
 595
 596     if(list0 && list1){
 597         /* don't optimize for luma-only case, since B-frames usually
 598          * use implicit weights => chroma too. */
 599         uint8_t *tmp_cb = s->obmc_scratchpad;
 600         uint8_t *tmp_cr = s->obmc_scratchpad + (16 << pixel_shift);
 601         uint8_t *tmp_y  = s->obmc_scratchpad + 16*h->mb_uvlinesize;
 602         int refn0 = h->ref_cache[0][ scan8[n] ];
 603         int refn1 = h->ref_cache[1][ scan8[n] ];
 604
 605         mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
 606                     dest_y, dest_cb, dest_cr,
 607                     x_offset, y_offset, qpix_put, chroma_put, pixel_shift, chroma444);
 608         mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
 609                     tmp_y, tmp_cb, tmp_cr,
 610                     x_offset, y_offset, qpix_put, chroma_put, pixel_shift, chroma444);
 611
 612         if(h->use_weight == 2){
 613             int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
 614             int weight1 = 64 - weight0;
 615             luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
 616             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
 617             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
 618         }else{
 619             luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
 620                             h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
 621                             h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
 622             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 623                             h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
 624                             h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
 625             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 626                             h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
 627                             h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
 628         }
 629     }else{
 630         int list = list1 ? 1 : 0;
 631         int refn = h->ref_cache[list][ scan8[n] ];
 632         Picture *ref= &h->ref_list[list][refn];
 633         mc_dir_part(h, ref, n, square, chroma_height, delta, list,
 634                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
 635                     qpix_put, chroma_put, pixel_shift, chroma444);
 636
 637         luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
 638                        h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
 639         if(h->use_weight_chroma){
 640             chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 641                              h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
 642             chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 643                              h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
 644         }
 645     }
 646 }
 647
 648 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
 649                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 650                            int x_offset, int y_offset,
 651                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 652                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
 653                            h264_weight_func *weight_op, h264_biweight_func *weight_avg,
 654                            int list0, int list1, int pixel_shift, int chroma444){
 655     if((h->use_weight==2 && list0 && list1
 656         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
 657        || h->use_weight==1)
 658         mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
 659                          x_offset, y_offset, qpix_put, chroma_put,
 660                          weight_op[0], weight_op[3], weight_avg[0],
 661                          weight_avg[3], list0, list1, pixel_shift, chroma444);
 662     else
 663         mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
 664                     x_offset, y_offset, qpix_put, chroma_put, qpix_avg,
 665                     chroma_avg, list0, list1, pixel_shift, chroma444);
 666 }
 667
 668 static inline void prefetch_motion(H264Context *h, int list, int pixel_shift, int chroma444){
 669     /* fetch pixels for estimated mv 4 macroblocks ahead
 670      * optimized for 64byte cache lines */
 671     MpegEncContext * const s = &h->s;
 672     const int refn = h->ref_cache[list][scan8[0]];
 673     if(refn >= 0){
 674         const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
 675         const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
 676         uint8_t **src= h->ref_list[list][refn].data;
 677         int off= (mx << pixel_shift) + (my + (s->mb_x&3)*4)*h->mb_linesize + (64 << pixel_shift);
 678         s->dsp.prefetch(src[0]+off, s->linesize, 4);
 679         if(chroma444){
 680             s->dsp.prefetch(src[1]+off, s->linesize, 4);
 681             s->dsp.prefetch(src[2]+off, s->linesize, 4);
 682         }else{
 683             off= ((mx>>1) << pixel_shift) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + (64 << pixel_shift);
 684             s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
 685         }
 686     }
 687 }
 688
 689 static av_always_inline void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 690                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
 691                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
 692                       h264_weight_func *weight_op, h264_biweight_func *weight_avg,
 693                       int pixel_shift, int chroma444){
 694     MpegEncContext * const s = &h->s;
 695     const int mb_xy= h->mb_xy;
 696     const int mb_type= s->current_picture.mb_type[mb_xy];
 697
 698     assert(IS_INTER(mb_type));
 699
 700     if(HAVE_PTHREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
 701         await_references(h);
 702     prefetch_motion(h, 0, pixel_shift, chroma444);
 703
 704     if(IS_16X16(mb_type)){
 705         mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
 706                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
 707                 weight_op, weight_avg,
 708                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
 709                 pixel_shift, chroma444);
 710     }else if(IS_16X8(mb_type)){
 711         mc_part(h, 0, 0, 4, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 0,
 712                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
 713                 &weight_op[1], &weight_avg[1],
 714                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
 715                 pixel_shift, chroma444);
 716         mc_part(h, 8, 0, 4, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 4,
 717                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
 718                 &weight_op[1], &weight_avg[1],
 719                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
 720                 pixel_shift, chroma444);
 721     }else if(IS_8X16(mb_type)){
 722         mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
 723                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 724                 &weight_op[2], &weight_avg[2],
 725                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
 726                 pixel_shift, chroma444);
 727         mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
 728                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 729                 &weight_op[2], &weight_avg[2],
 730                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
 731                 pixel_shift, chroma444);
 732     }else{
 733         int i;
 734
 735         assert(IS_8X8(mb_type));
 736
 737         for(i=0; i<4; i++){
 738             const int sub_mb_type= h->sub_mb_type[i];
 739             const int n= 4*i;
 740             int x_offset= (i&1)<<2;
 741             int y_offset= (i&2)<<1;
 742
 743             if(IS_SUB_8X8(sub_mb_type)){
 744                 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 745                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 746                     &weight_op[3], &weight_avg[3],
 747                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
 748                     pixel_shift, chroma444);
 749             }else if(IS_SUB_8X4(sub_mb_type)){
 750                 mc_part(h, n  , 0, 2, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 751                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
 752                     &weight_op[4], &weight_avg[4],
 753                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
 754                     pixel_shift, chroma444);
 755                 mc_part(h, n+2, 0, 2, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
 756                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
 757                     &weight_op[4], &weight_avg[4],
 758                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
 759                     pixel_shift, chroma444);
 760             }else if(IS_SUB_4X8(sub_mb_type)){
 761                 mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 762                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 763                     &weight_op[5], &weight_avg[5],
 764                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
 765                     pixel_shift, chroma444);
 766                 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
 767                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 768                     &weight_op[5], &weight_avg[5],
 769                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
 770                     pixel_shift, chroma444);
 771             }else{
 772                 int j;
 773                 assert(IS_SUB_4X4(sub_mb_type));
 774                 for(j=0; j<4; j++){
 775                     int sub_x_offset= x_offset + 2*(j&1);
 776                     int sub_y_offset= y_offset +   (j&2);
 777                     mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
 778                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 779                         &weight_op[6], &weight_avg[6],
 780                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
 781                         pixel_shift, chroma444);
 782                 }
 783             }
 784         }
 785     }
 786
 787     prefetch_motion(h, 1, pixel_shift, chroma444);
 788 }
 789
 790 #define hl_motion_fn(sh, bits) \
 791 static av_always_inline void hl_motion_ ## bits(H264Context *h, \
 792                                        uint8_t *dest_y, \
 793                                        uint8_t *dest_cb, uint8_t *dest_cr, \
 794                                        qpel_mc_func (*qpix_put)[16], \
 795                                        h264_chroma_mc_func (*chroma_put), \
 796                                        qpel_mc_func (*qpix_avg)[16], \
 797                                        h264_chroma_mc_func (*chroma_avg), \
 798                                        h264_weight_func *weight_op, \
 799                                        h264_biweight_func *weight_avg, \
 800                                        int chroma444) \
 801 { \
 802     hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put, \
 803               qpix_avg, chroma_avg, weight_op, weight_avg, sh, chroma444); \
 804 }
 805 hl_motion_fn(0, 8);
 806 hl_motion_fn(1, 16);
 807
 808 static void free_tables(H264Context *h, int free_rbsp){
 809     int i;
 810     H264Context *hx;
 811
 812     av_freep(&h->intra4x4_pred_mode);
 813     av_freep(&h->chroma_pred_mode_table);
 814     av_freep(&h->cbp_table);
 815     av_freep(&h->mvd_table[0]);
 816     av_freep(&h->mvd_table[1]);
 817     av_freep(&h->direct_table);
 818     av_freep(&h->non_zero_count);
 819     av_freep(&h->slice_table_base);
 820     h->slice_table= NULL;
 821     av_freep(&h->list_counts);
 822
 823     av_freep(&h->mb2b_xy);
 824     av_freep(&h->mb2br_xy);
 825
 826     for(i = 0; i < MAX_THREADS; i++) {
 827         hx = h->thread_context[i];
 828         if(!hx) continue;
 829         av_freep(&hx->top_borders[1]);
 830         av_freep(&hx->top_borders[0]);
 831         av_freep(&hx->s.obmc_scratchpad);
 832         if (free_rbsp){
 833             av_freep(&hx->rbsp_buffer[1]);
 834             av_freep(&hx->rbsp_buffer[0]);
 835             hx->rbsp_buffer_size[0] = 0;
 836             hx->rbsp_buffer_size[1] = 0;
 837         }
 838         if (i) av_freep(&h->thread_context[i]);
 839     }
 840 }
 841
 842 static void init_dequant8_coeff_table(H264Context *h){
 843     int i,j,q,x;
 844     const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
 845
 846     for(i=0; i<6; i++ ){
 847         h->dequant8_coeff[i] = h->dequant8_buffer[i];
 848         for(j=0; j<i; j++){
 849             if(!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i], 64*sizeof(uint8_t))){
 850                 h->dequant8_coeff[i] = h->dequant8_buffer[j];
 851                 break;
 852             }
 853         }
 854         if(j<i)
 855             continue;
 856
 857         for(q=0; q<max_qp+1; q++){
 858             int shift = div6[q];
 859             int idx = rem6[q];
 860             for(x=0; x<64; x++)
 861                 h->dequant8_coeff[i][q][(x>>3)|((x&7)<<3)] =
 862                     ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
 863                     h->pps.scaling_matrix8[i][x]) << shift;
 864         }
 865     }
 866 }
 867
 868 static void init_dequant4_coeff_table(H264Context *h){
 869     int i,j,q,x;
 870     const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
 871     for(i=0; i<6; i++ ){
 872         h->dequant4_coeff[i] = h->dequant4_buffer[i];
 873         for(j=0; j<i; j++){
 874             if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
 875                 h->dequant4_coeff[i] = h->dequant4_buffer[j];
 876                 break;
 877             }
 878         }
 879         if(j<i)
 880             continue;
 881
 882         for(q=0; q<max_qp+1; q++){
 883             int shift = div6[q] + 2;
 884             int idx = rem6[q];
 885             for(x=0; x<16; x++)
 886                 h->dequant4_coeff[i][q][(x>>2)|((x<<2)&0xF)] =
 887                     ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
 888                     h->pps.scaling_matrix4[i][x]) << shift;
 889         }
 890     }
 891 }
 892
 893 static void init_dequant_tables(H264Context *h){
 894     int i,x;
 895     init_dequant4_coeff_table(h);
 896     if(h->pps.transform_8x8_mode)
 897         init_dequant8_coeff_table(h);
 898     if(h->sps.transform_bypass){
 899         for(i=0; i<6; i++)
 900             for(x=0; x<16; x++)
 901                 h->dequant4_coeff[i][0][x] = 1<<6;
 902         if(h->pps.transform_8x8_mode)
 903             for(i=0; i<6; i++)
 904                 for(x=0; x<64; x++)
 905                     h->dequant8_coeff[i][0][x] = 1<<6;
 906     }
 907 }
 908
 909
 910 int ff_h264_alloc_tables(H264Context *h){
 911     MpegEncContext * const s = &h->s;
 912     const int big_mb_num= s->mb_stride * (s->mb_height+1);
 913     const int row_mb_num= 2*s->mb_stride*s->avctx->thread_count;
 914     int x,y;
 915
 916     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8  * sizeof(uint8_t), fail)
 917
 918     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 48 * sizeof(uint8_t), fail)
 919     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
 920     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
 921
 922     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
 923     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
 924     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
 925     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
 926     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
 927
 928     memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
 929     h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
 930
 931     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy  , big_mb_num * sizeof(uint32_t), fail);
 932     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
 933     for(y=0; y<s->mb_height; y++){
 934         for(x=0; x<s->mb_width; x++){
 935             const int mb_xy= x + y*s->mb_stride;
 936             const int b_xy = 4*x + 4*y*h->b_stride;
 937
 938             h->mb2b_xy [mb_xy]= b_xy;
 939             h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
 940         }
 941     }
 942
 943     s->obmc_scratchpad = NULL;
 944
 945     if(!h->dequant4_coeff[0])
 946         init_dequant_tables(h);
 947
 948     return 0;
 949 fail:
 950     free_tables(h, 1);
 951     return -1;
 952 }
 953
 954 /**
 955  * Mimic alloc_tables(), but for every context thread.
 956  */
 957 static void clone_tables(H264Context *dst, H264Context *src, int i){
 958     MpegEncContext * const s = &src->s;
 959     dst->intra4x4_pred_mode       = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
 960     dst->non_zero_count           = src->non_zero_count;
 961     dst->slice_table              = src->slice_table;
 962     dst->cbp_table                = src->cbp_table;
 963     dst->mb2b_xy                  = src->mb2b_xy;
 964     dst->mb2br_xy                 = src->mb2br_xy;
 965     dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
 966     dst->mvd_table[0]             = src->mvd_table[0] + i*8*2*s->mb_stride;
 967     dst->mvd_table[1]             = src->mvd_table[1] + i*8*2*s->mb_stride;
 968     dst->direct_table             = src->direct_table;
 969     dst->list_counts              = src->list_counts;
 970
 971     dst->s.obmc_scratchpad = NULL;
 972     ff_h264_pred_init(&dst->hpc, src->s.codec_id, src->sps.bit_depth_luma);
 973 }
 974
 975 /**
 976  * Init context
 977  * Allocate buffers which are not shared amongst multiple threads.
 978  */
 979 static int context_init(H264Context *h){
 980     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
 981     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
 982
 983     h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
 984     h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
 985
 986     return 0;
 987 fail:
 988     return -1; // free_tables will clean up for us
 989 }
 990
 991 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
 992
 993 static av_cold void common_init(H264Context *h){
 994     MpegEncContext * const s = &h->s;
 995
 996     s->width = s->avctx->width;
 997     s->height = s->avctx->height;
 998     s->codec_id= s->avctx->codec->id;
 999
1000     ff_h264dsp_init(&h->h264dsp, 8);
1001     ff_h264_pred_init(&h->hpc, s->codec_id, 8);
1002
1003     h->dequant_coeff_pps= -1;
1004     s->unrestricted_mv=1;
1005     s->decode=1; //FIXME
1006
1007     dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
1008
1009     memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
1010     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
1011 }
1012
1013 int ff_h264_decode_extradata(H264Context *h)
1014 {
1015     AVCodecContext *avctx = h->s.avctx;
1016
1017     if(avctx->extradata[0] == 1){
1018         int i, cnt, nalsize;
1019         unsigned char *p = avctx->extradata;
1020
1021         h->is_avc = 1;
1022
1023         if(avctx->extradata_size < 7) {
1024             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
1025             return -1;
1026         }
1027         /* sps and pps in the avcC always have length coded with 2 bytes,
1028            so put a fake nal_length_size = 2 while parsing them */
1029         h->nal_length_size = 2;
1030         // Decode sps from avcC
1031         cnt = *(p+5) & 0x1f; // Number of sps
1032         p += 6;
1033         for (i = 0; i < cnt; i++) {
1034             nalsize = AV_RB16(p) + 2;
1035             if(decode_nal_units(h, p, nalsize) < 0) {
1036                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
1037                 return -1;
1038             }
1039             p += nalsize;
1040         }
1041         // Decode pps from avcC
1042         cnt = *(p++); // Number of pps
1043         for (i = 0; i < cnt; i++) {
1044             nalsize = AV_RB16(p) + 2;
1045             if (decode_nal_units(h, p, nalsize) < 0) {
1046                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
1047                 return -1;
1048             }
1049             p += nalsize;
1050         }
1051         // Now store right nal length size, that will be use to parse all other nals
1052         h->nal_length_size = (avctx->extradata[4] & 0x03) + 1;
1053     } else {
1054         h->is_avc = 0;
1055         if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
1056             return -1;
1057     }
1058     return 0;
1059 }
1060
1061 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
1062     H264Context *h= avctx->priv_data;
1063     MpegEncContext * const s = &h->s;
1064
1065     MPV_decode_defaults(s);
1066
1067     s->avctx = avctx;
1068     common_init(h);
1069
1070     s->out_format = FMT_H264;
1071     s->workaround_bugs= avctx->workaround_bugs;
1072
1073     // set defaults
1074 //    s->decode_mb= ff_h263_decode_mb;
1075     s->quarter_sample = 1;
1076     if(!avctx->has_b_frames)
1077     s->low_delay= 1;
1078
1079     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
1080
1081     ff_h264_decode_init_vlc();
1082
1083     h->pixel_shift = 0;
1084     h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
1085
1086     h->thread_context[0] = h;
1087     h->outputed_poc = h->next_outputed_poc = INT_MIN;
1088     h->prev_poc_msb= 1<<16;
1089     h->x264_build = -1;
1090     ff_h264_reset_sei(h);
1091     if(avctx->codec_id == CODEC_ID_H264){
1092         if(avctx->ticks_per_frame == 1){
1093             s->avctx->time_base.den *=2;
1094         }
1095         avctx->ticks_per_frame = 2;
1096     }
1097
1098     if(avctx->extradata_size > 0 && avctx->extradata &&
1099         ff_h264_decode_extradata(h))
1100         return -1;
1101
1102     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
1103         s->avctx->has_b_frames = h->sps.num_reorder_frames;
1104         s->low_delay = 0;
1105     }
1106
1107     return 0;
1108 }
1109
1110 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b)+(size))))
1111 static void copy_picture_range(Picture **to, Picture **from, int count, MpegEncContext *new_base, MpegEncContext *old_base)
1112 {
1113     int i;
1114
1115     for (i=0; i<count; i++){
1116         assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
1117                 IN_RANGE(from[i], old_base->picture, sizeof(Picture) * old_base->picture_count) ||
1118                 !from[i]));
1119         to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1120     }
1121 }
1122
1123 static void copy_parameter_set(void **to, void **from, int count, int size)
1124 {
1125     int i;
1126
1127     for (i=0; i<count; i++){
1128         if (to[i] && !from[i]) av_freep(&to[i]);
1129         else if (from[i] && !to[i]) to[i] = av_malloc(size);
1130
1131         if (from[i]) memcpy(to[i], from[i], size);
1132     }
1133 }
1134
1135 static int decode_init_thread_copy(AVCodecContext *avctx){
1136     H264Context *h= avctx->priv_data;
1137
1138     if (!avctx->is_copy) return 0;
1139     memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1140     memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1141
1142     return 0;
1143 }
1144
1145 #define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
1146 static int decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src){
1147     H264Context *h= dst->priv_data, *h1= src->priv_data;
1148     MpegEncContext * const s = &h->s, * const s1 = &h1->s;
1149     int inited = s->context_initialized, err;
1150     int i;
1151
1152     if(dst == src || !s1->context_initialized) return 0;
1153
1154     err = ff_mpeg_update_thread_context(dst, src);
1155     if(err) return err;
1156
1157     //FIXME handle width/height changing
1158     if(!inited){
1159         for(i = 0; i < MAX_SPS_COUNT; i++)
1160             av_freep(h->sps_buffers + i);
1161
1162         for(i = 0; i < MAX_PPS_COUNT; i++)
1163             av_freep(h->pps_buffers + i);
1164
1165         memcpy(&h->s + 1, &h1->s + 1, sizeof(H264Context) - sizeof(MpegEncContext)); //copy all fields after MpegEnc
1166         memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1167         memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1168         if (ff_h264_alloc_tables(h) < 0) {
1169             av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
1170             return AVERROR(ENOMEM);
1171         }
1172         context_init(h);
1173
1174         for(i=0; i<2; i++){
1175             h->rbsp_buffer[i] = NULL;
1176             h->rbsp_buffer_size[i] = 0;
1177         }
1178
1179         h->thread_context[0] = h;
1180
1181         // frame_start may not be called for the next thread (if it's decoding a bottom field)
1182         // so this has to be allocated here
1183         h->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
1184
1185         s->dsp.clear_blocks(h->mb);
1186         s->dsp.clear_blocks(h->mb+(24*16<<h->pixel_shift));
1187     }
1188
1189     //extradata/NAL handling
1190     h->is_avc          = h1->is_avc;
1191
1192     //SPS/PPS
1193     copy_parameter_set((void**)h->sps_buffers, (void**)h1->sps_buffers, MAX_SPS_COUNT, sizeof(SPS));
1194     h->sps             = h1->sps;
1195     copy_parameter_set((void**)h->pps_buffers, (void**)h1->pps_buffers, MAX_PPS_COUNT, sizeof(PPS));
1196     h->pps             = h1->pps;
1197
1198     //Dequantization matrices
1199     //FIXME these are big - can they be only copied when PPS changes?
1200     copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1201
1202     for(i=0; i<6; i++)
1203         h->dequant4_coeff[i] = h->dequant4_buffer[0] + (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
1204
1205     for(i=0; i<6; i++)
1206         h->dequant8_coeff[i] = h->dequant8_buffer[0] + (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
1207
1208     h->dequant_coeff_pps = h1->dequant_coeff_pps;
1209
1210     //POC timing
1211     copy_fields(h, h1, poc_lsb, redundant_pic_count);
1212
1213     //reference lists
1214     copy_fields(h, h1, ref_count, list_count);
1215     copy_fields(h, h1, ref_list,  intra_gb);
1216     copy_fields(h, h1, short_ref, cabac_init_idc);
1217
1218     copy_picture_range(h->short_ref,   h1->short_ref,   32, s, s1);
1219     copy_picture_range(h->long_ref,    h1->long_ref,    32, s, s1);
1220     copy_picture_range(h->delayed_pic, h1->delayed_pic, MAX_DELAYED_PIC_COUNT+2, s, s1);
1221
1222     h->last_slice_type = h1->last_slice_type;
1223
1224     if(!s->current_picture_ptr) return 0;
1225
1226     if(!s->dropable) {
1227         ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1228         h->prev_poc_msb     = h->poc_msb;
1229         h->prev_poc_lsb     = h->poc_lsb;
1230     }
1231     h->prev_frame_num_offset= h->frame_num_offset;
1232     h->prev_frame_num       = h->frame_num;
1233     h->outputed_poc         = h->next_outputed_poc;
1234
1235     return 0;
1236 }
1237
1238 int ff_h264_frame_start(H264Context *h){
1239     MpegEncContext * const s = &h->s;
1240     int i;
1241     const int pixel_shift = h->pixel_shift;
1242     int thread_count = (s->avctx->active_thread_type & FF_THREAD_SLICE) ? s->avctx->thread_count : 1;
1243
1244     if(MPV_frame_start(s, s->avctx) < 0)
1245         return -1;
1246     ff_er_frame_start(s);
1247     /*
1248      * MPV_frame_start uses pict_type to derive key_frame.
1249      * This is incorrect for H.264; IDR markings must be used.
1250      * Zero here; IDR markings per slice in frame or fields are ORed in later.
1251      * See decode_nal_units().
1252      */
1253     s->current_picture_ptr->key_frame= 0;
1254     s->current_picture_ptr->mmco_reset= 0;
1255
1256     assert(s->linesize && s->uvlinesize);
1257
1258     for(i=0; i<16; i++){
1259         h->block_offset[i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
1260         h->block_offset[48+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
1261     }
1262     for(i=0; i<16; i++){
1263         h->block_offset[16+i]=
1264         h->block_offset[32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1265         h->block_offset[48+16+i]=
1266         h->block_offset[48+32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1267     }
1268
1269     /* can't be in alloc_tables because linesize isn't known there.
1270      * FIXME: redo bipred weight to not require extra buffer? */
1271     for(i = 0; i < thread_count; i++)
1272         if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
1273             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
1274
1275     /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
1276     memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
1277
1278 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
1279
1280     // We mark the current picture as non-reference after allocating it, so
1281     // that if we break out due to an error it can be released automatically
1282     // in the next MPV_frame_start().
1283     // SVQ3 as well as most other codecs have only last/next/current and thus
1284     // get released even with set reference, besides SVQ3 and others do not
1285     // mark frames as reference later "naturally".
1286     if(s->codec_id != CODEC_ID_SVQ3)
1287         s->current_picture_ptr->reference= 0;
1288
1289     s->current_picture_ptr->field_poc[0]=
1290     s->current_picture_ptr->field_poc[1]= INT_MAX;
1291
1292     h->next_output_pic = NULL;
1293
1294     assert(s->current_picture_ptr->long_ref==0);
1295
1296     return 0;
1297 }
1298
1299 /**
1300   * Run setup operations that must be run after slice header decoding.
1301   * This includes finding the next displayed frame.
1302   *
1303   * @param h h264 master context
1304   * @param setup_finished enough NALs have been read that we can call
1305   * ff_thread_finish_setup()
1306   */
1307 static void decode_postinit(H264Context *h, int setup_finished){
1308     MpegEncContext * const s = &h->s;
1309     Picture *out = s->current_picture_ptr;
1310     Picture *cur = s->current_picture_ptr;
1311     int i, pics, out_of_order, out_idx;
1312
1313     s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1314     s->current_picture_ptr->pict_type= s->pict_type;
1315
1316     if (h->next_output_pic) return;
1317
1318     if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
1319         //FIXME: if we have two PAFF fields in one packet, we can't start the next thread here.
1320         //If we have one field per packet, we can. The check in decode_nal_units() is not good enough
1321         //to find this yet, so we assume the worst for now.
1322         //if (setup_finished)
1323         //    ff_thread_finish_setup(s->avctx);
1324         return;
1325     }
1326
1327     cur->interlaced_frame = 0;
1328     cur->repeat_pict = 0;
1329
1330     /* Signal interlacing information externally. */
1331     /* Prioritize picture timing SEI information over used decoding process if it exists. */
1332
1333     if(h->sps.pic_struct_present_flag){
1334         switch (h->sei_pic_struct)
1335         {
1336         case SEI_PIC_STRUCT_FRAME:
1337             break;
1338         case SEI_PIC_STRUCT_TOP_FIELD:
1339         case SEI_PIC_STRUCT_BOTTOM_FIELD:
1340             cur->interlaced_frame = 1;
1341             break;
1342         case SEI_PIC_STRUCT_TOP_BOTTOM:
1343         case SEI_PIC_STRUCT_BOTTOM_TOP:
1344             if (FIELD_OR_MBAFF_PICTURE)
1345                 cur->interlaced_frame = 1;
1346             else
1347                 // try to flag soft telecine progressive
1348                 cur->interlaced_frame = h->prev_interlaced_frame;
1349             break;
1350         case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1351         case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
1352             // Signal the possibility of telecined film externally (pic_struct 5,6)
1353             // From these hints, let the applications decide if they apply deinterlacing.
1354             cur->repeat_pict = 1;
1355             break;
1356         case SEI_PIC_STRUCT_FRAME_DOUBLING:
1357             // Force progressive here, as doubling interlaced frame is a bad idea.
1358             cur->repeat_pict = 2;
1359             break;
1360         case SEI_PIC_STRUCT_FRAME_TRIPLING:
1361             cur->repeat_pict = 4;
1362             break;
1363         }
1364
1365         if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
1366             cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
1367     }else{
1368         /* Derive interlacing flag from used decoding process. */
1369         cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
1370     }
1371     h->prev_interlaced_frame = cur->interlaced_frame;
1372
1373     if (cur->field_poc[0] != cur->field_poc[1]){
1374         /* Derive top_field_first from field pocs. */
1375         cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
1376     }else{
1377         if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
1378             /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
1379             if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
1380               || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
1381                 cur->top_field_first = 1;
1382             else
1383                 cur->top_field_first = 0;
1384         }else{
1385             /* Most likely progressive */
1386             cur->top_field_first = 0;
1387         }
1388     }
1389
1390     //FIXME do something with unavailable reference frames
1391
1392     /* Sort B-frames into display order */
1393
1394     if(h->sps.bitstream_restriction_flag
1395        && s->avctx->has_b_frames < h->sps.num_reorder_frames){
1396         s->avctx->has_b_frames = h->sps.num_reorder_frames;
1397         s->low_delay = 0;
1398     }
1399
1400     if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
1401        && !h->sps.bitstream_restriction_flag){
1402         s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
1403         s->low_delay= 0;
1404     }
1405
1406     pics = 0;
1407     while(h->delayed_pic[pics]) pics++;
1408
1409     assert(pics <= MAX_DELAYED_PIC_COUNT);
1410
1411     h->delayed_pic[pics++] = cur;
1412     if(cur->reference == 0)
1413         cur->reference = DELAYED_PIC_REF;
1414
1415     out = h->delayed_pic[0];
1416     out_idx = 0;
1417     for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
1418         if(h->delayed_pic[i]->poc < out->poc){
1419             out = h->delayed_pic[i];
1420             out_idx = i;
1421         }
1422     if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
1423         h->next_outputed_poc= INT_MIN;
1424     out_of_order = out->poc < h->next_outputed_poc;
1425
1426     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
1427         { }
1428     else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
1429        || (s->low_delay &&
1430         ((h->next_outputed_poc != INT_MIN && out->poc > h->next_outputed_poc + 2)
1431          || cur->pict_type == AV_PICTURE_TYPE_B)))
1432     {
1433         s->low_delay = 0;
1434         s->avctx->has_b_frames++;
1435     }
1436
1437     if(out_of_order || pics > s->avctx->has_b_frames){
1438         out->reference &= ~DELAYED_PIC_REF;
1439         out->owner2 = s; // for frame threading, the owner must be the second field's thread
1440                          // or else the first thread can release the picture and reuse it unsafely
1441         for(i=out_idx; h->delayed_pic[i]; i++)
1442             h->delayed_pic[i] = h->delayed_pic[i+1];
1443     }
1444     if(!out_of_order && pics > s->avctx->has_b_frames){
1445         h->next_output_pic = out;
1446         if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
1447             h->next_outputed_poc = INT_MIN;
1448         } else
1449             h->next_outputed_poc = out->poc;
1450     }else{
1451         av_log(s->avctx, AV_LOG_DEBUG, "no picture\n");
1452     }
1453
1454     if (setup_finished)
1455         ff_thread_finish_setup(s->avctx);
1456 }
1457
1458 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 chroma444, int simple){
1459     MpegEncContext * const s = &h->s;
1460     uint8_t *top_border;
1461     int top_idx = 1;
1462     const int pixel_shift = h->pixel_shift;
1463
1464     src_y  -=   linesize;
1465     src_cb -= uvlinesize;
1466     src_cr -= uvlinesize;
1467
1468     if(!simple && FRAME_MBAFF){
1469         if(s->mb_y&1){
1470             if(!MB_MBAFF){
1471                 top_border = h->top_borders[0][s->mb_x];
1472                 AV_COPY128(top_border, src_y + 15*linesize);
1473                 if (pixel_shift)
1474                     AV_COPY128(top_border+16, src_y+15*linesize+16);
1475                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1476                     if(chroma444){
1477                         if (pixel_shift){
1478                             AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
1479                             AV_COPY128(top_border+48, src_cb + 15*uvlinesize+16);
1480                             AV_COPY128(top_border+64, src_cr + 15*uvlinesize);
1481                             AV_COPY128(top_border+80, src_cr + 15*uvlinesize+16);
1482                         } else {
1483                             AV_COPY128(top_border+16, src_cb + 15*uvlinesize);
1484                             AV_COPY128(top_border+32, src_cr + 15*uvlinesize);
1485                         }
1486                     } else {
1487                         if (pixel_shift) {
1488                             AV_COPY128(top_border+32, src_cb+7*uvlinesize);
1489                             AV_COPY128(top_border+48, src_cr+7*uvlinesize);
1490                         } else {
1491                             AV_COPY64(top_border+16, src_cb+7*uvlinesize);
1492                             AV_COPY64(top_border+24, src_cr+7*uvlinesize);
1493                         }
1494                     }
1495                 }
1496             }
1497         }else if(MB_MBAFF){
1498             top_idx = 0;
1499         }else
1500             return;
1501     }
1502
1503     top_border = h->top_borders[top_idx][s->mb_x];
1504     // There are two lines saved, the line above the the top macroblock of a pair,
1505     // and the line above the bottom macroblock
1506     AV_COPY128(top_border, src_y + 16*linesize);
1507     if (pixel_shift)
1508         AV_COPY128(top_border+16, src_y+16*linesize+16);
1509
1510     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1511         if(chroma444){
1512             if (pixel_shift){
1513                 AV_COPY128(top_border+32, src_cb + 16*linesize);
1514                 AV_COPY128(top_border+48, src_cb + 16*linesize+16);
1515                 AV_COPY128(top_border+64, src_cr + 16*linesize);
1516                 AV_COPY128(top_border+80, src_cr + 16*linesize+16);
1517             } else {
1518                 AV_COPY128(top_border+16, src_cb + 16*linesize);
1519                 AV_COPY128(top_border+32, src_cr + 16*linesize);
1520             }
1521         } else {
1522             if (pixel_shift) {
1523                 AV_COPY128(top_border+32, src_cb+8*uvlinesize);
1524                 AV_COPY128(top_border+48, src_cr+8*uvlinesize);
1525             } else {
1526                 AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1527                 AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1528             }
1529         }
1530     }
1531 }
1532
1533 static inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
1534                                   uint8_t *src_cb, uint8_t *src_cr,
1535                                   int linesize, int uvlinesize,
1536                                   int xchg, int chroma444,
1537                                   int simple, int pixel_shift){
1538     MpegEncContext * const s = &h->s;
1539     int deblock_topleft;
1540     int deblock_top;
1541     int top_idx = 1;
1542     uint8_t *top_border_m1;
1543     uint8_t *top_border;
1544
1545     if(!simple && FRAME_MBAFF){
1546         if(s->mb_y&1){
1547             if(!MB_MBAFF)
1548                 return;
1549         }else{
1550             top_idx = MB_MBAFF ? 0 : 1;
1551         }
1552     }
1553
1554     if(h->deblocking_filter == 2) {
1555         deblock_topleft = h->slice_table[h->mb_xy - 1 - s->mb_stride] == h->slice_num;
1556         deblock_top     = h->top_type;
1557     } else {
1558         deblock_topleft = (s->mb_x > 0);
1559         deblock_top     = (s->mb_y > !!MB_FIELD);
1560     }
1561
1562     src_y  -=   linesize + 1 + pixel_shift;
1563     src_cb -= uvlinesize + 1 + pixel_shift;
1564     src_cr -= uvlinesize + 1 + pixel_shift;
1565
1566     top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1567     top_border    = h->top_borders[top_idx][s->mb_x];
1568
1569 #define XCHG(a,b,xchg)\
1570     if (pixel_shift) {\
1571         if (xchg) {\
1572             AV_SWAP64(b+0,a+0);\
1573             AV_SWAP64(b+8,a+8);\
1574         } else {\
1575             AV_COPY128(b,a); \
1576         }\
1577     } else \
1578 if (xchg) AV_SWAP64(b,a);\
1579 else      AV_COPY64(b,a);
1580
1581     if(deblock_top){
1582         if(deblock_topleft){
1583             XCHG(top_border_m1 + (8 << pixel_shift), src_y - (7 << pixel_shift), 1);
1584         }
1585         XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
1586         XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
1587         if(s->mb_x+1 < s->mb_width){
1588             XCHG(h->top_borders[top_idx][s->mb_x+1], src_y + (17 << pixel_shift), 1);
1589         }
1590     }
1591     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1592         if(chroma444){
1593             if(deblock_topleft){
1594                 XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1595                 XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1596             }
1597             XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
1598             XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
1599             XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
1600             XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
1601             if(s->mb_x+1 < s->mb_width){
1602                 XCHG(h->top_borders[top_idx][s->mb_x+1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
1603                 XCHG(h->top_borders[top_idx][s->mb_x+1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
1604             }
1605         } else {
1606             if(deblock_top){
1607                 if(deblock_topleft){
1608                     XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1609                     XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1610                 }
1611                 XCHG(top_border + (16 << pixel_shift), src_cb+1+pixel_shift, 1);
1612                 XCHG(top_border + (24 << pixel_shift), src_cr+1+pixel_shift, 1);
1613             }
1614         }
1615     }
1616 }
1617
1618 static av_always_inline int dctcoef_get(DCTELEM *mb, int high_bit_depth, int index) {
1619     if (high_bit_depth) {
1620         return AV_RN32A(((int32_t*)mb) + index);
1621     } else
1622         return AV_RN16A(mb + index);
1623 }
1624
1625 static av_always_inline void dctcoef_set(DCTELEM *mb, int high_bit_depth, int index, int value) {
1626     if (high_bit_depth) {
1627         AV_WN32A(((int32_t*)mb) + index, value);
1628     } else
1629         AV_WN16A(mb + index, value);
1630 }
1631
1632 static av_always_inline void hl_decode_mb_predict_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1633                                                        int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1634 {
1635     MpegEncContext * const s = &h->s;
1636     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1637     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1638     int i;
1639     int qscale = p == 0 ? s->qscale : h->chroma_qp[p-1];
1640     block_offset += 16*p;
1641     if(IS_INTRA4x4(mb_type)){
1642         if(simple || !s->encoding){
1643             if(IS_8x8DCT(mb_type)){
1644                 if(transform_bypass){
1645                     idct_dc_add =
1646                     idct_add    = s->dsp.add_pixels8;
1647                 }else{
1648                     idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1649                     idct_add    = h->h264dsp.h264_idct8_add;
1650                 }
1651                 for(i=0; i<16; i+=4){
1652                     uint8_t * const ptr= dest_y + block_offset[i];
1653                     const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1654                     if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1655                         h->hpc.pred8x8l_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1656                     }else{
1657                         const int nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1658                         h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1659                                                     (h->topright_samples_available<<i)&0x4000, linesize);
1660                         if(nnz){
1661                             if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1662                                 idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1663                             else
1664                                 idct_add   (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1665                         }
1666                     }
1667                 }
1668             }else{
1669                 if(transform_bypass){
1670                     idct_dc_add =
1671                     idct_add    = s->dsp.add_pixels4;
1672                 }else{
1673                     idct_dc_add = h->h264dsp.h264_idct_dc_add;
1674                     idct_add    = h->h264dsp.h264_idct_add;
1675                 }
1676                 for(i=0; i<16; i++){
1677                     uint8_t * const ptr= dest_y + block_offset[i];
1678                     const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1679
1680                     if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1681                         h->hpc.pred4x4_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1682                     }else{
1683                         uint8_t *topright;
1684                         int nnz, tr;
1685                         uint64_t tr_high;
1686                         if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1687                             const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1688                             assert(s->mb_y || linesize <= block_offset[i]);
1689                             if(!topright_avail){
1690                                 if (pixel_shift) {
1691                                     tr_high= ((uint16_t*)ptr)[3 - linesize/2]*0x0001000100010001ULL;
1692                                     topright= (uint8_t*) &tr_high;
1693                                 } else {
1694                                     tr= ptr[3 - linesize]*0x01010101;
1695                                     topright= (uint8_t*) &tr;
1696                                 }
1697                             }else
1698                                 topright= ptr + (4 << pixel_shift) - linesize;
1699                         }else
1700                             topright= NULL;
1701
1702                         h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1703                         nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1704                         if(nnz){
1705                             if(is_h264){
1706                                 if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1707                                     idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1708                                 else
1709                                     idct_add   (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1710                             }else
1711                                 ff_svq3_add_idct_c(ptr, h->mb + i*16+p*256, linesize, qscale, 0);
1712                         }
1713                     }
1714                 }
1715             }
1716         }
1717     }else{
1718         h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1719         if(is_h264){
1720             if(h->non_zero_count_cache[ scan8[LUMA_DC_BLOCK_INDEX+p] ]){
1721                 if(!transform_bypass)
1722                     h->h264dsp.h264_luma_dc_dequant_idct(h->mb+(p*256 << pixel_shift), h->mb_luma_dc[p], h->dequant4_coeff[p][qscale][0]);
1723                 else{
1724                     static const uint8_t dc_mapping[16] = { 0*16, 1*16, 4*16, 5*16, 2*16, 3*16, 6*16, 7*16,
1725                                                             8*16, 9*16,12*16,13*16,10*16,11*16,14*16,15*16};
1726                     for(i = 0; i < 16; i++)
1727                         dctcoef_set(h->mb+p*256, pixel_shift, dc_mapping[i], dctcoef_get(h->mb_luma_dc[p], pixel_shift, i));
1728                 }
1729             }
1730         }else
1731             ff_svq3_luma_dc_dequant_idct_c(h->mb+p*256, h->mb_luma_dc[p], qscale);
1732     }
1733 }
1734
1735 static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1736                                                     int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1737 {
1738     MpegEncContext * const s = &h->s;
1739     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1740     int i;
1741     block_offset += 16*p;
1742     if(!IS_INTRA4x4(mb_type)){
1743         if(is_h264){
1744             if(IS_INTRA16x16(mb_type)){
1745                 if(transform_bypass){
1746                     if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1747                         h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize);
1748                     }else{
1749                         for(i=0; i<16; i++){
1750                             if(h->non_zero_count_cache[ scan8[i+p*16] ] || dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1751                                 s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1752                         }
1753                     }
1754                 }else{
1755                     h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1756                 }
1757             }else if(h->cbp&15){
1758                 if(transform_bypass){
1759                     const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1760                     idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1761                     for(i=0; i<16; i+=di){
1762                         if(h->non_zero_count_cache[ scan8[i+p*16] ]){
1763                             idct_add(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1764                         }
1765                     }
1766                 }else{
1767                     if(IS_8x8DCT(mb_type)){
1768                         h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1769                     }else{
1770                         h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1771                     }
1772                 }
1773             }
1774         }else{
1775             for(i=0; i<16; i++){
1776                 if(h->non_zero_count_cache[ scan8[i+p*16] ] || h->mb[i*16+p*256]){ //FIXME benchmark weird rule, & below
1777                     uint8_t * const ptr= dest_y + block_offset[i];
1778                     ff_svq3_add_idct_c(ptr, h->mb + i*16 + p*256, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1779                 }
1780             }
1781         }
1782     }
1783 }
1784
1785 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, int pixel_shift){
1786     MpegEncContext * const s = &h->s;
1787     const int mb_x= s->mb_x;
1788     const int mb_y= s->mb_y;
1789     const int mb_xy= h->mb_xy;
1790     const int mb_type= s->current_picture.mb_type[mb_xy];
1791     uint8_t  *dest_y, *dest_cb, *dest_cr;
1792     int linesize, uvlinesize /*dct_offset*/;
1793     int i, j;
1794     int *block_offset = &h->block_offset[0];
1795     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1796     /* is_h264 should always be true if SVQ3 is disabled. */
1797     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1798     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1799
1800     dest_y  = s->current_picture.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize  ) * 16;
1801     dest_cb = s->current_picture.data[1] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * 8;
1802     dest_cr = s->current_picture.data[2] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * 8;
1803
1804     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
1805     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + (64 << pixel_shift), dest_cr - dest_cb, 2);
1806
1807     h->list_counts[mb_xy]= h->list_count;
1808
1809     if (!simple && MB_FIELD) {
1810         linesize   = h->mb_linesize   = s->linesize * 2;
1811         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1812         block_offset = &h->block_offset[48];
1813         if(mb_y&1){ //FIXME move out of this function?
1814             dest_y -= s->linesize*15;
1815             dest_cb-= s->uvlinesize*7;
1816             dest_cr-= s->uvlinesize*7;
1817         }
1818         if(FRAME_MBAFF) {
1819             int list;
1820             for(list=0; list<h->list_count; list++){
1821                 if(!USES_LIST(mb_type, list))
1822                     continue;
1823                 if(IS_16X16(mb_type)){
1824                     int8_t *ref = &h->ref_cache[list][scan8[0]];
1825                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1826                 }else{
1827                     for(i=0; i<16; i+=4){
1828                         int ref = h->ref_cache[list][scan8[i]];
1829                         if(ref >= 0)
1830                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1831                     }
1832                 }
1833             }
1834         }
1835     } else {
1836         linesize   = h->mb_linesize   = s->linesize;
1837         uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1838 //        dct_offset = s->linesize * 16;
1839     }
1840
1841     if (!simple && IS_INTRA_PCM(mb_type)) {
1842         if (pixel_shift) {
1843             const int bit_depth = h->sps.bit_depth_luma;
1844             int j;
1845             GetBitContext gb;
1846             init_get_bits(&gb, (uint8_t*)h->mb, 384*bit_depth);
1847
1848             for (i = 0; i < 16; i++) {
1849                 uint16_t *tmp_y  = (uint16_t*)(dest_y  + i*linesize);
1850                 for (j = 0; j < 16; j++)
1851                     tmp_y[j] = get_bits(&gb, bit_depth);
1852             }
1853             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1854                 if (!h->sps.chroma_format_idc) {
1855                     for (i = 0; i < 8; i++) {
1856                         uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1857                         for (j = 0; j < 8; j++) {
1858                             tmp_cb[j] = 1 << (bit_depth - 1);
1859                         }
1860                     }
1861                     for (i = 0; i < 8; i++) {
1862                         uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1863                         for (j = 0; j < 8; j++) {
1864                             tmp_cr[j] = 1 << (bit_depth - 1);
1865                         }
1866                     }
1867                 } else {
1868                     for (i = 0; i < 8; i++) {
1869                         uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1870                         for (j = 0; j < 8; j++)
1871                             tmp_cb[j] = get_bits(&gb, bit_depth);
1872                     }
1873                     for (i = 0; i < 8; i++) {
1874                         uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1875                         for (j = 0; j < 8; j++)
1876                             tmp_cr[j] = get_bits(&gb, bit_depth);
1877                     }
1878                 }
1879             }
1880         } else {
1881             for (i=0; i<16; i++) {
1882                 memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1883             }
1884             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1885                 if (!h->sps.chroma_format_idc) {
1886                     for (i = 0; i < 8; i++) {
1887                         memset(dest_cb + i*uvlinesize, 128, 8);
1888                         memset(dest_cr + i*uvlinesize, 128, 8);
1889                     }
1890                 } else {
1891                     for (i = 0; i < 8; i++) {
1892                         memcpy(dest_cb + i*uvlinesize, h->mb + 128 + i*4,  8);
1893                         memcpy(dest_cr + i*uvlinesize, h->mb + 160 + i*4,  8);
1894                     }
1895                 }
1896             }
1897         }
1898     } else {
1899         if(IS_INTRA(mb_type)){
1900             if(h->deblocking_filter)
1901                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, 0, simple, pixel_shift);
1902
1903             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1904                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1905                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1906             }
1907
1908             hl_decode_mb_predict_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
1909
1910             if(h->deblocking_filter)
1911                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, 0, simple, pixel_shift);
1912         }else if(is_h264){
1913             if (pixel_shift) {
1914                 hl_motion_16(h, dest_y, dest_cb, dest_cr,
1915                              s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1916                              s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1917                              h->h264dsp.weight_h264_pixels_tab,
1918                              h->h264dsp.biweight_h264_pixels_tab, 0);
1919             } else
1920                 hl_motion_8(h, dest_y, dest_cb, dest_cr,
1921                             s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1922                             s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1923                             h->h264dsp.weight_h264_pixels_tab,
1924                             h->h264dsp.biweight_h264_pixels_tab, 0);
1925         }
1926
1927         hl_decode_mb_idct_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
1928
1929         if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1930             uint8_t *dest[2] = {dest_cb, dest_cr};
1931             if(transform_bypass){
1932                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1933                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + (16*16*1 << pixel_shift), uvlinesize);
1934                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 32, h->mb + (16*16*2 << pixel_shift), uvlinesize);
1935                 }else{
1936                     idct_add = s->dsp.add_pixels4;
1937                     for(j=1; j<3; j++){
1938                         for(i=j*16; i<j*16+4; i++){
1939                             if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
1940                                 idct_add   (dest[j-1] + block_offset[i], h->mb + (i*16 << pixel_shift), uvlinesize);
1941                         }
1942                     }
1943                 }
1944             }else{
1945                 if(is_h264){
1946                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+0] ])
1947                         h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16*1 << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
1948                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+1] ])
1949                         h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16*2 << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
1950                     h->h264dsp.h264_idct_add8(dest, block_offset,
1951                                               h->mb, uvlinesize,
1952                                               h->non_zero_count_cache);
1953                 }else{
1954                     h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16*1, h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
1955                     h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16*2, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
1956                     for(j=1; j<3; j++){
1957                         for(i=j*16; i<j*16+4; i++){
1958                             if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1959                                 uint8_t * const ptr= dest[j-1] + block_offset[i];
1960                                 ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[0][s->qscale + 12] - 12, 2);
1961                             }
1962                         }
1963                     }
1964                 }
1965             }
1966         }
1967     }
1968     if(h->cbp || IS_INTRA(mb_type))
1969     {
1970         s->dsp.clear_blocks(h->mb);
1971         s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
1972     }
1973 }
1974
1975 static av_always_inline void hl_decode_mb_444_internal(H264Context *h, int simple, int pixel_shift){
1976     MpegEncContext * const s = &h->s;
1977     const int mb_x= s->mb_x;
1978     const int mb_y= s->mb_y;
1979     const int mb_xy= h->mb_xy;
1980     const int mb_type= s->current_picture.mb_type[mb_xy];
1981     uint8_t  *dest[3];
1982     int linesize;
1983     int i, j, p;
1984     int *block_offset = &h->block_offset[0];
1985     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1986     const int plane_count = (simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) ? 3 : 1;
1987
1988     for (p = 0; p < plane_count; p++)
1989     {
1990         dest[p] = s->current_picture.data[p] + ((mb_x << pixel_shift) + mb_y * s->linesize) * 16;
1991         s->dsp.prefetch(dest[p] + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
1992     }
1993
1994     h->list_counts[mb_xy]= h->list_count;
1995
1996     if (!simple && MB_FIELD) {
1997         linesize   = h->mb_linesize = h->mb_uvlinesize = s->linesize * 2;
1998         block_offset = &h->block_offset[48];
1999         if(mb_y&1) //FIXME move out of this function?
2000             for (p = 0; p < 3; p++)
2001                 dest[p] -= s->linesize*15;
2002         if(FRAME_MBAFF) {
2003             int list;
2004             for(list=0; list<h->list_count; list++){
2005                 if(!USES_LIST(mb_type, list))
2006                     continue;
2007                 if(IS_16X16(mb_type)){
2008                     int8_t *ref = &h->ref_cache[list][scan8[0]];
2009                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
2010                 }else{
2011                     for(i=0; i<16; i+=4){
2012                         int ref = h->ref_cache[list][scan8[i]];
2013                         if(ref >= 0)
2014                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
2015                     }
2016                 }
2017             }
2018         }
2019     } else {
2020         linesize   = h->mb_linesize = h->mb_uvlinesize = s->linesize;
2021     }
2022
2023     if (!simple && IS_INTRA_PCM(mb_type)) {
2024         if (pixel_shift) {
2025             const int bit_depth = h->sps.bit_depth_luma;
2026             GetBitContext gb;
2027             init_get_bits(&gb, (uint8_t*)h->mb, 768*bit_depth);
2028
2029             for (p = 0; p < plane_count; p++) {
2030                 for (i = 0; i < 16; i++) {
2031                     uint16_t *tmp = (uint16_t*)(dest[p] + i*linesize);
2032                     for (j = 0; j < 16; j++)
2033                         tmp[j] = get_bits(&gb, bit_depth);
2034                 }
2035             }
2036         } else {
2037             for (p = 0; p < plane_count; p++) {
2038                 for (i = 0; i < 16; i++) {
2039                     memcpy(dest[p] + i*linesize, h->mb + p*128 + i*8, 16);
2040                 }
2041             }
2042         }
2043     } else {
2044         if(IS_INTRA(mb_type)){
2045             if(h->deblocking_filter)
2046                 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 1, 1, simple, pixel_shift);
2047
2048             for (p = 0; p < plane_count; p++)
2049                 hl_decode_mb_predict_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2050
2051             if(h->deblocking_filter)
2052                 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 0, 1, simple, pixel_shift);
2053         }else{
2054             if (pixel_shift) {
2055                 hl_motion_16(h, dest[0], dest[1], dest[2],
2056                              s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2057                              s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2058                              h->h264dsp.weight_h264_pixels_tab,
2059                              h->h264dsp.biweight_h264_pixels_tab, 1);
2060             } else
2061                 hl_motion_8(h, dest[0], dest[1], dest[2],
2062                             s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2063                             s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2064                             h->h264dsp.weight_h264_pixels_tab,
2065                             h->h264dsp.biweight_h264_pixels_tab, 1);
2066         }
2067
2068         for (p = 0; p < plane_count; p++)
2069             hl_decode_mb_idct_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2070     }
2071     if(h->cbp || IS_INTRA(mb_type))
2072     {
2073         s->dsp.clear_blocks(h->mb);
2074         s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
2075     }
2076 }
2077
2078 /**
2079  * Process a macroblock; this case avoids checks for expensive uncommon cases.
2080  */
2081 #define hl_decode_mb_simple(sh, bits) \
2082 static void hl_decode_mb_simple_ ## bits(H264Context *h){ \
2083     hl_decode_mb_internal(h, 1, sh); \
2084 }
2085 hl_decode_mb_simple(0, 8);
2086 hl_decode_mb_simple(1, 16);
2087
2088 /**
2089  * Process a macroblock; this handles edge cases, such as interlacing.
2090  */
2091 static void av_noinline hl_decode_mb_complex(H264Context *h){
2092     hl_decode_mb_internal(h, 0, h->pixel_shift);
2093 }
2094
2095 static void av_noinline hl_decode_mb_444_complex(H264Context *h){
2096     hl_decode_mb_444_internal(h, 0, h->pixel_shift);
2097 }
2098
2099 static void av_noinline hl_decode_mb_444_simple(H264Context *h){
2100     hl_decode_mb_444_internal(h, 1, 0);
2101 }
2102
2103 void ff_h264_hl_decode_mb(H264Context *h){
2104     MpegEncContext * const s = &h->s;
2105     const int mb_xy= h->mb_xy;
2106     const int mb_type= s->current_picture.mb_type[mb_xy];
2107     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
2108
2109     if (CHROMA444) {
2110         if(is_complex || h->pixel_shift)
2111             hl_decode_mb_444_complex(h);
2112         else
2113             hl_decode_mb_444_simple(h);
2114     } else if (is_complex) {
2115         hl_decode_mb_complex(h);
2116     } else if (h->pixel_shift) {
2117         hl_decode_mb_simple_16(h);
2118     } else
2119         hl_decode_mb_simple_8(h);
2120 }
2121
2122 static int pred_weight_table(H264Context *h){
2123     MpegEncContext * const s = &h->s;
2124     int list, i;
2125     int luma_def, chroma_def;
2126
2127     h->use_weight= 0;
2128     h->use_weight_chroma= 0;
2129     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2130     if(h->sps.chroma_format_idc)
2131         h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2132     luma_def = 1<<h->luma_log2_weight_denom;
2133     chroma_def = 1<<h->chroma_log2_weight_denom;
2134
2135     for(list=0; list<2; list++){
2136         h->luma_weight_flag[list]   = 0;
2137         h->chroma_weight_flag[list] = 0;
2138         for(i=0; i<h->ref_count[list]; i++){
2139             int luma_weight_flag, chroma_weight_flag;
2140
2141             luma_weight_flag= get_bits1(&s->gb);
2142             if(luma_weight_flag){
2143                 h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
2144                 h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
2145                 if(   h->luma_weight[i][list][0] != luma_def
2146                    || h->luma_weight[i][list][1] != 0) {
2147                     h->use_weight= 1;
2148                     h->luma_weight_flag[list]= 1;
2149                 }
2150             }else{
2151                 h->luma_weight[i][list][0]= luma_def;
2152                 h->luma_weight[i][list][1]= 0;
2153             }
2154
2155             if(h->sps.chroma_format_idc){
2156                 chroma_weight_flag= get_bits1(&s->gb);
2157                 if(chroma_weight_flag){
2158                     int j;
2159                     for(j=0; j<2; j++){
2160                         h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
2161                         h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
2162                         if(   h->chroma_weight[i][list][j][0] != chroma_def
2163                            || h->chroma_weight[i][list][j][1] != 0) {
2164                             h->use_weight_chroma= 1;
2165                             h->chroma_weight_flag[list]= 1;
2166                         }
2167                     }
2168                 }else{
2169                     int j;
2170                     for(j=0; j<2; j++){
2171                         h->chroma_weight[i][list][j][0]= chroma_def;
2172                         h->chroma_weight[i][list][j][1]= 0;
2173                     }
2174                 }
2175             }
2176         }
2177         if(h->slice_type_nos != AV_PICTURE_TYPE_B) break;
2178     }
2179     h->use_weight= h->use_weight || h->use_weight_chroma;
2180     return 0;
2181 }
2182
2183 /**
2184  * Initialize implicit_weight table.
2185  * @param field  0/1 initialize the weight for interlaced MBAFF
2186  *                -1 initializes the rest
2187  */
2188 static void implicit_weight_table(H264Context *h, int field){
2189     MpegEncContext * const s = &h->s;
2190     int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
2191
2192     for (i = 0; i < 2; i++) {
2193         h->luma_weight_flag[i]   = 0;
2194         h->chroma_weight_flag[i] = 0;
2195     }
2196
2197     if(field < 0){
2198         cur_poc = s->current_picture_ptr->poc;
2199     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
2200        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
2201         h->use_weight= 0;
2202         h->use_weight_chroma= 0;
2203         return;
2204     }
2205         ref_start= 0;
2206         ref_count0= h->ref_count[0];
2207         ref_count1= h->ref_count[1];
2208     }else{
2209         cur_poc = s->current_picture_ptr->field_poc[field];
2210         ref_start= 16;
2211         ref_count0= 16+2*h->ref_count[0];
2212         ref_count1= 16+2*h->ref_count[1];
2213     }
2214
2215     h->use_weight= 2;
2216     h->use_weight_chroma= 2;
2217     h->luma_log2_weight_denom= 5;
2218     h->chroma_log2_weight_denom= 5;
2219
2220     for(ref0=ref_start; ref0 < ref_count0; ref0++){
2221         int poc0 = h->ref_list[0][ref0].poc;
2222         for(ref1=ref_start; ref1 < ref_count1; ref1++){
2223             int w = 32;
2224             if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
2225                 int poc1 = h->ref_list[1][ref1].poc;
2226                 int td = av_clip(poc1 - poc0, -128, 127);
2227                 if(td){
2228                     int tb = av_clip(cur_poc - poc0, -128, 127);
2229                     int tx = (16384 + (FFABS(td) >> 1)) / td;
2230                     int dist_scale_factor = (tb*tx + 32) >> 8;
2231                     if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
2232                         w = 64 - dist_scale_factor;
2233                 }
2234             }
2235             if(field<0){
2236                 h->implicit_weight[ref0][ref1][0]=
2237                 h->implicit_weight[ref0][ref1][1]= w;
2238             }else{
2239                 h->implicit_weight[ref0][ref1][field]=w;
2240             }
2241         }
2242     }
2243 }
2244
2245 /**
2246  * instantaneous decoder refresh.
2247  */
2248 static void idr(H264Context *h){
2249     ff_h264_remove_all_refs(h);
2250     h->prev_frame_num= 0;
2251     h->prev_frame_num_offset= 0;
2252     h->prev_poc_msb=
2253     h->prev_poc_lsb= 0;
2254 }
2255
2256 /* forget old pics after a seek */
2257 static void flush_dpb(AVCodecContext *avctx){
2258     H264Context *h= avctx->priv_data;
2259     int i;
2260     for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
2261         if(h->delayed_pic[i])
2262             h->delayed_pic[i]->reference= 0;
2263         h->delayed_pic[i]= NULL;
2264     }
2265     h->outputed_poc=h->next_outputed_poc= INT_MIN;
2266     h->prev_interlaced_frame = 1;
2267     idr(h);
2268     if(h->s.current_picture_ptr)
2269         h->s.current_picture_ptr->reference= 0;
2270     h->s.first_field= 0;
2271     ff_h264_reset_sei(h);
2272     ff_mpeg_flush(avctx);
2273 }
2274
2275 static int init_poc(H264Context *h){
2276     MpegEncContext * const s = &h->s;
2277     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2278     int field_poc[2];
2279     Picture *cur = s->current_picture_ptr;
2280
2281     h->frame_num_offset= h->prev_frame_num_offset;
2282     if(h->frame_num < h->prev_frame_num)
2283         h->frame_num_offset += max_frame_num;
2284
2285     if(h->sps.poc_type==0){
2286         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2287
2288         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2289             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2290         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2291             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2292         else
2293             h->poc_msb = h->prev_poc_msb;
2294 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2295         field_poc[0] =
2296         field_poc[1] = h->poc_msb + h->poc_lsb;
2297         if(s->picture_structure == PICT_FRAME)
2298             field_poc[1] += h->delta_poc_bottom;
2299     }else if(h->sps.poc_type==1){
2300         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2301         int i;
2302
2303         if(h->sps.poc_cycle_length != 0)
2304             abs_frame_num = h->frame_num_offset + h->frame_num;
2305         else
2306             abs_frame_num = 0;
2307
2308         if(h->nal_ref_idc==0 && abs_frame_num > 0)
2309             abs_frame_num--;
2310
2311         expected_delta_per_poc_cycle = 0;
2312         for(i=0; i < h->sps.poc_cycle_length; i++)
2313             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2314
2315         if(abs_frame_num > 0){
2316             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2317             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2318
2319             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2320             for(i = 0; i <= frame_num_in_poc_cycle; i++)
2321                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2322         } else
2323             expectedpoc = 0;
2324
2325         if(h->nal_ref_idc == 0)
2326             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2327
2328         field_poc[0] = expectedpoc + h->delta_poc[0];
2329         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2330
2331         if(s->picture_structure == PICT_FRAME)
2332             field_poc[1] += h->delta_poc[1];
2333     }else{
2334         int poc= 2*(h->frame_num_offset + h->frame_num);
2335
2336         if(!h->nal_ref_idc)
2337             poc--;
2338
2339         field_poc[0]= poc;
2340         field_poc[1]= poc;
2341     }
2342
2343     if(s->picture_structure != PICT_BOTTOM_FIELD)
2344         s->current_picture_ptr->field_poc[0]= field_poc[0];
2345     if(s->picture_structure != PICT_TOP_FIELD)
2346         s->current_picture_ptr->field_poc[1]= field_poc[1];
2347     cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
2348
2349     return 0;
2350 }
2351
2352
2353 /**
2354  * initialize scan tables
2355  */
2356 static void init_scan_tables(H264Context *h){
2357     int i;
2358     for(i=0; i<16; i++){
2359 #define T(x) (x>>2) | ((x<<2) & 0xF)
2360         h->zigzag_scan[i] = T(zigzag_scan[i]);
2361         h-> field_scan[i] = T( field_scan[i]);
2362 #undef T
2363     }
2364     for(i=0; i<64; i++){
2365 #define T(x) (x>>3) | ((x&7)<<3)
2366         h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
2367         h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
2368         h->field_scan8x8[i]        = T(field_scan8x8[i]);
2369         h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
2370 #undef T
2371     }
2372     if(h->sps.transform_bypass){ //FIXME same ugly
2373         h->zigzag_scan_q0          = zigzag_scan;
2374         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
2375         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
2376         h->field_scan_q0           = field_scan;
2377         h->field_scan8x8_q0        = field_scan8x8;
2378         h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
2379     }else{
2380         h->zigzag_scan_q0          = h->zigzag_scan;
2381         h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
2382         h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
2383         h->field_scan_q0           = h->field_scan;
2384         h->field_scan8x8_q0        = h->field_scan8x8;
2385         h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
2386     }
2387 }
2388
2389 static void field_end(H264Context *h, int in_setup){
2390     MpegEncContext * const s = &h->s;
2391     AVCodecContext * const avctx= s->avctx;
2392     s->mb_y= 0;
2393
2394     if (!in_setup && !s->dropable)
2395         ff_thread_report_progress((AVFrame*)s->current_picture_ptr, (16*s->mb_height >> FIELD_PICTURE) - 1,
2396                                  s->picture_structure==PICT_BOTTOM_FIELD);
2397
2398     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2399         ff_vdpau_h264_set_reference_frames(s);
2400
2401     if(in_setup || !(avctx->active_thread_type&FF_THREAD_FRAME)){
2402         if(!s->dropable) {
2403             ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
2404             h->prev_poc_msb= h->poc_msb;
2405             h->prev_poc_lsb= h->poc_lsb;
2406         }
2407         h->prev_frame_num_offset= h->frame_num_offset;
2408         h->prev_frame_num= h->frame_num;
2409         h->outputed_poc = h->next_outputed_poc;
2410     }
2411
2412     if (avctx->hwaccel) {
2413         if (avctx->hwaccel->end_frame(avctx) < 0)
2414             av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
2415     }
2416
2417     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2418         ff_vdpau_h264_picture_complete(s);
2419
2420     /*
2421      * FIXME: Error handling code does not seem to support interlaced
2422      * when slices span multiple rows
2423      * The ff_er_add_slice calls don't work right for bottom
2424      * fields; they cause massive erroneous error concealing
2425      * Error marking covers both fields (top and bottom).
2426      * This causes a mismatched s->error_count
2427      * and a bad error table. Further, the error count goes to
2428      * INT_MAX when called for bottom field, because mb_y is
2429      * past end by one (callers fault) and resync_mb_y != 0
2430      * causes problems for the first MB line, too.
2431      */
2432     if (!FIELD_PICTURE)
2433         ff_er_frame_end(s);
2434
2435     MPV_frame_end(s);
2436
2437     h->current_slice=0;
2438 }
2439
2440 /**
2441  * Replicate H264 "master" context to thread contexts.
2442  */
2443 static void clone_slice(H264Context *dst, H264Context *src)
2444 {
2445     memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
2446     dst->s.current_picture_ptr  = src->s.current_picture_ptr;
2447     dst->s.current_picture      = src->s.current_picture;
2448     dst->s.linesize             = src->s.linesize;
2449     dst->s.uvlinesize           = src->s.uvlinesize;
2450     dst->s.first_field          = src->s.first_field;
2451
2452     dst->prev_poc_msb           = src->prev_poc_msb;
2453     dst->prev_poc_lsb           = src->prev_poc_lsb;
2454     dst->prev_frame_num_offset  = src->prev_frame_num_offset;
2455     dst->prev_frame_num         = src->prev_frame_num;
2456     dst->short_ref_count        = src->short_ref_count;
2457
2458     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
2459     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
2460     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
2461     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
2462
2463     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
2464     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
2465 }
2466
2467 /**
2468  * computes profile from profile_idc and constraint_set?_flags
2469  *
2470  * @param sps SPS
2471  *
2472  * @return profile as defined by FF_PROFILE_H264_*
2473  */
2474 int ff_h264_get_profile(SPS *sps)
2475 {
2476     int profile = sps->profile_idc;
2477
2478     switch(sps->profile_idc) {
2479     case FF_PROFILE_H264_BASELINE:
2480         // constraint_set1_flag set to 1
2481         profile |= (sps->constraint_set_flags & 1<<1) ? FF_PROFILE_H264_CONSTRAINED : 0;
2482         break;
2483     case FF_PROFILE_H264_HIGH_10:
2484     case FF_PROFILE_H264_HIGH_422:
2485     case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
2486         // constraint_set3_flag set to 1
2487         profile |= (sps->constraint_set_flags & 1<<3) ? FF_PROFILE_H264_INTRA : 0;
2488         break;
2489     }
2490
2491     return profile;
2492 }
2493
2494 /**
2495  * decodes a slice header.
2496  * This will also call MPV_common_init() and frame_start() as needed.
2497  *
2498  * @param h h264context
2499  * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
2500  *
2501  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
2502  */
2503 static int decode_slice_header(H264Context *h, H264Context *h0){
2504     MpegEncContext * const s = &h->s;
2505     MpegEncContext * const s0 = &h0->s;
2506     unsigned int first_mb_in_slice;
2507     unsigned int pps_id;
2508     int num_ref_idx_active_override_flag;
2509     unsigned int slice_type, tmp, i, j;
2510     int default_ref_list_done = 0;
2511     int last_pic_structure;
2512
2513     s->dropable= h->nal_ref_idc == 0;
2514
2515     /* FIXME: 2tap qpel isn't implemented for high bit depth. */
2516     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc && !h->pixel_shift){
2517         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
2518         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
2519     }else{
2520         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
2521         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
2522     }
2523
2524     first_mb_in_slice= get_ue_golomb(&s->gb);
2525
2526     if(first_mb_in_slice == 0){ //FIXME better field boundary detection
2527         if(h0->current_slice && FIELD_PICTURE){
2528             field_end(h, 1);
2529         }
2530
2531         h0->current_slice = 0;
2532         if (!s0->first_field)
2533             s->current_picture_ptr= NULL;
2534     }
2535
2536     slice_type= get_ue_golomb_31(&s->gb);
2537     if(slice_type > 9){
2538         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);
2539         return -1;
2540     }
2541     if(slice_type > 4){
2542         slice_type -= 5;
2543         h->slice_type_fixed=1;
2544     }else
2545         h->slice_type_fixed=0;
2546
2547     slice_type= golomb_to_pict_type[ slice_type ];
2548     if (slice_type == AV_PICTURE_TYPE_I
2549         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
2550         default_ref_list_done = 1;
2551     }
2552     h->slice_type= slice_type;
2553     h->slice_type_nos= slice_type & 3;
2554
2555     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
2556
2557     pps_id= get_ue_golomb(&s->gb);
2558     if(pps_id>=MAX_PPS_COUNT){
2559         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
2560         return -1;
2561     }
2562     if(!h0->pps_buffers[pps_id]) {
2563         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
2564         return -1;
2565     }
2566     h->pps= *h0->pps_buffers[pps_id];
2567
2568     if(!h0->sps_buffers[h->pps.sps_id]) {
2569         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
2570         return -1;
2571     }
2572     h->sps = *h0->sps_buffers[h->pps.sps_id];
2573
2574     s->avctx->profile = ff_h264_get_profile(&h->sps);
2575     s->avctx->level   = h->sps.level_idc;
2576     s->avctx->refs    = h->sps.ref_frame_count;
2577
2578     if(h == h0 && h->dequant_coeff_pps != pps_id){
2579         h->dequant_coeff_pps = pps_id;
2580         init_dequant_tables(h);
2581     }
2582
2583     s->mb_width= h->sps.mb_width;
2584     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
2585
2586     h->b_stride=  s->mb_width*4;
2587
2588     s->width = 16*s->mb_width - (2>>CHROMA444)*FFMIN(h->sps.crop_right, (8<<CHROMA444)-1);
2589     if(h->sps.frame_mbs_only_flag)
2590         s->height= 16*s->mb_height - (2>>CHROMA444)*FFMIN(h->sps.crop_bottom, (8<<CHROMA444)-1);
2591     else
2592         s->height= 16*s->mb_height - (4>>CHROMA444)*FFMIN(h->sps.crop_bottom, (8<<CHROMA444)-1);
2593
2594     if (s->context_initialized
2595         && (   s->width != s->avctx->width || s->height != s->avctx->height
2596             || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
2597         if(h != h0) {
2598             av_log_missing_feature(s->avctx, "Width/height changing with threads is", 0);
2599             return -1;   // width / height changed during parallelized decoding
2600         }
2601         free_tables(h, 0);
2602         flush_dpb(s->avctx);
2603         MPV_common_end(s);
2604     }
2605     if (!s->context_initialized) {
2606         if (h != h0) {
2607             av_log(h->s.avctx, AV_LOG_ERROR, "Cannot (re-)initialize context during parallel decoding.\n");
2608             return -1;
2609         }
2610
2611         avcodec_set_dimensions(s->avctx, s->width, s->height);
2612         s->avctx->sample_aspect_ratio= h->sps.sar;
2613         av_assert0(s->avctx->sample_aspect_ratio.den);
2614
2615         h->s.avctx->coded_width = 16*s->mb_width;
2616         h->s.avctx->coded_height = 16*s->mb_height;
2617
2618         if(h->sps.video_signal_type_present_flag){
2619             s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
2620             if(h->sps.colour_description_present_flag){
2621                 s->avctx->color_primaries = h->sps.color_primaries;
2622                 s->avctx->color_trc       = h->sps.color_trc;
2623                 s->avctx->colorspace      = h->sps.colorspace;
2624             }
2625         }
2626
2627         if(h->sps.timing_info_present_flag){
2628             int64_t den= h->sps.time_scale;
2629             if(h->x264_build < 44U)
2630                 den *= 2;
2631             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
2632                       h->sps.num_units_in_tick, den, 1<<30);
2633         }
2634
2635         switch (h->sps.bit_depth_luma) {
2636             case 9 :
2637                 s->avctx->pix_fmt = CHROMA444 ? PIX_FMT_YUV444P9 : PIX_FMT_YUV420P9;
2638                 break;
2639             case 10 :
2640                 s->avctx->pix_fmt = CHROMA444 ? PIX_FMT_YUV444P10 : PIX_FMT_YUV420P10;
2641                 break;
2642             default:
2643                 if (CHROMA444){
2644                     s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ444P : PIX_FMT_YUV444P;
2645                 }else{
2646                     s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
2647                                                              s->avctx->codec->pix_fmts ?
2648                                                              s->avctx->codec->pix_fmts :
2649                                                              s->avctx->color_range == AVCOL_RANGE_JPEG ?
2650                                                              hwaccel_pixfmt_list_h264_jpeg_420 :
2651                                                              ff_hwaccel_pixfmt_list_420);
2652                 }
2653         }
2654
2655         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
2656
2657         if (MPV_common_init(s) < 0) {
2658             av_log(h->s.avctx, AV_LOG_ERROR, "MPV_common_init() failed.\n");
2659             return -1;
2660         }
2661         s->first_field = 0;
2662         h->prev_interlaced_frame = 1;
2663
2664         init_scan_tables(h);
2665         if (ff_h264_alloc_tables(h) < 0) {
2666             av_log(h->s.avctx, AV_LOG_ERROR, "Could not allocate memory for h264\n");
2667             return AVERROR(ENOMEM);
2668         }
2669
2670         if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_SLICE)) {
2671             if (context_init(h) < 0) {
2672                 av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2673                 return -1;
2674             }
2675         } else {
2676             for(i = 1; i < s->avctx->thread_count; i++) {
2677                 H264Context *c;
2678                 c = h->thread_context[i] = av_malloc(sizeof(H264Context));
2679                 memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
2680                 memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
2681                 c->h264dsp = h->h264dsp;
2682                 c->sps = h->sps;
2683                 c->pps = h->pps;
2684                 c->pixel_shift = h->pixel_shift;
2685                 init_scan_tables(c);
2686                 clone_tables(c, h, i);
2687             }
2688
2689             for(i = 0; i < s->avctx->thread_count; i++)
2690                 if (context_init(h->thread_context[i]) < 0) {
2691                     av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2692                     return -1;
2693                 }
2694         }
2695     }
2696
2697     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
2698
2699     h->mb_mbaff = 0;
2700     h->mb_aff_frame = 0;
2701     last_pic_structure = s0->picture_structure;
2702     if(h->sps.frame_mbs_only_flag){
2703         s->picture_structure= PICT_FRAME;
2704     }else{
2705         if(get_bits1(&s->gb)) { //field_pic_flag
2706             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
2707         } else {
2708             s->picture_structure= PICT_FRAME;
2709             h->mb_aff_frame = h->sps.mb_aff;
2710         }
2711     }
2712     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
2713
2714     if(h0->current_slice == 0){
2715         // Shorten frame num gaps so we don't have to allocate reference frames just to throw them away
2716         if(h->frame_num != h->prev_frame_num) {
2717             int unwrap_prev_frame_num = h->prev_frame_num, max_frame_num = 1<<h->sps.log2_max_frame_num;
2718
2719             if (unwrap_prev_frame_num > h->frame_num) unwrap_prev_frame_num -= max_frame_num;
2720
2721             if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
2722                 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
2723                 if (unwrap_prev_frame_num < 0)
2724                     unwrap_prev_frame_num += max_frame_num;
2725
2726                 h->prev_frame_num = unwrap_prev_frame_num;
2727             }
2728         }
2729
2730         while(h->frame_num !=  h->prev_frame_num &&
2731               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
2732             Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
2733             av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
2734             if (ff_h264_frame_start(h) < 0)
2735                 return -1;
2736             h->prev_frame_num++;
2737             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
2738             s->current_picture_ptr->frame_num= h->prev_frame_num;
2739             ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 0);
2740             ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 1);
2741             ff_generate_sliding_window_mmcos(h);
2742             ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
2743             /* Error concealment: if a ref is missing, copy the previous ref in its place.
2744              * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
2745              * about there being no actual duplicates.
2746              * FIXME: this doesn't copy padding for out-of-frame motion vectors.  Given we're
2747              * concealing a lost frame, this probably isn't noticable by comparison, but it should
2748              * be fixed. */
2749             if (h->short_ref_count) {
2750                 if (prev) {
2751                     av_image_copy(h->short_ref[0]->data, h->short_ref[0]->linesize,
2752                                   (const uint8_t**)prev->data, prev->linesize,
2753                                   s->avctx->pix_fmt, s->mb_width*16, s->mb_height*16);
2754                     h->short_ref[0]->poc = prev->poc+2;
2755                 }
2756                 h->short_ref[0]->frame_num = h->prev_frame_num;
2757             }
2758         }
2759
2760         /* See if we have a decoded first field looking for a pair... */
2761         if (s0->first_field) {
2762             assert(s0->current_picture_ptr);
2763             assert(s0->current_picture_ptr->data[0]);
2764             assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
2765
2766             /* figure out if we have a complementary field pair */
2767             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2768                 /*
2769                  * Previous field is unmatched. Don't display it, but let it
2770                  * remain for reference if marked as such.
2771                  */
2772                 s0->current_picture_ptr = NULL;
2773                 s0->first_field = FIELD_PICTURE;
2774
2775             } else {
2776                 if (h->nal_ref_idc &&
2777                         s0->current_picture_ptr->reference &&
2778                         s0->current_picture_ptr->frame_num != h->frame_num) {
2779                     /*
2780                      * This and previous field were reference, but had
2781                      * different frame_nums. Consider this field first in
2782                      * pair. Throw away previous field except for reference
2783                      * purposes.
2784                      */
2785                     s0->first_field = 1;
2786                     s0->current_picture_ptr = NULL;
2787
2788                 } else {
2789                     /* Second field in complementary pair */
2790                     s0->first_field = 0;
2791                 }
2792             }
2793
2794         } else {
2795             /* Frame or first field in a potentially complementary pair */
2796             assert(!s0->current_picture_ptr);
2797             s0->first_field = FIELD_PICTURE;
2798         }
2799
2800         if(!FIELD_PICTURE || s0->first_field) {
2801             if (ff_h264_frame_start(h) < 0) {
2802                 s0->first_field = 0;
2803                 return -1;
2804             }
2805         } else {
2806             ff_release_unused_pictures(s, 0);
2807         }
2808     }
2809     if(h != h0)
2810         clone_slice(h, h0);
2811
2812     s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
2813
2814     assert(s->mb_num == s->mb_width * s->mb_height);
2815     if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
2816        first_mb_in_slice                    >= s->mb_num){
2817         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
2818         return -1;
2819     }
2820     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
2821     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
2822     if (s->picture_structure == PICT_BOTTOM_FIELD)
2823         s->resync_mb_y = s->mb_y = s->mb_y + 1;
2824     assert(s->mb_y < s->mb_height);
2825
2826     if(s->picture_structure==PICT_FRAME){
2827         h->curr_pic_num=   h->frame_num;
2828         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
2829     }else{
2830         h->curr_pic_num= 2*h->frame_num + 1;
2831         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
2832     }
2833
2834     if(h->nal_unit_type == NAL_IDR_SLICE){
2835         get_ue_golomb(&s->gb); /* idr_pic_id */
2836     }
2837
2838     if(h->sps.poc_type==0){
2839         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2840
2841         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
2842             h->delta_poc_bottom= get_se_golomb(&s->gb);
2843         }
2844     }
2845
2846     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
2847         h->delta_poc[0]= get_se_golomb(&s->gb);
2848
2849         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2850             h->delta_poc[1]= get_se_golomb(&s->gb);
2851     }
2852
2853     init_poc(h);
2854
2855     if(h->pps.redundant_pic_cnt_present){
2856         h->redundant_pic_count= get_ue_golomb(&s->gb);
2857     }
2858
2859     //set defaults, might be overridden a few lines later
2860     h->ref_count[0]= h->pps.ref_count[0];
2861     h->ref_count[1]= h->pps.ref_count[1];
2862
2863     if(h->slice_type_nos != AV_PICTURE_TYPE_I){
2864         if(h->slice_type_nos == AV_PICTURE_TYPE_B){
2865             h->direct_spatial_mv_pred= get_bits1(&s->gb);
2866         }
2867         num_ref_idx_active_override_flag= get_bits1(&s->gb);
2868
2869         if(num_ref_idx_active_override_flag){
2870             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2871             if(h->slice_type_nos==AV_PICTURE_TYPE_B)
2872                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2873
2874             if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
2875                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
2876                 h->ref_count[0]= h->ref_count[1]= 1;
2877                 return -1;
2878             }
2879         }
2880         if(h->slice_type_nos == AV_PICTURE_TYPE_B)
2881             h->list_count= 2;
2882         else
2883             h->list_count= 1;
2884     }else
2885         h->list_count= 0;
2886
2887     if(!default_ref_list_done){
2888         ff_h264_fill_default_ref_list(h);
2889     }
2890
2891     if(h->slice_type_nos!=AV_PICTURE_TYPE_I && ff_h264_decode_ref_pic_list_reordering(h) < 0)
2892         return -1;
2893
2894     if(h->slice_type_nos!=AV_PICTURE_TYPE_I){
2895         s->last_picture_ptr= &h->ref_list[0][0];
2896         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
2897     }
2898     if(h->slice_type_nos==AV_PICTURE_TYPE_B){
2899         s->next_picture_ptr= &h->ref_list[1][0];
2900         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
2901     }
2902
2903     if(   (h->pps.weighted_pred          && h->slice_type_nos == AV_PICTURE_TYPE_P )
2904        ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== AV_PICTURE_TYPE_B ) )
2905         pred_weight_table(h);
2906     else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
2907         implicit_weight_table(h, -1);
2908     }else {
2909         h->use_weight = 0;
2910         for (i = 0; i < 2; i++) {
2911             h->luma_weight_flag[i]   = 0;
2912             h->chroma_weight_flag[i] = 0;
2913         }
2914     }
2915
2916     if(h->nal_ref_idc)
2917         ff_h264_decode_ref_pic_marking(h0, &s->gb);
2918
2919     if(FRAME_MBAFF){
2920         ff_h264_fill_mbaff_ref_list(h);
2921
2922         if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
2923             implicit_weight_table(h, 0);
2924             implicit_weight_table(h, 1);
2925         }
2926     }
2927
2928     if(h->slice_type_nos==AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
2929         ff_h264_direct_dist_scale_factor(h);
2930     ff_h264_direct_ref_list_init(h);
2931
2932     if( h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac ){
2933         tmp = get_ue_golomb_31(&s->gb);
2934         if(tmp > 2){
2935             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2936             return -1;
2937         }
2938         h->cabac_init_idc= tmp;
2939     }
2940
2941     h->last_qscale_diff = 0;
2942     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2943     if(tmp>51+6*(h->sps.bit_depth_luma-8)){
2944         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2945         return -1;
2946     }
2947     s->qscale= tmp;
2948     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2949     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2950     //FIXME qscale / qp ... stuff
2951     if(h->slice_type == AV_PICTURE_TYPE_SP){
2952         get_bits1(&s->gb); /* sp_for_switch_flag */
2953     }
2954     if(h->slice_type==AV_PICTURE_TYPE_SP || h->slice_type == AV_PICTURE_TYPE_SI){
2955         get_se_golomb(&s->gb); /* slice_qs_delta */
2956     }
2957
2958     h->deblocking_filter = 1;
2959     h->slice_alpha_c0_offset = 52;
2960     h->slice_beta_offset = 52;
2961     if( h->pps.deblocking_filter_parameters_present ) {
2962         tmp= get_ue_golomb_31(&s->gb);
2963         if(tmp > 2){
2964             av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2965             return -1;
2966         }
2967         h->deblocking_filter= tmp;
2968         if(h->deblocking_filter < 2)
2969             h->deblocking_filter^= 1; // 1<->0
2970
2971         if( h->deblocking_filter ) {
2972             h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2973             h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
2974             if(   h->slice_alpha_c0_offset > 104U
2975                || h->slice_beta_offset     > 104U){
2976                 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);
2977                 return -1;
2978             }
2979         }
2980     }
2981
2982     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
2983        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != AV_PICTURE_TYPE_I)
2984        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == AV_PICTURE_TYPE_B)
2985        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2986         h->deblocking_filter= 0;
2987
2988     if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2989         if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2990             /* Cheat slightly for speed:
2991                Do not bother to deblock across slices. */
2992             h->deblocking_filter = 2;
2993         } else {
2994             h0->max_contexts = 1;
2995             if(!h0->single_decode_warning) {
2996                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2997                 h0->single_decode_warning = 1;
2998             }
2999             if (h != h0) {
3000                 av_log(h->s.avctx, AV_LOG_ERROR, "Deblocking switched inside frame.\n");
3001                 return 1;
3002             }
3003         }
3004     }
3005     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]);
3006
3007 #if 0 //FMO
3008     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3009         slice_group_change_cycle= get_bits(&s->gb, ?);
3010 #endif
3011
3012     h0->last_slice_type = slice_type;
3013     h->slice_num = ++h0->current_slice;
3014     if(h->slice_num >= MAX_SLICES){
3015         av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
3016     }
3017
3018     for(j=0; j<2; j++){
3019         int id_list[16];
3020         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
3021         for(i=0; i<16; i++){
3022             id_list[i]= 60;
3023             if(h->ref_list[j][i].data[0]){
3024                 int k;
3025                 uint8_t *base= h->ref_list[j][i].base[0];
3026                 for(k=0; k<h->short_ref_count; k++)
3027                     if(h->short_ref[k]->base[0] == base){
3028                         id_list[i]= k;
3029                         break;
3030                     }
3031                 for(k=0; k<h->long_ref_count; k++)
3032                     if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
3033                         id_list[i]= h->short_ref_count + k;
3034                         break;
3035                     }
3036             }
3037         }
3038
3039         ref2frm[0]=
3040         ref2frm[1]= -1;
3041         for(i=0; i<16; i++)
3042             ref2frm[i+2]= 4*id_list[i]
3043                           +(h->ref_list[j][i].reference&3);
3044         ref2frm[18+0]=
3045         ref2frm[18+1]= -1;
3046         for(i=16; i<48; i++)
3047             ref2frm[i+4]= 4*id_list[(i-16)>>1]
3048                           +(h->ref_list[j][i].reference&3);
3049     }
3050
3051     //FIXME: fix draw_edges+PAFF+frame threads
3052     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE || (!h->sps.frame_mbs_only_flag && s->avctx->active_thread_type)) ? 0 : 16;
3053     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
3054
3055     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3056         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",
3057                h->slice_num,
3058                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
3059                first_mb_in_slice,
3060                av_get_picture_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
3061                pps_id, h->frame_num,
3062                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3063                h->ref_count[0], h->ref_count[1],
3064                s->qscale,
3065                h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
3066                h->use_weight,
3067                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
3068                h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
3069                );
3070     }
3071
3072     return 0;
3073 }
3074
3075 int ff_h264_get_slice_type(const H264Context *h)
3076 {
3077     switch (h->slice_type) {
3078     case AV_PICTURE_TYPE_P:  return 0;
3079     case AV_PICTURE_TYPE_B:  return 1;
3080     case AV_PICTURE_TYPE_I:  return 2;
3081     case AV_PICTURE_TYPE_SP: return 3;
3082     case AV_PICTURE_TYPE_SI: return 4;
3083     default:         return -1;
3084     }
3085 }
3086
3087 /**
3088  *
3089  * @return non zero if the loop filter can be skiped
3090  */
3091 static int fill_filter_caches(H264Context *h, int mb_type){
3092     MpegEncContext * const s = &h->s;
3093     const int mb_xy= h->mb_xy;
3094     int top_xy, left_xy[2];
3095     int top_type, left_type[2];
3096
3097     top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
3098
3099     //FIXME deblocking could skip the intra and nnz parts.
3100
3101     /* Wow, what a mess, why didn't they simplify the interlacing & intra
3102      * stuff, I can't imagine that these complex rules are worth it. */
3103
3104     left_xy[1] = left_xy[0] = mb_xy-1;
3105     if(FRAME_MBAFF){
3106         const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
3107         const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
3108         if(s->mb_y&1){
3109             if (left_mb_field_flag != curr_mb_field_flag) {
3110                 left_xy[0] -= s->mb_stride;
3111             }
3112         }else{
3113             if(curr_mb_field_flag){
3114                 top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
3115             }
3116             if (left_mb_field_flag != curr_mb_field_flag) {
3117                 left_xy[1] += s->mb_stride;
3118             }
3119         }
3120     }
3121
3122     h->top_mb_xy = top_xy;
3123     h->left_mb_xy[0] = left_xy[0];
3124     h->left_mb_xy[1] = left_xy[1];
3125     {
3126         //for sufficiently low qp, filtering wouldn't do anything
3127         //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
3128         int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
3129         int qp = s->current_picture.qscale_table[mb_xy];
3130         if(qp <= qp_thresh
3131            && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
3132            && (top_xy   < 0 || ((qp + s->current_picture.qscale_table[top_xy    ] + 1)>>1) <= qp_thresh)){
3133             if(!FRAME_MBAFF)
3134                 return 1;
3135             if(   (left_xy[0]< 0            || ((qp + s->current_picture.qscale_table[left_xy[1]             ] + 1)>>1) <= qp_thresh)
3136                && (top_xy    < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy    -s->mb_stride] + 1)>>1) <= qp_thresh))
3137                 return 1;
3138         }
3139     }
3140
3141     top_type     = s->current_picture.mb_type[top_xy]    ;
3142     left_type[0] = s->current_picture.mb_type[left_xy[0]];
3143     left_type[1] = s->current_picture.mb_type[left_xy[1]];
3144     if(h->deblocking_filter == 2){
3145         if(h->slice_table[top_xy     ] != h->slice_num) top_type= 0;
3146         if(h->slice_table[left_xy[0] ] != h->slice_num) left_type[0]= left_type[1]= 0;
3147     }else{
3148         if(h->slice_table[top_xy     ] == 0xFFFF) top_type= 0;
3149         if(h->slice_table[left_xy[0] ] == 0xFFFF) left_type[0]= left_type[1] =0;
3150     }
3151     h->top_type    = top_type    ;
3152     h->left_type[0]= left_type[0];
3153     h->left_type[1]= left_type[1];
3154
3155     if(IS_INTRA(mb_type))
3156         return 0;
3157
3158     AV_COPY32(&h->non_zero_count_cache[4+8* 1], &h->non_zero_count[mb_xy][ 0]);
3159     AV_COPY32(&h->non_zero_count_cache[4+8* 2], &h->non_zero_count[mb_xy][ 4]);
3160     AV_COPY32(&h->non_zero_count_cache[4+8* 3], &h->non_zero_count[mb_xy][ 8]);
3161     AV_COPY32(&h->non_zero_count_cache[4+8* 4], &h->non_zero_count[mb_xy][12]);
3162
3163     h->cbp= h->cbp_table[mb_xy];
3164
3165     {
3166         int list;
3167         for(list=0; list<h->list_count; list++){
3168             int8_t *ref;
3169             int y, b_stride;
3170             int16_t (*mv_dst)[2];
3171             int16_t (*mv_src)[2];
3172
3173             if(!USES_LIST(mb_type, list)){
3174                 fill_rectangle(  h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
3175                 AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3176                 AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3177                 AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3178                 AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3179                 continue;
3180             }
3181
3182             ref = &s->current_picture.ref_index[list][4*mb_xy];
3183             {
3184                 int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3185                 AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
3186                 AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
3187                 ref += 2;
3188                 AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
3189                 AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
3190             }
3191
3192             b_stride = h->b_stride;
3193             mv_dst   = &h->mv_cache[list][scan8[0]];
3194             mv_src   = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
3195             for(y=0; y<4; y++){
3196                 AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
3197             }
3198
3199         }
3200     }
3201
3202
3203 /*
3204 0 . T T. T T T T
3205 1 L . .L . . . .
3206 2 L . .L . . . .
3207 3 . T TL . . . .
3208 4 L . .L . . . .
3209 5 L . .. . . . .
3210 */
3211 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
3212     if(top_type){
3213         AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][3*4]);
3214     }
3215
3216     if(left_type[0]){
3217         h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][3+0*4];
3218         h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][3+1*4];
3219         h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][3+2*4];
3220         h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][3+3*4];
3221     }
3222
3223     // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
3224     if(!CABAC && h->pps.transform_8x8_mode){
3225         if(IS_8x8DCT(top_type)){
3226             h->non_zero_count_cache[4+8*0]=
3227             h->non_zero_count_cache[5+8*0]= (h->cbp_table[top_xy] & 0x4000) >> 12;
3228             h->non_zero_count_cache[6+8*0]=
3229             h->non_zero_count_cache[7+8*0]= (h->cbp_table[top_xy] & 0x8000) >> 12;
3230         }
3231         if(IS_8x8DCT(left_type[0])){
3232             h->non_zero_count_cache[3+8*1]=
3233             h->non_zero_count_cache[3+8*2]= (h->cbp_table[left_xy[0]]&0x2000) >> 12; //FIXME check MBAFF
3234         }
3235         if(IS_8x8DCT(left_type[1])){
3236             h->non_zero_count_cache[3+8*3]=
3237             h->non_zero_count_cache[3+8*4]= (h->cbp_table[left_xy[1]]&0x8000) >> 12; //FIXME check MBAFF
3238         }
3239
3240         if(IS_8x8DCT(mb_type)){
3241             h->non_zero_count_cache[scan8[0   ]]= h->non_zero_count_cache[scan8[1   ]]=
3242             h->non_zero_count_cache[scan8[2   ]]= h->non_zero_count_cache[scan8[3   ]]= (h->cbp & 0x1000) >> 12;
3243
3244             h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
3245             h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= (h->cbp & 0x2000) >> 12;
3246
3247             h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
3248             h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= (h->cbp & 0x4000) >> 12;
3249
3250             h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
3251             h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= (h->cbp & 0x8000) >> 12;
3252         }
3253     }
3254
3255     if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
3256         int list;
3257         for(list=0; list<h->list_count; list++){
3258             if(USES_LIST(top_type, list)){
3259                 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
3260                 const int b8_xy= 4*top_xy + 2;
3261                 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3262                 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
3263                 h->ref_cache[list][scan8[0] + 0 - 1*8]=
3264                 h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
3265                 h->ref_cache[list][scan8[0] + 2 - 1*8]=
3266                 h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
3267             }else{
3268                 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
3269                 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3270             }
3271
3272             if(!IS_INTERLACED(mb_type^left_type[0])){
3273                 if(USES_LIST(left_type[0], list)){
3274                     const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
3275                     const int b8_xy= 4*left_xy[0] + 1;
3276                     int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3277                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
3278                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
3279                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
3280                     AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
3281                     h->ref_cache[list][scan8[0] - 1 + 0 ]=
3282                     h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]];
3283                     h->ref_cache[list][scan8[0] - 1 +16 ]=
3284                     h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]];
3285                 }else{
3286                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
3287                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
3288                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
3289                     AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
3290                     h->ref_cache[list][scan8[0] - 1 + 0  ]=
3291                     h->ref_cache[list][scan8[0] - 1 + 8  ]=
3292                     h->ref_cache[list][scan8[0] - 1 + 16 ]=
3293                     h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
3294                 }
3295             }
3296         }
3297     }
3298
3299     return 0;
3300 }
3301
3302 static void loop_filter(H264Context *h, int start_x, int end_x){
3303     MpegEncContext * const s = &h->s;
3304     uint8_t  *dest_y, *dest_cb, *dest_cr;
3305     int linesize, uvlinesize, mb_x, mb_y;
3306     const int end_mb_y= s->mb_y + FRAME_MBAFF;
3307     const int old_slice_type= h->slice_type;
3308     const int pixel_shift = h->pixel_shift;
3309
3310     if(h->deblocking_filter) {
3311         for(mb_x= start_x; mb_x<end_x; mb_x++){
3312             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
3313                 int mb_xy, mb_type;
3314                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
3315                 h->slice_num= h->slice_table[mb_xy];
3316                 mb_type= s->current_picture.mb_type[mb_xy];
3317                 h->list_count= h->list_counts[mb_xy];
3318
3319                 if(FRAME_MBAFF)
3320                     h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
3321
3322                 s->mb_x= mb_x;
3323                 s->mb_y= mb_y;
3324                 dest_y  = s->current_picture.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize  ) * 16;
3325                 dest_cb = s->current_picture.data[1] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * (8 << CHROMA444);
3326                 dest_cr = s->current_picture.data[2] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * (8 << CHROMA444);
3327                     //FIXME simplify above
3328
3329                 if (MB_FIELD) {
3330                     linesize   = h->mb_linesize   = s->linesize * 2;
3331                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
3332                     if(mb_y&1){ //FIXME move out of this function?
3333                         dest_y -= s->linesize*15;
3334                         dest_cb-= s->uvlinesize*((8 << CHROMA444)-1);
3335                         dest_cr-= s->uvlinesize*((8 << CHROMA444)-1);
3336                     }
3337                 } else {
3338                     linesize   = h->mb_linesize   = s->linesize;
3339                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
3340                 }
3341                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, CHROMA444, 0);
3342                 if(fill_filter_caches(h, mb_type))
3343                     continue;
3344                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
3345                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
3346
3347                 if (FRAME_MBAFF) {
3348                     ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3349                 } else {
3350                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3351                 }
3352             }
3353         }
3354     }
3355     h->slice_type= old_slice_type;
3356     s->mb_x= end_x;
3357     s->mb_y= end_mb_y - FRAME_MBAFF;
3358     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3359     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3360 }
3361
3362 static void predict_field_decoding_flag(H264Context *h){
3363     MpegEncContext * const s = &h->s;
3364     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3365     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
3366                 ? s->current_picture.mb_type[mb_xy-1]
3367                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
3368                 ? s->current_picture.mb_type[mb_xy-s->mb_stride]
3369                 : 0;
3370     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
3371 }
3372
3373 /**
3374  * Draw edges and report progress for the last MB row.
3375  */
3376 static void decode_finish_row(H264Context *h){
3377     MpegEncContext * const s = &h->s;
3378     int top = 16*(s->mb_y >> FIELD_PICTURE);
3379     int height = 16 << FRAME_MBAFF;
3380     int deblock_border = (16 + 4) << FRAME_MBAFF;
3381     int pic_height = 16*s->mb_height >> FIELD_PICTURE;
3382
3383     if (h->deblocking_filter) {
3384         if((top + height) >= pic_height)
3385             height += deblock_border;
3386
3387         top -= deblock_border;
3388     }
3389
3390     if (top >= pic_height || (top + height) < h->emu_edge_height)
3391         return;
3392
3393     height = FFMIN(height, pic_height - top);
3394     if (top < h->emu_edge_height) {
3395         height = top+height;
3396         top = 0;
3397     }
3398
3399     ff_draw_horiz_band(s, top, height);
3400
3401     if (s->dropable) return;
3402
3403     ff_thread_report_progress((AVFrame*)s->current_picture_ptr, top + height - 1,
3404                              s->picture_structure==PICT_BOTTOM_FIELD);
3405 }
3406
3407 static int decode_slice(struct AVCodecContext *avctx, void *arg){
3408     H264Context *h = *(void**)arg;
3409     MpegEncContext * const s = &h->s;
3410     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
3411     int lf_x_start = s->mb_x;
3412
3413     s->mb_skip_run= -1;
3414
3415     h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
3416                     (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
3417
3418     if( h->pps.cabac ) {
3419         /* realign */
3420         align_get_bits( &s->gb );
3421
3422         /* init cabac */
3423         ff_init_cabac_states( &h->cabac);
3424         ff_init_cabac_decoder( &h->cabac,
3425                                s->gb.buffer + get_bits_count(&s->gb)/8,
3426                                (get_bits_left(&s->gb) + 7)/8);
3427
3428         ff_h264_init_cabac_states(h);
3429
3430         for(;;){
3431 //START_TIMER
3432             int ret = ff_h264_decode_mb_cabac(h);
3433             int eos;
3434 //STOP_TIMER("decode_mb_cabac")
3435
3436             if(ret>=0) ff_h264_hl_decode_mb(h);
3437
3438             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
3439                 s->mb_y++;
3440
3441                 ret = ff_h264_decode_mb_cabac(h);
3442
3443                 if(ret>=0) ff_h264_hl_decode_mb(h);
3444                 s->mb_y--;
3445             }
3446             eos = get_cabac_terminate( &h->cabac );
3447
3448             if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
3449                 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);
3450                 if (s->mb_x >= lf_x_start) loop_filter(h, lf_x_start, s->mb_x + 1);
3451                 return 0;
3452             }
3453             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
3454                 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);
3455                 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);
3456                 return -1;
3457             }
3458
3459             if( ++s->mb_x >= s->mb_width ) {
3460                 loop_filter(h, lf_x_start, s->mb_x);
3461                 s->mb_x = lf_x_start = 0;
3462                 decode_finish_row(h);
3463                 ++s->mb_y;
3464                 if(FIELD_OR_MBAFF_PICTURE) {
3465                     ++s->mb_y;
3466                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
3467                         predict_field_decoding_flag(h);
3468                 }
3469             }
3470
3471             if( eos || s->mb_y >= s->mb_height ) {
3472                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3473                 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);
3474                 if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
3475                 return 0;
3476             }
3477         }
3478
3479     } else {
3480         for(;;){
3481             int ret = ff_h264_decode_mb_cavlc(h);
3482
3483             if(ret>=0) ff_h264_hl_decode_mb(h);
3484
3485             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
3486                 s->mb_y++;
3487                 ret = ff_h264_decode_mb_cavlc(h);
3488
3489                 if(ret>=0) ff_h264_hl_decode_mb(h);
3490                 s->mb_y--;
3491             }
3492
3493             if(ret<0){
3494                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3495                 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);
3496                 return -1;
3497             }
3498
3499             if(++s->mb_x >= s->mb_width){
3500                 loop_filter(h, lf_x_start, s->mb_x);
3501                 s->mb_x = lf_x_start = 0;
3502                 decode_finish_row(h);
3503                 ++s->mb_y;
3504                 if(FIELD_OR_MBAFF_PICTURE) {
3505                     ++s->mb_y;
3506                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
3507                         predict_field_decoding_flag(h);
3508                 }
3509                 if(s->mb_y >= s->mb_height){
3510                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3511
3512                     if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
3513                         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);
3514
3515                         return 0;
3516                     }else{
3517                         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);
3518
3519                         return -1;
3520                     }
3521                 }
3522             }
3523
3524             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
3525                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3526                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
3527                     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);
3528                     if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
3529
3530                     return 0;
3531                 }else{
3532                     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);
3533
3534                     return -1;
3535                 }
3536             }
3537         }
3538     }
3539
3540 #if 0
3541     for(;s->mb_y < s->mb_height; s->mb_y++){
3542         for(;s->mb_x < s->mb_width; s->mb_x++){
3543             int ret= decode_mb(h);
3544
3545             ff_h264_hl_decode_mb(h);
3546
3547             if(ret<0){
3548                 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3549                 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);
3550
3551                 return -1;
3552             }
3553
3554             if(++s->mb_x >= s->mb_width){
3555                 s->mb_x=0;
3556                 if(++s->mb_y >= s->mb_height){
3557                     if(get_bits_count(s->gb) == s->gb.size_in_bits){
3558                         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);
3559
3560                         return 0;
3561                     }else{
3562                         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);
3563
3564                         return -1;
3565                     }
3566                 }
3567             }
3568
3569             if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
3570                 if(get_bits_count(s->gb) == s->gb.size_in_bits){
3571                     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);
3572
3573                     return 0;
3574                 }else{
3575                     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);
3576
3577                     return -1;
3578                 }
3579             }
3580         }
3581         s->mb_x=0;
3582         ff_draw_horiz_band(s, 16*s->mb_y, 16);
3583     }
3584 #endif
3585     return -1; //not reached
3586 }
3587
3588 /**
3589  * Call decode_slice() for each context.
3590  *
3591  * @param h h264 master context
3592  * @param context_count number of contexts to execute
3593  */
3594 static void execute_decode_slices(H264Context *h, int context_count){
3595     MpegEncContext * const s = &h->s;
3596     AVCodecContext * const avctx= s->avctx;
3597     H264Context *hx;
3598     int i;
3599
3600     if (s->avctx->hwaccel)
3601         return;
3602     if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
3603         return;
3604     if(context_count == 1) {
3605         decode_slice(avctx, &h);
3606     } else {
3607         for(i = 1; i < context_count; i++) {
3608             hx = h->thread_context[i];
3609             hx->s.error_recognition = avctx->error_recognition;
3610             hx->s.error_count = 0;
3611         }
3612
3613         avctx->execute(avctx, (void *)decode_slice,
3614                        h->thread_context, NULL, context_count, sizeof(void*));
3615
3616         /* pull back stuff from slices to master context */
3617         hx = h->thread_context[context_count - 1];
3618         s->mb_x = hx->s.mb_x;
3619         s->mb_y = hx->s.mb_y;
3620         s->dropable = hx->s.dropable;
3621         s->picture_structure = hx->s.picture_structure;
3622         for(i = 1; i < context_count; i++)
3623             h->s.error_count += h->thread_context[i]->s.error_count;
3624     }
3625 }
3626
3627
3628 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
3629     MpegEncContext * const s = &h->s;
3630     AVCodecContext * const avctx= s->avctx;
3631     H264Context *hx; ///< thread context
3632     int buf_index;
3633     int context_count;
3634     int next_avc;
3635     int pass = !(avctx->active_thread_type & FF_THREAD_FRAME);
3636     int nals_needed=0; ///< number of NALs that need decoding before the next frame thread starts
3637     int nal_index;
3638
3639     h->max_contexts = (HAVE_THREADS && (s->avctx->active_thread_type&FF_THREAD_SLICE)) ? avctx->thread_count : 1;
3640     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
3641         h->current_slice = 0;
3642         if (!s->first_field)
3643             s->current_picture_ptr= NULL;
3644         ff_h264_reset_sei(h);
3645     }
3646
3647     for(;pass <= 1;pass++){
3648         buf_index = 0;
3649         context_count = 0;
3650         next_avc = h->is_avc ? 0 : buf_size;
3651         nal_index = 0;
3652     for(;;){
3653         int consumed;
3654         int dst_length;
3655         int bit_length;
3656         const uint8_t *ptr;
3657         int i, nalsize = 0;
3658         int err;
3659
3660         if(buf_index >= next_avc) {
3661             if(buf_index >= buf_size) break;
3662             nalsize = 0;
3663             for(i = 0; i < h->nal_length_size; i++)
3664                 nalsize = (nalsize << 8) | buf[buf_index++];
3665             if(nalsize <= 0 || nalsize > buf_size - buf_index){
3666                 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
3667                 break;
3668             }
3669             next_avc= buf_index + nalsize;
3670         } else {
3671             // start code prefix search
3672             for(; buf_index + 3 < next_avc; buf_index++){
3673                 // This should always succeed in the first iteration.
3674                 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
3675                     break;
3676             }
3677
3678             if(buf_index+3 >= buf_size) break;
3679
3680             buf_index+=3;
3681             if(buf_index >= next_avc) continue;
3682         }
3683
3684         hx = h->thread_context[context_count];
3685
3686         ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
3687         if (ptr==NULL || dst_length < 0){
3688             return -1;
3689         }
3690         i= buf_index + consumed;
3691         if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
3692            buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
3693             s->workaround_bugs |= FF_BUG_TRUNCATED;
3694
3695         if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
3696         while(ptr[dst_length - 1] == 0 && dst_length > 0)
3697             dst_length--;
3698         }
3699         bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
3700
3701         if(s->avctx->debug&FF_DEBUG_STARTCODE){
3702             av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d/%d length %d\n", hx->nal_unit_type, buf_index, buf_size, dst_length);
3703         }
3704
3705         if (h->is_avc && (nalsize != consumed) && nalsize){
3706             av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
3707         }
3708
3709         buf_index += consumed;
3710         nal_index++;
3711
3712         if(pass == 0) {
3713             // packets can sometimes contain multiple PPS/SPS
3714             // e.g. two PAFF field pictures in one packet, or a demuxer which splits NALs strangely
3715             // if so, when frame threading we can't start the next thread until we've read all of them
3716             switch (hx->nal_unit_type) {
3717                 case NAL_SPS:
3718                 case NAL_PPS:
3719                 case NAL_IDR_SLICE:
3720                 case NAL_SLICE:
3721                     nals_needed = nal_index;
3722             }
3723             continue;
3724         }
3725
3726         //FIXME do not discard SEI id
3727         if(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0)
3728             continue;
3729
3730       again:
3731         err = 0;
3732         switch(hx->nal_unit_type){
3733         case NAL_IDR_SLICE:
3734             if (h->nal_unit_type != NAL_IDR_SLICE) {
3735                 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
3736                 return -1;
3737             }
3738             idr(h); //FIXME ensure we don't loose some frames if there is reordering
3739         case NAL_SLICE:
3740             init_get_bits(&hx->s.gb, ptr, bit_length);
3741             hx->intra_gb_ptr=
3742             hx->inter_gb_ptr= &hx->s.gb;
3743             hx->s.data_partitioning = 0;
3744
3745             if((err = decode_slice_header(hx, h)))
3746                break;
3747
3748             s->current_picture_ptr->key_frame |=
3749                     (hx->nal_unit_type == NAL_IDR_SLICE) ||
3750                     (h->sei_recovery_frame_cnt >= 0);
3751
3752             if (h->current_slice == 1) {
3753                 if(!(s->flags2 & CODEC_FLAG2_CHUNKS)) {
3754                     decode_postinit(h, nal_index >= nals_needed);
3755                 }
3756
3757                 if (s->avctx->hwaccel && s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
3758                     return -1;
3759                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
3760                     ff_vdpau_h264_picture_start(s);
3761             }
3762
3763             if(hx->redundant_pic_count==0
3764                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
3765                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
3766                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
3767                && avctx->skip_frame < AVDISCARD_ALL){
3768                 if(avctx->hwaccel) {
3769                     if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
3770                         return -1;
3771                 }else
3772                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
3773                     static const uint8_t start_code[] = {0x00, 0x00, 0x01};
3774                     ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
3775                     ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
3776                 }else
3777                     context_count++;
3778             }
3779             break;
3780         case NAL_DPA:
3781             init_get_bits(&hx->s.gb, ptr, bit_length);
3782             hx->intra_gb_ptr=
3783             hx->inter_gb_ptr= NULL;
3784
3785             if ((err = decode_slice_header(hx, h)) < 0)
3786                 break;
3787
3788             hx->s.data_partitioning = 1;
3789
3790             break;
3791         case NAL_DPB:
3792             init_get_bits(&hx->intra_gb, ptr, bit_length);
3793             hx->intra_gb_ptr= &hx->intra_gb;
3794             break;
3795         case NAL_DPC:
3796             init_get_bits(&hx->inter_gb, ptr, bit_length);
3797             hx->inter_gb_ptr= &hx->inter_gb;
3798
3799             if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
3800                && s->context_initialized
3801                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
3802                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
3803                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
3804                && avctx->skip_frame < AVDISCARD_ALL)
3805                 context_count++;
3806             break;
3807         case NAL_SEI:
3808             init_get_bits(&s->gb, ptr, bit_length);
3809             ff_h264_decode_sei(h);
3810             break;
3811         case NAL_SPS:
3812             init_get_bits(&s->gb, ptr, bit_length);
3813             ff_h264_decode_seq_parameter_set(h);
3814
3815             if (s->flags& CODEC_FLAG_LOW_DELAY ||
3816                 (h->sps.bitstream_restriction_flag && !h->sps.num_reorder_frames))
3817                 s->low_delay=1;
3818
3819             if(avctx->has_b_frames < 2)
3820                 avctx->has_b_frames= !s->low_delay;
3821
3822             if (avctx->bits_per_raw_sample != h->sps.bit_depth_luma) {
3823                 if (h->sps.bit_depth_luma >= 8 && h->sps.bit_depth_luma <= 10) {
3824                     avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
3825                     h->pixel_shift = h->sps.bit_depth_luma > 8;
3826
3827                     ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma);
3828                     ff_h264_pred_init(&h->hpc, s->codec_id, h->sps.bit_depth_luma);
3829                     dsputil_init(&s->dsp, s->avctx);
3830                 } else {
3831                     av_log(avctx, AV_LOG_DEBUG, "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
3832                     return -1;
3833                 }
3834             }
3835             break;
3836         case NAL_PPS:
3837             init_get_bits(&s->gb, ptr, bit_length);
3838
3839             ff_h264_decode_picture_parameter_set(h, bit_length);
3840
3841             break;
3842         case NAL_AUD:
3843         case NAL_END_SEQUENCE:
3844         case NAL_END_STREAM:
3845         case NAL_FILLER_DATA:
3846         case NAL_SPS_EXT:
3847         case NAL_AUXILIARY_SLICE:
3848             break;
3849         default:
3850             av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
3851         }
3852
3853         if(context_count == h->max_contexts) {
3854             execute_decode_slices(h, context_count);
3855             context_count = 0;
3856         }
3857
3858         if (err < 0)
3859             av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
3860         else if(err == 1) {
3861             /* Slice could not be decoded in parallel mode, copy down
3862              * NAL unit stuff to context 0 and restart. Note that
3863              * rbsp_buffer is not transferred, but since we no longer
3864              * run in parallel mode this should not be an issue. */
3865             h->nal_unit_type = hx->nal_unit_type;
3866             h->nal_ref_idc   = hx->nal_ref_idc;
3867             hx = h;
3868             goto again;
3869         }
3870     }
3871     }
3872     if(context_count)
3873         execute_decode_slices(h, context_count);
3874     return buf_index;
3875 }
3876
3877 /**
3878  * returns the number of bytes consumed for building the current frame
3879  */
3880 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
3881         if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
3882         if(pos+10>buf_size) pos=buf_size; // oops ;)
3883
3884         return pos;
3885 }
3886
3887 static int decode_frame(AVCodecContext *avctx,
3888                              void *data, int *data_size,
3889                              AVPacket *avpkt)
3890 {
3891     const uint8_t *buf = avpkt->data;
3892     int buf_size = avpkt->size;
3893     H264Context *h = avctx->priv_data;
3894     MpegEncContext *s = &h->s;
3895     AVFrame *pict = data;
3896     int buf_index;
3897
3898     s->flags= avctx->flags;
3899     s->flags2= avctx->flags2;
3900
3901    /* end of stream, output what is still in the buffers */
3902  out:
3903     if (buf_size == 0) {
3904         Picture *out;
3905         int i, out_idx;
3906
3907         s->current_picture_ptr = NULL;
3908
3909 //FIXME factorize this with the output code below
3910         out = h->delayed_pic[0];
3911         out_idx = 0;
3912         for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
3913             if(h->delayed_pic[i]->poc < out->poc){
3914                 out = h->delayed_pic[i];
3915                 out_idx = i;
3916             }
3917
3918         for(i=out_idx; h->delayed_pic[i]; i++)
3919             h->delayed_pic[i] = h->delayed_pic[i+1];
3920
3921         if(out){
3922             *data_size = sizeof(AVFrame);
3923             *pict= *(AVFrame*)out;
3924         }
3925
3926         return 0;
3927     }
3928
3929     buf_index=decode_nal_units(h, buf, buf_size);
3930     if(buf_index < 0)
3931         return -1;
3932
3933     if (!s->current_picture_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
3934         buf_size = 0;
3935         goto out;
3936     }
3937
3938     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
3939         if (avctx->skip_frame >= AVDISCARD_NONREF)
3940             return 0;
3941         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
3942         return -1;
3943     }
3944
3945     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
3946
3947         if(s->flags2 & CODEC_FLAG2_CHUNKS) decode_postinit(h, 1);
3948
3949         field_end(h, 0);
3950
3951         if (!h->next_output_pic) {
3952             /* Wait for second field. */
3953             *data_size = 0;
3954
3955         } else {
3956             *data_size = sizeof(AVFrame);
3957             *pict = *(AVFrame*)h->next_output_pic;
3958         }
3959     }
3960
3961     assert(pict->data[0] || !*data_size);
3962     ff_print_debug_info(s, pict);
3963 //printf("out %d\n", (int)pict->data[0]);
3964
3965     return get_consumed_bytes(s, buf_index, buf_size);
3966 }
3967 #if 0
3968 static inline void fill_mb_avail(H264Context *h){
3969     MpegEncContext * const s = &h->s;
3970     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3971
3972     if(s->mb_y){
3973         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
3974         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
3975         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
3976     }else{
3977         h->mb_avail[0]=
3978         h->mb_avail[1]=
3979         h->mb_avail[2]= 0;
3980     }
3981     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
3982     h->mb_avail[4]= 1; //FIXME move out
3983     h->mb_avail[5]= 0; //FIXME move out
3984 }
3985 #endif
3986
3987 #ifdef TEST
3988 #undef printf
3989 #undef random
3990 #define COUNT 8000
3991 #define SIZE (COUNT*40)
3992 int main(void){
3993     int i;
3994     uint8_t temp[SIZE];
3995     PutBitContext pb;
3996     GetBitContext gb;
3997 //    int int_temp[10000];
3998     DSPContext dsp;
3999     AVCodecContext avctx;
4000
4001     dsputil_init(&dsp, &avctx);
4002
4003     init_put_bits(&pb, temp, SIZE);
4004     printf("testing unsigned exp golomb\n");
4005     for(i=0; i<COUNT; i++){
4006         START_TIMER
4007         set_ue_golomb(&pb, i);
4008         STOP_TIMER("set_ue_golomb");
4009     }
4010     flush_put_bits(&pb);
4011
4012     init_get_bits(&gb, temp, 8*SIZE);
4013     for(i=0; i<COUNT; i++){
4014         int j, s;
4015
4016         s= show_bits(&gb, 24);
4017
4018         START_TIMER
4019         j= get_ue_golomb(&gb);
4020         if(j != i){
4021             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4022 //            return -1;
4023         }
4024         STOP_TIMER("get_ue_golomb");
4025     }
4026
4027
4028     init_put_bits(&pb, temp, SIZE);
4029     printf("testing signed exp golomb\n");
4030     for(i=0; i<COUNT; i++){
4031         START_TIMER
4032         set_se_golomb(&pb, i - COUNT/2);
4033         STOP_TIMER("set_se_golomb");
4034     }
4035     flush_put_bits(&pb);
4036
4037     init_get_bits(&gb, temp, 8*SIZE);
4038     for(i=0; i<COUNT; i++){
4039         int j, s;
4040
4041         s= show_bits(&gb, 24);
4042
4043         START_TIMER
4044         j= get_se_golomb(&gb);
4045         if(j != i - COUNT/2){
4046             printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4047 //            return -1;
4048         }
4049         STOP_TIMER("get_se_golomb");
4050     }
4051
4052 #if 0
4053     printf("testing 4x4 (I)DCT\n");
4054
4055     DCTELEM block[16];
4056     uint8_t src[16], ref[16];
4057     uint64_t error= 0, max_error=0;
4058
4059     for(i=0; i<COUNT; i++){
4060         int j;
4061 //        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
4062         for(j=0; j<16; j++){
4063             ref[j]= random()%255;
4064             src[j]= random()%255;
4065         }
4066
4067         h264_diff_dct_c(block, src, ref, 4);
4068
4069         //normalize
4070         for(j=0; j<16; j++){
4071 //            printf("%d ", block[j]);
4072             block[j]= block[j]*4;
4073             if(j&1) block[j]= (block[j]*4 + 2)/5;
4074             if(j&4) block[j]= (block[j]*4 + 2)/5;
4075         }
4076 //        printf("\n");
4077
4078         h->h264dsp.h264_idct_add(ref, block, 4);
4079 /*        for(j=0; j<16; j++){
4080             printf("%d ", ref[j]);
4081         }
4082         printf("\n");*/
4083
4084         for(j=0; j<16; j++){
4085             int diff= FFABS(src[j] - ref[j]);
4086
4087             error+= diff*diff;
4088             max_error= FFMAX(max_error, diff);
4089         }
4090     }
4091     printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
4092     printf("testing quantizer\n");
4093     for(qp=0; qp<52; qp++){
4094         for(i=0; i<16; i++)
4095             src1_block[i]= src2_block[i]= random()%255;
4096
4097     }
4098     printf("Testing NAL layer\n");
4099
4100     uint8_t bitstream[COUNT];
4101     uint8_t nal[COUNT*2];
4102     H264Context h;
4103     memset(&h, 0, sizeof(H264Context));
4104
4105     for(i=0; i<COUNT; i++){
4106         int zeros= i;
4107         int nal_length;
4108         int consumed;
4109         int out_length;
4110         uint8_t *out;
4111         int j;
4112
4113         for(j=0; j<COUNT; j++){
4114             bitstream[j]= (random() % 255) + 1;
4115         }
4116
4117         for(j=0; j<zeros; j++){
4118             int pos= random() % COUNT;
4119             while(bitstream[pos] == 0){
4120                 pos++;
4121                 pos %= COUNT;
4122             }
4123             bitstream[pos]=0;
4124         }
4125
4126         START_TIMER
4127
4128         nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
4129         if(nal_length<0){
4130             printf("encoding failed\n");
4131             return -1;
4132         }
4133
4134         out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
4135
4136         STOP_TIMER("NAL")
4137
4138         if(out_length != COUNT){
4139             printf("incorrect length %d %d\n", out_length, COUNT);
4140             return -1;
4141         }
4142
4143         if(consumed != nal_length){
4144             printf("incorrect consumed length %d %d\n", nal_length, consumed);
4145             return -1;
4146         }
4147
4148         if(memcmp(bitstream, out, COUNT)){
4149             printf("mismatch\n");
4150             return -1;
4151         }
4152     }
4153 #endif
4154
4155     printf("Testing RBSP\n");
4156
4157
4158     return 0;
4159 }
4160 #endif /* TEST */
4161
4162
4163 av_cold void ff_h264_free_context(H264Context *h)
4164 {
4165     int i;
4166
4167     free_tables(h, 1); //FIXME cleanup init stuff perhaps
4168
4169     for(i = 0; i < MAX_SPS_COUNT; i++)
4170         av_freep(h->sps_buffers + i);
4171
4172     for(i = 0; i < MAX_PPS_COUNT; i++)
4173         av_freep(h->pps_buffers + i);
4174 }
4175
4176 av_cold int ff_h264_decode_end(AVCodecContext *avctx)
4177 {
4178     H264Context *h = avctx->priv_data;
4179     MpegEncContext *s = &h->s;
4180
4181     ff_h264_free_context(h);
4182
4183     MPV_common_end(s);
4184
4185 //    memset(h, 0, sizeof(H264Context));
4186
4187     return 0;
4188 }
4189
4190 static const AVProfile profiles[] = {
4191     { FF_PROFILE_H264_BASELINE,             "Baseline"              },
4192     { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline"  },
4193     { FF_PROFILE_H264_MAIN,                 "Main"                  },
4194     { FF_PROFILE_H264_EXTENDED,             "Extended"              },
4195     { FF_PROFILE_H264_HIGH,                 "High"                  },
4196     { FF_PROFILE_H264_HIGH_10,              "High 10"               },
4197     { FF_PROFILE_H264_HIGH_10_INTRA,        "High 10 Intra"         },
4198     { FF_PROFILE_H264_HIGH_422,             "High 4:2:2"            },
4199     { FF_PROFILE_H264_HIGH_422_INTRA,       "High 4:2:2 Intra"      },
4200     { FF_PROFILE_H264_HIGH_444,             "High 4:4:4"            },
4201     { FF_PROFILE_H264_HIGH_444_PREDICTIVE,  "High 4:4:4 Predictive" },
4202     { FF_PROFILE_H264_HIGH_444_INTRA,       "High 4:4:4 Intra"      },
4203     { FF_PROFILE_H264_CAVLC_444,            "CAVLC 4:4:4"           },
4204     { FF_PROFILE_UNKNOWN },
4205 };
4206
4207 AVCodec ff_h264_decoder = {
4208     "h264",
4209     AVMEDIA_TYPE_VIDEO,
4210     CODEC_ID_H264,
4211     sizeof(H264Context),
4212     ff_h264_decode_init,
4213     NULL,
4214     ff_h264_decode_end,
4215     decode_frame,
4216     /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY |
4217         CODEC_CAP_SLICE_THREADS | CODEC_CAP_FRAME_THREADS,
4218     .flush= flush_dpb,
4219     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
4220     .init_thread_copy      = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
4221     .update_thread_context = ONLY_IF_THREADS_ENABLED(decode_update_thread_context),
4222     .profiles = NULL_IF_CONFIG_SMALL(profiles),
4223 };
4224
4225 #if CONFIG_H264_VDPAU_DECODER
4226 AVCodec ff_h264_vdpau_decoder = {
4227     "h264_vdpau",
4228     AVMEDIA_TYPE_VIDEO,
4229     CODEC_ID_H264,
4230     sizeof(H264Context),
4231     ff_h264_decode_init,
4232     NULL,
4233     ff_h264_decode_end,
4234     decode_frame,
4235     CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
4236     .flush= flush_dpb,
4237     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
4238     .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
4239     .profiles = NULL_IF_CONFIG_SMALL(profiles),
4240 };
4241 #endif