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

   1 /*
   2  * H.26L/H.264/AVC/JVT/14496-10/... decoder
   3  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
   4  *
   5  * This file is part of FFmpeg.
   6  *
   7  * FFmpeg is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2.1 of the License, or (at your option) any later version.
  11  *
  12  * FFmpeg is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with FFmpeg; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 /**
  23  * @file
  24  * H.264 / AVC / MPEG4 part10 codec.
  25  * @author Michael Niedermayer <michaelni@gmx.at>
  26  */
  27
  28 #define UNCHECKED_BITSTREAM_READER 1
  29
  30 #include "libavutil/imgutils.h"
  31 #include "libavutil/opt.h"
  32 #include "internal.h"
  33 #include "cabac.h"
  34 #include "cabac_functions.h"
  35 #include "dsputil.h"
  36 #include "avcodec.h"
  37 #include "mpegvideo.h"
  38 #include "h264.h"
  39 #include "h264data.h"
  40 #include "h264_mvpred.h"
  41 #include "golomb.h"
  42 #include "mathops.h"
  43 #include "rectangle.h"
  44 #include "thread.h"
  45 #include "vdpau_internal.h"
  46 #include "libavutil/avassert.h"
  47
  48 //#undef NDEBUG
  49 #include <assert.h>
  50
  51 static const uint8_t rem6[QP_MAX_NUM+1]={
  52 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,
  53 };
  54
  55 static const uint8_t div6[QP_MAX_NUM+1]={
  56 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,
  57 };
  58
  59 static const enum PixelFormat hwaccel_pixfmt_list_h264_jpeg_420[] = {
  60     PIX_FMT_DXVA2_VLD,
  61     PIX_FMT_VAAPI_VLD,
  62     PIX_FMT_VDA_VLD,
  63     PIX_FMT_YUVJ420P,
  64     PIX_FMT_NONE
  65 };
  66
  67 /**
  68  * Check if the top & left blocks are available if needed and
  69  * change the dc mode so it only uses the available blocks.
  70  */
  71 int ff_h264_check_intra4x4_pred_mode(H264Context *h){
  72     MpegEncContext * const s = &h->s;
  73     static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
  74     static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
  75     int i;
  76
  77     if(!(h->top_samples_available&0x8000)){
  78         for(i=0; i<4; i++){
  79             int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
  80             if(status<0){
  81                 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);
  82                 return -1;
  83             } else if(status){
  84                 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
  85             }
  86         }
  87     }
  88
  89     if((h->left_samples_available&0x8888)!=0x8888){
  90         static const int mask[4]={0x8000,0x2000,0x80,0x20};
  91         for(i=0; i<4; i++){
  92             if(!(h->left_samples_available&mask[i])){
  93                 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
  94                 if(status<0){
  95                     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);
  96                     return -1;
  97                 } else if(status){
  98                     h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
  99                 }
 100             }
 101         }
 102     }
 103
 104     return 0;
 105 } //FIXME cleanup like check_intra_pred_mode
 106
 107 int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma){
 108     MpegEncContext * const s = &h->s;
 109     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
 110     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
 111
 112     if(mode > 6U) {
 113         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);
 114         return -1;
 115     }
 116
 117     if(!(h->top_samples_available&0x8000)){
 118         mode= top[ mode ];
 119         if(mode<0){
 120             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);
 121             return -1;
 122         }
 123     }
 124
 125     if((h->left_samples_available&0x8080) != 0x8080){
 126         mode= left[ mode ];
 127         if(is_chroma && (h->left_samples_available&0x8080)){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
 128             mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
 129         }
 130         if(mode<0){
 131             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);
 132             return -1;
 133         }
 134     }
 135
 136     return mode;
 137 }
 138
 139
 140 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
 141     int i, si, di;
 142     uint8_t *dst;
 143     int bufidx;
 144
 145 //    src[0]&0x80;                //forbidden bit
 146     h->nal_ref_idc= src[0]>>5;
 147     h->nal_unit_type= src[0]&0x1F;
 148
 149     src++; length--;
 150
 151 #if HAVE_FAST_UNALIGNED
 152 # if HAVE_FAST_64BIT
 153 #   define RS 7
 154     for(i=0; i+1<length; i+=9){
 155         if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
 156 # else
 157 #   define RS 3
 158     for(i=0; i+1<length; i+=5){
 159         if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
 160 # endif
 161             continue;
 162         if(i>0 && !src[i]) i--;
 163         while(src[i]) i++;
 164 #else
 165 #   define RS 0
 166     for(i=0; i+1<length; i+=2){
 167         if(src[i]) continue;
 168         if(i>0 && src[i-1]==0) i--;
 169 #endif
 170         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
 171             if(src[i+2]!=3){
 172                 /* startcode, so we must be past the end */
 173                 length=i;
 174             }
 175             break;
 176         }
 177         i-= RS;
 178     }
 179
 180     bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
 181     si=h->rbsp_buffer_size[bufidx];
 182     av_fast_padded_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+MAX_MBPAIR_SIZE);
 183     dst= h->rbsp_buffer[bufidx];
 184
 185     if (dst == NULL){
 186         return NULL;
 187     }
 188
 189     if(i>=length-1){ //no escaped 0
 190         *dst_length= length;
 191         *consumed= length+1; //+1 for the header
 192         if(h->s.avctx->flags2 & CODEC_FLAG2_FAST){
 193             return src;
 194         }else{
 195             memcpy(dst, src, length);
 196             return dst;
 197         }
 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->f.thread_opaque != s->current_picture.f.thread_opaque ||
 274            (ref->f.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->f.thread_opaque != s->current_picture.f.thread_opaque ||
 286            (ref->f.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.f.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->f.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 av_always_inline void
 451 mc_dir_part(H264Context *h, Picture *pic, int n, int square,
 452             int height, int delta, int list,
 453             uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 454             int src_x_offset, int src_y_offset,
 455             qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op,
 456             int pixel_shift, int chroma_idc)
 457 {
 458     MpegEncContext * const s = &h->s;
 459     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
 460     int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
 461     const int luma_xy= (mx&3) + ((my&3)<<2);
 462     int offset = ((mx>>2) << pixel_shift) + (my>>2)*h->mb_linesize;
 463     uint8_t * src_y = pic->f.data[0] + offset;
 464     uint8_t * src_cb, * src_cr;
 465     int extra_width= h->emu_edge_width;
 466     int extra_height= h->emu_edge_height;
 467     int emu=0;
 468     const int full_mx= mx>>2;
 469     const int full_my= my>>2;
 470     const int pic_width  = 16*s->mb_width;
 471     const int pic_height = 16*s->mb_height >> MB_FIELD;
 472     int ysh;
 473
 474     if(mx&7) extra_width -= 3;
 475     if(my&7) extra_height -= 3;
 476
 477     if(   full_mx < 0-extra_width
 478        || full_my < 0-extra_height
 479        || full_mx + 16/*FIXME*/ > pic_width + extra_width
 480        || full_my + 16/*FIXME*/ > pic_height + extra_height){
 481         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_y - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
 482                                 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
 483             src_y= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
 484         emu=1;
 485     }
 486
 487     qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
 488     if(!square){
 489         qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
 490     }
 491
 492     if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
 493
 494     if(chroma_idc == 3 /* yuv444 */){
 495         src_cb = pic->f.data[1] + offset;
 496         if(emu){
 497             s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
 498                                     16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
 499             src_cb= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
 500         }
 501         qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); //FIXME try variable height perhaps?
 502         if(!square){
 503             qpix_op[luma_xy](dest_cb + delta, src_cb + delta, h->mb_linesize);
 504         }
 505
 506         src_cr = pic->f.data[2] + offset;
 507         if(emu){
 508             s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
 509                                     16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
 510             src_cr= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
 511         }
 512         qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); //FIXME try variable height perhaps?
 513         if(!square){
 514             qpix_op[luma_xy](dest_cr + delta, src_cr + delta, h->mb_linesize);
 515         }
 516         return;
 517     }
 518
 519     ysh = 3 - (chroma_idc == 2 /* yuv422 */);
 520     if(chroma_idc == 1 /* yuv420 */ && MB_FIELD){
 521         // chroma offset when predicting from a field of opposite parity
 522         my += 2 * ((s->mb_y & 1) - (pic->f.reference - 1));
 523         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
 524     }
 525
 526     src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) + (my >> ysh) * h->mb_uvlinesize;
 527     src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) + (my >> ysh) * h->mb_uvlinesize;
 528
 529     if(emu){
 530         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize,
 531                                 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
 532                                 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
 533             src_cb= s->edge_emu_buffer;
 534     }
 535     chroma_op(dest_cb, src_cb, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
 536               mx&7, (my << (chroma_idc == 2 /* yuv422 */)) &7);
 537
 538     if(emu){
 539         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize,
 540                                 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
 541                                 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
 542             src_cr= s->edge_emu_buffer;
 543     }
 544     chroma_op(dest_cr, src_cr, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
 545               mx&7, (my << (chroma_idc == 2 /* yuv422 */)) &7);
 546 }
 547
 548 static av_always_inline void
 549 mc_part_std(H264Context *h, int n, int square, int height, int delta,
 550             uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 551             int x_offset, int y_offset,
 552             qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 553             qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
 554             int list0, int list1, int pixel_shift, int chroma_idc)
 555 {
 556     MpegEncContext * const s = &h->s;
 557     qpel_mc_func *qpix_op=  qpix_put;
 558     h264_chroma_mc_func chroma_op= chroma_put;
 559
 560     dest_y  += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
 561     if (chroma_idc == 3 /* yuv444 */) {
 562         dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
 563         dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
 564     } else if (chroma_idc == 2 /* yuv422 */) {
 565         dest_cb += (  x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
 566         dest_cr += (  x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
 567     } else /* yuv420 */ {
 568         dest_cb += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
 569         dest_cr += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
 570     }
 571     x_offset += 8*s->mb_x;
 572     y_offset += 8*(s->mb_y >> MB_FIELD);
 573
 574     if(list0){
 575         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
 576         mc_dir_part(h, ref, n, square, height, delta, 0,
 577                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
 578                            qpix_op, chroma_op, pixel_shift, chroma_idc);
 579
 580         qpix_op=  qpix_avg;
 581         chroma_op= chroma_avg;
 582     }
 583
 584     if(list1){
 585         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
 586         mc_dir_part(h, ref, n, square, height, delta, 1,
 587                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
 588                            qpix_op, chroma_op, pixel_shift, chroma_idc);
 589     }
 590 }
 591
 592 static av_always_inline void
 593 mc_part_weighted(H264Context *h, int n, int square, int height, int delta,
 594                  uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 595                  int x_offset, int y_offset,
 596                  qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 597                  h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
 598                  h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
 599                  int list0, int list1, int pixel_shift, int chroma_idc){
 600     MpegEncContext * const s = &h->s;
 601     int chroma_height;
 602
 603     dest_y += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
 604     if (chroma_idc == 3 /* yuv444 */) {
 605         chroma_height = height;
 606         chroma_weight_avg = luma_weight_avg;
 607         chroma_weight_op = luma_weight_op;
 608         dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
 609         dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
 610     } else if (chroma_idc == 2 /* yuv422 */) {
 611         chroma_height = height;
 612         dest_cb += (  x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
 613         dest_cr += (  x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
 614     } else /* yuv420 */ {
 615         chroma_height = height >> 1;
 616         dest_cb += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
 617         dest_cr += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
 618     }
 619     x_offset += 8*s->mb_x;
 620     y_offset += 8*(s->mb_y >> MB_FIELD);
 621
 622     if(list0 && list1){
 623         /* don't optimize for luma-only case, since B-frames usually
 624          * use implicit weights => chroma too. */
 625         uint8_t *tmp_cb = s->obmc_scratchpad;
 626         uint8_t *tmp_cr = s->obmc_scratchpad + (16 << pixel_shift);
 627         uint8_t *tmp_y  = s->obmc_scratchpad + 16*h->mb_uvlinesize;
 628         int refn0 = h->ref_cache[0][ scan8[n] ];
 629         int refn1 = h->ref_cache[1][ scan8[n] ];
 630
 631         mc_dir_part(h, &h->ref_list[0][refn0], n, square, height, delta, 0,
 632                     dest_y, dest_cb, dest_cr,
 633                     x_offset, y_offset, qpix_put, chroma_put,
 634                     pixel_shift, chroma_idc);
 635         mc_dir_part(h, &h->ref_list[1][refn1], n, square, height, delta, 1,
 636                     tmp_y, tmp_cb, tmp_cr,
 637                     x_offset, y_offset, qpix_put, chroma_put,
 638                     pixel_shift, chroma_idc);
 639
 640         if(h->use_weight == 2){
 641             int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
 642             int weight1 = 64 - weight0;
 643             luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize,
 644                               height,        5, weight0, weight1, 0);
 645             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize,
 646                               chroma_height, 5, weight0, weight1, 0);
 647             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize,
 648                               chroma_height, 5, weight0, weight1, 0);
 649         }else{
 650             luma_weight_avg(dest_y, tmp_y, h->mb_linesize, height, h->luma_log2_weight_denom,
 651                             h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
 652                             h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
 653             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
 654                             h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
 655                             h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
 656             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
 657                             h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
 658                             h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
 659         }
 660     }else{
 661         int list = list1 ? 1 : 0;
 662         int refn = h->ref_cache[list][ scan8[n] ];
 663         Picture *ref= &h->ref_list[list][refn];
 664         mc_dir_part(h, ref, n, square, height, delta, list,
 665                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
 666                     qpix_put, chroma_put, pixel_shift, chroma_idc);
 667
 668         luma_weight_op(dest_y, h->mb_linesize, height, h->luma_log2_weight_denom,
 669                        h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
 670         if(h->use_weight_chroma){
 671             chroma_weight_op(dest_cb, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
 672                              h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
 673             chroma_weight_op(dest_cr, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
 674                              h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
 675         }
 676     }
 677 }
 678
 679 static av_always_inline void
 680 mc_part(H264Context *h, int n, int square, int height, int delta,
 681         uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 682         int x_offset, int y_offset,
 683         qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
 684         qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
 685         h264_weight_func *weight_op, h264_biweight_func *weight_avg,
 686         int list0, int list1, int pixel_shift, int chroma_idc)
 687 {
 688     if((h->use_weight==2 && list0 && list1
 689         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
 690        || h->use_weight==1)
 691         mc_part_weighted(h, n, square, height, delta, dest_y, dest_cb, dest_cr,
 692                          x_offset, y_offset, qpix_put, chroma_put,
 693                          weight_op[0], weight_op[1], weight_avg[0],
 694                          weight_avg[1], list0, list1, pixel_shift, chroma_idc);
 695     else
 696         mc_part_std(h, n, square, height, delta, dest_y, dest_cb, dest_cr,
 697                     x_offset, y_offset, qpix_put, chroma_put, qpix_avg,
 698                     chroma_avg, list0, list1, pixel_shift, chroma_idc);
 699 }
 700
 701 static av_always_inline void
 702 prefetch_motion(H264Context *h, int list, int pixel_shift, int chroma_idc)
 703 {
 704     /* fetch pixels for estimated mv 4 macroblocks ahead
 705      * optimized for 64byte cache lines */
 706     MpegEncContext * const s = &h->s;
 707     const int refn = h->ref_cache[list][scan8[0]];
 708     if(refn >= 0){
 709         const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
 710         const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
 711         uint8_t **src = h->ref_list[list][refn].f.data;
 712         int off= (mx << pixel_shift) + (my + (s->mb_x&3)*4)*h->mb_linesize + (64 << pixel_shift);
 713         s->dsp.prefetch(src[0]+off, s->linesize, 4);
 714         if (chroma_idc == 3 /* yuv444 */) {
 715             s->dsp.prefetch(src[1]+off, s->linesize, 4);
 716             s->dsp.prefetch(src[2]+off, s->linesize, 4);
 717         }else{
 718             off= (((mx>>1)+64)<<pixel_shift) + ((my>>1) + (s->mb_x&7))*s->uvlinesize;
 719             s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
 720         }
 721     }
 722 }
 723
 724 static av_always_inline void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 725                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
 726                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
 727                       h264_weight_func *weight_op, h264_biweight_func *weight_avg,
 728                       int pixel_shift, int chroma_idc)
 729 {
 730     MpegEncContext * const s = &h->s;
 731     const int mb_xy= h->mb_xy;
 732     const int mb_type = s->current_picture.f.mb_type[mb_xy];
 733
 734     assert(IS_INTER(mb_type));
 735
 736     if(HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
 737         await_references(h);
 738     prefetch_motion(h, 0, pixel_shift, chroma_idc);
 739
 740     if(IS_16X16(mb_type)){
 741         mc_part(h, 0, 1, 16, 0, dest_y, dest_cb, dest_cr, 0, 0,
 742                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
 743                 weight_op, weight_avg,
 744                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
 745                 pixel_shift, chroma_idc);
 746     }else if(IS_16X8(mb_type)){
 747         mc_part(h, 0, 0, 8, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 0,
 748                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
 749                 weight_op, weight_avg,
 750                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
 751                 pixel_shift, chroma_idc);
 752         mc_part(h, 8, 0, 8, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 4,
 753                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
 754                 weight_op, weight_avg,
 755                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
 756                 pixel_shift, chroma_idc);
 757     }else if(IS_8X16(mb_type)){
 758         mc_part(h, 0, 0, 16, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
 759                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 760                 &weight_op[1], &weight_avg[1],
 761                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
 762                 pixel_shift, chroma_idc);
 763         mc_part(h, 4, 0, 16, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
 764                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 765                 &weight_op[1], &weight_avg[1],
 766                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
 767                 pixel_shift, chroma_idc);
 768     }else{
 769         int i;
 770
 771         assert(IS_8X8(mb_type));
 772
 773         for(i=0; i<4; i++){
 774             const int sub_mb_type= h->sub_mb_type[i];
 775             const int n= 4*i;
 776             int x_offset= (i&1)<<2;
 777             int y_offset= (i&2)<<1;
 778
 779             if(IS_SUB_8X8(sub_mb_type)){
 780                 mc_part(h, n, 1, 8, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 781                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 782                     &weight_op[1], &weight_avg[1],
 783                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
 784                     pixel_shift, chroma_idc);
 785             }else if(IS_SUB_8X4(sub_mb_type)){
 786                 mc_part(h, n  , 0, 4, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 787                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
 788                     &weight_op[1], &weight_avg[1],
 789                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
 790                     pixel_shift, chroma_idc);
 791                 mc_part(h, n+2, 0, 4, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
 792                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
 793                     &weight_op[1], &weight_avg[1],
 794                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
 795                     pixel_shift, chroma_idc);
 796             }else if(IS_SUB_4X8(sub_mb_type)){
 797                 mc_part(h, n  , 0, 8, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 798                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 799                     &weight_op[2], &weight_avg[2],
 800                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
 801                     pixel_shift, chroma_idc);
 802                 mc_part(h, n+1, 0, 8, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
 803                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 804                     &weight_op[2], &weight_avg[2],
 805                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
 806                     pixel_shift, chroma_idc);
 807             }else{
 808                 int j;
 809                 assert(IS_SUB_4X4(sub_mb_type));
 810                 for(j=0; j<4; j++){
 811                     int sub_x_offset= x_offset + 2*(j&1);
 812                     int sub_y_offset= y_offset +   (j&2);
 813                     mc_part(h, n+j, 1, 4, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
 814                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 815                         &weight_op[2], &weight_avg[2],
 816                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
 817                         pixel_shift, chroma_idc);
 818                 }
 819             }
 820         }
 821     }
 822
 823     prefetch_motion(h, 1, pixel_shift, chroma_idc);
 824 }
 825
 826 static av_always_inline void
 827 hl_motion_420(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 828               qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
 829               qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
 830               h264_weight_func *weight_op, h264_biweight_func *weight_avg,
 831               int pixel_shift)
 832 {
 833     hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put,
 834               qpix_avg, chroma_avg, weight_op, weight_avg, pixel_shift, 1);
 835 }
 836
 837 static av_always_inline void
 838 hl_motion_422(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 839               qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
 840               qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
 841               h264_weight_func *weight_op, h264_biweight_func *weight_avg,
 842               int pixel_shift)
 843 {
 844     hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put,
 845               qpix_avg, chroma_avg, weight_op, weight_avg, pixel_shift, 2);
 846 }
 847
 848 static void free_tables(H264Context *h, int free_rbsp){
 849     int i;
 850     H264Context *hx;
 851
 852     av_freep(&h->intra4x4_pred_mode);
 853     av_freep(&h->chroma_pred_mode_table);
 854     av_freep(&h->cbp_table);
 855     av_freep(&h->mvd_table[0]);
 856     av_freep(&h->mvd_table[1]);
 857     av_freep(&h->direct_table);
 858     av_freep(&h->non_zero_count);
 859     av_freep(&h->slice_table_base);
 860     h->slice_table= NULL;
 861     av_freep(&h->list_counts);
 862
 863     av_freep(&h->mb2b_xy);
 864     av_freep(&h->mb2br_xy);
 865
 866     for(i = 0; i < MAX_THREADS; i++) {
 867         hx = h->thread_context[i];
 868         if(!hx) continue;
 869         av_freep(&hx->top_borders[1]);
 870         av_freep(&hx->top_borders[0]);
 871         av_freep(&hx->s.obmc_scratchpad);
 872         if (free_rbsp){
 873             av_freep(&hx->rbsp_buffer[1]);
 874             av_freep(&hx->rbsp_buffer[0]);
 875             hx->rbsp_buffer_size[0] = 0;
 876             hx->rbsp_buffer_size[1] = 0;
 877         }
 878         if (i) av_freep(&h->thread_context[i]);
 879     }
 880 }
 881
 882 static void init_dequant8_coeff_table(H264Context *h){
 883     int i,j,q,x;
 884     const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
 885
 886     for(i=0; i<6; i++ ){
 887         h->dequant8_coeff[i] = h->dequant8_buffer[i];
 888         for(j=0; j<i; j++){
 889             if(!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i], 64*sizeof(uint8_t))){
 890                 h->dequant8_coeff[i] = h->dequant8_buffer[j];
 891                 break;
 892             }
 893         }
 894         if(j<i)
 895             continue;
 896
 897         for(q=0; q<max_qp+1; q++){
 898             int shift = div6[q];
 899             int idx = rem6[q];
 900             for(x=0; x<64; x++)
 901                 h->dequant8_coeff[i][q][(x>>3)|((x&7)<<3)] =
 902                     ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
 903                     h->pps.scaling_matrix8[i][x]) << shift;
 904         }
 905     }
 906 }
 907
 908 static void init_dequant4_coeff_table(H264Context *h){
 909     int i,j,q,x;
 910     const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
 911     for(i=0; i<6; i++ ){
 912         h->dequant4_coeff[i] = h->dequant4_buffer[i];
 913         for(j=0; j<i; j++){
 914             if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
 915                 h->dequant4_coeff[i] = h->dequant4_buffer[j];
 916                 break;
 917             }
 918         }
 919         if(j<i)
 920             continue;
 921
 922         for(q=0; q<max_qp+1; q++){
 923             int shift = div6[q] + 2;
 924             int idx = rem6[q];
 925             for(x=0; x<16; x++)
 926                 h->dequant4_coeff[i][q][(x>>2)|((x<<2)&0xF)] =
 927                     ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
 928                     h->pps.scaling_matrix4[i][x]) << shift;
 929         }
 930     }
 931 }
 932
 933 static void init_dequant_tables(H264Context *h){
 934     int i,x;
 935     init_dequant4_coeff_table(h);
 936     if(h->pps.transform_8x8_mode)
 937         init_dequant8_coeff_table(h);
 938     if(h->sps.transform_bypass){
 939         for(i=0; i<6; i++)
 940             for(x=0; x<16; x++)
 941                 h->dequant4_coeff[i][0][x] = 1<<6;
 942         if(h->pps.transform_8x8_mode)
 943             for(i=0; i<6; i++)
 944                 for(x=0; x<64; x++)
 945                     h->dequant8_coeff[i][0][x] = 1<<6;
 946     }
 947 }
 948
 949
 950 int ff_h264_alloc_tables(H264Context *h){
 951     MpegEncContext * const s = &h->s;
 952     const int big_mb_num= s->mb_stride * (s->mb_height+1);
 953     const int row_mb_num= 2*s->mb_stride*FFMAX(s->avctx->thread_count, 1);
 954     int x,y;
 955
 956     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8  * sizeof(uint8_t), fail)
 957
 958     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 48 * sizeof(uint8_t), fail)
 959     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
 960     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
 961
 962     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
 963     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
 964     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
 965     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
 966     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
 967
 968     memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
 969     h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
 970
 971     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy  , big_mb_num * sizeof(uint32_t), fail);
 972     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
 973     for(y=0; y<s->mb_height; y++){
 974         for(x=0; x<s->mb_width; x++){
 975             const int mb_xy= x + y*s->mb_stride;
 976             const int b_xy = 4*x + 4*y*h->b_stride;
 977
 978             h->mb2b_xy [mb_xy]= b_xy;
 979             h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
 980         }
 981     }
 982
 983     s->obmc_scratchpad = NULL;
 984
 985     if(!h->dequant4_coeff[0])
 986         init_dequant_tables(h);
 987
 988     return 0;
 989 fail:
 990     free_tables(h, 1);
 991     return -1;
 992 }
 993
 994 /**
 995  * Mimic alloc_tables(), but for every context thread.
 996  */
 997 static void clone_tables(H264Context *dst, H264Context *src, int i){
 998     MpegEncContext * const s = &src->s;
 999     dst->intra4x4_pred_mode       = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
1000     dst->non_zero_count           = src->non_zero_count;
1001     dst->slice_table              = src->slice_table;
1002     dst->cbp_table                = src->cbp_table;
1003     dst->mb2b_xy                  = src->mb2b_xy;
1004     dst->mb2br_xy                 = src->mb2br_xy;
1005     dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
1006     dst->mvd_table[0]             = src->mvd_table[0] + i*8*2*s->mb_stride;
1007     dst->mvd_table[1]             = src->mvd_table[1] + i*8*2*s->mb_stride;
1008     dst->direct_table             = src->direct_table;
1009     dst->list_counts              = src->list_counts;
1010
1011     dst->s.obmc_scratchpad = NULL;
1012     ff_h264_pred_init(&dst->hpc, src->s.codec_id, src->sps.bit_depth_luma, src->sps.chroma_format_idc);
1013 }
1014
1015 /**
1016  * Init context
1017  * Allocate buffers which are not shared amongst multiple threads.
1018  */
1019 static int context_init(H264Context *h){
1020     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
1021     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
1022
1023     h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
1024     h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
1025
1026     return 0;
1027 fail:
1028     return -1; // free_tables will clean up for us
1029 }
1030
1031 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
1032
1033 static av_cold void common_init(H264Context *h){
1034     MpegEncContext * const s = &h->s;
1035
1036     s->width = s->avctx->width;
1037     s->height = s->avctx->height;
1038     s->codec_id= s->avctx->codec->id;
1039
1040     s->avctx->bits_per_raw_sample = 8;
1041     h->cur_chroma_format_idc = 1;
1042
1043     ff_h264dsp_init(&h->h264dsp,
1044                     s->avctx->bits_per_raw_sample, h->cur_chroma_format_idc);
1045     ff_h264_pred_init(&h->hpc, s->codec_id,
1046                       s->avctx->bits_per_raw_sample, h->cur_chroma_format_idc);
1047
1048     h->dequant_coeff_pps= -1;
1049     s->unrestricted_mv=1;
1050
1051     s->dsp.dct_bits = 16;
1052     ff_dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
1053
1054     memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
1055     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
1056 }
1057
1058 int ff_h264_decode_extradata(H264Context *h, const uint8_t *buf, int size)
1059 {
1060     AVCodecContext *avctx = h->s.avctx;
1061
1062     if(!buf || size <= 0)
1063         return -1;
1064
1065     if(buf[0] == 1){
1066         int i, cnt, nalsize;
1067         const unsigned char *p = buf;
1068
1069         h->is_avc = 1;
1070
1071         if(size < 7) {
1072             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
1073             return -1;
1074         }
1075         /* sps and pps in the avcC always have length coded with 2 bytes,
1076            so put a fake nal_length_size = 2 while parsing them */
1077         h->nal_length_size = 2;
1078         // Decode sps from avcC
1079         cnt = *(p+5) & 0x1f; // Number of sps
1080         p += 6;
1081         for (i = 0; i < cnt; i++) {
1082             nalsize = AV_RB16(p) + 2;
1083             if(nalsize > size - (p-buf))
1084                 return -1;
1085             if(decode_nal_units(h, p, nalsize) < 0) {
1086                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
1087                 return -1;
1088             }
1089             p += nalsize;
1090         }
1091         // Decode pps from avcC
1092         cnt = *(p++); // Number of pps
1093         for (i = 0; i < cnt; i++) {
1094             nalsize = AV_RB16(p) + 2;
1095             if(nalsize > size - (p-buf))
1096                 return -1;
1097             if (decode_nal_units(h, p, nalsize) < 0) {
1098                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
1099                 return -1;
1100             }
1101             p += nalsize;
1102         }
1103         // Now store right nal length size, that will be use to parse all other nals
1104         h->nal_length_size = (buf[4] & 0x03) + 1;
1105     } else {
1106         h->is_avc = 0;
1107         if(decode_nal_units(h, buf, size) < 0)
1108             return -1;
1109     }
1110     return size;
1111 }
1112
1113 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
1114     H264Context *h= avctx->priv_data;
1115     MpegEncContext * const s = &h->s;
1116     int i;
1117
1118     ff_MPV_decode_defaults(s);
1119
1120     s->avctx = avctx;
1121     common_init(h);
1122
1123     s->out_format = FMT_H264;
1124     s->workaround_bugs= avctx->workaround_bugs;
1125
1126     // set defaults
1127 //    s->decode_mb= ff_h263_decode_mb;
1128     s->quarter_sample = 1;
1129     if(!avctx->has_b_frames)
1130     s->low_delay= 1;
1131
1132     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
1133
1134     ff_h264_decode_init_vlc();
1135
1136     h->pixel_shift = 0;
1137     h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
1138
1139     h->thread_context[0] = h;
1140     h->outputed_poc = h->next_outputed_poc = INT_MIN;
1141     for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1142         h->last_pocs[i] = INT_MIN;
1143     h->prev_poc_msb= 1<<16;
1144     h->prev_frame_num= -1;
1145     h->x264_build = -1;
1146     ff_h264_reset_sei(h);
1147     if(avctx->codec_id == CODEC_ID_H264){
1148         if(avctx->ticks_per_frame == 1){
1149             s->avctx->time_base.den *=2;
1150         }
1151         avctx->ticks_per_frame = 2;
1152     }
1153
1154     if(avctx->extradata_size > 0 && avctx->extradata &&
1155         ff_h264_decode_extradata(h, avctx->extradata, avctx->extradata_size)<0)
1156         return -1;
1157
1158     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
1159         s->avctx->has_b_frames = h->sps.num_reorder_frames;
1160         s->low_delay = 0;
1161     }
1162
1163     return 0;
1164 }
1165
1166 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b)+(size))))
1167 static void copy_picture_range(Picture **to, Picture **from, int count, MpegEncContext *new_base, MpegEncContext *old_base)
1168 {
1169     int i;
1170
1171     for (i=0; i<count; i++){
1172         assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
1173                 IN_RANGE(from[i], old_base->picture, sizeof(Picture) * old_base->picture_count) ||
1174                 !from[i]));
1175         to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1176     }
1177 }
1178
1179 static void copy_parameter_set(void **to, void **from, int count, int size)
1180 {
1181     int i;
1182
1183     for (i=0; i<count; i++){
1184         if (to[i] && !from[i]) av_freep(&to[i]);
1185         else if (from[i] && !to[i]) to[i] = av_malloc(size);
1186
1187         if (from[i]) memcpy(to[i], from[i], size);
1188     }
1189 }
1190
1191 static int decode_init_thread_copy(AVCodecContext *avctx){
1192     H264Context *h= avctx->priv_data;
1193
1194     if (!avctx->internal->is_copy)
1195         return 0;
1196     memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1197     memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1198
1199     return 0;
1200 }
1201
1202 #define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
1203 static int decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src){
1204     H264Context *h= dst->priv_data, *h1= src->priv_data;
1205     MpegEncContext * const s = &h->s, * const s1 = &h1->s;
1206     int inited = s->context_initialized, err;
1207     int i;
1208
1209     if(dst == src) return 0;
1210
1211     err = ff_mpeg_update_thread_context(dst, src);
1212     if(err) return err;
1213
1214     //FIXME handle width/height changing
1215     if(!inited){
1216         for(i = 0; i < MAX_SPS_COUNT; i++)
1217             av_freep(h->sps_buffers + i);
1218
1219         for(i = 0; i < MAX_PPS_COUNT; i++)
1220             av_freep(h->pps_buffers + i);
1221
1222         memcpy(&h->s + 1, &h1->s + 1, sizeof(H264Context) - sizeof(MpegEncContext)); //copy all fields after MpegEnc
1223         memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1224         memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1225
1226         if (s1->context_initialized) {
1227         if (ff_h264_alloc_tables(h) < 0) {
1228             av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
1229             return AVERROR(ENOMEM);
1230         }
1231         context_init(h);
1232
1233         // frame_start may not be called for the next thread (if it's decoding a bottom field)
1234         // so this has to be allocated here
1235         h->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
1236         }
1237
1238         for(i=0; i<2; i++){
1239             h->rbsp_buffer[i] = NULL;
1240             h->rbsp_buffer_size[i] = 0;
1241         }
1242
1243         h->thread_context[0] = h;
1244
1245         s->dsp.clear_blocks(h->mb);
1246         s->dsp.clear_blocks(h->mb+(24*16<<h->pixel_shift));
1247     }
1248
1249     //extradata/NAL handling
1250     h->is_avc          = h1->is_avc;
1251
1252     //SPS/PPS
1253     copy_parameter_set((void**)h->sps_buffers, (void**)h1->sps_buffers, MAX_SPS_COUNT, sizeof(SPS));
1254     h->sps             = h1->sps;
1255     copy_parameter_set((void**)h->pps_buffers, (void**)h1->pps_buffers, MAX_PPS_COUNT, sizeof(PPS));
1256     h->pps             = h1->pps;
1257
1258     //Dequantization matrices
1259     //FIXME these are big - can they be only copied when PPS changes?
1260     copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1261
1262     for(i=0; i<6; i++)
1263         h->dequant4_coeff[i] = h->dequant4_buffer[0] + (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
1264
1265     for(i=0; i<6; i++)
1266         h->dequant8_coeff[i] = h->dequant8_buffer[0] + (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
1267
1268     h->dequant_coeff_pps = h1->dequant_coeff_pps;
1269
1270     //POC timing
1271     copy_fields(h, h1, poc_lsb, redundant_pic_count);
1272
1273     //reference lists
1274     copy_fields(h, h1, ref_count, list_count);
1275     copy_fields(h, h1, ref_list,  intra_gb);
1276     copy_fields(h, h1, short_ref, cabac_init_idc);
1277
1278     copy_picture_range(h->short_ref,   h1->short_ref,   32, s, s1);
1279     copy_picture_range(h->long_ref,    h1->long_ref,    32, s, s1);
1280     copy_picture_range(h->delayed_pic, h1->delayed_pic, MAX_DELAYED_PIC_COUNT+2, s, s1);
1281
1282     h->last_slice_type = h1->last_slice_type;
1283     h->sync            = h1->sync;
1284
1285     if(!s->current_picture_ptr) return 0;
1286
1287     if(!s->dropable) {
1288         err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1289         h->prev_poc_msb     = h->poc_msb;
1290         h->prev_poc_lsb     = h->poc_lsb;
1291     }
1292     h->prev_frame_num_offset= h->frame_num_offset;
1293     h->prev_frame_num       = h->frame_num;
1294     h->outputed_poc         = h->next_outputed_poc;
1295
1296     return err;
1297 }
1298
1299 int ff_h264_frame_start(H264Context *h){
1300     MpegEncContext * const s = &h->s;
1301     int i;
1302     const int pixel_shift = h->pixel_shift;
1303
1304     if(ff_MPV_frame_start(s, s->avctx) < 0)
1305         return -1;
1306     ff_er_frame_start(s);
1307     /*
1308      * ff_MPV_frame_start uses pict_type to derive key_frame.
1309      * This is incorrect for H.264; IDR markings must be used.
1310      * Zero here; IDR markings per slice in frame or fields are ORed in later.
1311      * See decode_nal_units().
1312      */
1313     s->current_picture_ptr->f.key_frame = 0;
1314     s->current_picture_ptr->sync = 0;
1315     s->current_picture_ptr->mmco_reset= 0;
1316
1317     assert(s->linesize && s->uvlinesize);
1318
1319     for(i=0; i<16; i++){
1320         h->block_offset[i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
1321         h->block_offset[48+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
1322     }
1323     for(i=0; i<16; i++){
1324         h->block_offset[16+i]=
1325         h->block_offset[32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1326         h->block_offset[48+16+i]=
1327         h->block_offset[48+32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1328     }
1329
1330     /* can't be in alloc_tables because linesize isn't known there.
1331      * FIXME: redo bipred weight to not require extra buffer? */
1332     for(i = 0; i < s->slice_context_count; i++)
1333         if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
1334             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
1335
1336     /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
1337     memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
1338
1339 //    s->decode = (s->flags & CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.f.reference /*|| h->contains_intra*/ || 1;
1340
1341     // We mark the current picture as non-reference after allocating it, so
1342     // that if we break out due to an error it can be released automatically
1343     // in the next ff_MPV_frame_start().
1344     // SVQ3 as well as most other codecs have only last/next/current and thus
1345     // get released even with set reference, besides SVQ3 and others do not
1346     // mark frames as reference later "naturally".
1347     if(s->codec_id != CODEC_ID_SVQ3)
1348         s->current_picture_ptr->f.reference = 0;
1349
1350     s->current_picture_ptr->field_poc[0]=
1351     s->current_picture_ptr->field_poc[1]= INT_MAX;
1352
1353     h->next_output_pic = NULL;
1354
1355     assert(s->current_picture_ptr->long_ref==0);
1356
1357     return 0;
1358 }
1359
1360 /**
1361   * Run setup operations that must be run after slice header decoding.
1362   * This includes finding the next displayed frame.
1363   *
1364   * @param h h264 master context
1365   * @param setup_finished enough NALs have been read that we can call
1366   * ff_thread_finish_setup()
1367   */
1368 static void decode_postinit(H264Context *h, int setup_finished){
1369     MpegEncContext * const s = &h->s;
1370     Picture *out = s->current_picture_ptr;
1371     Picture *cur = s->current_picture_ptr;
1372     int i, pics, out_of_order, out_idx;
1373
1374     s->current_picture_ptr->f.qscale_type = FF_QSCALE_TYPE_H264;
1375     s->current_picture_ptr->f.pict_type   = s->pict_type;
1376
1377     if (h->next_output_pic) return;
1378
1379     if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
1380         //FIXME: if we have two PAFF fields in one packet, we can't start the next thread here.
1381         //If we have one field per packet, we can. The check in decode_nal_units() is not good enough
1382         //to find this yet, so we assume the worst for now.
1383         //if (setup_finished)
1384         //    ff_thread_finish_setup(s->avctx);
1385         return;
1386     }
1387
1388     cur->f.interlaced_frame = 0;
1389     cur->f.repeat_pict      = 0;
1390
1391     /* Signal interlacing information externally. */
1392     /* Prioritize picture timing SEI information over used decoding process if it exists. */
1393
1394     if(h->sps.pic_struct_present_flag){
1395         switch (h->sei_pic_struct)
1396         {
1397         case SEI_PIC_STRUCT_FRAME:
1398             break;
1399         case SEI_PIC_STRUCT_TOP_FIELD:
1400         case SEI_PIC_STRUCT_BOTTOM_FIELD:
1401             cur->f.interlaced_frame = 1;
1402             break;
1403         case SEI_PIC_STRUCT_TOP_BOTTOM:
1404         case SEI_PIC_STRUCT_BOTTOM_TOP:
1405             if (FIELD_OR_MBAFF_PICTURE)
1406                 cur->f.interlaced_frame = 1;
1407             else
1408                 // try to flag soft telecine progressive
1409                 cur->f.interlaced_frame = h->prev_interlaced_frame;
1410             break;
1411         case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1412         case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
1413             // Signal the possibility of telecined film externally (pic_struct 5,6)
1414             // From these hints, let the applications decide if they apply deinterlacing.
1415             cur->f.repeat_pict = 1;
1416             break;
1417         case SEI_PIC_STRUCT_FRAME_DOUBLING:
1418             // Force progressive here, as doubling interlaced frame is a bad idea.
1419             cur->f.repeat_pict = 2;
1420             break;
1421         case SEI_PIC_STRUCT_FRAME_TRIPLING:
1422             cur->f.repeat_pict = 4;
1423             break;
1424         }
1425
1426         if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
1427             cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
1428     }else{
1429         /* Derive interlacing flag from used decoding process. */
1430         cur->f.interlaced_frame = FIELD_OR_MBAFF_PICTURE;
1431     }
1432     h->prev_interlaced_frame = cur->f.interlaced_frame;
1433
1434     if (cur->field_poc[0] != cur->field_poc[1]){
1435         /* Derive top_field_first from field pocs. */
1436         cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1];
1437     }else{
1438         if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) {
1439             /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
1440             if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
1441               || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
1442                 cur->f.top_field_first = 1;
1443             else
1444                 cur->f.top_field_first = 0;
1445         }else{
1446             /* Most likely progressive */
1447             cur->f.top_field_first = 0;
1448         }
1449     }
1450
1451     cur->mmco_reset = h->mmco_reset;
1452     h->mmco_reset = 0;
1453     //FIXME do something with unavailable reference frames
1454
1455     /* Sort B-frames into display order */
1456
1457     if(h->sps.bitstream_restriction_flag
1458        && s->avctx->has_b_frames < h->sps.num_reorder_frames){
1459         s->avctx->has_b_frames = h->sps.num_reorder_frames;
1460         s->low_delay = 0;
1461     }
1462
1463     if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
1464        && !h->sps.bitstream_restriction_flag){
1465         s->avctx->has_b_frames = MAX_DELAYED_PIC_COUNT - 1;
1466         s->low_delay= 0;
1467     }
1468
1469     for (i = 0; 1; i++) {
1470         if(i == MAX_DELAYED_PIC_COUNT || cur->poc < h->last_pocs[i]){
1471             if(i)
1472                 h->last_pocs[i-1] = cur->poc;
1473             break;
1474         } else if(i) {
1475             h->last_pocs[i-1]= h->last_pocs[i];
1476         }
1477     }
1478     out_of_order = MAX_DELAYED_PIC_COUNT - i;
1479     if(   cur->f.pict_type == AV_PICTURE_TYPE_B
1480        || (h->last_pocs[MAX_DELAYED_PIC_COUNT-2] > INT_MIN && h->last_pocs[MAX_DELAYED_PIC_COUNT-1] - h->last_pocs[MAX_DELAYED_PIC_COUNT-2] > 2))
1481         out_of_order = FFMAX(out_of_order, 1);
1482     if(s->avctx->has_b_frames < out_of_order && !h->sps.bitstream_restriction_flag){
1483         av_log(s->avctx, AV_LOG_WARNING, "Increasing reorder buffer to %d\n", out_of_order);
1484         s->avctx->has_b_frames = out_of_order;
1485         s->low_delay = 0;
1486     }
1487
1488     pics = 0;
1489     while(h->delayed_pic[pics]) pics++;
1490
1491     av_assert0(pics <= MAX_DELAYED_PIC_COUNT);
1492
1493     h->delayed_pic[pics++] = cur;
1494     if (cur->f.reference == 0)
1495         cur->f.reference = DELAYED_PIC_REF;
1496
1497     out = h->delayed_pic[0];
1498     out_idx = 0;
1499     for (i = 1; h->delayed_pic[i] && !h->delayed_pic[i]->f.key_frame && !h->delayed_pic[i]->mmco_reset; i++)
1500         if(h->delayed_pic[i]->poc < out->poc){
1501             out = h->delayed_pic[i];
1502             out_idx = i;
1503         }
1504     if (s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset))
1505         h->next_outputed_poc= INT_MIN;
1506     out_of_order = out->poc < h->next_outputed_poc;
1507
1508     if(out_of_order || pics > s->avctx->has_b_frames){
1509         out->f.reference &= ~DELAYED_PIC_REF;
1510         out->owner2 = s; // for frame threading, the owner must be the second field's thread
1511                          // or else the first thread can release the picture and reuse it unsafely
1512         for(i=out_idx; h->delayed_pic[i]; i++)
1513             h->delayed_pic[i] = h->delayed_pic[i+1];
1514     }
1515     if(!out_of_order && pics > s->avctx->has_b_frames){
1516         h->next_output_pic = out;
1517         if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset)) {
1518             h->next_outputed_poc = INT_MIN;
1519         } else
1520             h->next_outputed_poc = out->poc;
1521     }else{
1522         av_log(s->avctx, AV_LOG_DEBUG, "no picture %s\n", out_of_order ? "ooo" : "");
1523     }
1524
1525     if (h->next_output_pic && h->next_output_pic->sync) {
1526         h->sync |= 2;
1527     }
1528
1529     if (setup_finished)
1530         ff_thread_finish_setup(s->avctx);
1531 }
1532
1533 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
1534                                               uint8_t *src_cb, uint8_t *src_cr,
1535                                               int linesize, int uvlinesize, int simple)
1536 {
1537     MpegEncContext * const s = &h->s;
1538     uint8_t *top_border;
1539     int top_idx = 1;
1540     const int pixel_shift = h->pixel_shift;
1541     int chroma444 = CHROMA444;
1542     int chroma422 = CHROMA422;
1543
1544     src_y  -=   linesize;
1545     src_cb -= uvlinesize;
1546     src_cr -= uvlinesize;
1547
1548     if(!simple && FRAME_MBAFF){
1549         if(s->mb_y&1){
1550             if(!MB_MBAFF){
1551                 top_border = h->top_borders[0][s->mb_x];
1552                 AV_COPY128(top_border, src_y + 15*linesize);
1553                 if (pixel_shift)
1554                     AV_COPY128(top_border+16, src_y+15*linesize+16);
1555                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1556                     if(chroma444){
1557                         if (pixel_shift){
1558                             AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
1559                             AV_COPY128(top_border+48, src_cb + 15*uvlinesize+16);
1560                             AV_COPY128(top_border+64, src_cr + 15*uvlinesize);
1561                             AV_COPY128(top_border+80, src_cr + 15*uvlinesize+16);
1562                         } else {
1563                             AV_COPY128(top_border+16, src_cb + 15*uvlinesize);
1564                             AV_COPY128(top_border+32, src_cr + 15*uvlinesize);
1565                         }
1566                     } else if(chroma422){
1567                         if (pixel_shift) {
1568                             AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
1569                             AV_COPY128(top_border+48, src_cr + 15*uvlinesize);
1570                         } else {
1571                             AV_COPY64(top_border+16, src_cb +  15*uvlinesize);
1572                             AV_COPY64(top_border+24, src_cr +  15*uvlinesize);
1573                         }
1574                     } else {
1575                         if (pixel_shift) {
1576                             AV_COPY128(top_border+32, src_cb+7*uvlinesize);
1577                             AV_COPY128(top_border+48, src_cr+7*uvlinesize);
1578                         } else {
1579                             AV_COPY64(top_border+16, src_cb+7*uvlinesize);
1580                             AV_COPY64(top_border+24, src_cr+7*uvlinesize);
1581                         }
1582                     }
1583                 }
1584             }
1585         }else if(MB_MBAFF){
1586             top_idx = 0;
1587         }else
1588             return;
1589     }
1590
1591     top_border = h->top_borders[top_idx][s->mb_x];
1592     // There are two lines saved, the line above the the top macroblock of a pair,
1593     // and the line above the bottom macroblock
1594     AV_COPY128(top_border, src_y + 16*linesize);
1595     if (pixel_shift)
1596         AV_COPY128(top_border+16, src_y+16*linesize+16);
1597
1598     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1599         if(chroma444){
1600             if (pixel_shift){
1601                 AV_COPY128(top_border+32, src_cb + 16*linesize);
1602                 AV_COPY128(top_border+48, src_cb + 16*linesize+16);
1603                 AV_COPY128(top_border+64, src_cr + 16*linesize);
1604                 AV_COPY128(top_border+80, src_cr + 16*linesize+16);
1605             } else {
1606                 AV_COPY128(top_border+16, src_cb + 16*linesize);
1607                 AV_COPY128(top_border+32, src_cr + 16*linesize);
1608             }
1609         } else if(chroma422) {
1610             if (pixel_shift) {
1611                 AV_COPY128(top_border+32, src_cb+16*uvlinesize);
1612                 AV_COPY128(top_border+48, src_cr+16*uvlinesize);
1613             } else {
1614                 AV_COPY64(top_border+16, src_cb+16*uvlinesize);
1615                 AV_COPY64(top_border+24, src_cr+16*uvlinesize);
1616             }
1617         } else {
1618             if (pixel_shift) {
1619                 AV_COPY128(top_border+32, src_cb+8*uvlinesize);
1620                 AV_COPY128(top_border+48, src_cr+8*uvlinesize);
1621             } else {
1622                 AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1623                 AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1624             }
1625         }
1626     }
1627 }
1628
1629 static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
1630                                   uint8_t *src_cb, uint8_t *src_cr,
1631                                   int linesize, int uvlinesize,
1632                                   int xchg, int chroma444,
1633                                   int simple, int pixel_shift){
1634     MpegEncContext * const s = &h->s;
1635     int deblock_topleft;
1636     int deblock_top;
1637     int top_idx = 1;
1638     uint8_t *top_border_m1;
1639     uint8_t *top_border;
1640
1641     if(!simple && FRAME_MBAFF){
1642         if(s->mb_y&1){
1643             if(!MB_MBAFF)
1644                 return;
1645         }else{
1646             top_idx = MB_MBAFF ? 0 : 1;
1647         }
1648     }
1649
1650     if(h->deblocking_filter == 2) {
1651         deblock_topleft = h->slice_table[h->mb_xy - 1 - s->mb_stride] == h->slice_num;
1652         deblock_top     = h->top_type;
1653     } else {
1654         deblock_topleft = (s->mb_x > 0);
1655         deblock_top     = (s->mb_y > !!MB_FIELD);
1656     }
1657
1658     src_y  -=   linesize + 1 + pixel_shift;
1659     src_cb -= uvlinesize + 1 + pixel_shift;
1660     src_cr -= uvlinesize + 1 + pixel_shift;
1661
1662     top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1663     top_border    = h->top_borders[top_idx][s->mb_x];
1664
1665 #define XCHG(a,b,xchg)\
1666     if (pixel_shift) {\
1667         if (xchg) {\
1668             AV_SWAP64(b+0,a+0);\
1669             AV_SWAP64(b+8,a+8);\
1670         } else {\
1671             AV_COPY128(b,a); \
1672         }\
1673     } else \
1674 if (xchg) AV_SWAP64(b,a);\
1675 else      AV_COPY64(b,a);
1676
1677     if(deblock_top){
1678         if(deblock_topleft){
1679             XCHG(top_border_m1 + (8 << pixel_shift), src_y - (7 << pixel_shift), 1);
1680         }
1681         XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
1682         XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
1683         if(s->mb_x+1 < s->mb_width){
1684             XCHG(h->top_borders[top_idx][s->mb_x+1], src_y + (17 << pixel_shift), 1);
1685         }
1686     }
1687     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1688         if(chroma444){
1689             if(deblock_topleft){
1690                 XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1691                 XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1692             }
1693             XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
1694             XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
1695             XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
1696             XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
1697             if(s->mb_x+1 < s->mb_width){
1698                 XCHG(h->top_borders[top_idx][s->mb_x+1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
1699                 XCHG(h->top_borders[top_idx][s->mb_x+1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
1700             }
1701         } else {
1702             if(deblock_top){
1703                 if(deblock_topleft){
1704                     XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1705                     XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1706                 }
1707                 XCHG(top_border + (16 << pixel_shift), src_cb+1+pixel_shift, 1);
1708                 XCHG(top_border + (24 << pixel_shift), src_cr+1+pixel_shift, 1);
1709             }
1710         }
1711     }
1712 }
1713
1714 static av_always_inline int dctcoef_get(DCTELEM *mb, int high_bit_depth, int index) {
1715     if (high_bit_depth) {
1716         return AV_RN32A(((int32_t*)mb) + index);
1717     } else
1718         return AV_RN16A(mb + index);
1719 }
1720
1721 static av_always_inline void dctcoef_set(DCTELEM *mb, int high_bit_depth, int index, int value) {
1722     if (high_bit_depth) {
1723         AV_WN32A(((int32_t*)mb) + index, value);
1724     } else
1725         AV_WN16A(mb + index, value);
1726 }
1727
1728 static av_always_inline void hl_decode_mb_predict_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1729                                                        int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1730 {
1731     MpegEncContext * const s = &h->s;
1732     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1733     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1734     int i;
1735     int qscale = p == 0 ? s->qscale : h->chroma_qp[p-1];
1736     block_offset += 16*p;
1737     if(IS_INTRA4x4(mb_type)){
1738         if(simple || !s->encoding){
1739             if(IS_8x8DCT(mb_type)){
1740                 if(transform_bypass){
1741                     idct_dc_add =
1742                     idct_add    = s->dsp.add_pixels8;
1743                 }else{
1744                     idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1745                     idct_add    = h->h264dsp.h264_idct8_add;
1746                 }
1747                 for(i=0; i<16; i+=4){
1748                     uint8_t * const ptr= dest_y + block_offset[i];
1749                     const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1750                     if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1751                         h->hpc.pred8x8l_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1752                     }else{
1753                         const int nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1754                         h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1755                                                     (h->topright_samples_available<<i)&0x4000, linesize);
1756                         if(nnz){
1757                             if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1758                                 idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1759                             else
1760                                 idct_add   (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1761                         }
1762                     }
1763                 }
1764             }else{
1765                 if(transform_bypass){
1766                     idct_dc_add =
1767                     idct_add    = s->dsp.add_pixels4;
1768                 }else{
1769                     idct_dc_add = h->h264dsp.h264_idct_dc_add;
1770                     idct_add    = h->h264dsp.h264_idct_add;
1771                 }
1772                 for(i=0; i<16; i++){
1773                     uint8_t * const ptr= dest_y + block_offset[i];
1774                     const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1775
1776                     if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1777                         h->hpc.pred4x4_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1778                     }else{
1779                         uint8_t *topright;
1780                         int nnz, tr;
1781                         uint64_t tr_high;
1782                         if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1783                             const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1784                             assert(s->mb_y || linesize <= block_offset[i]);
1785                             if(!topright_avail){
1786                                 if (pixel_shift) {
1787                                     tr_high= ((uint16_t*)ptr)[3 - linesize/2]*0x0001000100010001ULL;
1788                                     topright= (uint8_t*) &tr_high;
1789                                 } else {
1790                                     tr= ptr[3 - linesize]*0x01010101u;
1791                                     topright= (uint8_t*) &tr;
1792                                 }
1793                             }else
1794                                 topright= ptr + (4 << pixel_shift) - linesize;
1795                         }else
1796                             topright= NULL;
1797
1798                         h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1799                         nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1800                         if(nnz){
1801                             if(is_h264){
1802                                 if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1803                                     idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1804                                 else
1805                                     idct_add   (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1806                             } else if (CONFIG_SVQ3_DECODER)
1807                                 ff_svq3_add_idct_c(ptr, h->mb + i*16+p*256, linesize, qscale, 0);
1808                         }
1809                     }
1810                 }
1811             }
1812         }
1813     }else{
1814         h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1815         if(is_h264){
1816             if(h->non_zero_count_cache[ scan8[LUMA_DC_BLOCK_INDEX+p] ]){
1817                 if(!transform_bypass)
1818                     h->h264dsp.h264_luma_dc_dequant_idct(h->mb+(p*256 << pixel_shift), h->mb_luma_dc[p], h->dequant4_coeff[p][qscale][0]);
1819                 else{
1820                     static const uint8_t dc_mapping[16] = { 0*16, 1*16, 4*16, 5*16, 2*16, 3*16, 6*16, 7*16,
1821                                                             8*16, 9*16,12*16,13*16,10*16,11*16,14*16,15*16};
1822                     for(i = 0; i < 16; i++)
1823                         dctcoef_set(h->mb+(p*256 << pixel_shift), pixel_shift, dc_mapping[i], dctcoef_get(h->mb_luma_dc[p], pixel_shift, i));
1824                 }
1825             }
1826         } else if (CONFIG_SVQ3_DECODER)
1827             ff_svq3_luma_dc_dequant_idct_c(h->mb+p*256, h->mb_luma_dc[p], qscale);
1828     }
1829 }
1830
1831 static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1832                                                     int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1833 {
1834     MpegEncContext * const s = &h->s;
1835     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1836     int i;
1837     block_offset += 16*p;
1838     if(!IS_INTRA4x4(mb_type)){
1839         if(is_h264){
1840             if(IS_INTRA16x16(mb_type)){
1841                 if(transform_bypass){
1842                     if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1843                         h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize);
1844                     }else{
1845                         for(i=0; i<16; i++){
1846                             if(h->non_zero_count_cache[ scan8[i+p*16] ] || dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1847                                 s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1848                         }
1849                     }
1850                 }else{
1851                     h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1852                 }
1853             }else if(h->cbp&15){
1854                 if(transform_bypass){
1855                     const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1856                     idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1857                     for(i=0; i<16; i+=di){
1858                         if(h->non_zero_count_cache[ scan8[i+p*16] ]){
1859                             idct_add(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1860                         }
1861                     }
1862                 }else{
1863                     if(IS_8x8DCT(mb_type)){
1864                         h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1865                     }else{
1866                         h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1867                     }
1868                 }
1869             }
1870         } else if (CONFIG_SVQ3_DECODER) {
1871             for(i=0; i<16; i++){
1872                 if(h->non_zero_count_cache[ scan8[i+p*16] ] || h->mb[i*16+p*256]){ //FIXME benchmark weird rule, & below
1873                     uint8_t * const ptr= dest_y + block_offset[i];
1874                     ff_svq3_add_idct_c(ptr, h->mb + i*16 + p*256, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1875                 }
1876             }
1877         }
1878     }
1879 }
1880
1881 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, int pixel_shift)
1882 {
1883     MpegEncContext * const s = &h->s;
1884     const int mb_x= s->mb_x;
1885     const int mb_y= s->mb_y;
1886     const int mb_xy= h->mb_xy;
1887     const int mb_type = s->current_picture.f.mb_type[mb_xy];
1888     uint8_t  *dest_y, *dest_cb, *dest_cr;
1889     int linesize, uvlinesize /*dct_offset*/;
1890     int i, j;
1891     int *block_offset = &h->block_offset[0];
1892     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1893     /* is_h264 should always be true if SVQ3 is disabled. */
1894     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1895     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1896     const int block_h = 16 >> s->chroma_y_shift;
1897     const int chroma422 = CHROMA422;
1898
1899     dest_y  = s->current_picture.f.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize  ) * 16;
1900     dest_cb = s->current_picture.f.data[1] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
1901     dest_cr = s->current_picture.f.data[2] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
1902
1903     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
1904     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + (64 << pixel_shift), dest_cr - dest_cb, 2);
1905
1906     h->list_counts[mb_xy]= h->list_count;
1907
1908     if (!simple && MB_FIELD) {
1909         linesize   = h->mb_linesize   = s->linesize * 2;
1910         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1911         block_offset = &h->block_offset[48];
1912         if(mb_y&1){ //FIXME move out of this function?
1913             dest_y -= s->linesize*15;
1914             dest_cb-= s->uvlinesize * (block_h - 1);
1915             dest_cr-= s->uvlinesize * (block_h - 1);
1916         }
1917         if(FRAME_MBAFF) {
1918             int list;
1919             for(list=0; list<h->list_count; list++){
1920                 if(!USES_LIST(mb_type, list))
1921                     continue;
1922                 if(IS_16X16(mb_type)){
1923                     int8_t *ref = &h->ref_cache[list][scan8[0]];
1924                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1925                 }else{
1926                     for(i=0; i<16; i+=4){
1927                         int ref = h->ref_cache[list][scan8[i]];
1928                         if(ref >= 0)
1929                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1930                     }
1931                 }
1932             }
1933         }
1934     } else {
1935         linesize   = h->mb_linesize   = s->linesize;
1936         uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1937 //        dct_offset = s->linesize * 16;
1938     }
1939
1940     if (!simple && IS_INTRA_PCM(mb_type)) {
1941         const int bit_depth = h->sps.bit_depth_luma;
1942         if (pixel_shift) {
1943             int j;
1944             GetBitContext gb;
1945             init_get_bits(&gb, (uint8_t*)h->mb, 384*bit_depth);
1946
1947             for (i = 0; i < 16; i++) {
1948                 uint16_t *tmp_y  = (uint16_t*)(dest_y  + i*linesize);
1949                 for (j = 0; j < 16; j++)
1950                     tmp_y[j] = get_bits(&gb, bit_depth);
1951             }
1952             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1953                 if (!h->sps.chroma_format_idc) {
1954                     for (i = 0; i < block_h; i++) {
1955                         uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1956                         uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1957                         for (j = 0; j < 8; j++) {
1958                             tmp_cb[j] = tmp_cr[j] = 1 << (bit_depth - 1);
1959                         }
1960                     }
1961                 } else {
1962                     for (i = 0; i < block_h; i++) {
1963                         uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1964                         for (j = 0; j < 8; j++)
1965                             tmp_cb[j] = get_bits(&gb, bit_depth);
1966                     }
1967                     for (i = 0; i < block_h; i++) {
1968                         uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1969                         for (j = 0; j < 8; j++)
1970                             tmp_cr[j] = get_bits(&gb, bit_depth);
1971                     }
1972                 }
1973             }
1974         } else {
1975             for (i=0; i<16; i++) {
1976                 memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1977             }
1978             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1979                 if (!h->sps.chroma_format_idc) {
1980                     for (i=0; i<8; i++) {
1981                         memset(dest_cb + i*uvlinesize, 1 << (bit_depth - 1), 8);
1982                         memset(dest_cr + i*uvlinesize, 1 << (bit_depth - 1), 8);
1983                     }
1984                 } else {
1985                     for (i=0; i<block_h; i++) {
1986                         memcpy(dest_cb + i*uvlinesize, h->mb + 128 + i*4,  8);
1987                         memcpy(dest_cr + i*uvlinesize, h->mb + 160 + i*4,  8);
1988                     }
1989                 }
1990             }
1991         }
1992     } else {
1993         if(IS_INTRA(mb_type)){
1994             if(h->deblocking_filter)
1995                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, 0, simple, pixel_shift);
1996
1997             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1998                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1999                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2000             }
2001
2002             hl_decode_mb_predict_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
2003
2004             if(h->deblocking_filter)
2005                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, 0, simple, pixel_shift);
2006         }else if(is_h264){
2007             if (chroma422) {
2008                 hl_motion_422(h, dest_y, dest_cb, dest_cr,
2009                               s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2010                               s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2011                               h->h264dsp.weight_h264_pixels_tab,
2012                               h->h264dsp.biweight_h264_pixels_tab,
2013                               pixel_shift);
2014             } else {
2015                 hl_motion_420(h, dest_y, dest_cb, dest_cr,
2016                               s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2017                               s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2018                               h->h264dsp.weight_h264_pixels_tab,
2019                               h->h264dsp.biweight_h264_pixels_tab,
2020                               pixel_shift);
2021             }
2022         }
2023
2024         hl_decode_mb_idct_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
2025
2026         if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
2027             uint8_t *dest[2] = {dest_cb, dest_cr};
2028             if(transform_bypass){
2029                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
2030                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + (16*16*1 << pixel_shift), uvlinesize);
2031                     h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 32, h->mb + (16*16*2 << pixel_shift), uvlinesize);
2032                 }else{
2033                     idct_add = s->dsp.add_pixels4;
2034                     for(j=1; j<3; j++){
2035                         for(i=j*16; i<j*16+4; i++){
2036                             if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
2037                                 idct_add   (dest[j-1] + block_offset[i], h->mb + (i*16 << pixel_shift), uvlinesize);
2038                         }
2039                         if (chroma422) {
2040                             for(i=j*16+4; i<j*16+8; i++){
2041                                 if(h->non_zero_count_cache[ scan8[i+4] ] || dctcoef_get(h->mb, pixel_shift, i*16))
2042                                     idct_add   (dest[j-1] + block_offset[i+4], h->mb + (i*16 << pixel_shift), uvlinesize);
2043                             }
2044                         }
2045                     }
2046                 }
2047             }else{
2048                 if(is_h264){
2049                     int qp[2];
2050                     if (chroma422) {
2051                         qp[0] = h->chroma_qp[0] + 3;
2052                         qp[1] = h->chroma_qp[1] + 3;
2053                     } else {
2054                         qp[0] = h->chroma_qp[0];
2055                         qp[1] = h->chroma_qp[1];
2056                     }
2057                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+0] ])
2058                         h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16*1 << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][qp[0]][0]);
2059                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+1] ])
2060                         h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16*2 << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][qp[1]][0]);
2061                     h->h264dsp.h264_idct_add8(dest, block_offset,
2062                                               h->mb, uvlinesize,
2063                                               h->non_zero_count_cache);
2064                 } else if (CONFIG_SVQ3_DECODER) {
2065                     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]);
2066                     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]);
2067                     for(j=1; j<3; j++){
2068                         for(i=j*16; i<j*16+4; i++){
2069                             if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2070                                 uint8_t * const ptr= dest[j-1] + block_offset[i];
2071                                 ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[0][s->qscale + 12] - 12, 2);
2072                             }
2073                         }
2074                     }
2075                 }
2076             }
2077         }
2078     }
2079     if(h->cbp || IS_INTRA(mb_type))
2080     {
2081         s->dsp.clear_blocks(h->mb);
2082         s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
2083     }
2084 }
2085
2086 static av_always_inline void hl_decode_mb_444_internal(H264Context *h, int simple, int pixel_shift){
2087     MpegEncContext * const s = &h->s;
2088     const int mb_x= s->mb_x;
2089     const int mb_y= s->mb_y;
2090     const int mb_xy= h->mb_xy;
2091     const int mb_type = s->current_picture.f.mb_type[mb_xy];
2092     uint8_t  *dest[3];
2093     int linesize;
2094     int i, j, p;
2095     int *block_offset = &h->block_offset[0];
2096     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
2097     const int plane_count = (simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) ? 3 : 1;
2098
2099     for (p = 0; p < plane_count; p++)
2100     {
2101         dest[p] = s->current_picture.f.data[p] + ((mb_x << pixel_shift) + mb_y * s->linesize) * 16;
2102         s->dsp.prefetch(dest[p] + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
2103     }
2104
2105     h->list_counts[mb_xy]= h->list_count;
2106
2107     if (!simple && MB_FIELD) {
2108         linesize   = h->mb_linesize = h->mb_uvlinesize = s->linesize * 2;
2109         block_offset = &h->block_offset[48];
2110         if(mb_y&1) //FIXME move out of this function?
2111             for (p = 0; p < 3; p++)
2112                 dest[p] -= s->linesize*15;
2113         if(FRAME_MBAFF) {
2114             int list;
2115             for(list=0; list<h->list_count; list++){
2116                 if(!USES_LIST(mb_type, list))
2117                     continue;
2118                 if(IS_16X16(mb_type)){
2119                     int8_t *ref = &h->ref_cache[list][scan8[0]];
2120                     fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
2121                 }else{
2122                     for(i=0; i<16; i+=4){
2123                         int ref = h->ref_cache[list][scan8[i]];
2124                         if(ref >= 0)
2125                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
2126                     }
2127                 }
2128             }
2129         }
2130     } else {
2131         linesize   = h->mb_linesize = h->mb_uvlinesize = s->linesize;
2132     }
2133
2134     if (!simple && IS_INTRA_PCM(mb_type)) {
2135         if (pixel_shift) {
2136             const int bit_depth = h->sps.bit_depth_luma;
2137             GetBitContext gb;
2138             init_get_bits(&gb, (uint8_t*)h->mb, 768*bit_depth);
2139
2140             for (p = 0; p < plane_count; p++) {
2141                 for (i = 0; i < 16; i++) {
2142                     uint16_t *tmp = (uint16_t*)(dest[p] + i*linesize);
2143                     for (j = 0; j < 16; j++)
2144                         tmp[j] = get_bits(&gb, bit_depth);
2145                 }
2146             }
2147         } else {
2148             for (p = 0; p < plane_count; p++) {
2149                 for (i = 0; i < 16; i++) {
2150                     memcpy(dest[p] + i*linesize, h->mb + p*128 + i*8, 16);
2151                 }
2152             }
2153         }
2154     } else {
2155         if(IS_INTRA(mb_type)){
2156             if(h->deblocking_filter)
2157                 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 1, 1, simple, pixel_shift);
2158
2159             for (p = 0; p < plane_count; p++)
2160                 hl_decode_mb_predict_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2161
2162             if(h->deblocking_filter)
2163                 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 0, 1, simple, pixel_shift);
2164         }else{
2165             hl_motion(h, dest[0], dest[1], dest[2],
2166                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2167                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2168                       h->h264dsp.weight_h264_pixels_tab,
2169                       h->h264dsp.biweight_h264_pixels_tab, pixel_shift, 3);
2170         }
2171
2172         for (p = 0; p < plane_count; p++)
2173             hl_decode_mb_idct_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2174     }
2175     if(h->cbp || IS_INTRA(mb_type))
2176     {
2177         s->dsp.clear_blocks(h->mb);
2178         s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
2179     }
2180 }
2181
2182 /**
2183  * Process a macroblock; this case avoids checks for expensive uncommon cases.
2184  */
2185 #define hl_decode_mb_simple(sh, bits) \
2186 static void hl_decode_mb_simple_ ## bits(H264Context *h){ \
2187     hl_decode_mb_internal(h, 1, sh); \
2188 }
2189 hl_decode_mb_simple(0, 8)
2190 hl_decode_mb_simple(1, 16)
2191
2192 /**
2193  * Process a macroblock; this handles edge cases, such as interlacing.
2194  */
2195 static void av_noinline hl_decode_mb_complex(H264Context *h){
2196     hl_decode_mb_internal(h, 0, h->pixel_shift);
2197 }
2198
2199 static void av_noinline hl_decode_mb_444_complex(H264Context *h){
2200     hl_decode_mb_444_internal(h, 0, h->pixel_shift);
2201 }
2202
2203 static void av_noinline hl_decode_mb_444_simple(H264Context *h){
2204     hl_decode_mb_444_internal(h, 1, 0);
2205 }
2206
2207 void ff_h264_hl_decode_mb(H264Context *h){
2208     MpegEncContext * const s = &h->s;
2209     const int mb_xy= h->mb_xy;
2210     const int mb_type = s->current_picture.f.mb_type[mb_xy];
2211     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
2212
2213     if (CHROMA444) {
2214         if(is_complex || h->pixel_shift)
2215             hl_decode_mb_444_complex(h);
2216         else
2217             hl_decode_mb_444_simple(h);
2218     } else if (is_complex) {
2219         hl_decode_mb_complex(h);
2220     } else if (h->pixel_shift) {
2221         hl_decode_mb_simple_16(h);
2222     } else
2223         hl_decode_mb_simple_8(h);
2224 }
2225
2226 static int pred_weight_table(H264Context *h){
2227     MpegEncContext * const s = &h->s;
2228     int list, i;
2229     int luma_def, chroma_def;
2230
2231     h->use_weight= 0;
2232     h->use_weight_chroma= 0;
2233     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2234     if(h->sps.chroma_format_idc)
2235         h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2236     luma_def = 1<<h->luma_log2_weight_denom;
2237     chroma_def = 1<<h->chroma_log2_weight_denom;
2238
2239     for(list=0; list<2; list++){
2240         h->luma_weight_flag[list]   = 0;
2241         h->chroma_weight_flag[list] = 0;
2242         for(i=0; i<h->ref_count[list]; i++){
2243             int luma_weight_flag, chroma_weight_flag;
2244
2245             luma_weight_flag= get_bits1(&s->gb);
2246             if(luma_weight_flag){
2247                 h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
2248                 h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
2249                 if(   h->luma_weight[i][list][0] != luma_def
2250                    || h->luma_weight[i][list][1] != 0) {
2251                     h->use_weight= 1;
2252                     h->luma_weight_flag[list]= 1;
2253                 }
2254             }else{
2255                 h->luma_weight[i][list][0]= luma_def;
2256                 h->luma_weight[i][list][1]= 0;
2257             }
2258
2259             if(h->sps.chroma_format_idc){
2260                 chroma_weight_flag= get_bits1(&s->gb);
2261                 if(chroma_weight_flag){
2262                     int j;
2263                     for(j=0; j<2; j++){
2264                         h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
2265                         h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
2266                         if(   h->chroma_weight[i][list][j][0] != chroma_def
2267                            || h->chroma_weight[i][list][j][1] != 0) {
2268                             h->use_weight_chroma= 1;
2269                             h->chroma_weight_flag[list]= 1;
2270                         }
2271                     }
2272                 }else{
2273                     int j;
2274                     for(j=0; j<2; j++){
2275                         h->chroma_weight[i][list][j][0]= chroma_def;
2276                         h->chroma_weight[i][list][j][1]= 0;
2277                     }
2278                 }
2279             }
2280         }
2281         if(h->slice_type_nos != AV_PICTURE_TYPE_B) break;
2282     }
2283     h->use_weight= h->use_weight || h->use_weight_chroma;
2284     return 0;
2285 }
2286
2287 /**
2288  * Initialize implicit_weight table.
2289  * @param field  0/1 initialize the weight for interlaced MBAFF
2290  *                -1 initializes the rest
2291  */
2292 static void implicit_weight_table(H264Context *h, int field){
2293     MpegEncContext * const s = &h->s;
2294     int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
2295
2296     for (i = 0; i < 2; i++) {
2297         h->luma_weight_flag[i]   = 0;
2298         h->chroma_weight_flag[i] = 0;
2299     }
2300
2301     if(field < 0){
2302         if (s->picture_structure == PICT_FRAME) {
2303             cur_poc = s->current_picture_ptr->poc;
2304         } else {
2305             cur_poc = s->current_picture_ptr->field_poc[s->picture_structure - 1];
2306         }
2307     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
2308        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
2309         h->use_weight= 0;
2310         h->use_weight_chroma= 0;
2311         return;
2312     }
2313         ref_start= 0;
2314         ref_count0= h->ref_count[0];
2315         ref_count1= h->ref_count[1];
2316     }else{
2317         cur_poc = s->current_picture_ptr->field_poc[field];
2318         ref_start= 16;
2319         ref_count0= 16+2*h->ref_count[0];
2320         ref_count1= 16+2*h->ref_count[1];
2321     }
2322
2323     h->use_weight= 2;
2324     h->use_weight_chroma= 2;
2325     h->luma_log2_weight_denom= 5;
2326     h->chroma_log2_weight_denom= 5;
2327
2328     for(ref0=ref_start; ref0 < ref_count0; ref0++){
2329         int poc0 = h->ref_list[0][ref0].poc;
2330         for(ref1=ref_start; ref1 < ref_count1; ref1++){
2331             int w = 32;
2332             if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
2333                 int poc1 = h->ref_list[1][ref1].poc;
2334                 int td = av_clip(poc1 - poc0, -128, 127);
2335                 if(td){
2336                     int tb = av_clip(cur_poc - poc0, -128, 127);
2337                     int tx = (16384 + (FFABS(td) >> 1)) / td;
2338                     int dist_scale_factor = (tb*tx + 32) >> 8;
2339                     if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
2340                         w = 64 - dist_scale_factor;
2341                 }
2342             }
2343             if(field<0){
2344                 h->implicit_weight[ref0][ref1][0]=
2345                 h->implicit_weight[ref0][ref1][1]= w;
2346             }else{
2347                 h->implicit_weight[ref0][ref1][field]=w;
2348             }
2349         }
2350     }
2351 }
2352
2353 /**
2354  * instantaneous decoder refresh.
2355  */
2356 static void idr(H264Context *h){
2357     int i;
2358     ff_h264_remove_all_refs(h);
2359     h->prev_frame_num= 0;
2360     h->prev_frame_num_offset= 0;
2361     h->prev_poc_msb= 1<<16;
2362     h->prev_poc_lsb= 0;
2363     for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
2364         h->last_pocs[i] = INT_MIN;
2365 }
2366
2367 /* forget old pics after a seek */
2368 static void flush_dpb(AVCodecContext *avctx){
2369     H264Context *h= avctx->priv_data;
2370     int i;
2371     for(i=0; i<=MAX_DELAYED_PIC_COUNT; i++) {
2372         if(h->delayed_pic[i])
2373             h->delayed_pic[i]->f.reference = 0;
2374         h->delayed_pic[i]= NULL;
2375     }
2376     h->outputed_poc=h->next_outputed_poc= INT_MIN;
2377     h->prev_interlaced_frame = 1;
2378     idr(h);
2379     h->prev_frame_num= -1;
2380     if(h->s.current_picture_ptr)
2381         h->s.current_picture_ptr->f.reference = 0;
2382     h->s.first_field= 0;
2383     ff_h264_reset_sei(h);
2384     ff_mpeg_flush(avctx);
2385     h->recovery_frame= -1;
2386     h->sync= 0;
2387 }
2388
2389 static int init_poc(H264Context *h){
2390     MpegEncContext * const s = &h->s;
2391     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2392     int field_poc[2];
2393     Picture *cur = s->current_picture_ptr;
2394
2395     h->frame_num_offset= h->prev_frame_num_offset;
2396     if(h->frame_num < h->prev_frame_num)
2397         h->frame_num_offset += max_frame_num;
2398
2399     if(h->sps.poc_type==0){
2400         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2401
2402         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2403             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2404         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2405             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2406         else
2407             h->poc_msb = h->prev_poc_msb;
2408 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2409         field_poc[0] =
2410         field_poc[1] = h->poc_msb + h->poc_lsb;
2411         if(s->picture_structure == PICT_FRAME)
2412             field_poc[1] += h->delta_poc_bottom;
2413     }else if(h->sps.poc_type==1){
2414         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2415         int i;
2416
2417         if(h->sps.poc_cycle_length != 0)
2418             abs_frame_num = h->frame_num_offset + h->frame_num;
2419         else
2420             abs_frame_num = 0;
2421
2422         if(h->nal_ref_idc==0 && abs_frame_num > 0)
2423             abs_frame_num--;
2424
2425         expected_delta_per_poc_cycle = 0;
2426         for(i=0; i < h->sps.poc_cycle_length; i++)
2427             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2428
2429         if(abs_frame_num > 0){
2430             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2431             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2432
2433             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2434             for(i = 0; i <= frame_num_in_poc_cycle; i++)
2435                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2436         } else
2437             expectedpoc = 0;
2438
2439         if(h->nal_ref_idc == 0)
2440             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2441
2442         field_poc[0] = expectedpoc + h->delta_poc[0];
2443         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2444
2445         if(s->picture_structure == PICT_FRAME)
2446             field_poc[1] += h->delta_poc[1];
2447     }else{
2448         int poc= 2*(h->frame_num_offset + h->frame_num);
2449
2450         if(!h->nal_ref_idc)
2451             poc--;
2452
2453         field_poc[0]= poc;
2454         field_poc[1]= poc;
2455     }
2456
2457     if(s->picture_structure != PICT_BOTTOM_FIELD)
2458         s->current_picture_ptr->field_poc[0]= field_poc[0];
2459     if(s->picture_structure != PICT_TOP_FIELD)
2460         s->current_picture_ptr->field_poc[1]= field_poc[1];
2461     cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
2462
2463     return 0;
2464 }
2465
2466
2467 /**
2468  * initialize scan tables
2469  */
2470 static void init_scan_tables(H264Context *h){
2471     int i;
2472     for(i=0; i<16; i++){
2473 #define T(x) (x>>2) | ((x<<2) & 0xF)
2474         h->zigzag_scan[i] = T(zigzag_scan[i]);
2475         h-> field_scan[i] = T( field_scan[i]);
2476 #undef T
2477     }
2478     for(i=0; i<64; i++){
2479 #define T(x) (x>>3) | ((x&7)<<3)
2480         h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
2481         h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
2482         h->field_scan8x8[i]        = T(field_scan8x8[i]);
2483         h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
2484 #undef T
2485     }
2486     if(h->sps.transform_bypass){ //FIXME same ugly
2487         h->zigzag_scan_q0          = zigzag_scan;
2488         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
2489         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
2490         h->field_scan_q0           = field_scan;
2491         h->field_scan8x8_q0        = field_scan8x8;
2492         h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
2493     }else{
2494         h->zigzag_scan_q0          = h->zigzag_scan;
2495         h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
2496         h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
2497         h->field_scan_q0           = h->field_scan;
2498         h->field_scan8x8_q0        = h->field_scan8x8;
2499         h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
2500     }
2501 }
2502
2503 static int field_end(H264Context *h, int in_setup){
2504     MpegEncContext * const s = &h->s;
2505     AVCodecContext * const avctx= s->avctx;
2506     int err = 0;
2507     s->mb_y= 0;
2508
2509     if (!in_setup && !s->dropable)
2510         ff_thread_report_progress((AVFrame*)s->current_picture_ptr, (16*s->mb_height >> FIELD_PICTURE) - 1,
2511                                  s->picture_structure==PICT_BOTTOM_FIELD);
2512
2513     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2514         ff_vdpau_h264_set_reference_frames(s);
2515
2516     if(in_setup || !(avctx->active_thread_type&FF_THREAD_FRAME)){
2517         if(!s->dropable) {
2518             err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
2519             h->prev_poc_msb= h->poc_msb;
2520             h->prev_poc_lsb= h->poc_lsb;
2521         }
2522         h->prev_frame_num_offset= h->frame_num_offset;
2523         h->prev_frame_num= h->frame_num;
2524         h->outputed_poc = h->next_outputed_poc;
2525     }
2526
2527     if (avctx->hwaccel) {
2528         if (avctx->hwaccel->end_frame(avctx) < 0)
2529             av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
2530     }
2531
2532     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2533         ff_vdpau_h264_picture_complete(s);
2534
2535     /*
2536      * FIXME: Error handling code does not seem to support interlaced
2537      * when slices span multiple rows
2538      * The ff_er_add_slice calls don't work right for bottom
2539      * fields; they cause massive erroneous error concealing
2540      * Error marking covers both fields (top and bottom).
2541      * This causes a mismatched s->error_count
2542      * and a bad error table. Further, the error count goes to
2543      * INT_MAX when called for bottom field, because mb_y is
2544      * past end by one (callers fault) and resync_mb_y != 0
2545      * causes problems for the first MB line, too.
2546      */
2547     if (!FIELD_PICTURE)
2548         ff_er_frame_end(s);
2549
2550     ff_MPV_frame_end(s);
2551
2552     h->current_slice=0;
2553
2554     return err;
2555 }
2556
2557 /**
2558  * Replicate H264 "master" context to thread contexts.
2559  */
2560 static void clone_slice(H264Context *dst, H264Context *src)
2561 {
2562     memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
2563     dst->s.current_picture_ptr  = src->s.current_picture_ptr;
2564     dst->s.current_picture      = src->s.current_picture;
2565     dst->s.linesize             = src->s.linesize;
2566     dst->s.uvlinesize           = src->s.uvlinesize;
2567     dst->s.first_field          = src->s.first_field;
2568
2569     dst->prev_poc_msb           = src->prev_poc_msb;
2570     dst->prev_poc_lsb           = src->prev_poc_lsb;
2571     dst->prev_frame_num_offset  = src->prev_frame_num_offset;
2572     dst->prev_frame_num         = src->prev_frame_num;
2573     dst->short_ref_count        = src->short_ref_count;
2574
2575     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
2576     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
2577     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
2578     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
2579
2580     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
2581     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
2582 }
2583
2584 /**
2585  * Compute profile from profile_idc and constraint_set?_flags.
2586  *
2587  * @param sps SPS
2588  *
2589  * @return profile as defined by FF_PROFILE_H264_*
2590  */
2591 int ff_h264_get_profile(SPS *sps)
2592 {
2593     int profile = sps->profile_idc;
2594
2595     switch(sps->profile_idc) {
2596     case FF_PROFILE_H264_BASELINE:
2597         // constraint_set1_flag set to 1
2598         profile |= (sps->constraint_set_flags & 1<<1) ? FF_PROFILE_H264_CONSTRAINED : 0;
2599         break;
2600     case FF_PROFILE_H264_HIGH_10:
2601     case FF_PROFILE_H264_HIGH_422:
2602     case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
2603         // constraint_set3_flag set to 1
2604         profile |= (sps->constraint_set_flags & 1<<3) ? FF_PROFILE_H264_INTRA : 0;
2605         break;
2606     }
2607
2608     return profile;
2609 }
2610
2611 /**
2612  * Decode a slice header.
2613  * This will also call ff_MPV_common_init() and frame_start() as needed.
2614  *
2615  * @param h h264context
2616  * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
2617  *
2618  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
2619  */
2620 static int decode_slice_header(H264Context *h, H264Context *h0){
2621     MpegEncContext * const s = &h->s;
2622     MpegEncContext * const s0 = &h0->s;
2623     unsigned int first_mb_in_slice;
2624     unsigned int pps_id;
2625     int num_ref_idx_active_override_flag;
2626     unsigned int slice_type, tmp, i, j;
2627     int default_ref_list_done = 0;
2628     int last_pic_structure;
2629
2630     s->dropable= h->nal_ref_idc == 0;
2631
2632     /* FIXME: 2tap qpel isn't implemented for high bit depth. */
2633     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc && !h->pixel_shift){
2634         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
2635         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
2636     }else{
2637         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
2638         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
2639     }
2640
2641     first_mb_in_slice= get_ue_golomb_long(&s->gb);
2642
2643     if(first_mb_in_slice == 0){ //FIXME better field boundary detection
2644         if(h0->current_slice && FIELD_PICTURE){
2645             field_end(h, 1);
2646         }
2647
2648         h0->current_slice = 0;
2649         if (!s0->first_field)
2650             s->current_picture_ptr= NULL;
2651     }
2652
2653     slice_type= get_ue_golomb_31(&s->gb);
2654     if(slice_type > 9){
2655         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);
2656         return -1;
2657     }
2658     if(slice_type > 4){
2659         slice_type -= 5;
2660         h->slice_type_fixed=1;
2661     }else
2662         h->slice_type_fixed=0;
2663
2664     slice_type= golomb_to_pict_type[ slice_type ];
2665     if (slice_type == AV_PICTURE_TYPE_I
2666         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
2667         default_ref_list_done = 1;
2668     }
2669     h->slice_type= slice_type;
2670     h->slice_type_nos= slice_type & 3;
2671
2672     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
2673
2674     pps_id= get_ue_golomb(&s->gb);
2675     if(pps_id>=MAX_PPS_COUNT){
2676         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
2677         return -1;
2678     }
2679     if(!h0->pps_buffers[pps_id]) {
2680         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
2681         return -1;
2682     }
2683     h->pps= *h0->pps_buffers[pps_id];
2684
2685     if(!h0->sps_buffers[h->pps.sps_id]) {
2686         av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
2687         return -1;
2688     }
2689     h->sps = *h0->sps_buffers[h->pps.sps_id];
2690
2691     s->avctx->profile = ff_h264_get_profile(&h->sps);
2692     s->avctx->level   = h->sps.level_idc;
2693     s->avctx->refs    = h->sps.ref_frame_count;
2694
2695     s->mb_width= h->sps.mb_width;
2696     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
2697
2698     h->b_stride=  s->mb_width*4;
2699
2700     s->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
2701
2702     s->width = 16*s->mb_width;
2703     s->height= 16*s->mb_height;
2704
2705     if (s->context_initialized
2706         && (   s->width != s->avctx->coded_width || s->height != s->avctx->coded_height
2707             || s->avctx->bits_per_raw_sample != h->sps.bit_depth_luma
2708             || h->cur_chroma_format_idc != h->sps.chroma_format_idc
2709             || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
2710         if(h != h0 || (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
2711             av_log_missing_feature(s->avctx, "Width/height/bit depth/chroma idc changing with threads is", 0);
2712             return -1;   // width / height changed during parallelized decoding
2713         }
2714         free_tables(h, 0);
2715         flush_dpb(s->avctx);
2716         ff_MPV_common_end(s);
2717         h->list_count = 0;
2718     }
2719     if (!s->context_initialized) {
2720         if (h != h0) {
2721             av_log(h->s.avctx, AV_LOG_ERROR, "Cannot (re-)initialize context during parallel decoding.\n");
2722             return -1;
2723         }
2724         avcodec_set_dimensions(s->avctx, s->width, s->height);
2725         s->avctx->width  -= (2>>CHROMA444)*FFMIN(h->sps.crop_right, (8<<CHROMA444)-1);
2726         s->avctx->height -= (1<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1) * (2 - h->sps.frame_mbs_only_flag);
2727         s->avctx->sample_aspect_ratio= h->sps.sar;
2728         av_assert0(s->avctx->sample_aspect_ratio.den);
2729
2730         if (s->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
2731             h->cur_chroma_format_idc != h->sps.chroma_format_idc) {
2732             if (h->sps.bit_depth_luma >= 8 && h->sps.bit_depth_luma <= 10 &&
2733                 (h->sps.bit_depth_luma != 9 || !CHROMA422)) {
2734                 s->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
2735                 h->cur_chroma_format_idc = h->sps.chroma_format_idc;
2736                 h->pixel_shift = h->sps.bit_depth_luma > 8;
2737
2738                 ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma, h->sps.chroma_format_idc);
2739                 ff_h264_pred_init(&h->hpc, s->codec_id, h->sps.bit_depth_luma, h->sps.chroma_format_idc);
2740                 s->dsp.dct_bits = h->sps.bit_depth_luma > 8 ? 32 : 16;
2741                 ff_dsputil_init(&s->dsp, s->avctx);
2742             } else {
2743                 av_log(s->avctx, AV_LOG_ERROR, "Unsupported bit depth: %d chroma_idc: %d\n",
2744                        h->sps.bit_depth_luma, h->sps.chroma_format_idc);
2745                 return -1;
2746             }
2747         }
2748
2749         if(h->sps.video_signal_type_present_flag){
2750             s->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
2751             if(h->sps.colour_description_present_flag){
2752                 s->avctx->color_primaries = h->sps.color_primaries;
2753                 s->avctx->color_trc       = h->sps.color_trc;
2754                 s->avctx->colorspace      = h->sps.colorspace;
2755             }
2756         }
2757
2758         if(h->sps.timing_info_present_flag){
2759             int64_t den= h->sps.time_scale;
2760             if(h->x264_build < 44U)
2761                 den *= 2;
2762             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
2763                       h->sps.num_units_in_tick, den, 1<<30);
2764         }
2765
2766         switch (h->sps.bit_depth_luma) {
2767             case 9 :
2768                 if (CHROMA444) {
2769                     if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2770                         s->avctx->pix_fmt = PIX_FMT_GBRP9;
2771                     } else
2772                         s->avctx->pix_fmt = PIX_FMT_YUV444P9;
2773                 } else if (CHROMA422)
2774                     s->avctx->pix_fmt = PIX_FMT_YUV422P9;
2775                 else
2776                     s->avctx->pix_fmt = PIX_FMT_YUV420P9;
2777                 break;
2778             case 10 :
2779                 if (CHROMA444) {
2780                     if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2781                         s->avctx->pix_fmt = PIX_FMT_GBRP10;
2782                     } else
2783                         s->avctx->pix_fmt = PIX_FMT_YUV444P10;
2784                 } else if (CHROMA422)
2785                     s->avctx->pix_fmt = PIX_FMT_YUV422P10;
2786                 else
2787                     s->avctx->pix_fmt = PIX_FMT_YUV420P10;
2788                 break;
2789             case 8:
2790                 if (CHROMA444){
2791                     s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ444P : PIX_FMT_YUV444P;
2792                     if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2793                        s->avctx->pix_fmt = PIX_FMT_GBR24P;
2794                        av_log(h->s.avctx, AV_LOG_DEBUG, "Detected GBR colorspace.\n");
2795                     } else if (s->avctx->colorspace == AVCOL_SPC_YCGCO) {
2796                         av_log(h->s.avctx, AV_LOG_WARNING, "Detected unsupported YCgCo colorspace.\n");
2797                     }
2798                 } else if (CHROMA422) {
2799                     s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ422P : PIX_FMT_YUV422P;
2800                 }else{
2801                     s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
2802                                                              s->avctx->codec->pix_fmts ?
2803                                                              s->avctx->codec->pix_fmts :
2804                                                              s->avctx->color_range == AVCOL_RANGE_JPEG ?
2805                                                              hwaccel_pixfmt_list_h264_jpeg_420 :
2806                                                              ff_hwaccel_pixfmt_list_420);
2807                 }
2808                 break;
2809             default:
2810                 av_log(s->avctx, AV_LOG_ERROR,
2811                        "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
2812                 return AVERROR_INVALIDDATA;
2813         }
2814
2815         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
2816
2817         if (ff_MPV_common_init(s) < 0) {
2818             av_log(h->s.avctx, AV_LOG_ERROR, "ff_MPV_common_init() failed.\n");
2819             return -1;
2820         }
2821         s->first_field = 0;
2822         h->prev_interlaced_frame = 1;
2823
2824         init_scan_tables(h);
2825         if (ff_h264_alloc_tables(h) < 0) {
2826             av_log(h->s.avctx, AV_LOG_ERROR, "Could not allocate memory for h264\n");
2827             return AVERROR(ENOMEM);
2828         }
2829
2830         if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_SLICE)) {
2831             if (context_init(h) < 0) {
2832                 av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2833                 return -1;
2834             }
2835         } else {
2836             for(i = 1; i < s->slice_context_count; i++) {
2837                 H264Context *c;
2838                 c = h->thread_context[i] = av_malloc(sizeof(H264Context));
2839                 memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
2840                 memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
2841                 c->h264dsp = h->h264dsp;
2842                 c->sps = h->sps;
2843                 c->pps = h->pps;
2844                 c->pixel_shift = h->pixel_shift;
2845                 c->cur_chroma_format_idc = h->cur_chroma_format_idc;
2846                 init_scan_tables(c);
2847                 clone_tables(c, h, i);
2848             }
2849
2850             for(i = 0; i < s->slice_context_count; i++)
2851                 if (context_init(h->thread_context[i]) < 0) {
2852                     av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2853                     return -1;
2854                 }
2855         }
2856     }
2857
2858     if(h == h0 && h->dequant_coeff_pps != pps_id){
2859         h->dequant_coeff_pps = pps_id;
2860         init_dequant_tables(h);
2861     }
2862
2863     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
2864
2865     h->mb_mbaff = 0;
2866     h->mb_aff_frame = 0;
2867     last_pic_structure = s0->picture_structure;
2868     if(h->sps.frame_mbs_only_flag){
2869         s->picture_structure= PICT_FRAME;
2870     }else{
2871         if(!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B){
2872             av_log(h->s.avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
2873             return -1;
2874         }
2875         if(get_bits1(&s->gb)) { //field_pic_flag
2876             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
2877         } else {
2878             s->picture_structure= PICT_FRAME;
2879             h->mb_aff_frame = h->sps.mb_aff;
2880         }
2881     }
2882     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
2883
2884     if(h0->current_slice == 0){
2885         // Shorten frame num gaps so we don't have to allocate reference frames just to throw them away
2886         if(h->frame_num != h->prev_frame_num && h->prev_frame_num >= 0) {
2887             int unwrap_prev_frame_num = h->prev_frame_num, max_frame_num = 1<<h->sps.log2_max_frame_num;
2888
2889             if (unwrap_prev_frame_num > h->frame_num) unwrap_prev_frame_num -= max_frame_num;
2890
2891             if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
2892                 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
2893                 if (unwrap_prev_frame_num < 0)
2894                     unwrap_prev_frame_num += max_frame_num;
2895
2896                 h->prev_frame_num = unwrap_prev_frame_num;
2897             }
2898         }
2899
2900         while(h->frame_num !=  h->prev_frame_num && h->prev_frame_num >= 0 &&
2901               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
2902             Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
2903             av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
2904             if (ff_h264_frame_start(h) < 0)
2905                 return -1;
2906             h->prev_frame_num++;
2907             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
2908             s->current_picture_ptr->frame_num= h->prev_frame_num;
2909             ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 0);
2910             ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 1);
2911             ff_generate_sliding_window_mmcos(h);
2912             if (ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index) < 0 &&
2913                 (s->avctx->err_recognition & AV_EF_EXPLODE))
2914                 return AVERROR_INVALIDDATA;
2915             /* Error concealment: if a ref is missing, copy the previous ref in its place.
2916              * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
2917              * about there being no actual duplicates.
2918              * FIXME: this doesn't copy padding for out-of-frame motion vectors.  Given we're
2919              * concealing a lost frame, this probably isn't noticeable by comparison, but it should
2920              * be fixed. */
2921             if (h->short_ref_count) {
2922                 if (prev) {
2923                     av_image_copy(h->short_ref[0]->f.data, h->short_ref[0]->f.linesize,
2924                                   (const uint8_t**)prev->f.data, prev->f.linesize,
2925                                   s->avctx->pix_fmt, s->mb_width*16, s->mb_height*16);
2926                     h->short_ref[0]->poc = prev->poc+2;
2927                 }
2928                 h->short_ref[0]->frame_num = h->prev_frame_num;
2929             }
2930         }
2931
2932         /* See if we have a decoded first field looking for a pair... */
2933         if (s0->first_field) {
2934             assert(s0->current_picture_ptr);
2935             assert(s0->current_picture_ptr->f.data[0]);
2936             assert(s0->current_picture_ptr->f.reference != DELAYED_PIC_REF);
2937
2938             /* figure out if we have a complementary field pair */
2939             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2940                 /*
2941                  * Previous field is unmatched. Don't display it, but let it
2942                  * remain for reference if marked as such.
2943                  */
2944                 s0->current_picture_ptr = NULL;
2945                 s0->first_field = FIELD_PICTURE;
2946
2947             } else {
2948                 if (s0->current_picture_ptr->frame_num != h->frame_num) {
2949                     ff_thread_report_progress((AVFrame*)s0->current_picture_ptr, INT_MAX,
2950                                               s0->picture_structure==PICT_BOTTOM_FIELD);
2951                     /*
2952                      * This and previous field had
2953                      * different frame_nums. Consider this field first in
2954                      * pair. Throw away previous field except for reference
2955                      * purposes.
2956                      */
2957                     s0->first_field = 1;
2958                     s0->current_picture_ptr = NULL;
2959
2960                 } else {
2961                     /* Second field in complementary pair */
2962                     s0->first_field = 0;
2963                 }
2964             }
2965
2966         } else {
2967             /* Frame or first field in a potentially complementary pair */
2968             assert(!s0->current_picture_ptr);
2969             s0->first_field = FIELD_PICTURE;
2970         }
2971
2972         if(!FIELD_PICTURE || s0->first_field) {
2973             if (ff_h264_frame_start(h) < 0) {
2974                 s0->first_field = 0;
2975                 return -1;
2976             }
2977         } else {
2978             ff_release_unused_pictures(s, 0);
2979         }
2980     }
2981     if(h != h0)
2982         clone_slice(h, h0);
2983
2984     s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
2985
2986     assert(s->mb_num == s->mb_width * s->mb_height);
2987     if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
2988        first_mb_in_slice                    >= s->mb_num){
2989         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
2990         return -1;
2991     }
2992     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
2993     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
2994     if (s->picture_structure == PICT_BOTTOM_FIELD)
2995         s->resync_mb_y = s->mb_y = s->mb_y + 1;
2996     assert(s->mb_y < s->mb_height);
2997
2998     if(s->picture_structure==PICT_FRAME){
2999         h->curr_pic_num=   h->frame_num;
3000         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3001     }else{
3002         h->curr_pic_num= 2*h->frame_num + 1;
3003         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3004     }
3005
3006     if(h->nal_unit_type == NAL_IDR_SLICE){
3007         get_ue_golomb(&s->gb); /* idr_pic_id */
3008     }
3009
3010     if(h->sps.poc_type==0){
3011         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3012
3013         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3014             h->delta_poc_bottom= get_se_golomb(&s->gb);
3015         }
3016     }
3017
3018     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3019         h->delta_poc[0]= get_se_golomb(&s->gb);
3020
3021         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3022             h->delta_poc[1]= get_se_golomb(&s->gb);
3023     }
3024
3025     init_poc(h);
3026
3027     if(h->pps.redundant_pic_cnt_present){
3028         h->redundant_pic_count= get_ue_golomb(&s->gb);
3029     }
3030
3031     //set defaults, might be overridden a few lines later
3032     h->ref_count[0]= h->pps.ref_count[0];
3033     h->ref_count[1]= h->pps.ref_count[1];
3034
3035     if(h->slice_type_nos != AV_PICTURE_TYPE_I){
3036         unsigned max= (16<<(s->picture_structure != PICT_FRAME))-1;
3037         if(h->slice_type_nos == AV_PICTURE_TYPE_B){
3038             h->direct_spatial_mv_pred= get_bits1(&s->gb);
3039         }
3040         num_ref_idx_active_override_flag= get_bits1(&s->gb);
3041
3042         if(num_ref_idx_active_override_flag){
3043             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3044             if(h->slice_type_nos==AV_PICTURE_TYPE_B)
3045                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3046
3047         }
3048         if(h->ref_count[0]-1 > max || h->ref_count[1]-1 > max){
3049             av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3050             h->ref_count[0]= h->ref_count[1]= 1;
3051             return -1;
3052         }
3053         if(h->slice_type_nos == AV_PICTURE_TYPE_B)
3054             h->list_count= 2;
3055         else
3056             h->list_count= 1;
3057     }else
3058         h->ref_count[1]= h->ref_count[0]= h->list_count= 0;
3059
3060     if(!default_ref_list_done){
3061         ff_h264_fill_default_ref_list(h);
3062     }
3063
3064     if(h->slice_type_nos!=AV_PICTURE_TYPE_I && ff_h264_decode_ref_pic_list_reordering(h) < 0) {
3065         h->ref_count[1]= h->ref_count[0]= 0;
3066         return -1;
3067     }
3068
3069     if(h->slice_type_nos!=AV_PICTURE_TYPE_I){
3070         s->last_picture_ptr= &h->ref_list[0][0];
3071         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
3072     }
3073     if(h->slice_type_nos==AV_PICTURE_TYPE_B){
3074         s->next_picture_ptr= &h->ref_list[1][0];
3075         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
3076     }
3077
3078     if(   (h->pps.weighted_pred          && h->slice_type_nos == AV_PICTURE_TYPE_P )
3079        ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== AV_PICTURE_TYPE_B ) )
3080         pred_weight_table(h);
3081     else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
3082         implicit_weight_table(h, -1);
3083     }else {
3084         h->use_weight = 0;
3085         for (i = 0; i < 2; i++) {
3086             h->luma_weight_flag[i]   = 0;
3087             h->chroma_weight_flag[i] = 0;
3088         }
3089     }
3090
3091     if(h->nal_ref_idc && ff_h264_decode_ref_pic_marking(h0, &s->gb) < 0 &&
3092        (s->avctx->err_recognition & AV_EF_EXPLODE))
3093         return AVERROR_INVALIDDATA;
3094
3095     if(FRAME_MBAFF){
3096         ff_h264_fill_mbaff_ref_list(h);
3097
3098         if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
3099             implicit_weight_table(h, 0);
3100             implicit_weight_table(h, 1);
3101         }
3102     }
3103
3104     if(h->slice_type_nos==AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
3105         ff_h264_direct_dist_scale_factor(h);
3106     ff_h264_direct_ref_list_init(h);
3107
3108     if( h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac ){
3109         tmp = get_ue_golomb_31(&s->gb);
3110         if(tmp > 2){
3111             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3112             return -1;
3113         }
3114         h->cabac_init_idc= tmp;
3115     }
3116
3117     h->last_qscale_diff = 0;
3118     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
3119     if(tmp>51+6*(h->sps.bit_depth_luma-8)){
3120         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3121         return -1;
3122     }
3123     s->qscale= tmp;
3124     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3125     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3126     //FIXME qscale / qp ... stuff
3127     if(h->slice_type == AV_PICTURE_TYPE_SP){
3128         get_bits1(&s->gb); /* sp_for_switch_flag */
3129     }
3130     if(h->slice_type==AV_PICTURE_TYPE_SP || h->slice_type == AV_PICTURE_TYPE_SI){
3131         get_se_golomb(&s->gb); /* slice_qs_delta */
3132     }
3133
3134     h->deblocking_filter = 1;
3135     h->slice_alpha_c0_offset = 52;
3136     h->slice_beta_offset = 52;
3137     if( h->pps.deblocking_filter_parameters_present ) {
3138         tmp= get_ue_golomb_31(&s->gb);
3139         if(tmp > 2){
3140             av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
3141             return -1;
3142         }
3143         h->deblocking_filter= tmp;
3144         if(h->deblocking_filter < 2)
3145             h->deblocking_filter^= 1; // 1<->0
3146
3147         if( h->deblocking_filter ) {
3148             h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
3149             h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
3150             if(   h->slice_alpha_c0_offset > 104U
3151                || h->slice_beta_offset     > 104U){
3152                 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);
3153                 return -1;
3154             }
3155         }
3156     }
3157
3158     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
3159        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != AV_PICTURE_TYPE_I)
3160        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == AV_PICTURE_TYPE_B)
3161        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
3162         h->deblocking_filter= 0;
3163
3164     if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
3165         if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
3166             /* Cheat slightly for speed:
3167                Do not bother to deblock across slices. */
3168             h->deblocking_filter = 2;
3169         } else {
3170             h0->max_contexts = 1;
3171             if(!h0->single_decode_warning) {
3172                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
3173                 h0->single_decode_warning = 1;
3174             }
3175             if (h != h0) {
3176                 av_log(h->s.avctx, AV_LOG_ERROR, "Deblocking switched inside frame.\n");
3177                 return 1;
3178             }
3179         }
3180     }
3181     h->qp_thresh = 15 + 52 - FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset)
3182                  - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1])
3183                  + 6 * (h->sps.bit_depth_luma - 8);
3184
3185 #if 0 //FMO
3186     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3187         slice_group_change_cycle= get_bits(&s->gb, ?);
3188 #endif
3189
3190     h0->last_slice_type = slice_type;
3191     h->slice_num = ++h0->current_slice;
3192
3193     if(h->slice_num)
3194         h0->slice_row[(h->slice_num-1)&(MAX_SLICES-1)]= s->resync_mb_y;
3195     if (   h0->slice_row[h->slice_num&(MAX_SLICES-1)] + 3 >= s->resync_mb_y
3196         && h0->slice_row[h->slice_num&(MAX_SLICES-1)] <= s->resync_mb_y
3197         && h->slice_num >= MAX_SLICES) {
3198         //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
3199         av_log(s->avctx, AV_LOG_WARNING, "Possibly too many slices (%d >= %d), increase MAX_SLICES and recompile if there are artifacts\n", h->slice_num, MAX_SLICES);
3200     }
3201
3202     for(j=0; j<2; j++){
3203         int id_list[16];
3204         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
3205         for(i=0; i<16; i++){
3206             id_list[i]= 60;
3207             if (h->ref_list[j][i].f.data[0]) {
3208                 int k;
3209                 uint8_t *base = h->ref_list[j][i].f.base[0];
3210                 for(k=0; k<h->short_ref_count; k++)
3211                     if (h->short_ref[k]->f.base[0] == base) {
3212                         id_list[i]= k;
3213                         break;
3214                     }
3215                 for(k=0; k<h->long_ref_count; k++)
3216                     if (h->long_ref[k] && h->long_ref[k]->f.base[0] == base) {
3217                         id_list[i]= h->short_ref_count + k;
3218                         break;
3219                     }
3220             }
3221         }
3222
3223         ref2frm[0]=
3224         ref2frm[1]= -1;
3225         for(i=0; i<16; i++)
3226             ref2frm[i+2]= 4*id_list[i]
3227                           + (h->ref_list[j][i].f.reference & 3);
3228         ref2frm[18+0]=
3229         ref2frm[18+1]= -1;
3230         for(i=16; i<48; i++)
3231             ref2frm[i+4]= 4*id_list[(i-16)>>1]
3232                           + (h->ref_list[j][i].f.reference & 3);
3233     }
3234
3235     //FIXME: fix draw_edges+PAFF+frame threads
3236     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE || (!h->sps.frame_mbs_only_flag && s->avctx->active_thread_type)) ? 0 : 16;
3237     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
3238
3239     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3240         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",
3241                h->slice_num,
3242                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
3243                first_mb_in_slice,
3244                av_get_picture_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
3245                pps_id, h->frame_num,
3246                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3247                h->ref_count[0], h->ref_count[1],
3248                s->qscale,
3249                h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
3250                h->use_weight,
3251                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
3252                h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
3253                );
3254     }
3255
3256     return 0;
3257 }
3258
3259 int ff_h264_get_slice_type(const H264Context *h)
3260 {
3261     switch (h->slice_type) {
3262     case AV_PICTURE_TYPE_P:  return 0;
3263     case AV_PICTURE_TYPE_B:  return 1;
3264     case AV_PICTURE_TYPE_I:  return 2;
3265     case AV_PICTURE_TYPE_SP: return 3;
3266     case AV_PICTURE_TYPE_SI: return 4;
3267     default:         return -1;
3268     }
3269 }
3270
3271 static av_always_inline void fill_filter_caches_inter(H264Context *h, MpegEncContext * const s, int mb_type, int top_xy,
3272                                                       int left_xy[LEFT_MBS], int top_type, int left_type[LEFT_MBS], int mb_xy, int list)
3273 {
3274     int b_stride = h->b_stride;
3275     int16_t (*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
3276     int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
3277     if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
3278         if(USES_LIST(top_type, list)){
3279             const int b_xy= h->mb2b_xy[top_xy] + 3*b_stride;
3280             const int b8_xy= 4*top_xy + 2;
3281             int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3282             AV_COPY128(mv_dst - 1*8, s->current_picture.f.motion_val[list][b_xy + 0]);
3283             ref_cache[0 - 1*8]=
3284             ref_cache[1 - 1*8]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 0]];
3285             ref_cache[2 - 1*8]=
3286             ref_cache[3 - 1*8]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 1]];
3287         }else{
3288             AV_ZERO128(mv_dst - 1*8);
3289             AV_WN32A(&ref_cache[0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3290         }
3291
3292         if(!IS_INTERLACED(mb_type^left_type[LTOP])){
3293             if(USES_LIST(left_type[LTOP], list)){
3294                 const int b_xy= h->mb2b_xy[left_xy[LTOP]] + 3;
3295                 const int b8_xy= 4*left_xy[LTOP] + 1;
3296                 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[LTOP]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3297                 AV_COPY32(mv_dst - 1 +  0, s->current_picture.f.motion_val[list][b_xy + b_stride*0]);
3298                 AV_COPY32(mv_dst - 1 +  8, s->current_picture.f.motion_val[list][b_xy + b_stride*1]);
3299                 AV_COPY32(mv_dst - 1 + 16, s->current_picture.f.motion_val[list][b_xy + b_stride*2]);
3300                 AV_COPY32(mv_dst - 1 + 24, s->current_picture.f.motion_val[list][b_xy + b_stride*3]);
3301                 ref_cache[-1 +  0]=
3302                 ref_cache[-1 +  8]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2*0]];
3303                 ref_cache[-1 + 16]=
3304                 ref_cache[-1 + 24]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2*1]];
3305             }else{
3306                 AV_ZERO32(mv_dst - 1 + 0);
3307                 AV_ZERO32(mv_dst - 1 + 8);
3308                 AV_ZERO32(mv_dst - 1 +16);
3309                 AV_ZERO32(mv_dst - 1 +24);
3310                 ref_cache[-1 +  0]=
3311                 ref_cache[-1 +  8]=
3312                 ref_cache[-1 + 16]=
3313                 ref_cache[-1 + 24]= LIST_NOT_USED;
3314             }
3315         }
3316     }
3317
3318     if(!USES_LIST(mb_type, list)){
3319         fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0,0), 4);
3320         AV_WN32A(&ref_cache[0*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3321         AV_WN32A(&ref_cache[1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3322         AV_WN32A(&ref_cache[2*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3323         AV_WN32A(&ref_cache[3*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3324         return;
3325     }
3326
3327     {
3328         int8_t *ref = &s->current_picture.f.ref_index[list][4*mb_xy];
3329         int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3330         uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101;
3331         uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]],ref2frm[list][ref[3]])&0x00FF00FF)*0x0101;
3332         AV_WN32A(&ref_cache[0*8], ref01);
3333         AV_WN32A(&ref_cache[1*8], ref01);
3334         AV_WN32A(&ref_cache[2*8], ref23);
3335         AV_WN32A(&ref_cache[3*8], ref23);
3336     }
3337
3338     {
3339         int16_t (*mv_src)[2] = &s->current_picture.f.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
3340         AV_COPY128(mv_dst + 8*0, mv_src + 0*b_stride);
3341         AV_COPY128(mv_dst + 8*1, mv_src + 1*b_stride);
3342         AV_COPY128(mv_dst + 8*2, mv_src + 2*b_stride);
3343         AV_COPY128(mv_dst + 8*3, mv_src + 3*b_stride);
3344     }
3345 }
3346
3347 /**
3348  *
3349  * @return non zero if the loop filter can be skipped
3350  */
3351 static int fill_filter_caches(H264Context *h, int mb_type){
3352     MpegEncContext * const s = &h->s;
3353     const int mb_xy= h->mb_xy;
3354     int top_xy, left_xy[LEFT_MBS];
3355     int top_type, left_type[LEFT_MBS];
3356     uint8_t *nnz;
3357     uint8_t *nnz_cache;
3358
3359     top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
3360
3361     /* Wow, what a mess, why didn't they simplify the interlacing & intra
3362      * stuff, I can't imagine that these complex rules are worth it. */
3363
3364     left_xy[LBOT] = left_xy[LTOP] = mb_xy-1;
3365     if(FRAME_MBAFF){
3366         const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.f.mb_type[mb_xy - 1]);
3367         const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
3368         if(s->mb_y&1){
3369             if (left_mb_field_flag != curr_mb_field_flag) {
3370                 left_xy[LTOP] -= s->mb_stride;
3371             }
3372         }else{
3373             if(curr_mb_field_flag){
3374                 top_xy += s->mb_stride & (((s->current_picture.f.mb_type[top_xy] >> 7) & 1) - 1);
3375             }
3376             if (left_mb_field_flag != curr_mb_field_flag) {
3377                 left_xy[LBOT] += s->mb_stride;
3378             }
3379         }
3380     }
3381
3382     h->top_mb_xy = top_xy;
3383     h->left_mb_xy[LTOP] = left_xy[LTOP];
3384     h->left_mb_xy[LBOT] = left_xy[LBOT];
3385     {
3386         //for sufficiently low qp, filtering wouldn't do anything
3387         //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
3388         int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
3389         int qp = s->current_picture.f.qscale_table[mb_xy];
3390         if(qp <= qp_thresh
3391            && (left_xy[LTOP] < 0 || ((qp + s->current_picture.f.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh)
3392            && (top_xy        < 0 || ((qp + s->current_picture.f.qscale_table[top_xy       ] + 1) >> 1) <= qp_thresh)) {
3393             if(!FRAME_MBAFF)
3394                 return 1;
3395             if ((left_xy[LTOP] < 0            || ((qp + s->current_picture.f.qscale_table[left_xy[LBOT]        ] + 1) >> 1) <= qp_thresh) &&
3396                 (top_xy        < s->mb_stride || ((qp + s->current_picture.f.qscale_table[top_xy - s->mb_stride] + 1) >> 1) <= qp_thresh))
3397                 return 1;
3398         }
3399     }
3400
3401     top_type        = s->current_picture.f.mb_type[top_xy];
3402     left_type[LTOP] = s->current_picture.f.mb_type[left_xy[LTOP]];
3403     left_type[LBOT] = s->current_picture.f.mb_type[left_xy[LBOT]];
3404     if(h->deblocking_filter == 2){
3405         if(h->slice_table[top_xy       ] != h->slice_num) top_type= 0;
3406         if(h->slice_table[left_xy[LBOT]] != h->slice_num) left_type[LTOP]= left_type[LBOT]= 0;
3407     }else{
3408         if(h->slice_table[top_xy       ] == 0xFFFF) top_type= 0;
3409         if(h->slice_table[left_xy[LBOT]] == 0xFFFF) left_type[LTOP]= left_type[LBOT] =0;
3410     }
3411     h->top_type       = top_type;
3412     h->left_type[LTOP]= left_type[LTOP];
3413     h->left_type[LBOT]= left_type[LBOT];
3414
3415     if(IS_INTRA(mb_type))
3416         return 0;
3417
3418     fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy, top_type, left_type, mb_xy, 0);
3419     if(h->list_count == 2)
3420         fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy, top_type, left_type, mb_xy, 1);
3421
3422     nnz = h->non_zero_count[mb_xy];
3423     nnz_cache = h->non_zero_count_cache;
3424     AV_COPY32(&nnz_cache[4+8*1], &nnz[ 0]);
3425     AV_COPY32(&nnz_cache[4+8*2], &nnz[ 4]);
3426     AV_COPY32(&nnz_cache[4+8*3], &nnz[ 8]);
3427     AV_COPY32(&nnz_cache[4+8*4], &nnz[12]);
3428     h->cbp= h->cbp_table[mb_xy];
3429
3430     if(top_type){
3431         nnz = h->non_zero_count[top_xy];
3432         AV_COPY32(&nnz_cache[4+8*0], &nnz[3*4]);
3433     }
3434
3435     if(left_type[LTOP]){
3436         nnz = h->non_zero_count[left_xy[LTOP]];
3437         nnz_cache[3+8*1]= nnz[3+0*4];
3438         nnz_cache[3+8*2]= nnz[3+1*4];
3439         nnz_cache[3+8*3]= nnz[3+2*4];
3440         nnz_cache[3+8*4]= nnz[3+3*4];
3441     }
3442
3443     // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
3444     if(!CABAC && h->pps.transform_8x8_mode){
3445         if(IS_8x8DCT(top_type)){
3446             nnz_cache[4+8*0]=
3447             nnz_cache[5+8*0]= (h->cbp_table[top_xy] & 0x4000) >> 12;
3448             nnz_cache[6+8*0]=
3449             nnz_cache[7+8*0]= (h->cbp_table[top_xy] & 0x8000) >> 12;
3450         }
3451         if(IS_8x8DCT(left_type[LTOP])){
3452             nnz_cache[3+8*1]=
3453             nnz_cache[3+8*2]= (h->cbp_table[left_xy[LTOP]]&0x2000) >> 12; //FIXME check MBAFF
3454         }
3455         if(IS_8x8DCT(left_type[LBOT])){
3456             nnz_cache[3+8*3]=
3457             nnz_cache[3+8*4]= (h->cbp_table[left_xy[LBOT]]&0x8000) >> 12; //FIXME check MBAFF
3458         }
3459
3460         if(IS_8x8DCT(mb_type)){
3461             nnz_cache[scan8[0   ]]= nnz_cache[scan8[1   ]]=
3462             nnz_cache[scan8[2   ]]= nnz_cache[scan8[3   ]]= (h->cbp & 0x1000) >> 12;
3463
3464             nnz_cache[scan8[0+ 4]]= nnz_cache[scan8[1+ 4]]=
3465             nnz_cache[scan8[2+ 4]]= nnz_cache[scan8[3+ 4]]= (h->cbp & 0x2000) >> 12;
3466
3467             nnz_cache[scan8[0+ 8]]= nnz_cache[scan8[1+ 8]]=
3468             nnz_cache[scan8[2+ 8]]= nnz_cache[scan8[3+ 8]]= (h->cbp & 0x4000) >> 12;
3469
3470             nnz_cache[scan8[0+12]]= nnz_cache[scan8[1+12]]=
3471             nnz_cache[scan8[2+12]]= nnz_cache[scan8[3+12]]= (h->cbp & 0x8000) >> 12;
3472         }
3473     }
3474
3475     return 0;
3476 }
3477
3478 static void loop_filter(H264Context *h, int start_x, int end_x){
3479     MpegEncContext * const s = &h->s;
3480     uint8_t  *dest_y, *dest_cb, *dest_cr;
3481     int linesize, uvlinesize, mb_x, mb_y;
3482     const int end_mb_y= s->mb_y + FRAME_MBAFF;
3483     const int old_slice_type= h->slice_type;
3484     const int pixel_shift = h->pixel_shift;
3485     const int block_h = 16 >> s->chroma_y_shift;
3486
3487     if(h->deblocking_filter) {
3488         for(mb_x= start_x; mb_x<end_x; mb_x++){
3489             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
3490                 int mb_xy, mb_type;
3491                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
3492                 h->slice_num= h->slice_table[mb_xy];
3493                 mb_type = s->current_picture.f.mb_type[mb_xy];
3494                 h->list_count= h->list_counts[mb_xy];
3495
3496                 if(FRAME_MBAFF)
3497                     h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
3498
3499                 s->mb_x= mb_x;
3500                 s->mb_y= mb_y;
3501                 dest_y  = s->current_picture.f.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize  ) * 16;
3502                 dest_cb = s->current_picture.f.data[1] + (mb_x << pixel_shift) * (8 << CHROMA444) + mb_y * s->uvlinesize * block_h;
3503                 dest_cr = s->current_picture.f.data[2] + (mb_x << pixel_shift) * (8 << CHROMA444) + mb_y * s->uvlinesize * block_h;
3504                     //FIXME simplify above
3505
3506                 if (MB_FIELD) {
3507                     linesize   = h->mb_linesize   = s->linesize * 2;
3508                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
3509                     if(mb_y&1){ //FIXME move out of this function?
3510                         dest_y -= s->linesize*15;
3511                         dest_cb-= s->uvlinesize * (block_h - 1);
3512                         dest_cr-= s->uvlinesize * (block_h - 1);
3513                     }
3514                 } else {
3515                     linesize   = h->mb_linesize   = s->linesize;
3516                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
3517                 }
3518                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
3519                 if(fill_filter_caches(h, mb_type))
3520                     continue;
3521                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.f.qscale_table[mb_xy]);
3522                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.f.qscale_table[mb_xy]);
3523
3524                 if (FRAME_MBAFF) {
3525                     ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3526                 } else {
3527                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3528                 }
3529             }
3530         }
3531     }
3532     h->slice_type= old_slice_type;
3533     s->mb_x= end_x;
3534     s->mb_y= end_mb_y - FRAME_MBAFF;
3535     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3536     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3537 }
3538
3539 static void predict_field_decoding_flag(H264Context *h){
3540     MpegEncContext * const s = &h->s;
3541     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3542     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
3543                 ? s->current_picture.f.mb_type[mb_xy - 1]
3544                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
3545                 ? s->current_picture.f.mb_type[mb_xy - s->mb_stride]
3546                 : 0;
3547     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
3548 }
3549
3550 /**
3551  * Draw edges and report progress for the last MB row.
3552  */
3553 static void decode_finish_row(H264Context *h){
3554     MpegEncContext * const s = &h->s;
3555     int top = 16*(s->mb_y >> FIELD_PICTURE);
3556     int height = 16 << FRAME_MBAFF;
3557     int deblock_border = (16 + 4) << FRAME_MBAFF;
3558     int pic_height = 16*s->mb_height >> FIELD_PICTURE;
3559
3560     if (h->deblocking_filter) {
3561         if((top + height) >= pic_height)
3562             height += deblock_border;
3563
3564         top -= deblock_border;
3565     }
3566
3567     if (top >= pic_height || (top + height) < h->emu_edge_height)
3568         return;
3569
3570     height = FFMIN(height, pic_height - top);
3571     if (top < h->emu_edge_height) {
3572         height = top+height;
3573         top = 0;
3574     }
3575
3576     ff_draw_horiz_band(s, top, height);
3577
3578     if (s->dropable) return;
3579
3580     ff_thread_report_progress((AVFrame*)s->current_picture_ptr, top + height - 1,
3581                              s->picture_structure==PICT_BOTTOM_FIELD);
3582 }
3583
3584 static int decode_slice(struct AVCodecContext *avctx, void *arg){
3585     H264Context *h = *(void**)arg;
3586     MpegEncContext * const s = &h->s;
3587     const int part_mask= s->partitioned_frame ? (ER_AC_END|ER_AC_ERROR) : 0x7F;
3588     int lf_x_start = s->mb_x;
3589
3590     s->mb_skip_run= -1;
3591
3592     h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
3593                     (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
3594
3595     if( h->pps.cabac ) {
3596         /* realign */
3597         align_get_bits( &s->gb );
3598
3599         /* init cabac */
3600         ff_init_cabac_states( &h->cabac);
3601         ff_init_cabac_decoder( &h->cabac,
3602                                s->gb.buffer + get_bits_count(&s->gb)/8,
3603                                (get_bits_left(&s->gb) + 7)/8);
3604
3605         ff_h264_init_cabac_states(h);
3606
3607         for(;;){
3608 //START_TIMER
3609             int ret = ff_h264_decode_mb_cabac(h);
3610             int eos;
3611 //STOP_TIMER("decode_mb_cabac")
3612
3613             if(ret>=0) ff_h264_hl_decode_mb(h);
3614
3615             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
3616                 s->mb_y++;
3617
3618                 ret = ff_h264_decode_mb_cabac(h);
3619
3620                 if(ret>=0) ff_h264_hl_decode_mb(h);
3621                 s->mb_y--;
3622             }
3623             eos = get_cabac_terminate( &h->cabac );
3624
3625             if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
3626                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END&part_mask);
3627                 if (s->mb_x >= lf_x_start) loop_filter(h, lf_x_start, s->mb_x + 1);
3628                 return 0;
3629             }
3630             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
3631                 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);
3632                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, ER_MB_ERROR&part_mask);
3633                 return -1;
3634             }
3635
3636             if( ++s->mb_x >= s->mb_width ) {
3637                 loop_filter(h, lf_x_start, s->mb_x);
3638                 s->mb_x = lf_x_start = 0;
3639                 decode_finish_row(h);
3640                 ++s->mb_y;
3641                 if(FIELD_OR_MBAFF_PICTURE) {
3642                     ++s->mb_y;
3643                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
3644                         predict_field_decoding_flag(h);
3645                 }
3646             }
3647
3648             if( eos || s->mb_y >= s->mb_height ) {
3649                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3650                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END&part_mask);
3651                 if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
3652                 return 0;
3653             }
3654         }
3655
3656     } else {
3657         for(;;){
3658             int ret = ff_h264_decode_mb_cavlc(h);
3659
3660             if(ret>=0) ff_h264_hl_decode_mb(h);
3661
3662             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
3663                 s->mb_y++;
3664                 ret = ff_h264_decode_mb_cavlc(h);
3665
3666                 if(ret>=0) ff_h264_hl_decode_mb(h);
3667                 s->mb_y--;
3668             }
3669
3670             if(ret<0){
3671                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3672                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, ER_MB_ERROR&part_mask);
3673                 return -1;
3674             }
3675
3676             if(++s->mb_x >= s->mb_width){
3677                 loop_filter(h, lf_x_start, s->mb_x);
3678                 s->mb_x = lf_x_start = 0;
3679                 decode_finish_row(h);
3680                 ++s->mb_y;
3681                 if(FIELD_OR_MBAFF_PICTURE) {
3682                     ++s->mb_y;
3683                     if(FRAME_MBAFF && s->mb_y < s->mb_height)
3684                         predict_field_decoding_flag(h);
3685                 }
3686                 if(s->mb_y >= s->mb_height){
3687                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3688
3689                     if(   get_bits_count(&s->gb) == s->gb.size_in_bits
3690                        || get_bits_count(&s->gb) <  s->gb.size_in_bits && !(s->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
3691                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END&part_mask);
3692
3693                         return 0;
3694                     }else{
3695                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, ER_MB_END&part_mask);
3696
3697                         return -1;
3698                     }
3699                 }
3700             }
3701
3702             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
3703                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3704                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
3705                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END&part_mask);
3706                     if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
3707
3708                     return 0;
3709                 }else{
3710                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, ER_MB_ERROR&part_mask);
3711
3712                     return -1;
3713                 }
3714             }
3715         }
3716     }
3717 }
3718
3719 /**
3720  * Call decode_slice() for each context.
3721  *
3722  * @param h h264 master context
3723  * @param context_count number of contexts to execute
3724  */
3725 static int execute_decode_slices(H264Context *h, int context_count){
3726     MpegEncContext * const s = &h->s;
3727     AVCodecContext * const avctx= s->avctx;
3728     H264Context *hx;
3729     int i;
3730
3731     if (s->avctx->hwaccel || s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
3732         return 0;
3733     if(context_count == 1) {
3734         return decode_slice(avctx, &h);
3735     } else {
3736         for(i = 1; i < context_count; i++) {
3737             hx = h->thread_context[i];
3738             hx->s.err_recognition = avctx->err_recognition;
3739             hx->s.error_count = 0;
3740             hx->x264_build= h->x264_build;
3741         }
3742
3743         avctx->execute(avctx, decode_slice,
3744                        h->thread_context, NULL, context_count, sizeof(void*));
3745
3746         /* pull back stuff from slices to master context */
3747         hx = h->thread_context[context_count - 1];
3748         s->mb_x = hx->s.mb_x;
3749         s->mb_y = hx->s.mb_y;
3750         s->dropable = hx->s.dropable;
3751         s->picture_structure = hx->s.picture_structure;
3752         for(i = 1; i < context_count; i++)
3753             h->s.error_count += h->thread_context[i]->s.error_count;
3754     }
3755
3756     return 0;
3757 }
3758
3759
3760 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
3761     MpegEncContext * const s = &h->s;
3762     AVCodecContext * const avctx= s->avctx;
3763     H264Context *hx; ///< thread context
3764     int buf_index;
3765     int context_count;
3766     int next_avc;
3767     int pass = !(avctx->active_thread_type & FF_THREAD_FRAME);
3768     int nals_needed=0; ///< number of NALs that need decoding before the next frame thread starts
3769     int nal_index;
3770
3771     h->nal_unit_type= 0;
3772
3773     if(!s->slice_context_count)
3774          s->slice_context_count= 1;
3775     h->max_contexts = s->slice_context_count;
3776
3777     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
3778         h->current_slice = 0;
3779         if (!s->first_field)
3780             s->current_picture_ptr= NULL;
3781         ff_h264_reset_sei(h);
3782     }
3783
3784     for(;pass <= 1;pass++){
3785         buf_index = 0;
3786         context_count = 0;
3787         next_avc = h->is_avc ? 0 : buf_size;
3788         nal_index = 0;
3789     for(;;){
3790         int consumed;
3791         int dst_length;
3792         int bit_length;
3793         uint8_t *ptr;
3794         int i, nalsize = 0;
3795         int err;
3796
3797         if(buf_index >= next_avc) {
3798             if (buf_index >= buf_size - h->nal_length_size) break;
3799             nalsize = 0;
3800             for(i = 0; i < h->nal_length_size; i++)
3801                 nalsize = (nalsize << 8) | buf[buf_index++];
3802             if(nalsize <= 0 || nalsize > buf_size - buf_index){
3803                 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
3804                 break;
3805             }
3806             next_avc= buf_index + nalsize;
3807         } else {
3808             // start code prefix search
3809             for(; buf_index + 3 < next_avc; buf_index++){
3810                 // This should always succeed in the first iteration.
3811                 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
3812                     break;
3813             }
3814
3815             if(buf_index+3 >= buf_size) break;
3816
3817             buf_index+=3;
3818             if(buf_index >= next_avc) continue;
3819         }
3820
3821         hx = h->thread_context[context_count];
3822
3823         ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
3824         if (ptr==NULL || dst_length < 0){
3825             return -1;
3826         }
3827         i= buf_index + consumed;
3828         if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
3829            buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
3830             s->workaround_bugs |= FF_BUG_TRUNCATED;
3831
3832         if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
3833         while(dst_length > 0 && ptr[dst_length - 1] == 0)
3834             dst_length--;
3835         }
3836         bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
3837
3838         if(s->avctx->debug&FF_DEBUG_STARTCODE){
3839             av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d/%d at %d/%d length %d pass %d\n", hx->nal_unit_type, hx->nal_ref_idc, buf_index, buf_size, dst_length, pass);
3840         }
3841
3842         if (h->is_avc && (nalsize != consumed) && nalsize){
3843             av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
3844         }
3845
3846         buf_index += consumed;
3847         nal_index++;
3848
3849         if(pass == 0) {
3850             // packets can sometimes contain multiple PPS/SPS
3851             // e.g. two PAFF field pictures in one packet, or a demuxer which splits NALs strangely
3852             // if so, when frame threading we can't start the next thread until we've read all of them
3853             switch (hx->nal_unit_type) {
3854                 case NAL_SPS:
3855                 case NAL_PPS:
3856                     nals_needed = nal_index;
3857                     break;
3858                 case NAL_IDR_SLICE:
3859                 case NAL_SLICE:
3860                     init_get_bits(&hx->s.gb, ptr, bit_length);
3861                     if (!get_ue_golomb(&hx->s.gb))
3862                         nals_needed = nal_index;
3863             }
3864             continue;
3865         }
3866
3867         //FIXME do not discard SEI id
3868         if(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0)
3869             continue;
3870
3871       again:
3872         err = 0;
3873         switch(hx->nal_unit_type){
3874         case NAL_IDR_SLICE:
3875             if (h->nal_unit_type != NAL_IDR_SLICE) {
3876                 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices\n");
3877                 return -1;
3878             }
3879             idr(h); // FIXME ensure we don't lose some frames if there is reordering
3880         case NAL_SLICE:
3881             init_get_bits(&hx->s.gb, ptr, bit_length);
3882             hx->intra_gb_ptr=
3883             hx->inter_gb_ptr= &hx->s.gb;
3884             hx->s.data_partitioning = 0;
3885
3886             if((err = decode_slice_header(hx, h)))
3887                break;
3888
3889             if (   h->sei_recovery_frame_cnt >= 0
3890                 && (   h->recovery_frame<0
3891                     || ((h->recovery_frame - h->frame_num) & ((1 << h->sps.log2_max_frame_num)-1)) > h->sei_recovery_frame_cnt)) {
3892                 h->recovery_frame = (h->frame_num + h->sei_recovery_frame_cnt) %
3893                                     (1 << h->sps.log2_max_frame_num);
3894             }
3895
3896             s->current_picture_ptr->f.key_frame |=
3897                     (hx->nal_unit_type == NAL_IDR_SLICE);
3898
3899             if (h->recovery_frame == h->frame_num) {
3900                 s->current_picture_ptr->sync |= 1;
3901                 h->recovery_frame = -1;
3902             }
3903
3904             h->sync |= !!s->current_picture_ptr->f.key_frame;
3905             h->sync |= 3*!!(s->flags2 & CODEC_FLAG2_SHOW_ALL);
3906             s->current_picture_ptr->sync |= h->sync;
3907
3908             if (h->current_slice == 1) {
3909                 if(!(s->flags2 & CODEC_FLAG2_CHUNKS)) {
3910                     decode_postinit(h, nal_index >= nals_needed);
3911                 }
3912
3913                 if (s->avctx->hwaccel && s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
3914                     return -1;
3915                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
3916                     ff_vdpau_h264_picture_start(s);
3917             }
3918
3919             if(hx->redundant_pic_count==0
3920                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
3921                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
3922                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
3923                && avctx->skip_frame < AVDISCARD_ALL){
3924                 if(avctx->hwaccel) {
3925                     if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
3926                         return -1;
3927                 }else
3928                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
3929                     static const uint8_t start_code[] = {0x00, 0x00, 0x01};
3930                     ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
3931                     ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
3932                 }else
3933                     context_count++;
3934             }
3935             break;
3936         case NAL_DPA:
3937             init_get_bits(&hx->s.gb, ptr, bit_length);
3938             hx->intra_gb_ptr=
3939             hx->inter_gb_ptr= NULL;
3940
3941             if ((err = decode_slice_header(hx, h)) < 0)
3942                 break;
3943
3944             hx->s.data_partitioning = 1;
3945
3946             break;
3947         case NAL_DPB:
3948             init_get_bits(&hx->intra_gb, ptr, bit_length);
3949             hx->intra_gb_ptr= &hx->intra_gb;
3950             break;
3951         case NAL_DPC:
3952             init_get_bits(&hx->inter_gb, ptr, bit_length);
3953             hx->inter_gb_ptr= &hx->inter_gb;
3954
3955             if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
3956                && s->context_initialized
3957                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
3958                && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
3959                && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
3960                && avctx->skip_frame < AVDISCARD_ALL)
3961                 context_count++;
3962             break;
3963         case NAL_SEI:
3964             init_get_bits(&s->gb, ptr, bit_length);
3965             ff_h264_decode_sei(h);
3966             break;
3967         case NAL_SPS:
3968             init_get_bits(&s->gb, ptr, bit_length);
3969             if(ff_h264_decode_seq_parameter_set(h) < 0 && (h->is_avc ? (nalsize != consumed) && nalsize : 1)){
3970                 av_log(h->s.avctx, AV_LOG_DEBUG, "SPS decoding failure, trying alternative mode\n");
3971                 if(h->is_avc) av_assert0(next_avc - buf_index + consumed == nalsize);
3972                 init_get_bits(&s->gb, &buf[buf_index + 1 - consumed], 8*(next_avc - buf_index + consumed));
3973                 ff_h264_decode_seq_parameter_set(h);
3974             }
3975
3976             if (s->flags& CODEC_FLAG_LOW_DELAY ||
3977                 (h->sps.bitstream_restriction_flag && !h->sps.num_reorder_frames))
3978                 s->low_delay=1;
3979
3980             if(avctx->has_b_frames < 2)
3981                 avctx->has_b_frames= !s->low_delay;
3982             break;
3983         case NAL_PPS:
3984             init_get_bits(&s->gb, ptr, bit_length);
3985
3986             ff_h264_decode_picture_parameter_set(h, bit_length);
3987
3988             break;
3989         case NAL_AUD:
3990         case NAL_END_SEQUENCE:
3991         case NAL_END_STREAM:
3992         case NAL_FILLER_DATA:
3993         case NAL_SPS_EXT:
3994         case NAL_AUXILIARY_SLICE:
3995             break;
3996         default:
3997             av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
3998         }
3999
4000         if(context_count == h->max_contexts) {
4001             execute_decode_slices(h, context_count);
4002             context_count = 0;
4003         }
4004
4005         if (err < 0)
4006             av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
4007         else if(err == 1) {
4008             /* Slice could not be decoded in parallel mode, copy down
4009              * NAL unit stuff to context 0 and restart. Note that
4010              * rbsp_buffer is not transferred, but since we no longer
4011              * run in parallel mode this should not be an issue. */
4012             h->nal_unit_type = hx->nal_unit_type;
4013             h->nal_ref_idc   = hx->nal_ref_idc;
4014             hx = h;
4015             goto again;
4016         }
4017     }
4018     }
4019     if(context_count)
4020         execute_decode_slices(h, context_count);
4021     return buf_index;
4022 }
4023
4024 /**
4025  * Return the number of bytes consumed for building the current frame.
4026  */
4027 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
4028         if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
4029         if(pos+10>buf_size) pos=buf_size; // oops ;)
4030
4031         return pos;
4032 }
4033
4034 static int decode_frame(AVCodecContext *avctx,
4035                              void *data, int *data_size,
4036                              AVPacket *avpkt)
4037 {
4038     const uint8_t *buf = avpkt->data;
4039     int buf_size = avpkt->size;
4040     H264Context *h = avctx->priv_data;
4041     MpegEncContext *s = &h->s;
4042     AVFrame *pict = data;
4043     int buf_index = 0;
4044     Picture *out;
4045     int i, out_idx;
4046
4047     s->flags= avctx->flags;
4048     s->flags2= avctx->flags2;
4049
4050    /* end of stream, output what is still in the buffers */
4051     if (buf_size == 0) {
4052  out:
4053
4054         s->current_picture_ptr = NULL;
4055
4056 //FIXME factorize this with the output code below
4057         out = h->delayed_pic[0];
4058         out_idx = 0;
4059         for (i = 1; h->delayed_pic[i] && !h->delayed_pic[i]->f.key_frame && !h->delayed_pic[i]->mmco_reset; i++)
4060             if(h->delayed_pic[i]->poc < out->poc){
4061                 out = h->delayed_pic[i];
4062                 out_idx = i;
4063             }
4064
4065         for(i=out_idx; h->delayed_pic[i]; i++)
4066             h->delayed_pic[i] = h->delayed_pic[i+1];
4067
4068         if(out){
4069             *data_size = sizeof(AVFrame);
4070             *pict= *(AVFrame*)out;
4071         }
4072
4073         return buf_index;
4074     }
4075     if(h->is_avc && buf_size >= 9 && buf[0]==1 && buf[2]==0 && (buf[4]&0xFC)==0xFC && (buf[5]&0x1F) && buf[8]==0x67){
4076         int cnt= buf[5]&0x1f;
4077         uint8_t *p= buf+6;
4078         while(cnt--){
4079             int nalsize= AV_RB16(p) + 2;
4080             if(nalsize > buf_size - (p-buf) || p[2]!=0x67)
4081                 goto not_extra;
4082             p += nalsize;
4083         }
4084         cnt = *(p++);
4085         if(!cnt)
4086             goto not_extra;
4087         while(cnt--){
4088             int nalsize= AV_RB16(p) + 2;
4089             if(nalsize > buf_size - (p-buf) || p[2]!=0x68)
4090                 goto not_extra;
4091             p += nalsize;
4092         }
4093
4094         return ff_h264_decode_extradata(h, buf, buf_size);
4095     }
4096 not_extra:
4097
4098     buf_index=decode_nal_units(h, buf, buf_size);
4099     if(buf_index < 0)
4100         return -1;
4101
4102     if (!s->current_picture_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
4103         av_assert0(buf_index <= buf_size);
4104         goto out;
4105     }
4106
4107     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
4108         if (avctx->skip_frame >= AVDISCARD_NONREF ||
4109             buf_size >= 4 && !memcmp("Q264", buf, 4))
4110             return buf_size;
4111         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
4112         return -1;
4113     }
4114
4115     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
4116
4117         if(s->flags2 & CODEC_FLAG2_CHUNKS) decode_postinit(h, 1);
4118
4119         field_end(h, 0);
4120
4121         *data_size = 0; /* Wait for second field. */
4122         if (h->next_output_pic && (h->next_output_pic->sync || h->sync>1)) {
4123                 *data_size = sizeof(AVFrame);
4124                 *pict = *(AVFrame*)h->next_output_pic;
4125         }
4126     }
4127
4128     assert(pict->data[0] || !*data_size);
4129     ff_print_debug_info(s, pict);
4130 //printf("out %d\n", (int)pict->data[0]);
4131
4132     return get_consumed_bytes(s, buf_index, buf_size);
4133 }
4134 #if 0
4135 static inline void fill_mb_avail(H264Context *h){
4136     MpegEncContext * const s = &h->s;
4137     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4138
4139     if(s->mb_y){
4140         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
4141         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
4142         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
4143     }else{
4144         h->mb_avail[0]=
4145         h->mb_avail[1]=
4146         h->mb_avail[2]= 0;
4147     }
4148     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
4149     h->mb_avail[4]= 1; //FIXME move out
4150     h->mb_avail[5]= 0; //FIXME move out
4151 }
4152 #endif
4153
4154 av_cold void ff_h264_free_context(H264Context *h)
4155 {
4156     int i;
4157
4158     free_tables(h, 1); //FIXME cleanup init stuff perhaps
4159
4160     for(i = 0; i < MAX_SPS_COUNT; i++)
4161         av_freep(h->sps_buffers + i);
4162
4163     for(i = 0; i < MAX_PPS_COUNT; i++)
4164         av_freep(h->pps_buffers + i);
4165 }
4166
4167 av_cold int ff_h264_decode_end(AVCodecContext *avctx)
4168 {
4169     H264Context *h = avctx->priv_data;
4170     MpegEncContext *s = &h->s;
4171
4172     ff_h264_remove_all_refs(h);
4173     ff_h264_free_context(h);
4174
4175     ff_MPV_common_end(s);
4176
4177 //    memset(h, 0, sizeof(H264Context));
4178
4179     return 0;
4180 }
4181
4182 static const AVProfile profiles[] = {
4183     { FF_PROFILE_H264_BASELINE,             "Baseline"              },
4184     { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline"  },
4185     { FF_PROFILE_H264_MAIN,                 "Main"                  },
4186     { FF_PROFILE_H264_EXTENDED,             "Extended"              },
4187     { FF_PROFILE_H264_HIGH,                 "High"                  },
4188     { FF_PROFILE_H264_HIGH_10,              "High 10"               },
4189     { FF_PROFILE_H264_HIGH_10_INTRA,        "High 10 Intra"         },
4190     { FF_PROFILE_H264_HIGH_422,             "High 4:2:2"            },
4191     { FF_PROFILE_H264_HIGH_422_INTRA,       "High 4:2:2 Intra"      },
4192     { FF_PROFILE_H264_HIGH_444,             "High 4:4:4"            },
4193     { FF_PROFILE_H264_HIGH_444_PREDICTIVE,  "High 4:4:4 Predictive" },
4194     { FF_PROFILE_H264_HIGH_444_INTRA,       "High 4:4:4 Intra"      },
4195     { FF_PROFILE_H264_CAVLC_444,            "CAVLC 4:4:4"           },
4196     { FF_PROFILE_UNKNOWN },
4197 };
4198
4199 static const AVOption h264_options[] = {
4200     {"is_avc", "is avc", offsetof(H264Context, is_avc), FF_OPT_TYPE_INT, {.dbl = 0}, 0, 1, 0},
4201     {"nal_length_size", "nal_length_size", offsetof(H264Context, nal_length_size), FF_OPT_TYPE_INT, {.dbl = 0}, 0, 4, 0},
4202     {NULL}
4203 };
4204
4205 static const AVClass h264_class = {
4206     "H264 Decoder",
4207     av_default_item_name,
4208     h264_options,
4209     LIBAVUTIL_VERSION_INT,
4210 };
4211
4212 static const AVClass h264_vdpau_class = {
4213     "H264 VDPAU Decoder",
4214     av_default_item_name,
4215     h264_options,
4216     LIBAVUTIL_VERSION_INT,
4217 };
4218
4219 AVCodec ff_h264_decoder = {
4220     .name           = "h264",
4221     .type           = AVMEDIA_TYPE_VIDEO,
4222     .id             = CODEC_ID_H264,
4223     .priv_data_size = sizeof(H264Context),
4224     .init           = ff_h264_decode_init,
4225     .close          = ff_h264_decode_end,
4226     .decode         = decode_frame,
4227     .capabilities   = /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY |
4228                       CODEC_CAP_SLICE_THREADS | CODEC_CAP_FRAME_THREADS,
4229     .flush= flush_dpb,
4230     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
4231     .init_thread_copy      = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
4232     .update_thread_context = ONLY_IF_THREADS_ENABLED(decode_update_thread_context),
4233     .profiles = NULL_IF_CONFIG_SMALL(profiles),
4234     .priv_class     = &h264_class,
4235 };
4236
4237 #if CONFIG_H264_VDPAU_DECODER
4238 AVCodec ff_h264_vdpau_decoder = {
4239     .name           = "h264_vdpau",
4240     .type           = AVMEDIA_TYPE_VIDEO,
4241     .id             = CODEC_ID_H264,
4242     .priv_data_size = sizeof(H264Context),
4243     .init           = ff_h264_decode_init,
4244     .close          = ff_h264_decode_end,
4245     .decode         = decode_frame,
4246     .capabilities   = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
4247     .flush= flush_dpb,
4248     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
4249     .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
4250     .profiles = NULL_IF_CONFIG_SMALL(profiles),
4251     .priv_class     = &h264_vdpau_class,
4252 };
4253 #endif